| <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"> |
| <html lang="en"> |
| <head> |
| <title>Source code</title> |
| <link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style"> |
| </head> |
| <body> |
| <div class="sourceContainer"> |
| <pre><span class="sourceLineNo">001</span>/*<a name="line.1"></a> |
| <span class="sourceLineNo">002</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a> |
| <span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a> |
| <span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a> |
| <span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a> |
| <span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a> |
| <span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a> |
| <span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a> |
| <span class="sourceLineNo">009</span> *<a name="line.9"></a> |
| <span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a> |
| <span class="sourceLineNo">011</span> *<a name="line.11"></a> |
| <span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a> |
| <span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a> |
| <span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a> |
| <span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a> |
| <span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a> |
| <span class="sourceLineNo">017</span> */<a name="line.17"></a> |
| <span class="sourceLineNo">018</span>package org.apache.hadoop.hbase.util;<a name="line.18"></a> |
| <span class="sourceLineNo">019</span><a name="line.19"></a> |
| <span class="sourceLineNo">020</span>import java.io.Closeable;<a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import java.io.FileNotFoundException;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span>import java.io.IOException;<a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.io.InterruptedIOException;<a name="line.23"></a> |
| <span class="sourceLineNo">024</span>import java.io.PrintWriter;<a name="line.24"></a> |
| <span class="sourceLineNo">025</span>import java.io.StringWriter;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import java.net.InetAddress;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import java.net.URI;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span>import java.util.ArrayList;<a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import java.util.Collection;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import java.util.Collections;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import java.util.Comparator;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import java.util.EnumSet;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import java.util.HashMap;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import java.util.HashSet;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import java.util.Iterator;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import java.util.List;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import java.util.Locale;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import java.util.Map;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import java.util.Map.Entry;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import java.util.Objects;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span>import java.util.Optional;<a name="line.41"></a> |
| <span class="sourceLineNo">042</span>import java.util.Set;<a name="line.42"></a> |
| <span class="sourceLineNo">043</span>import java.util.SortedMap;<a name="line.43"></a> |
| <span class="sourceLineNo">044</span>import java.util.TreeMap;<a name="line.44"></a> |
| <span class="sourceLineNo">045</span>import java.util.Vector;<a name="line.45"></a> |
| <span class="sourceLineNo">046</span>import java.util.concurrent.Callable;<a name="line.46"></a> |
| <span class="sourceLineNo">047</span>import java.util.concurrent.ConcurrentSkipListMap;<a name="line.47"></a> |
| <span class="sourceLineNo">048</span>import java.util.concurrent.ExecutionException;<a name="line.48"></a> |
| <span class="sourceLineNo">049</span>import java.util.concurrent.ExecutorService;<a name="line.49"></a> |
| <span class="sourceLineNo">050</span>import java.util.concurrent.Executors;<a name="line.50"></a> |
| <span class="sourceLineNo">051</span>import java.util.concurrent.Future;<a name="line.51"></a> |
| <span class="sourceLineNo">052</span>import java.util.concurrent.FutureTask;<a name="line.52"></a> |
| <span class="sourceLineNo">053</span>import java.util.concurrent.ScheduledThreadPoolExecutor;<a name="line.53"></a> |
| <span class="sourceLineNo">054</span>import java.util.concurrent.TimeUnit;<a name="line.54"></a> |
| <span class="sourceLineNo">055</span>import java.util.concurrent.TimeoutException;<a name="line.55"></a> |
| <span class="sourceLineNo">056</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.56"></a> |
| <span class="sourceLineNo">057</span>import java.util.concurrent.atomic.AtomicInteger;<a name="line.57"></a> |
| <span class="sourceLineNo">058</span>import java.util.stream.Collectors;<a name="line.58"></a> |
| <span class="sourceLineNo">059</span>import org.apache.commons.io.IOUtils;<a name="line.59"></a> |
| <span class="sourceLineNo">060</span>import org.apache.commons.lang3.StringUtils;<a name="line.60"></a> |
| <span class="sourceLineNo">061</span>import org.apache.hadoop.conf.Configuration;<a name="line.61"></a> |
| <span class="sourceLineNo">062</span>import org.apache.hadoop.conf.Configured;<a name="line.62"></a> |
| <span class="sourceLineNo">063</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.63"></a> |
| <span class="sourceLineNo">064</span>import org.apache.hadoop.fs.FileStatus;<a name="line.64"></a> |
| <span class="sourceLineNo">065</span>import org.apache.hadoop.fs.FileSystem;<a name="line.65"></a> |
| <span class="sourceLineNo">066</span>import org.apache.hadoop.fs.Path;<a name="line.66"></a> |
| <span class="sourceLineNo">067</span>import org.apache.hadoop.fs.permission.FsAction;<a name="line.67"></a> |
| <span class="sourceLineNo">068</span>import org.apache.hadoop.fs.permission.FsPermission;<a name="line.68"></a> |
| <span class="sourceLineNo">069</span>import org.apache.hadoop.hbase.Abortable;<a name="line.69"></a> |
| <span class="sourceLineNo">070</span>import org.apache.hadoop.hbase.CatalogFamilyFormat;<a name="line.70"></a> |
| <span class="sourceLineNo">071</span>import org.apache.hadoop.hbase.Cell;<a name="line.71"></a> |
| <span class="sourceLineNo">072</span>import org.apache.hadoop.hbase.CellUtil;<a name="line.72"></a> |
| <span class="sourceLineNo">073</span>import org.apache.hadoop.hbase.ClientMetaTableAccessor;<a name="line.73"></a> |
| <span class="sourceLineNo">074</span>import org.apache.hadoop.hbase.ClusterMetrics;<a name="line.74"></a> |
| <span class="sourceLineNo">075</span>import org.apache.hadoop.hbase.ClusterMetrics.Option;<a name="line.75"></a> |
| <span class="sourceLineNo">076</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.76"></a> |
| <span class="sourceLineNo">077</span>import org.apache.hadoop.hbase.HBaseInterfaceAudience;<a name="line.77"></a> |
| <span class="sourceLineNo">078</span>import org.apache.hadoop.hbase.HConstants;<a name="line.78"></a> |
| <span class="sourceLineNo">079</span>import org.apache.hadoop.hbase.HRegionLocation;<a name="line.79"></a> |
| <span class="sourceLineNo">080</span>import org.apache.hadoop.hbase.KeyValue;<a name="line.80"></a> |
| <span class="sourceLineNo">081</span>import org.apache.hadoop.hbase.MasterNotRunningException;<a name="line.81"></a> |
| <span class="sourceLineNo">082</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.82"></a> |
| <span class="sourceLineNo">083</span>import org.apache.hadoop.hbase.RegionLocations;<a name="line.83"></a> |
| <span class="sourceLineNo">084</span>import org.apache.hadoop.hbase.ServerName;<a name="line.84"></a> |
| <span class="sourceLineNo">085</span>import org.apache.hadoop.hbase.TableName;<a name="line.85"></a> |
| <span class="sourceLineNo">086</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.86"></a> |
| <span class="sourceLineNo">087</span>import org.apache.hadoop.hbase.ZooKeeperConnectionException;<a name="line.87"></a> |
| <span class="sourceLineNo">088</span>import org.apache.hadoop.hbase.client.Admin;<a name="line.88"></a> |
| <span class="sourceLineNo">089</span>import org.apache.hadoop.hbase.client.ColumnFamilyDescriptor;<a name="line.89"></a> |
| <span class="sourceLineNo">090</span>import org.apache.hadoop.hbase.client.ColumnFamilyDescriptorBuilder;<a name="line.90"></a> |
| <span class="sourceLineNo">091</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.91"></a> |
| <span class="sourceLineNo">092</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.92"></a> |
| <span class="sourceLineNo">093</span>import org.apache.hadoop.hbase.client.Delete;<a name="line.93"></a> |
| <span class="sourceLineNo">094</span>import org.apache.hadoop.hbase.client.Get;<a name="line.94"></a> |
| <span class="sourceLineNo">095</span>import org.apache.hadoop.hbase.client.Put;<a name="line.95"></a> |
| <span class="sourceLineNo">096</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.96"></a> |
| <span class="sourceLineNo">097</span>import org.apache.hadoop.hbase.client.RegionInfoBuilder;<a name="line.97"></a> |
| <span class="sourceLineNo">098</span>import org.apache.hadoop.hbase.client.RegionLocator;<a name="line.98"></a> |
| <span class="sourceLineNo">099</span>import org.apache.hadoop.hbase.client.RegionReplicaUtil;<a name="line.99"></a> |
| <span class="sourceLineNo">100</span>import org.apache.hadoop.hbase.client.Result;<a name="line.100"></a> |
| <span class="sourceLineNo">101</span>import org.apache.hadoop.hbase.client.ResultScanner;<a name="line.101"></a> |
| <span class="sourceLineNo">102</span>import org.apache.hadoop.hbase.client.RowMutations;<a name="line.102"></a> |
| <span class="sourceLineNo">103</span>import org.apache.hadoop.hbase.client.Scan;<a name="line.103"></a> |
| <span class="sourceLineNo">104</span>import org.apache.hadoop.hbase.client.Table;<a name="line.104"></a> |
| <span class="sourceLineNo">105</span>import org.apache.hadoop.hbase.client.TableDescriptor;<a name="line.105"></a> |
| <span class="sourceLineNo">106</span>import org.apache.hadoop.hbase.client.TableDescriptorBuilder;<a name="line.106"></a> |
| <span class="sourceLineNo">107</span>import org.apache.hadoop.hbase.client.TableState;<a name="line.107"></a> |
| <span class="sourceLineNo">108</span>import org.apache.hadoop.hbase.io.FileLink;<a name="line.108"></a> |
| <span class="sourceLineNo">109</span>import org.apache.hadoop.hbase.io.HFileLink;<a name="line.109"></a> |
| <span class="sourceLineNo">110</span>import org.apache.hadoop.hbase.io.hfile.CacheConfig;<a name="line.110"></a> |
| <span class="sourceLineNo">111</span>import org.apache.hadoop.hbase.io.hfile.HFile;<a name="line.111"></a> |
| <span class="sourceLineNo">112</span>import org.apache.hadoop.hbase.master.RegionState;<a name="line.112"></a> |
| <span class="sourceLineNo">113</span>import org.apache.hadoop.hbase.regionserver.HRegion;<a name="line.113"></a> |
| <span class="sourceLineNo">114</span>import org.apache.hadoop.hbase.regionserver.HRegionFileSystem;<a name="line.114"></a> |
| <span class="sourceLineNo">115</span>import org.apache.hadoop.hbase.regionserver.StoreFileInfo;<a name="line.115"></a> |
| <span class="sourceLineNo">116</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.116"></a> |
| <span class="sourceLineNo">117</span>import org.apache.hadoop.hbase.replication.ReplicationPeerDescription;<a name="line.117"></a> |
| <span class="sourceLineNo">118</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.118"></a> |
| <span class="sourceLineNo">119</span>import org.apache.hadoop.hbase.replication.ReplicationStorageFactory;<a name="line.119"></a> |
| <span class="sourceLineNo">120</span>import org.apache.hadoop.hbase.security.UserProvider;<a name="line.120"></a> |
| <span class="sourceLineNo">121</span>import org.apache.hadoop.hbase.util.Bytes.ByteArrayComparator;<a name="line.121"></a> |
| <span class="sourceLineNo">122</span>import org.apache.hadoop.hbase.util.HbckErrorReporter.ERROR_CODE;<a name="line.122"></a> |
| <span class="sourceLineNo">123</span>import org.apache.hadoop.hbase.util.hbck.HFileCorruptionChecker;<a name="line.123"></a> |
| <span class="sourceLineNo">124</span>import org.apache.hadoop.hbase.util.hbck.ReplicationChecker;<a name="line.124"></a> |
| <span class="sourceLineNo">125</span>import org.apache.hadoop.hbase.util.hbck.TableIntegrityErrorHandler;<a name="line.125"></a> |
| <span class="sourceLineNo">126</span>import org.apache.hadoop.hbase.wal.WALSplitUtil;<a name="line.126"></a> |
| <span class="sourceLineNo">127</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.127"></a> |
| <span class="sourceLineNo">128</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.128"></a> |
| <span class="sourceLineNo">129</span>import org.apache.hadoop.hbase.zookeeper.ZNodePaths;<a name="line.129"></a> |
| <span class="sourceLineNo">130</span>import org.apache.hadoop.hdfs.protocol.AlreadyBeingCreatedException;<a name="line.130"></a> |
| <span class="sourceLineNo">131</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.131"></a> |
| <span class="sourceLineNo">132</span>import org.apache.hadoop.security.AccessControlException;<a name="line.132"></a> |
| <span class="sourceLineNo">133</span>import org.apache.hadoop.security.UserGroupInformation;<a name="line.133"></a> |
| <span class="sourceLineNo">134</span>import org.apache.hadoop.util.ReflectionUtils;<a name="line.134"></a> |
| <span class="sourceLineNo">135</span>import org.apache.hadoop.util.Tool;<a name="line.135"></a> |
| <span class="sourceLineNo">136</span>import org.apache.hadoop.util.ToolRunner;<a name="line.136"></a> |
| <span class="sourceLineNo">137</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.137"></a> |
| <span class="sourceLineNo">138</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.138"></a> |
| <span class="sourceLineNo">139</span>import org.apache.yetus.audience.InterfaceStability;<a name="line.139"></a> |
| <span class="sourceLineNo">140</span>import org.apache.zookeeper.KeeperException;<a name="line.140"></a> |
| <span class="sourceLineNo">141</span>import org.slf4j.Logger;<a name="line.141"></a> |
| <span class="sourceLineNo">142</span>import org.slf4j.LoggerFactory;<a name="line.142"></a> |
| <span class="sourceLineNo">143</span><a name="line.143"></a> |
| <span class="sourceLineNo">144</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.144"></a> |
| <span class="sourceLineNo">145</span>import org.apache.hbase.thirdparty.com.google.common.base.Joiner;<a name="line.145"></a> |
| <span class="sourceLineNo">146</span>import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;<a name="line.146"></a> |
| <span class="sourceLineNo">147</span>import org.apache.hbase.thirdparty.com.google.common.collect.Lists;<a name="line.147"></a> |
| <span class="sourceLineNo">148</span>import org.apache.hbase.thirdparty.com.google.common.collect.Sets;<a name="line.148"></a> |
| <span class="sourceLineNo">149</span><a name="line.149"></a> |
| <span class="sourceLineNo">150</span>/**<a name="line.150"></a> |
| <span class="sourceLineNo">151</span> * HBaseFsck (hbck) is a tool for checking and repairing region consistency and<a name="line.151"></a> |
| <span class="sourceLineNo">152</span> * table integrity problems in a corrupted HBase. This tool was written for hbase-1.x. It does not<a name="line.152"></a> |
| <span class="sourceLineNo">153</span> * work with hbase-2.x; it can read state but is not allowed to change state; i.e. effect 'repair'.<a name="line.153"></a> |
| <span class="sourceLineNo">154</span> * Even though it can 'read' state, given how so much has changed in how hbase1 and hbase2 operate,<a name="line.154"></a> |
| <span class="sourceLineNo">155</span> * it will often misread. See hbck2 (HBASE-19121) for a hbck tool for hbase2. This class is<a name="line.155"></a> |
| <span class="sourceLineNo">156</span> * deprecated.<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> *<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> * <p><a name="line.158"></a> |
| <span class="sourceLineNo">159</span> * Region consistency checks verify that hbase:meta, region deployment on region<a name="line.159"></a> |
| <span class="sourceLineNo">160</span> * servers and the state of data in HDFS (.regioninfo files) all are in<a name="line.160"></a> |
| <span class="sourceLineNo">161</span> * accordance.<a name="line.161"></a> |
| <span class="sourceLineNo">162</span> * <p><a name="line.162"></a> |
| <span class="sourceLineNo">163</span> * Table integrity checks verify that all possible row keys resolve to exactly<a name="line.163"></a> |
| <span class="sourceLineNo">164</span> * one region of a table. This means there are no individual degenerate<a name="line.164"></a> |
| <span class="sourceLineNo">165</span> * or backwards regions; no holes between regions; and that there are no<a name="line.165"></a> |
| <span class="sourceLineNo">166</span> * overlapping regions.<a name="line.166"></a> |
| <span class="sourceLineNo">167</span> * <p><a name="line.167"></a> |
| <span class="sourceLineNo">168</span> * The general repair strategy works in two phases:<a name="line.168"></a> |
| <span class="sourceLineNo">169</span> * <ol><a name="line.169"></a> |
| <span class="sourceLineNo">170</span> * <li> Repair Table Integrity on HDFS. (merge or fabricate regions)<a name="line.170"></a> |
| <span class="sourceLineNo">171</span> * <li> Repair Region Consistency with hbase:meta and assignments<a name="line.171"></a> |
| <span class="sourceLineNo">172</span> * </ol><a name="line.172"></a> |
| <span class="sourceLineNo">173</span> * <p><a name="line.173"></a> |
| <span class="sourceLineNo">174</span> * For table integrity repairs, the tables' region directories are scanned<a name="line.174"></a> |
| <span class="sourceLineNo">175</span> * for .regioninfo files. Each table's integrity is then verified. If there<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> * are any orphan regions (regions with no .regioninfo files) or holes, new<a name="line.176"></a> |
| <span class="sourceLineNo">177</span> * regions are fabricated. Backwards regions are sidelined as well as empty<a name="line.177"></a> |
| <span class="sourceLineNo">178</span> * degenerate (endkey==startkey) regions. If there are any overlapping regions,<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> * a new region is created and all data is merged into the new region.<a name="line.179"></a> |
| <span class="sourceLineNo">180</span> * <p><a name="line.180"></a> |
| <span class="sourceLineNo">181</span> * Table integrity repairs deal solely with HDFS and could potentially be done<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> * offline -- the hbase region servers or master do not need to be running.<a name="line.182"></a> |
| <span class="sourceLineNo">183</span> * This phase can eventually be used to completely reconstruct the hbase:meta table in<a name="line.183"></a> |
| <span class="sourceLineNo">184</span> * an offline fashion.<a name="line.184"></a> |
| <span class="sourceLineNo">185</span> * <p><a name="line.185"></a> |
| <span class="sourceLineNo">186</span> * Region consistency requires three conditions -- 1) valid .regioninfo file<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> * present in an HDFS region dir, 2) valid row with .regioninfo data in META,<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> * and 3) a region is deployed only at the regionserver that was assigned to<a name="line.188"></a> |
| <span class="sourceLineNo">189</span> * with proper state in the master.<a name="line.189"></a> |
| <span class="sourceLineNo">190</span> * <p><a name="line.190"></a> |
| <span class="sourceLineNo">191</span> * Region consistency repairs require hbase to be online so that hbck can<a name="line.191"></a> |
| <span class="sourceLineNo">192</span> * contact the HBase master and region servers. The hbck#connect() method must<a name="line.192"></a> |
| <span class="sourceLineNo">193</span> * first be called successfully. Much of the region consistency information<a name="line.193"></a> |
| <span class="sourceLineNo">194</span> * is transient and less risky to repair.<a name="line.194"></a> |
| <span class="sourceLineNo">195</span> * <p><a name="line.195"></a> |
| <span class="sourceLineNo">196</span> * If hbck is run from the command line, there are a handful of arguments that<a name="line.196"></a> |
| <span class="sourceLineNo">197</span> * can be used to limit the kinds of repairs hbck will do. See the code in<a name="line.197"></a> |
| <span class="sourceLineNo">198</span> * {@link #printUsageAndExit()} for more details.<a name="line.198"></a> |
| <span class="sourceLineNo">199</span> * @deprecated For removal in hbase-4.0.0. Use HBCK2 instead.<a name="line.199"></a> |
| <span class="sourceLineNo">200</span> */<a name="line.200"></a> |
| <span class="sourceLineNo">201</span>@Deprecated<a name="line.201"></a> |
| <span class="sourceLineNo">202</span>@InterfaceAudience.LimitedPrivate(HBaseInterfaceAudience.TOOLS)<a name="line.202"></a> |
| <span class="sourceLineNo">203</span>@InterfaceStability.Evolving<a name="line.203"></a> |
| <span class="sourceLineNo">204</span>public class HBaseFsck extends Configured implements Closeable {<a name="line.204"></a> |
| <span class="sourceLineNo">205</span> public static final long DEFAULT_TIME_LAG = 60000; // default value of 1 minute<a name="line.205"></a> |
| <span class="sourceLineNo">206</span> public static final long DEFAULT_SLEEP_BEFORE_RERUN = 10000;<a name="line.206"></a> |
| <span class="sourceLineNo">207</span> private static final int MAX_NUM_THREADS = 50; // #threads to contact regions<a name="line.207"></a> |
| <span class="sourceLineNo">208</span> private static boolean rsSupportsOffline = true;<a name="line.208"></a> |
| <span class="sourceLineNo">209</span> private static final int DEFAULT_OVERLAPS_TO_SIDELINE = 2;<a name="line.209"></a> |
| <span class="sourceLineNo">210</span> private static final int DEFAULT_MAX_MERGE = 5;<a name="line.210"></a> |
| <span class="sourceLineNo">211</span><a name="line.211"></a> |
| <span class="sourceLineNo">212</span> /**<a name="line.212"></a> |
| <span class="sourceLineNo">213</span> * Here is where hbase-1.x used to default the lock for hbck1.<a name="line.213"></a> |
| <span class="sourceLineNo">214</span> * It puts in place a lock when it goes to write/make changes.<a name="line.214"></a> |
| <span class="sourceLineNo">215</span> */<a name="line.215"></a> |
| <span class="sourceLineNo">216</span> @VisibleForTesting<a name="line.216"></a> |
| <span class="sourceLineNo">217</span> public static final String HBCK_LOCK_FILE = "hbase-hbck.lock";<a name="line.217"></a> |
| <span class="sourceLineNo">218</span> private static final int DEFAULT_MAX_LOCK_FILE_ATTEMPTS = 5;<a name="line.218"></a> |
| <span class="sourceLineNo">219</span> private static final int DEFAULT_LOCK_FILE_ATTEMPT_SLEEP_INTERVAL = 200; // milliseconds<a name="line.219"></a> |
| <span class="sourceLineNo">220</span> private static final int DEFAULT_LOCK_FILE_ATTEMPT_MAX_SLEEP_TIME = 5000; // milliseconds<a name="line.220"></a> |
| <span class="sourceLineNo">221</span> // We have to set the timeout value > HdfsConstants.LEASE_SOFTLIMIT_PERIOD.<a name="line.221"></a> |
| <span class="sourceLineNo">222</span> // In HADOOP-2.6 and later, the Namenode proxy now created with custom RetryPolicy for<a name="line.222"></a> |
| <span class="sourceLineNo">223</span> // AlreadyBeingCreatedException which is implies timeout on this operations up to<a name="line.223"></a> |
| <span class="sourceLineNo">224</span> // HdfsConstants.LEASE_SOFTLIMIT_PERIOD (60 seconds).<a name="line.224"></a> |
| <span class="sourceLineNo">225</span> private static final int DEFAULT_WAIT_FOR_LOCK_TIMEOUT = 80; // seconds<a name="line.225"></a> |
| <span class="sourceLineNo">226</span> private static final int DEFAULT_MAX_CREATE_ZNODE_ATTEMPTS = 5;<a name="line.226"></a> |
| <span class="sourceLineNo">227</span> private static final int DEFAULT_CREATE_ZNODE_ATTEMPT_SLEEP_INTERVAL = 200; // milliseconds<a name="line.227"></a> |
| <span class="sourceLineNo">228</span> private static final int DEFAULT_CREATE_ZNODE_ATTEMPT_MAX_SLEEP_TIME = 5000; // milliseconds<a name="line.228"></a> |
| <span class="sourceLineNo">229</span><a name="line.229"></a> |
| <span class="sourceLineNo">230</span> /**********************<a name="line.230"></a> |
| <span class="sourceLineNo">231</span> * Internal resources<a name="line.231"></a> |
| <span class="sourceLineNo">232</span> **********************/<a name="line.232"></a> |
| <span class="sourceLineNo">233</span> private static final Logger LOG = LoggerFactory.getLogger(HBaseFsck.class.getName());<a name="line.233"></a> |
| <span class="sourceLineNo">234</span> private ClusterMetrics status;<a name="line.234"></a> |
| <span class="sourceLineNo">235</span> private Connection connection;<a name="line.235"></a> |
| <span class="sourceLineNo">236</span> private Admin admin;<a name="line.236"></a> |
| <span class="sourceLineNo">237</span> private Table meta;<a name="line.237"></a> |
| <span class="sourceLineNo">238</span> // threads to do ||izable tasks: retrieve data from regionservers, handle overlapping regions<a name="line.238"></a> |
| <span class="sourceLineNo">239</span> protected ExecutorService executor;<a name="line.239"></a> |
| <span class="sourceLineNo">240</span> private long startMillis = EnvironmentEdgeManager.currentTime();<a name="line.240"></a> |
| <span class="sourceLineNo">241</span> private HFileCorruptionChecker hfcc;<a name="line.241"></a> |
| <span class="sourceLineNo">242</span> private int retcode = 0;<a name="line.242"></a> |
| <span class="sourceLineNo">243</span> private Path HBCK_LOCK_PATH;<a name="line.243"></a> |
| <span class="sourceLineNo">244</span> private FSDataOutputStream hbckOutFd;<a name="line.244"></a> |
| <span class="sourceLineNo">245</span> // This lock is to prevent cleanup of balancer resources twice between<a name="line.245"></a> |
| <span class="sourceLineNo">246</span> // ShutdownHook and the main code. We cleanup only if the connect() is<a name="line.246"></a> |
| <span class="sourceLineNo">247</span> // successful<a name="line.247"></a> |
| <span class="sourceLineNo">248</span> private final AtomicBoolean hbckLockCleanup = new AtomicBoolean(false);<a name="line.248"></a> |
| <span class="sourceLineNo">249</span><a name="line.249"></a> |
| <span class="sourceLineNo">250</span> // Unsupported options in HBase 2.0+<a name="line.250"></a> |
| <span class="sourceLineNo">251</span> private static final Set<String> unsupportedOptionsInV2 = Sets.newHashSet("-fix",<a name="line.251"></a> |
| <span class="sourceLineNo">252</span> "-fixAssignments", "-fixMeta", "-fixHdfsHoles", "-fixHdfsOrphans", "-fixTableOrphans",<a name="line.252"></a> |
| <span class="sourceLineNo">253</span> "-fixHdfsOverlaps", "-sidelineBigOverlaps", "-fixSplitParents", "-removeParents",<a name="line.253"></a> |
| <span class="sourceLineNo">254</span> "-fixEmptyMetaCells", "-repair", "-repairHoles", "-maxOverlapsToSideline", "-maxMerge");<a name="line.254"></a> |
| <span class="sourceLineNo">255</span><a name="line.255"></a> |
| <span class="sourceLineNo">256</span> /***********<a name="line.256"></a> |
| <span class="sourceLineNo">257</span> * Options<a name="line.257"></a> |
| <span class="sourceLineNo">258</span> ***********/<a name="line.258"></a> |
| <span class="sourceLineNo">259</span> private static boolean details = false; // do we display the full report<a name="line.259"></a> |
| <span class="sourceLineNo">260</span> private long timelag = DEFAULT_TIME_LAG; // tables whose modtime is older<a name="line.260"></a> |
| <span class="sourceLineNo">261</span> private static boolean forceExclusive = false; // only this hbck can modify HBase<a name="line.261"></a> |
| <span class="sourceLineNo">262</span> private boolean fixAssignments = false; // fix assignment errors?<a name="line.262"></a> |
| <span class="sourceLineNo">263</span> private boolean fixMeta = false; // fix meta errors?<a name="line.263"></a> |
| <span class="sourceLineNo">264</span> private boolean checkHdfs = true; // load and check fs consistency?<a name="line.264"></a> |
| <span class="sourceLineNo">265</span> private boolean fixHdfsHoles = false; // fix fs holes?<a name="line.265"></a> |
| <span class="sourceLineNo">266</span> private boolean fixHdfsOverlaps = false; // fix fs overlaps (risky)<a name="line.266"></a> |
| <span class="sourceLineNo">267</span> private boolean fixHdfsOrphans = false; // fix fs holes (missing .regioninfo)<a name="line.267"></a> |
| <span class="sourceLineNo">268</span> private boolean fixTableOrphans = false; // fix fs holes (missing .tableinfo)<a name="line.268"></a> |
| <span class="sourceLineNo">269</span> private boolean fixVersionFile = false; // fix missing hbase.version file in hdfs<a name="line.269"></a> |
| <span class="sourceLineNo">270</span> private boolean fixSplitParents = false; // fix lingering split parents<a name="line.270"></a> |
| <span class="sourceLineNo">271</span> private boolean removeParents = false; // remove split parents<a name="line.271"></a> |
| <span class="sourceLineNo">272</span> private boolean fixReferenceFiles = false; // fix lingering reference store file<a name="line.272"></a> |
| <span class="sourceLineNo">273</span> private boolean fixHFileLinks = false; // fix lingering HFileLinks<a name="line.273"></a> |
| <span class="sourceLineNo">274</span> private boolean fixEmptyMetaCells = false; // fix (remove) empty REGIONINFO_QUALIFIER rows<a name="line.274"></a> |
| <span class="sourceLineNo">275</span> private boolean fixReplication = false; // fix undeleted replication queues for removed peer<a name="line.275"></a> |
| <span class="sourceLineNo">276</span> private boolean cleanReplicationBarrier = false; // clean replication barriers of a table<a name="line.276"></a> |
| <span class="sourceLineNo">277</span> private boolean fixAny = false; // Set to true if any of the fix is required.<a name="line.277"></a> |
| <span class="sourceLineNo">278</span><a name="line.278"></a> |
| <span class="sourceLineNo">279</span> // limit checking/fixes to listed tables, if empty attempt to check/fix all<a name="line.279"></a> |
| <span class="sourceLineNo">280</span> // hbase:meta are always checked<a name="line.280"></a> |
| <span class="sourceLineNo">281</span> private Set<TableName> tablesIncluded = new HashSet<>();<a name="line.281"></a> |
| <span class="sourceLineNo">282</span> private TableName cleanReplicationBarrierTable;<a name="line.282"></a> |
| <span class="sourceLineNo">283</span> private int maxMerge = DEFAULT_MAX_MERGE; // maximum number of overlapping regions to merge<a name="line.283"></a> |
| <span class="sourceLineNo">284</span> // maximum number of overlapping regions to sideline<a name="line.284"></a> |
| <span class="sourceLineNo">285</span> private int maxOverlapsToSideline = DEFAULT_OVERLAPS_TO_SIDELINE;<a name="line.285"></a> |
| <span class="sourceLineNo">286</span> private boolean sidelineBigOverlaps = false; // sideline overlaps with >maxMerge regions<a name="line.286"></a> |
| <span class="sourceLineNo">287</span> private Path sidelineDir = null;<a name="line.287"></a> |
| <span class="sourceLineNo">288</span><a name="line.288"></a> |
| <span class="sourceLineNo">289</span> private boolean rerun = false; // if we tried to fix something, rerun hbck<a name="line.289"></a> |
| <span class="sourceLineNo">290</span> private static boolean summary = false; // if we want to print less output<a name="line.290"></a> |
| <span class="sourceLineNo">291</span> private boolean checkMetaOnly = false;<a name="line.291"></a> |
| <span class="sourceLineNo">292</span> private boolean checkRegionBoundaries = false;<a name="line.292"></a> |
| <span class="sourceLineNo">293</span> private boolean ignorePreCheckPermission = false; // if pre-check permission<a name="line.293"></a> |
| <span class="sourceLineNo">294</span><a name="line.294"></a> |
| <span class="sourceLineNo">295</span> /*********<a name="line.295"></a> |
| <span class="sourceLineNo">296</span> * State<a name="line.296"></a> |
| <span class="sourceLineNo">297</span> *********/<a name="line.297"></a> |
| <span class="sourceLineNo">298</span> final private HbckErrorReporter errors;<a name="line.298"></a> |
| <span class="sourceLineNo">299</span> int fixes = 0;<a name="line.299"></a> |
| <span class="sourceLineNo">300</span><a name="line.300"></a> |
| <span class="sourceLineNo">301</span> /**<a name="line.301"></a> |
| <span class="sourceLineNo">302</span> * This map contains the state of all hbck items. It maps from encoded region<a name="line.302"></a> |
| <span class="sourceLineNo">303</span> * name to HbckRegionInfo structure. The information contained in HbckRegionInfo is used<a name="line.303"></a> |
| <span class="sourceLineNo">304</span> * to detect and correct consistency (hdfs/meta/deployment) problems.<a name="line.304"></a> |
| <span class="sourceLineNo">305</span> */<a name="line.305"></a> |
| <span class="sourceLineNo">306</span> private TreeMap<String, HbckRegionInfo> regionInfoMap = new TreeMap<>();<a name="line.306"></a> |
| <span class="sourceLineNo">307</span> // Empty regioninfo qualifiers in hbase:meta<a name="line.307"></a> |
| <span class="sourceLineNo">308</span> private Set<Result> emptyRegionInfoQualifiers = new HashSet<>();<a name="line.308"></a> |
| <span class="sourceLineNo">309</span><a name="line.309"></a> |
| <span class="sourceLineNo">310</span> /**<a name="line.310"></a> |
| <span class="sourceLineNo">311</span> * This map from Tablename -> TableInfo contains the structures necessary to<a name="line.311"></a> |
| <span class="sourceLineNo">312</span> * detect table consistency problems (holes, dupes, overlaps). It is sorted<a name="line.312"></a> |
| <span class="sourceLineNo">313</span> * to prevent dupes.<a name="line.313"></a> |
| <span class="sourceLineNo">314</span> *<a name="line.314"></a> |
| <span class="sourceLineNo">315</span> * If tablesIncluded is empty, this map contains all tables.<a name="line.315"></a> |
| <span class="sourceLineNo">316</span> * Otherwise, it contains only meta tables and tables in tablesIncluded,<a name="line.316"></a> |
| <span class="sourceLineNo">317</span> * unless checkMetaOnly is specified, in which case, it contains only<a name="line.317"></a> |
| <span class="sourceLineNo">318</span> * the meta table<a name="line.318"></a> |
| <span class="sourceLineNo">319</span> */<a name="line.319"></a> |
| <span class="sourceLineNo">320</span> private SortedMap<TableName, HbckTableInfo> tablesInfo = new ConcurrentSkipListMap<>();<a name="line.320"></a> |
| <span class="sourceLineNo">321</span><a name="line.321"></a> |
| <span class="sourceLineNo">322</span> /**<a name="line.322"></a> |
| <span class="sourceLineNo">323</span> * When initially looking at HDFS, we attempt to find any orphaned data.<a name="line.323"></a> |
| <span class="sourceLineNo">324</span> */<a name="line.324"></a> |
| <span class="sourceLineNo">325</span> private List<HbckRegionInfo> orphanHdfsDirs = Collections.synchronizedList(new ArrayList<>());<a name="line.325"></a> |
| <span class="sourceLineNo">326</span><a name="line.326"></a> |
| <span class="sourceLineNo">327</span> private Map<TableName, Set<String>> orphanTableDirs = new HashMap<>();<a name="line.327"></a> |
| <span class="sourceLineNo">328</span> private Map<TableName, TableState> tableStates = new HashMap<>();<a name="line.328"></a> |
| <span class="sourceLineNo">329</span> private final RetryCounterFactory lockFileRetryCounterFactory;<a name="line.329"></a> |
| <span class="sourceLineNo">330</span> private final RetryCounterFactory createZNodeRetryCounterFactory;<a name="line.330"></a> |
| <span class="sourceLineNo">331</span><a name="line.331"></a> |
| <span class="sourceLineNo">332</span> private Map<TableName, Set<String>> skippedRegions = new HashMap<>();<a name="line.332"></a> |
| <span class="sourceLineNo">333</span><a name="line.333"></a> |
| <span class="sourceLineNo">334</span> private ZKWatcher zkw = null;<a name="line.334"></a> |
| <span class="sourceLineNo">335</span> private String hbckEphemeralNodePath = null;<a name="line.335"></a> |
| <span class="sourceLineNo">336</span> private boolean hbckZodeCreated = false;<a name="line.336"></a> |
| <span class="sourceLineNo">337</span><a name="line.337"></a> |
| <span class="sourceLineNo">338</span> /**<a name="line.338"></a> |
| <span class="sourceLineNo">339</span> * Constructor<a name="line.339"></a> |
| <span class="sourceLineNo">340</span> *<a name="line.340"></a> |
| <span class="sourceLineNo">341</span> * @param conf Configuration object<a name="line.341"></a> |
| <span class="sourceLineNo">342</span> * @throws MasterNotRunningException if the master is not running<a name="line.342"></a> |
| <span class="sourceLineNo">343</span> * @throws ZooKeeperConnectionException if unable to connect to ZooKeeper<a name="line.343"></a> |
| <span class="sourceLineNo">344</span> */<a name="line.344"></a> |
| <span class="sourceLineNo">345</span> public HBaseFsck(Configuration conf) throws IOException, ClassNotFoundException {<a name="line.345"></a> |
| <span class="sourceLineNo">346</span> this(conf, createThreadPool(conf));<a name="line.346"></a> |
| <span class="sourceLineNo">347</span> }<a name="line.347"></a> |
| <span class="sourceLineNo">348</span><a name="line.348"></a> |
| <span class="sourceLineNo">349</span> private static ExecutorService createThreadPool(Configuration conf) {<a name="line.349"></a> |
| <span class="sourceLineNo">350</span> int numThreads = conf.getInt("hbasefsck.numthreads", MAX_NUM_THREADS);<a name="line.350"></a> |
| <span class="sourceLineNo">351</span> return new ScheduledThreadPoolExecutor(numThreads,<a name="line.351"></a> |
| <span class="sourceLineNo">352</span> new ThreadFactoryBuilder().setNameFormat("hbasefsck-pool-%d").setDaemon(true)<a name="line.352"></a> |
| <span class="sourceLineNo">353</span> .setUncaughtExceptionHandler(Threads.LOGGING_EXCEPTION_HANDLER).build());<a name="line.353"></a> |
| <span class="sourceLineNo">354</span> }<a name="line.354"></a> |
| <span class="sourceLineNo">355</span><a name="line.355"></a> |
| <span class="sourceLineNo">356</span> /**<a name="line.356"></a> |
| <span class="sourceLineNo">357</span> * Constructor<a name="line.357"></a> |
| <span class="sourceLineNo">358</span> *<a name="line.358"></a> |
| <span class="sourceLineNo">359</span> * @param conf<a name="line.359"></a> |
| <span class="sourceLineNo">360</span> * Configuration object<a name="line.360"></a> |
| <span class="sourceLineNo">361</span> * @throws MasterNotRunningException<a name="line.361"></a> |
| <span class="sourceLineNo">362</span> * if the master is not running<a name="line.362"></a> |
| <span class="sourceLineNo">363</span> * @throws ZooKeeperConnectionException<a name="line.363"></a> |
| <span class="sourceLineNo">364</span> * if unable to connect to ZooKeeper<a name="line.364"></a> |
| <span class="sourceLineNo">365</span> */<a name="line.365"></a> |
| <span class="sourceLineNo">366</span> public HBaseFsck(Configuration conf, ExecutorService exec) throws MasterNotRunningException,<a name="line.366"></a> |
| <span class="sourceLineNo">367</span> ZooKeeperConnectionException, IOException, ClassNotFoundException {<a name="line.367"></a> |
| <span class="sourceLineNo">368</span> super(conf);<a name="line.368"></a> |
| <span class="sourceLineNo">369</span> errors = getErrorReporter(getConf());<a name="line.369"></a> |
| <span class="sourceLineNo">370</span> this.executor = exec;<a name="line.370"></a> |
| <span class="sourceLineNo">371</span> lockFileRetryCounterFactory = createLockRetryCounterFactory(getConf());<a name="line.371"></a> |
| <span class="sourceLineNo">372</span> createZNodeRetryCounterFactory = createZnodeRetryCounterFactory(getConf());<a name="line.372"></a> |
| <span class="sourceLineNo">373</span> zkw = createZooKeeperWatcher();<a name="line.373"></a> |
| <span class="sourceLineNo">374</span> }<a name="line.374"></a> |
| <span class="sourceLineNo">375</span><a name="line.375"></a> |
| <span class="sourceLineNo">376</span> /**<a name="line.376"></a> |
| <span class="sourceLineNo">377</span> * @return A retry counter factory configured for retrying lock file creation.<a name="line.377"></a> |
| <span class="sourceLineNo">378</span> */<a name="line.378"></a> |
| <span class="sourceLineNo">379</span> public static RetryCounterFactory createLockRetryCounterFactory(Configuration conf) {<a name="line.379"></a> |
| <span class="sourceLineNo">380</span> return new RetryCounterFactory(<a name="line.380"></a> |
| <span class="sourceLineNo">381</span> conf.getInt("hbase.hbck.lockfile.attempts", DEFAULT_MAX_LOCK_FILE_ATTEMPTS),<a name="line.381"></a> |
| <span class="sourceLineNo">382</span> conf.getInt("hbase.hbck.lockfile.attempt.sleep.interval",<a name="line.382"></a> |
| <span class="sourceLineNo">383</span> DEFAULT_LOCK_FILE_ATTEMPT_SLEEP_INTERVAL),<a name="line.383"></a> |
| <span class="sourceLineNo">384</span> conf.getInt("hbase.hbck.lockfile.attempt.maxsleeptime",<a name="line.384"></a> |
| <span class="sourceLineNo">385</span> DEFAULT_LOCK_FILE_ATTEMPT_MAX_SLEEP_TIME));<a name="line.385"></a> |
| <span class="sourceLineNo">386</span> }<a name="line.386"></a> |
| <span class="sourceLineNo">387</span><a name="line.387"></a> |
| <span class="sourceLineNo">388</span> /**<a name="line.388"></a> |
| <span class="sourceLineNo">389</span> * @return A retry counter factory configured for retrying znode creation.<a name="line.389"></a> |
| <span class="sourceLineNo">390</span> */<a name="line.390"></a> |
| <span class="sourceLineNo">391</span> private static RetryCounterFactory createZnodeRetryCounterFactory(Configuration conf) {<a name="line.391"></a> |
| <span class="sourceLineNo">392</span> return new RetryCounterFactory(<a name="line.392"></a> |
| <span class="sourceLineNo">393</span> conf.getInt("hbase.hbck.createznode.attempts", DEFAULT_MAX_CREATE_ZNODE_ATTEMPTS),<a name="line.393"></a> |
| <span class="sourceLineNo">394</span> conf.getInt("hbase.hbck.createznode.attempt.sleep.interval",<a name="line.394"></a> |
| <span class="sourceLineNo">395</span> DEFAULT_CREATE_ZNODE_ATTEMPT_SLEEP_INTERVAL),<a name="line.395"></a> |
| <span class="sourceLineNo">396</span> conf.getInt("hbase.hbck.createznode.attempt.maxsleeptime",<a name="line.396"></a> |
| <span class="sourceLineNo">397</span> DEFAULT_CREATE_ZNODE_ATTEMPT_MAX_SLEEP_TIME));<a name="line.397"></a> |
| <span class="sourceLineNo">398</span> }<a name="line.398"></a> |
| <span class="sourceLineNo">399</span><a name="line.399"></a> |
| <span class="sourceLineNo">400</span> /**<a name="line.400"></a> |
| <span class="sourceLineNo">401</span> * @return Return the tmp dir this tool writes too.<a name="line.401"></a> |
| <span class="sourceLineNo">402</span> */<a name="line.402"></a> |
| <span class="sourceLineNo">403</span> @VisibleForTesting<a name="line.403"></a> |
| <span class="sourceLineNo">404</span> public static Path getTmpDir(Configuration conf) throws IOException {<a name="line.404"></a> |
| <span class="sourceLineNo">405</span> return new Path(CommonFSUtils.getRootDir(conf), HConstants.HBASE_TEMP_DIRECTORY);<a name="line.405"></a> |
| <span class="sourceLineNo">406</span> }<a name="line.406"></a> |
| <span class="sourceLineNo">407</span><a name="line.407"></a> |
| <span class="sourceLineNo">408</span> private static class FileLockCallable implements Callable<FSDataOutputStream> {<a name="line.408"></a> |
| <span class="sourceLineNo">409</span> RetryCounter retryCounter;<a name="line.409"></a> |
| <span class="sourceLineNo">410</span> private final Configuration conf;<a name="line.410"></a> |
| <span class="sourceLineNo">411</span> private Path hbckLockPath = null;<a name="line.411"></a> |
| <span class="sourceLineNo">412</span><a name="line.412"></a> |
| <span class="sourceLineNo">413</span> public FileLockCallable(Configuration conf, RetryCounter retryCounter) {<a name="line.413"></a> |
| <span class="sourceLineNo">414</span> this.retryCounter = retryCounter;<a name="line.414"></a> |
| <span class="sourceLineNo">415</span> this.conf = conf;<a name="line.415"></a> |
| <span class="sourceLineNo">416</span> }<a name="line.416"></a> |
| <span class="sourceLineNo">417</span><a name="line.417"></a> |
| <span class="sourceLineNo">418</span> /**<a name="line.418"></a> |
| <span class="sourceLineNo">419</span> * @return Will be <code>null</code> unless you call {@link #call()}<a name="line.419"></a> |
| <span class="sourceLineNo">420</span> */<a name="line.420"></a> |
| <span class="sourceLineNo">421</span> Path getHbckLockPath() {<a name="line.421"></a> |
| <span class="sourceLineNo">422</span> return this.hbckLockPath;<a name="line.422"></a> |
| <span class="sourceLineNo">423</span> }<a name="line.423"></a> |
| <span class="sourceLineNo">424</span><a name="line.424"></a> |
| <span class="sourceLineNo">425</span> @Override<a name="line.425"></a> |
| <span class="sourceLineNo">426</span> public FSDataOutputStream call() throws IOException {<a name="line.426"></a> |
| <span class="sourceLineNo">427</span> try {<a name="line.427"></a> |
| <span class="sourceLineNo">428</span> FileSystem fs = CommonFSUtils.getCurrentFileSystem(this.conf);<a name="line.428"></a> |
| <span class="sourceLineNo">429</span> FsPermission defaultPerms =<a name="line.429"></a> |
| <span class="sourceLineNo">430</span> CommonFSUtils.getFilePermissions(fs, this.conf, HConstants.DATA_FILE_UMASK_KEY);<a name="line.430"></a> |
| <span class="sourceLineNo">431</span> Path tmpDir = getTmpDir(conf);<a name="line.431"></a> |
| <span class="sourceLineNo">432</span> this.hbckLockPath = new Path(tmpDir, HBCK_LOCK_FILE);<a name="line.432"></a> |
| <span class="sourceLineNo">433</span> fs.mkdirs(tmpDir);<a name="line.433"></a> |
| <span class="sourceLineNo">434</span> final FSDataOutputStream out = createFileWithRetries(fs, this.hbckLockPath, defaultPerms);<a name="line.434"></a> |
| <span class="sourceLineNo">435</span> out.writeBytes(InetAddress.getLocalHost().toString());<a name="line.435"></a> |
| <span class="sourceLineNo">436</span> // Add a note into the file we write on why hbase2 is writing out an hbck1 lock file.<a name="line.436"></a> |
| <span class="sourceLineNo">437</span> out.writeBytes(" Written by an hbase-2.x Master to block an " +<a name="line.437"></a> |
| <span class="sourceLineNo">438</span> "attempt by an hbase-1.x HBCK tool making modification to state. " +<a name="line.438"></a> |
| <span class="sourceLineNo">439</span> "See 'HBCK must match HBase server version' in the hbase refguide.");<a name="line.439"></a> |
| <span class="sourceLineNo">440</span> out.flush();<a name="line.440"></a> |
| <span class="sourceLineNo">441</span> return out;<a name="line.441"></a> |
| <span class="sourceLineNo">442</span> } catch(RemoteException e) {<a name="line.442"></a> |
| <span class="sourceLineNo">443</span> if(AlreadyBeingCreatedException.class.getName().equals(e.getClassName())){<a name="line.443"></a> |
| <span class="sourceLineNo">444</span> return null;<a name="line.444"></a> |
| <span class="sourceLineNo">445</span> } else {<a name="line.445"></a> |
| <span class="sourceLineNo">446</span> throw e;<a name="line.446"></a> |
| <span class="sourceLineNo">447</span> }<a name="line.447"></a> |
| <span class="sourceLineNo">448</span> }<a name="line.448"></a> |
| <span class="sourceLineNo">449</span> }<a name="line.449"></a> |
| <span class="sourceLineNo">450</span><a name="line.450"></a> |
| <span class="sourceLineNo">451</span> private FSDataOutputStream createFileWithRetries(final FileSystem fs,<a name="line.451"></a> |
| <span class="sourceLineNo">452</span> final Path hbckLockFilePath, final FsPermission defaultPerms)<a name="line.452"></a> |
| <span class="sourceLineNo">453</span> throws IOException {<a name="line.453"></a> |
| <span class="sourceLineNo">454</span> IOException exception = null;<a name="line.454"></a> |
| <span class="sourceLineNo">455</span> do {<a name="line.455"></a> |
| <span class="sourceLineNo">456</span> try {<a name="line.456"></a> |
| <span class="sourceLineNo">457</span> return CommonFSUtils.create(fs, hbckLockFilePath, defaultPerms, false);<a name="line.457"></a> |
| <span class="sourceLineNo">458</span> } catch (IOException ioe) {<a name="line.458"></a> |
| <span class="sourceLineNo">459</span> LOG.info("Failed to create lock file " + hbckLockFilePath.getName()<a name="line.459"></a> |
| <span class="sourceLineNo">460</span> + ", try=" + (retryCounter.getAttemptTimes() + 1) + " of "<a name="line.460"></a> |
| <span class="sourceLineNo">461</span> + retryCounter.getMaxAttempts());<a name="line.461"></a> |
| <span class="sourceLineNo">462</span> LOG.debug("Failed to create lock file " + hbckLockFilePath.getName(),<a name="line.462"></a> |
| <span class="sourceLineNo">463</span> ioe);<a name="line.463"></a> |
| <span class="sourceLineNo">464</span> try {<a name="line.464"></a> |
| <span class="sourceLineNo">465</span> exception = ioe;<a name="line.465"></a> |
| <span class="sourceLineNo">466</span> retryCounter.sleepUntilNextRetry();<a name="line.466"></a> |
| <span class="sourceLineNo">467</span> } catch (InterruptedException ie) {<a name="line.467"></a> |
| <span class="sourceLineNo">468</span> throw (InterruptedIOException) new InterruptedIOException(<a name="line.468"></a> |
| <span class="sourceLineNo">469</span> "Can't create lock file " + hbckLockFilePath.getName())<a name="line.469"></a> |
| <span class="sourceLineNo">470</span> .initCause(ie);<a name="line.470"></a> |
| <span class="sourceLineNo">471</span> }<a name="line.471"></a> |
| <span class="sourceLineNo">472</span> }<a name="line.472"></a> |
| <span class="sourceLineNo">473</span> } while (retryCounter.shouldRetry());<a name="line.473"></a> |
| <span class="sourceLineNo">474</span><a name="line.474"></a> |
| <span class="sourceLineNo">475</span> throw exception;<a name="line.475"></a> |
| <span class="sourceLineNo">476</span> }<a name="line.476"></a> |
| <span class="sourceLineNo">477</span> }<a name="line.477"></a> |
| <span class="sourceLineNo">478</span><a name="line.478"></a> |
| <span class="sourceLineNo">479</span> /**<a name="line.479"></a> |
| <span class="sourceLineNo">480</span> * This method maintains a lock using a file. If the creation fails we return null<a name="line.480"></a> |
| <span class="sourceLineNo">481</span> *<a name="line.481"></a> |
| <span class="sourceLineNo">482</span> * @return FSDataOutputStream object corresponding to the newly opened lock file<a name="line.482"></a> |
| <span class="sourceLineNo">483</span> * @throws IOException if IO failure occurs<a name="line.483"></a> |
| <span class="sourceLineNo">484</span> */<a name="line.484"></a> |
| <span class="sourceLineNo">485</span> public static Pair<Path, FSDataOutputStream> checkAndMarkRunningHbck(Configuration conf,<a name="line.485"></a> |
| <span class="sourceLineNo">486</span> RetryCounter retryCounter) throws IOException {<a name="line.486"></a> |
| <span class="sourceLineNo">487</span> FileLockCallable callable = new FileLockCallable(conf, retryCounter);<a name="line.487"></a> |
| <span class="sourceLineNo">488</span> ExecutorService executor = Executors.newFixedThreadPool(1);<a name="line.488"></a> |
| <span class="sourceLineNo">489</span> FutureTask<FSDataOutputStream> futureTask = new FutureTask<>(callable);<a name="line.489"></a> |
| <span class="sourceLineNo">490</span> executor.execute(futureTask);<a name="line.490"></a> |
| <span class="sourceLineNo">491</span> final int timeoutInSeconds = conf.getInt(<a name="line.491"></a> |
| <span class="sourceLineNo">492</span> "hbase.hbck.lockfile.maxwaittime", DEFAULT_WAIT_FOR_LOCK_TIMEOUT);<a name="line.492"></a> |
| <span class="sourceLineNo">493</span> FSDataOutputStream stream = null;<a name="line.493"></a> |
| <span class="sourceLineNo">494</span> try {<a name="line.494"></a> |
| <span class="sourceLineNo">495</span> stream = futureTask.get(timeoutInSeconds, TimeUnit.SECONDS);<a name="line.495"></a> |
| <span class="sourceLineNo">496</span> } catch (ExecutionException ee) {<a name="line.496"></a> |
| <span class="sourceLineNo">497</span> LOG.warn("Encountered exception when opening lock file", ee);<a name="line.497"></a> |
| <span class="sourceLineNo">498</span> } catch (InterruptedException ie) {<a name="line.498"></a> |
| <span class="sourceLineNo">499</span> LOG.warn("Interrupted when opening lock file", ie);<a name="line.499"></a> |
| <span class="sourceLineNo">500</span> Thread.currentThread().interrupt();<a name="line.500"></a> |
| <span class="sourceLineNo">501</span> } catch (TimeoutException exception) {<a name="line.501"></a> |
| <span class="sourceLineNo">502</span> // took too long to obtain lock<a name="line.502"></a> |
| <span class="sourceLineNo">503</span> LOG.warn("Took more than " + timeoutInSeconds + " seconds in obtaining lock");<a name="line.503"></a> |
| <span class="sourceLineNo">504</span> futureTask.cancel(true);<a name="line.504"></a> |
| <span class="sourceLineNo">505</span> } finally {<a name="line.505"></a> |
| <span class="sourceLineNo">506</span> executor.shutdownNow();<a name="line.506"></a> |
| <span class="sourceLineNo">507</span> }<a name="line.507"></a> |
| <span class="sourceLineNo">508</span> return new Pair<Path, FSDataOutputStream>(callable.getHbckLockPath(), stream);<a name="line.508"></a> |
| <span class="sourceLineNo">509</span> }<a name="line.509"></a> |
| <span class="sourceLineNo">510</span><a name="line.510"></a> |
| <span class="sourceLineNo">511</span> private void unlockHbck() {<a name="line.511"></a> |
| <span class="sourceLineNo">512</span> if (isExclusive() && hbckLockCleanup.compareAndSet(true, false)) {<a name="line.512"></a> |
| <span class="sourceLineNo">513</span> RetryCounter retryCounter = lockFileRetryCounterFactory.create();<a name="line.513"></a> |
| <span class="sourceLineNo">514</span> do {<a name="line.514"></a> |
| <span class="sourceLineNo">515</span> try {<a name="line.515"></a> |
| <span class="sourceLineNo">516</span> IOUtils.closeQuietly(hbckOutFd);<a name="line.516"></a> |
| <span class="sourceLineNo">517</span> CommonFSUtils.delete(CommonFSUtils.getCurrentFileSystem(getConf()), HBCK_LOCK_PATH, true);<a name="line.517"></a> |
| <span class="sourceLineNo">518</span> LOG.info("Finishing hbck");<a name="line.518"></a> |
| <span class="sourceLineNo">519</span> return;<a name="line.519"></a> |
| <span class="sourceLineNo">520</span> } catch (IOException ioe) {<a name="line.520"></a> |
| <span class="sourceLineNo">521</span> LOG.info("Failed to delete " + HBCK_LOCK_PATH + ", try="<a name="line.521"></a> |
| <span class="sourceLineNo">522</span> + (retryCounter.getAttemptTimes() + 1) + " of "<a name="line.522"></a> |
| <span class="sourceLineNo">523</span> + retryCounter.getMaxAttempts());<a name="line.523"></a> |
| <span class="sourceLineNo">524</span> LOG.debug("Failed to delete " + HBCK_LOCK_PATH, ioe);<a name="line.524"></a> |
| <span class="sourceLineNo">525</span> try {<a name="line.525"></a> |
| <span class="sourceLineNo">526</span> retryCounter.sleepUntilNextRetry();<a name="line.526"></a> |
| <span class="sourceLineNo">527</span> } catch (InterruptedException ie) {<a name="line.527"></a> |
| <span class="sourceLineNo">528</span> Thread.currentThread().interrupt();<a name="line.528"></a> |
| <span class="sourceLineNo">529</span> LOG.warn("Interrupted while deleting lock file" +<a name="line.529"></a> |
| <span class="sourceLineNo">530</span> HBCK_LOCK_PATH);<a name="line.530"></a> |
| <span class="sourceLineNo">531</span> return;<a name="line.531"></a> |
| <span class="sourceLineNo">532</span> }<a name="line.532"></a> |
| <span class="sourceLineNo">533</span> }<a name="line.533"></a> |
| <span class="sourceLineNo">534</span> } while (retryCounter.shouldRetry());<a name="line.534"></a> |
| <span class="sourceLineNo">535</span> }<a name="line.535"></a> |
| <span class="sourceLineNo">536</span> }<a name="line.536"></a> |
| <span class="sourceLineNo">537</span><a name="line.537"></a> |
| <span class="sourceLineNo">538</span> /**<a name="line.538"></a> |
| <span class="sourceLineNo">539</span> * To repair region consistency, one must call connect() in order to repair<a name="line.539"></a> |
| <span class="sourceLineNo">540</span> * online state.<a name="line.540"></a> |
| <span class="sourceLineNo">541</span> */<a name="line.541"></a> |
| <span class="sourceLineNo">542</span> public void connect() throws IOException {<a name="line.542"></a> |
| <span class="sourceLineNo">543</span><a name="line.543"></a> |
| <span class="sourceLineNo">544</span> if (isExclusive()) {<a name="line.544"></a> |
| <span class="sourceLineNo">545</span> // Grab the lock<a name="line.545"></a> |
| <span class="sourceLineNo">546</span> Pair<Path, FSDataOutputStream> pair =<a name="line.546"></a> |
| <span class="sourceLineNo">547</span> checkAndMarkRunningHbck(getConf(), this.lockFileRetryCounterFactory.create());<a name="line.547"></a> |
| <span class="sourceLineNo">548</span> HBCK_LOCK_PATH = pair.getFirst();<a name="line.548"></a> |
| <span class="sourceLineNo">549</span> this.hbckOutFd = pair.getSecond();<a name="line.549"></a> |
| <span class="sourceLineNo">550</span> if (hbckOutFd == null) {<a name="line.550"></a> |
| <span class="sourceLineNo">551</span> setRetCode(-1);<a name="line.551"></a> |
| <span class="sourceLineNo">552</span> LOG.error("Another instance of hbck is fixing HBase, exiting this instance. " +<a name="line.552"></a> |
| <span class="sourceLineNo">553</span> "[If you are sure no other instance is running, delete the lock file " +<a name="line.553"></a> |
| <span class="sourceLineNo">554</span> HBCK_LOCK_PATH + " and rerun the tool]");<a name="line.554"></a> |
| <span class="sourceLineNo">555</span> throw new IOException("Duplicate hbck - Abort");<a name="line.555"></a> |
| <span class="sourceLineNo">556</span> }<a name="line.556"></a> |
| <span class="sourceLineNo">557</span><a name="line.557"></a> |
| <span class="sourceLineNo">558</span> // Make sure to cleanup the lock<a name="line.558"></a> |
| <span class="sourceLineNo">559</span> hbckLockCleanup.set(true);<a name="line.559"></a> |
| <span class="sourceLineNo">560</span> }<a name="line.560"></a> |
| <span class="sourceLineNo">561</span><a name="line.561"></a> |
| <span class="sourceLineNo">562</span><a name="line.562"></a> |
| <span class="sourceLineNo">563</span> // Add a shutdown hook to this thread, in case user tries to<a name="line.563"></a> |
| <span class="sourceLineNo">564</span> // kill the hbck with a ctrl-c, we want to cleanup the lock so that<a name="line.564"></a> |
| <span class="sourceLineNo">565</span> // it is available for further calls<a name="line.565"></a> |
| <span class="sourceLineNo">566</span> Runtime.getRuntime().addShutdownHook(new Thread() {<a name="line.566"></a> |
| <span class="sourceLineNo">567</span> @Override<a name="line.567"></a> |
| <span class="sourceLineNo">568</span> public void run() {<a name="line.568"></a> |
| <span class="sourceLineNo">569</span> IOUtils.closeQuietly(HBaseFsck.this);<a name="line.569"></a> |
| <span class="sourceLineNo">570</span> cleanupHbckZnode();<a name="line.570"></a> |
| <span class="sourceLineNo">571</span> unlockHbck();<a name="line.571"></a> |
| <span class="sourceLineNo">572</span> }<a name="line.572"></a> |
| <span class="sourceLineNo">573</span> });<a name="line.573"></a> |
| <span class="sourceLineNo">574</span><a name="line.574"></a> |
| <span class="sourceLineNo">575</span> LOG.info("Launching hbck");<a name="line.575"></a> |
| <span class="sourceLineNo">576</span><a name="line.576"></a> |
| <span class="sourceLineNo">577</span> connection = ConnectionFactory.createConnection(getConf());<a name="line.577"></a> |
| <span class="sourceLineNo">578</span> admin = connection.getAdmin();<a name="line.578"></a> |
| <span class="sourceLineNo">579</span> meta = connection.getTable(TableName.META_TABLE_NAME);<a name="line.579"></a> |
| <span class="sourceLineNo">580</span> status = admin.getClusterMetrics(EnumSet.of(Option.LIVE_SERVERS,<a name="line.580"></a> |
| <span class="sourceLineNo">581</span> Option.DEAD_SERVERS, Option.MASTER, Option.BACKUP_MASTERS,<a name="line.581"></a> |
| <span class="sourceLineNo">582</span> Option.REGIONS_IN_TRANSITION, Option.HBASE_VERSION));<a name="line.582"></a> |
| <span class="sourceLineNo">583</span> }<a name="line.583"></a> |
| <span class="sourceLineNo">584</span><a name="line.584"></a> |
| <span class="sourceLineNo">585</span> /**<a name="line.585"></a> |
| <span class="sourceLineNo">586</span> * Get deployed regions according to the region servers.<a name="line.586"></a> |
| <span class="sourceLineNo">587</span> */<a name="line.587"></a> |
| <span class="sourceLineNo">588</span> private void loadDeployedRegions() throws IOException, InterruptedException {<a name="line.588"></a> |
| <span class="sourceLineNo">589</span> // From the master, get a list of all known live region servers<a name="line.589"></a> |
| <span class="sourceLineNo">590</span> Collection<ServerName> regionServers = status.getLiveServerMetrics().keySet();<a name="line.590"></a> |
| <span class="sourceLineNo">591</span> errors.print("Number of live region servers: " + regionServers.size());<a name="line.591"></a> |
| <span class="sourceLineNo">592</span> if (details) {<a name="line.592"></a> |
| <span class="sourceLineNo">593</span> for (ServerName rsinfo: regionServers) {<a name="line.593"></a> |
| <span class="sourceLineNo">594</span> errors.print(" " + rsinfo.getServerName());<a name="line.594"></a> |
| <span class="sourceLineNo">595</span> }<a name="line.595"></a> |
| <span class="sourceLineNo">596</span> }<a name="line.596"></a> |
| <span class="sourceLineNo">597</span><a name="line.597"></a> |
| <span class="sourceLineNo">598</span> // From the master, get a list of all dead region servers<a name="line.598"></a> |
| <span class="sourceLineNo">599</span> Collection<ServerName> deadRegionServers = status.getDeadServerNames();<a name="line.599"></a> |
| <span class="sourceLineNo">600</span> errors.print("Number of dead region servers: " + deadRegionServers.size());<a name="line.600"></a> |
| <span class="sourceLineNo">601</span> if (details) {<a name="line.601"></a> |
| <span class="sourceLineNo">602</span> for (ServerName name: deadRegionServers) {<a name="line.602"></a> |
| <span class="sourceLineNo">603</span> errors.print(" " + name);<a name="line.603"></a> |
| <span class="sourceLineNo">604</span> }<a name="line.604"></a> |
| <span class="sourceLineNo">605</span> }<a name="line.605"></a> |
| <span class="sourceLineNo">606</span><a name="line.606"></a> |
| <span class="sourceLineNo">607</span> // Print the current master name and state<a name="line.607"></a> |
| <span class="sourceLineNo">608</span> errors.print("Master: " + status.getMasterName());<a name="line.608"></a> |
| <span class="sourceLineNo">609</span><a name="line.609"></a> |
| <span class="sourceLineNo">610</span> // Print the list of all backup masters<a name="line.610"></a> |
| <span class="sourceLineNo">611</span> Collection<ServerName> backupMasters = status.getBackupMasterNames();<a name="line.611"></a> |
| <span class="sourceLineNo">612</span> errors.print("Number of backup masters: " + backupMasters.size());<a name="line.612"></a> |
| <span class="sourceLineNo">613</span> if (details) {<a name="line.613"></a> |
| <span class="sourceLineNo">614</span> for (ServerName name: backupMasters) {<a name="line.614"></a> |
| <span class="sourceLineNo">615</span> errors.print(" " + name);<a name="line.615"></a> |
| <span class="sourceLineNo">616</span> }<a name="line.616"></a> |
| <span class="sourceLineNo">617</span> }<a name="line.617"></a> |
| <span class="sourceLineNo">618</span><a name="line.618"></a> |
| <span class="sourceLineNo">619</span> errors.print("Average load: " + status.getAverageLoad());<a name="line.619"></a> |
| <span class="sourceLineNo">620</span> errors.print("Number of requests: " + status.getRequestCount());<a name="line.620"></a> |
| <span class="sourceLineNo">621</span> errors.print("Number of regions: " + status.getRegionCount());<a name="line.621"></a> |
| <span class="sourceLineNo">622</span><a name="line.622"></a> |
| <span class="sourceLineNo">623</span> List<RegionState> rits = status.getRegionStatesInTransition();<a name="line.623"></a> |
| <span class="sourceLineNo">624</span> errors.print("Number of regions in transition: " + rits.size());<a name="line.624"></a> |
| <span class="sourceLineNo">625</span> if (details) {<a name="line.625"></a> |
| <span class="sourceLineNo">626</span> for (RegionState state: rits) {<a name="line.626"></a> |
| <span class="sourceLineNo">627</span> errors.print(" " + state.toDescriptiveString());<a name="line.627"></a> |
| <span class="sourceLineNo">628</span> }<a name="line.628"></a> |
| <span class="sourceLineNo">629</span> }<a name="line.629"></a> |
| <span class="sourceLineNo">630</span><a name="line.630"></a> |
| <span class="sourceLineNo">631</span> // Determine what's deployed<a name="line.631"></a> |
| <span class="sourceLineNo">632</span> processRegionServers(regionServers);<a name="line.632"></a> |
| <span class="sourceLineNo">633</span> }<a name="line.633"></a> |
| <span class="sourceLineNo">634</span><a name="line.634"></a> |
| <span class="sourceLineNo">635</span> /**<a name="line.635"></a> |
| <span class="sourceLineNo">636</span> * Clear the current state of hbck.<a name="line.636"></a> |
| <span class="sourceLineNo">637</span> */<a name="line.637"></a> |
| <span class="sourceLineNo">638</span> private void clearState() {<a name="line.638"></a> |
| <span class="sourceLineNo">639</span> // Make sure regionInfo is empty before starting<a name="line.639"></a> |
| <span class="sourceLineNo">640</span> fixes = 0;<a name="line.640"></a> |
| <span class="sourceLineNo">641</span> regionInfoMap.clear();<a name="line.641"></a> |
| <span class="sourceLineNo">642</span> emptyRegionInfoQualifiers.clear();<a name="line.642"></a> |
| <span class="sourceLineNo">643</span> tableStates.clear();<a name="line.643"></a> |
| <span class="sourceLineNo">644</span> errors.clear();<a name="line.644"></a> |
| <span class="sourceLineNo">645</span> tablesInfo.clear();<a name="line.645"></a> |
| <span class="sourceLineNo">646</span> orphanHdfsDirs.clear();<a name="line.646"></a> |
| <span class="sourceLineNo">647</span> skippedRegions.clear();<a name="line.647"></a> |
| <span class="sourceLineNo">648</span> }<a name="line.648"></a> |
| <span class="sourceLineNo">649</span><a name="line.649"></a> |
| <span class="sourceLineNo">650</span> /**<a name="line.650"></a> |
| <span class="sourceLineNo">651</span> * This repair method analyzes hbase data in hdfs and repairs it to satisfy<a name="line.651"></a> |
| <span class="sourceLineNo">652</span> * the table integrity rules. HBase doesn't need to be online for this<a name="line.652"></a> |
| <span class="sourceLineNo">653</span> * operation to work.<a name="line.653"></a> |
| <span class="sourceLineNo">654</span> */<a name="line.654"></a> |
| <span class="sourceLineNo">655</span> public void offlineHdfsIntegrityRepair() throws IOException, InterruptedException {<a name="line.655"></a> |
| <span class="sourceLineNo">656</span> // Initial pass to fix orphans.<a name="line.656"></a> |
| <span class="sourceLineNo">657</span> if (shouldCheckHdfs() && (shouldFixHdfsOrphans() || shouldFixHdfsHoles()<a name="line.657"></a> |
| <span class="sourceLineNo">658</span> || shouldFixHdfsOverlaps() || shouldFixTableOrphans())) {<a name="line.658"></a> |
| <span class="sourceLineNo">659</span> LOG.info("Loading regioninfos HDFS");<a name="line.659"></a> |
| <span class="sourceLineNo">660</span> // if nothing is happening this should always complete in two iterations.<a name="line.660"></a> |
| <span class="sourceLineNo">661</span> int maxIterations = getConf().getInt("hbase.hbck.integrityrepair.iterations.max", 3);<a name="line.661"></a> |
| <span class="sourceLineNo">662</span> int curIter = 0;<a name="line.662"></a> |
| <span class="sourceLineNo">663</span> do {<a name="line.663"></a> |
| <span class="sourceLineNo">664</span> clearState(); // clears hbck state and reset fixes to 0 and.<a name="line.664"></a> |
| <span class="sourceLineNo">665</span> // repair what's on HDFS<a name="line.665"></a> |
| <span class="sourceLineNo">666</span> restoreHdfsIntegrity();<a name="line.666"></a> |
| <span class="sourceLineNo">667</span> curIter++;// limit the number of iterations.<a name="line.667"></a> |
| <span class="sourceLineNo">668</span> } while (fixes > 0 && curIter <= maxIterations);<a name="line.668"></a> |
| <span class="sourceLineNo">669</span><a name="line.669"></a> |
| <span class="sourceLineNo">670</span> // Repairs should be done in the first iteration and verification in the second.<a name="line.670"></a> |
| <span class="sourceLineNo">671</span> // If there are more than 2 passes, something funny has happened.<a name="line.671"></a> |
| <span class="sourceLineNo">672</span> if (curIter > 2) {<a name="line.672"></a> |
| <span class="sourceLineNo">673</span> if (curIter == maxIterations) {<a name="line.673"></a> |
| <span class="sourceLineNo">674</span> LOG.warn("Exiting integrity repairs after max " + curIter + " iterations. "<a name="line.674"></a> |
| <span class="sourceLineNo">675</span> + "Tables integrity may not be fully repaired!");<a name="line.675"></a> |
| <span class="sourceLineNo">676</span> } else {<a name="line.676"></a> |
| <span class="sourceLineNo">677</span> LOG.info("Successfully exiting integrity repairs after " + curIter + " iterations");<a name="line.677"></a> |
| <span class="sourceLineNo">678</span> }<a name="line.678"></a> |
| <span class="sourceLineNo">679</span> }<a name="line.679"></a> |
| <span class="sourceLineNo">680</span> }<a name="line.680"></a> |
| <span class="sourceLineNo">681</span> }<a name="line.681"></a> |
| <span class="sourceLineNo">682</span><a name="line.682"></a> |
| <span class="sourceLineNo">683</span> /**<a name="line.683"></a> |
| <span class="sourceLineNo">684</span> * This repair method requires the cluster to be online since it contacts<a name="line.684"></a> |
| <span class="sourceLineNo">685</span> * region servers and the masters. It makes each region's state in HDFS, in<a name="line.685"></a> |
| <span class="sourceLineNo">686</span> * hbase:meta, and deployments consistent.<a name="line.686"></a> |
| <span class="sourceLineNo">687</span> *<a name="line.687"></a> |
| <span class="sourceLineNo">688</span> * @return If &gt; 0 , number of errors detected, if &lt; 0 there was an unrecoverable<a name="line.688"></a> |
| <span class="sourceLineNo">689</span> * error. If 0, we have a clean hbase.<a name="line.689"></a> |
| <span class="sourceLineNo">690</span> */<a name="line.690"></a> |
| <span class="sourceLineNo">691</span> public int onlineConsistencyRepair() throws IOException, KeeperException,<a name="line.691"></a> |
| <span class="sourceLineNo">692</span> InterruptedException {<a name="line.692"></a> |
| <span class="sourceLineNo">693</span><a name="line.693"></a> |
| <span class="sourceLineNo">694</span> // get regions according to what is online on each RegionServer<a name="line.694"></a> |
| <span class="sourceLineNo">695</span> loadDeployedRegions();<a name="line.695"></a> |
| <span class="sourceLineNo">696</span> // check whether hbase:meta is deployed and online<a name="line.696"></a> |
| <span class="sourceLineNo">697</span> recordMetaRegion();<a name="line.697"></a> |
| <span class="sourceLineNo">698</span> // Check if hbase:meta is found only once and in the right place<a name="line.698"></a> |
| <span class="sourceLineNo">699</span> if (!checkMetaRegion()) {<a name="line.699"></a> |
| <span class="sourceLineNo">700</span> String errorMsg = "hbase:meta table is not consistent. ";<a name="line.700"></a> |
| <span class="sourceLineNo">701</span> if (shouldFixAssignments()) {<a name="line.701"></a> |
| <span class="sourceLineNo">702</span> errorMsg += "HBCK will try fixing it. Rerun once hbase:meta is back to consistent state.";<a name="line.702"></a> |
| <span class="sourceLineNo">703</span> } else {<a name="line.703"></a> |
| <span class="sourceLineNo">704</span> errorMsg += "Run HBCK with proper fix options to fix hbase:meta inconsistency.";<a name="line.704"></a> |
| <span class="sourceLineNo">705</span> }<a name="line.705"></a> |
| <span class="sourceLineNo">706</span> errors.reportError(errorMsg + " Exiting...");<a name="line.706"></a> |
| <span class="sourceLineNo">707</span> return -2;<a name="line.707"></a> |
| <span class="sourceLineNo">708</span> }<a name="line.708"></a> |
| <span class="sourceLineNo">709</span> // Not going with further consistency check for tables when hbase:meta itself is not consistent.<a name="line.709"></a> |
| <span class="sourceLineNo">710</span> LOG.info("Loading regionsinfo from the hbase:meta table");<a name="line.710"></a> |
| <span class="sourceLineNo">711</span> boolean success = loadMetaEntries();<a name="line.711"></a> |
| <span class="sourceLineNo">712</span> if (!success) return -1;<a name="line.712"></a> |
| <span class="sourceLineNo">713</span><a name="line.713"></a> |
| <span class="sourceLineNo">714</span> // Empty cells in hbase:meta?<a name="line.714"></a> |
| <span class="sourceLineNo">715</span> reportEmptyMetaCells();<a name="line.715"></a> |
| <span class="sourceLineNo">716</span><a name="line.716"></a> |
| <span class="sourceLineNo">717</span> // Check if we have to cleanup empty REGIONINFO_QUALIFIER rows from hbase:meta<a name="line.717"></a> |
| <span class="sourceLineNo">718</span> if (shouldFixEmptyMetaCells()) {<a name="line.718"></a> |
| <span class="sourceLineNo">719</span> fixEmptyMetaCells();<a name="line.719"></a> |
| <span class="sourceLineNo">720</span> }<a name="line.720"></a> |
| <span class="sourceLineNo">721</span><a name="line.721"></a> |
| <span class="sourceLineNo">722</span> // get a list of all tables that have not changed recently.<a name="line.722"></a> |
| <span class="sourceLineNo">723</span> if (!checkMetaOnly) {<a name="line.723"></a> |
| <span class="sourceLineNo">724</span> reportTablesInFlux();<a name="line.724"></a> |
| <span class="sourceLineNo">725</span> }<a name="line.725"></a> |
| <span class="sourceLineNo">726</span><a name="line.726"></a> |
| <span class="sourceLineNo">727</span> // Get disabled tables states<a name="line.727"></a> |
| <span class="sourceLineNo">728</span> loadTableStates();<a name="line.728"></a> |
| <span class="sourceLineNo">729</span><a name="line.729"></a> |
| <span class="sourceLineNo">730</span> // load regiondirs and regioninfos from HDFS<a name="line.730"></a> |
| <span class="sourceLineNo">731</span> if (shouldCheckHdfs()) {<a name="line.731"></a> |
| <span class="sourceLineNo">732</span> LOG.info("Loading region directories from HDFS");<a name="line.732"></a> |
| <span class="sourceLineNo">733</span> loadHdfsRegionDirs();<a name="line.733"></a> |
| <span class="sourceLineNo">734</span> LOG.info("Loading region information from HDFS");<a name="line.734"></a> |
| <span class="sourceLineNo">735</span> loadHdfsRegionInfos();<a name="line.735"></a> |
| <span class="sourceLineNo">736</span> }<a name="line.736"></a> |
| <span class="sourceLineNo">737</span><a name="line.737"></a> |
| <span class="sourceLineNo">738</span> // fix the orphan tables<a name="line.738"></a> |
| <span class="sourceLineNo">739</span> fixOrphanTables();<a name="line.739"></a> |
| <span class="sourceLineNo">740</span><a name="line.740"></a> |
| <span class="sourceLineNo">741</span> LOG.info("Checking and fixing region consistency");<a name="line.741"></a> |
| <span class="sourceLineNo">742</span> // Check and fix consistency<a name="line.742"></a> |
| <span class="sourceLineNo">743</span> checkAndFixConsistency();<a name="line.743"></a> |
| <span class="sourceLineNo">744</span><a name="line.744"></a> |
| <span class="sourceLineNo">745</span> // Check integrity (does not fix)<a name="line.745"></a> |
| <span class="sourceLineNo">746</span> checkIntegrity();<a name="line.746"></a> |
| <span class="sourceLineNo">747</span> return errors.getErrorList().size();<a name="line.747"></a> |
| <span class="sourceLineNo">748</span> }<a name="line.748"></a> |
| <span class="sourceLineNo">749</span><a name="line.749"></a> |
| <span class="sourceLineNo">750</span> /**<a name="line.750"></a> |
| <span class="sourceLineNo">751</span> * This method maintains an ephemeral znode. If the creation fails we return false or throw<a name="line.751"></a> |
| <span class="sourceLineNo">752</span> * exception<a name="line.752"></a> |
| <span class="sourceLineNo">753</span> *<a name="line.753"></a> |
| <span class="sourceLineNo">754</span> * @return true if creating znode succeeds; false otherwise<a name="line.754"></a> |
| <span class="sourceLineNo">755</span> * @throws IOException if IO failure occurs<a name="line.755"></a> |
| <span class="sourceLineNo">756</span> */<a name="line.756"></a> |
| <span class="sourceLineNo">757</span> private boolean setMasterInMaintenanceMode() throws IOException {<a name="line.757"></a> |
| <span class="sourceLineNo">758</span> RetryCounter retryCounter = createZNodeRetryCounterFactory.create();<a name="line.758"></a> |
| <span class="sourceLineNo">759</span> hbckEphemeralNodePath = ZNodePaths.joinZNode(<a name="line.759"></a> |
| <span class="sourceLineNo">760</span> zkw.getZNodePaths().masterMaintZNode,<a name="line.760"></a> |
| <span class="sourceLineNo">761</span> "hbck-" + Long.toString(EnvironmentEdgeManager.currentTime()));<a name="line.761"></a> |
| <span class="sourceLineNo">762</span> do {<a name="line.762"></a> |
| <span class="sourceLineNo">763</span> try {<a name="line.763"></a> |
| <span class="sourceLineNo">764</span> hbckZodeCreated = ZKUtil.createEphemeralNodeAndWatch(zkw, hbckEphemeralNodePath, null);<a name="line.764"></a> |
| <span class="sourceLineNo">765</span> if (hbckZodeCreated) {<a name="line.765"></a> |
| <span class="sourceLineNo">766</span> break;<a name="line.766"></a> |
| <span class="sourceLineNo">767</span> }<a name="line.767"></a> |
| <span class="sourceLineNo">768</span> } catch (KeeperException e) {<a name="line.768"></a> |
| <span class="sourceLineNo">769</span> if (retryCounter.getAttemptTimes() >= retryCounter.getMaxAttempts()) {<a name="line.769"></a> |
| <span class="sourceLineNo">770</span> throw new IOException("Can't create znode " + hbckEphemeralNodePath, e);<a name="line.770"></a> |
| <span class="sourceLineNo">771</span> }<a name="line.771"></a> |
| <span class="sourceLineNo">772</span> // fall through and retry<a name="line.772"></a> |
| <span class="sourceLineNo">773</span> }<a name="line.773"></a> |
| <span class="sourceLineNo">774</span><a name="line.774"></a> |
| <span class="sourceLineNo">775</span> LOG.warn("Fail to create znode " + hbckEphemeralNodePath + ", try=" +<a name="line.775"></a> |
| <span class="sourceLineNo">776</span> (retryCounter.getAttemptTimes() + 1) + " of " + retryCounter.getMaxAttempts());<a name="line.776"></a> |
| <span class="sourceLineNo">777</span><a name="line.777"></a> |
| <span class="sourceLineNo">778</span> try {<a name="line.778"></a> |
| <span class="sourceLineNo">779</span> retryCounter.sleepUntilNextRetry();<a name="line.779"></a> |
| <span class="sourceLineNo">780</span> } catch (InterruptedException ie) {<a name="line.780"></a> |
| <span class="sourceLineNo">781</span> throw (InterruptedIOException) new InterruptedIOException(<a name="line.781"></a> |
| <span class="sourceLineNo">782</span> "Can't create znode " + hbckEphemeralNodePath).initCause(ie);<a name="line.782"></a> |
| <span class="sourceLineNo">783</span> }<a name="line.783"></a> |
| <span class="sourceLineNo">784</span> } while (retryCounter.shouldRetry());<a name="line.784"></a> |
| <span class="sourceLineNo">785</span> return hbckZodeCreated;<a name="line.785"></a> |
| <span class="sourceLineNo">786</span> }<a name="line.786"></a> |
| <span class="sourceLineNo">787</span><a name="line.787"></a> |
| <span class="sourceLineNo">788</span> private void cleanupHbckZnode() {<a name="line.788"></a> |
| <span class="sourceLineNo">789</span> try {<a name="line.789"></a> |
| <span class="sourceLineNo">790</span> if (zkw != null && hbckZodeCreated) {<a name="line.790"></a> |
| <span class="sourceLineNo">791</span> ZKUtil.deleteNode(zkw, hbckEphemeralNodePath);<a name="line.791"></a> |
| <span class="sourceLineNo">792</span> hbckZodeCreated = false;<a name="line.792"></a> |
| <span class="sourceLineNo">793</span> }<a name="line.793"></a> |
| <span class="sourceLineNo">794</span> } catch (KeeperException e) {<a name="line.794"></a> |
| <span class="sourceLineNo">795</span> // Ignore<a name="line.795"></a> |
| <span class="sourceLineNo">796</span> if (!e.code().equals(KeeperException.Code.NONODE)) {<a name="line.796"></a> |
| <span class="sourceLineNo">797</span> LOG.warn("Delete HBCK znode " + hbckEphemeralNodePath + " failed ", e);<a name="line.797"></a> |
| <span class="sourceLineNo">798</span> }<a name="line.798"></a> |
| <span class="sourceLineNo">799</span> }<a name="line.799"></a> |
| <span class="sourceLineNo">800</span> }<a name="line.800"></a> |
| <span class="sourceLineNo">801</span><a name="line.801"></a> |
| <span class="sourceLineNo">802</span> /**<a name="line.802"></a> |
| <span class="sourceLineNo">803</span> * Contacts the master and prints out cluster-wide information<a name="line.803"></a> |
| <span class="sourceLineNo">804</span> * @return 0 on success, non-zero on failure<a name="line.804"></a> |
| <span class="sourceLineNo">805</span> */<a name="line.805"></a> |
| <span class="sourceLineNo">806</span> public int onlineHbck()<a name="line.806"></a> |
| <span class="sourceLineNo">807</span> throws IOException, KeeperException, InterruptedException, ReplicationException {<a name="line.807"></a> |
| <span class="sourceLineNo">808</span> // print hbase server version<a name="line.808"></a> |
| <span class="sourceLineNo">809</span> errors.print("Version: " + status.getHBaseVersion());<a name="line.809"></a> |
| <span class="sourceLineNo">810</span><a name="line.810"></a> |
| <span class="sourceLineNo">811</span> // Clean start<a name="line.811"></a> |
| <span class="sourceLineNo">812</span> clearState();<a name="line.812"></a> |
| <span class="sourceLineNo">813</span> // Do offline check and repair first<a name="line.813"></a> |
| <span class="sourceLineNo">814</span> offlineHdfsIntegrityRepair();<a name="line.814"></a> |
| <span class="sourceLineNo">815</span> offlineReferenceFileRepair();<a name="line.815"></a> |
| <span class="sourceLineNo">816</span> offlineHLinkFileRepair();<a name="line.816"></a> |
| <span class="sourceLineNo">817</span> // If Master runs maintenance tasks (such as balancer, catalog janitor, etc) during online<a name="line.817"></a> |
| <span class="sourceLineNo">818</span> // hbck, it is likely that hbck would be misled and report transient errors. Therefore, it<a name="line.818"></a> |
| <span class="sourceLineNo">819</span> // is better to set Master into maintenance mode during online hbck.<a name="line.819"></a> |
| <span class="sourceLineNo">820</span> //<a name="line.820"></a> |
| <span class="sourceLineNo">821</span> if (!setMasterInMaintenanceMode()) {<a name="line.821"></a> |
| <span class="sourceLineNo">822</span> LOG.warn("HBCK is running while master is not in maintenance mode, you might see transient "<a name="line.822"></a> |
| <span class="sourceLineNo">823</span> + "error. Please run HBCK multiple times to reduce the chance of transient error.");<a name="line.823"></a> |
| <span class="sourceLineNo">824</span> }<a name="line.824"></a> |
| <span class="sourceLineNo">825</span><a name="line.825"></a> |
| <span class="sourceLineNo">826</span> onlineConsistencyRepair();<a name="line.826"></a> |
| <span class="sourceLineNo">827</span><a name="line.827"></a> |
| <span class="sourceLineNo">828</span> if (checkRegionBoundaries) {<a name="line.828"></a> |
| <span class="sourceLineNo">829</span> checkRegionBoundaries();<a name="line.829"></a> |
| <span class="sourceLineNo">830</span> }<a name="line.830"></a> |
| <span class="sourceLineNo">831</span><a name="line.831"></a> |
| <span class="sourceLineNo">832</span> checkAndFixReplication();<a name="line.832"></a> |
| <span class="sourceLineNo">833</span><a name="line.833"></a> |
| <span class="sourceLineNo">834</span> cleanReplicationBarrier();<a name="line.834"></a> |
| <span class="sourceLineNo">835</span><a name="line.835"></a> |
| <span class="sourceLineNo">836</span> // Remove the hbck znode<a name="line.836"></a> |
| <span class="sourceLineNo">837</span> cleanupHbckZnode();<a name="line.837"></a> |
| <span class="sourceLineNo">838</span><a name="line.838"></a> |
| <span class="sourceLineNo">839</span> // Remove the hbck lock<a name="line.839"></a> |
| <span class="sourceLineNo">840</span> unlockHbck();<a name="line.840"></a> |
| <span class="sourceLineNo">841</span><a name="line.841"></a> |
| <span class="sourceLineNo">842</span> // Print table summary<a name="line.842"></a> |
| <span class="sourceLineNo">843</span> printTableSummary(tablesInfo);<a name="line.843"></a> |
| <span class="sourceLineNo">844</span> return errors.summarize();<a name="line.844"></a> |
| <span class="sourceLineNo">845</span> }<a name="line.845"></a> |
| <span class="sourceLineNo">846</span><a name="line.846"></a> |
| <span class="sourceLineNo">847</span> public static byte[] keyOnly(byte[] b) {<a name="line.847"></a> |
| <span class="sourceLineNo">848</span> if (b == null)<a name="line.848"></a> |
| <span class="sourceLineNo">849</span> return b;<a name="line.849"></a> |
| <span class="sourceLineNo">850</span> int rowlength = Bytes.toShort(b, 0);<a name="line.850"></a> |
| <span class="sourceLineNo">851</span> byte[] result = new byte[rowlength];<a name="line.851"></a> |
| <span class="sourceLineNo">852</span> System.arraycopy(b, Bytes.SIZEOF_SHORT, result, 0, rowlength);<a name="line.852"></a> |
| <span class="sourceLineNo">853</span> return result;<a name="line.853"></a> |
| <span class="sourceLineNo">854</span> }<a name="line.854"></a> |
| <span class="sourceLineNo">855</span><a name="line.855"></a> |
| <span class="sourceLineNo">856</span> @Override<a name="line.856"></a> |
| <span class="sourceLineNo">857</span> public void close() throws IOException {<a name="line.857"></a> |
| <span class="sourceLineNo">858</span> try {<a name="line.858"></a> |
| <span class="sourceLineNo">859</span> cleanupHbckZnode();<a name="line.859"></a> |
| <span class="sourceLineNo">860</span> unlockHbck();<a name="line.860"></a> |
| <span class="sourceLineNo">861</span> } catch (Exception io) {<a name="line.861"></a> |
| <span class="sourceLineNo">862</span> LOG.warn(io.toString(), io);<a name="line.862"></a> |
| <span class="sourceLineNo">863</span> } finally {<a name="line.863"></a> |
| <span class="sourceLineNo">864</span> if (zkw != null) {<a name="line.864"></a> |
| <span class="sourceLineNo">865</span> zkw.close();<a name="line.865"></a> |
| <span class="sourceLineNo">866</span> zkw = null;<a name="line.866"></a> |
| <span class="sourceLineNo">867</span> }<a name="line.867"></a> |
| <span class="sourceLineNo">868</span> IOUtils.closeQuietly(admin);<a name="line.868"></a> |
| <span class="sourceLineNo">869</span> IOUtils.closeQuietly(meta);<a name="line.869"></a> |
| <span class="sourceLineNo">870</span> IOUtils.closeQuietly(connection);<a name="line.870"></a> |
| <span class="sourceLineNo">871</span> }<a name="line.871"></a> |
| <span class="sourceLineNo">872</span> }<a name="line.872"></a> |
| <span class="sourceLineNo">873</span><a name="line.873"></a> |
| <span class="sourceLineNo">874</span> private static class RegionBoundariesInformation {<a name="line.874"></a> |
| <span class="sourceLineNo">875</span> public byte [] regionName;<a name="line.875"></a> |
| <span class="sourceLineNo">876</span> public byte [] metaFirstKey;<a name="line.876"></a> |
| <span class="sourceLineNo">877</span> public byte [] metaLastKey;<a name="line.877"></a> |
| <span class="sourceLineNo">878</span> public byte [] storesFirstKey;<a name="line.878"></a> |
| <span class="sourceLineNo">879</span> public byte [] storesLastKey;<a name="line.879"></a> |
| <span class="sourceLineNo">880</span> @Override<a name="line.880"></a> |
| <span class="sourceLineNo">881</span> public String toString () {<a name="line.881"></a> |
| <span class="sourceLineNo">882</span> return "regionName=" + Bytes.toStringBinary(regionName) +<a name="line.882"></a> |
| <span class="sourceLineNo">883</span> "\nmetaFirstKey=" + Bytes.toStringBinary(metaFirstKey) +<a name="line.883"></a> |
| <span class="sourceLineNo">884</span> "\nmetaLastKey=" + Bytes.toStringBinary(metaLastKey) +<a name="line.884"></a> |
| <span class="sourceLineNo">885</span> "\nstoresFirstKey=" + Bytes.toStringBinary(storesFirstKey) +<a name="line.885"></a> |
| <span class="sourceLineNo">886</span> "\nstoresLastKey=" + Bytes.toStringBinary(storesLastKey);<a name="line.886"></a> |
| <span class="sourceLineNo">887</span> }<a name="line.887"></a> |
| <span class="sourceLineNo">888</span> }<a name="line.888"></a> |
| <span class="sourceLineNo">889</span><a name="line.889"></a> |
| <span class="sourceLineNo">890</span> public void checkRegionBoundaries() {<a name="line.890"></a> |
| <span class="sourceLineNo">891</span> try {<a name="line.891"></a> |
| <span class="sourceLineNo">892</span> ByteArrayComparator comparator = new ByteArrayComparator();<a name="line.892"></a> |
| <span class="sourceLineNo">893</span> List<RegionInfo> regions = MetaTableAccessor.getAllRegions(connection, true);<a name="line.893"></a> |
| <span class="sourceLineNo">894</span> final RegionBoundariesInformation currentRegionBoundariesInformation =<a name="line.894"></a> |
| <span class="sourceLineNo">895</span> new RegionBoundariesInformation();<a name="line.895"></a> |
| <span class="sourceLineNo">896</span> Path hbaseRoot = CommonFSUtils.getRootDir(getConf());<a name="line.896"></a> |
| <span class="sourceLineNo">897</span> for (RegionInfo regionInfo : regions) {<a name="line.897"></a> |
| <span class="sourceLineNo">898</span> Path tableDir = CommonFSUtils.getTableDir(hbaseRoot, regionInfo.getTable());<a name="line.898"></a> |
| <span class="sourceLineNo">899</span> currentRegionBoundariesInformation.regionName = regionInfo.getRegionName();<a name="line.899"></a> |
| <span class="sourceLineNo">900</span> // For each region, get the start and stop key from the META and compare them to the<a name="line.900"></a> |
| <span class="sourceLineNo">901</span> // same information from the Stores.<a name="line.901"></a> |
| <span class="sourceLineNo">902</span> Path path = new Path(tableDir, regionInfo.getEncodedName());<a name="line.902"></a> |
| <span class="sourceLineNo">903</span> FileSystem fs = path.getFileSystem(getConf());<a name="line.903"></a> |
| <span class="sourceLineNo">904</span> FileStatus[] files = fs.listStatus(path);<a name="line.904"></a> |
| <span class="sourceLineNo">905</span> // For all the column families in this region...<a name="line.905"></a> |
| <span class="sourceLineNo">906</span> byte[] storeFirstKey = null;<a name="line.906"></a> |
| <span class="sourceLineNo">907</span> byte[] storeLastKey = null;<a name="line.907"></a> |
| <span class="sourceLineNo">908</span> for (FileStatus file : files) {<a name="line.908"></a> |
| <span class="sourceLineNo">909</span> String fileName = file.getPath().toString();<a name="line.909"></a> |
| <span class="sourceLineNo">910</span> fileName = fileName.substring(fileName.lastIndexOf("/") + 1);<a name="line.910"></a> |
| <span class="sourceLineNo">911</span> if (!fileName.startsWith(".") && !fileName.endsWith("recovered.edits")) {<a name="line.911"></a> |
| <span class="sourceLineNo">912</span> FileStatus[] storeFiles = fs.listStatus(file.getPath());<a name="line.912"></a> |
| <span class="sourceLineNo">913</span> // For all the stores in this column family.<a name="line.913"></a> |
| <span class="sourceLineNo">914</span> for (FileStatus storeFile : storeFiles) {<a name="line.914"></a> |
| <span class="sourceLineNo">915</span> HFile.Reader reader = HFile.createReader(fs, storeFile.getPath(),<a name="line.915"></a> |
| <span class="sourceLineNo">916</span> CacheConfig.DISABLED, true, getConf());<a name="line.916"></a> |
| <span class="sourceLineNo">917</span> if ((reader.getFirstKey() != null)<a name="line.917"></a> |
| <span class="sourceLineNo">918</span> && ((storeFirstKey == null) || (comparator.compare(storeFirstKey,<a name="line.918"></a> |
| <span class="sourceLineNo">919</span> ((KeyValue.KeyOnlyKeyValue) reader.getFirstKey().get()).getKey()) > 0))) {<a name="line.919"></a> |
| <span class="sourceLineNo">920</span> storeFirstKey = ((KeyValue.KeyOnlyKeyValue)reader.getFirstKey().get()).getKey();<a name="line.920"></a> |
| <span class="sourceLineNo">921</span> }<a name="line.921"></a> |
| <span class="sourceLineNo">922</span> if ((reader.getLastKey() != null)<a name="line.922"></a> |
| <span class="sourceLineNo">923</span> && ((storeLastKey == null) || (comparator.compare(storeLastKey,<a name="line.923"></a> |
| <span class="sourceLineNo">924</span> ((KeyValue.KeyOnlyKeyValue)reader.getLastKey().get()).getKey())) < 0)) {<a name="line.924"></a> |
| <span class="sourceLineNo">925</span> storeLastKey = ((KeyValue.KeyOnlyKeyValue)reader.getLastKey().get()).getKey();<a name="line.925"></a> |
| <span class="sourceLineNo">926</span> }<a name="line.926"></a> |
| <span class="sourceLineNo">927</span> reader.close();<a name="line.927"></a> |
| <span class="sourceLineNo">928</span> }<a name="line.928"></a> |
| <span class="sourceLineNo">929</span> }<a name="line.929"></a> |
| <span class="sourceLineNo">930</span> }<a name="line.930"></a> |
| <span class="sourceLineNo">931</span> currentRegionBoundariesInformation.metaFirstKey = regionInfo.getStartKey();<a name="line.931"></a> |
| <span class="sourceLineNo">932</span> currentRegionBoundariesInformation.metaLastKey = regionInfo.getEndKey();<a name="line.932"></a> |
| <span class="sourceLineNo">933</span> currentRegionBoundariesInformation.storesFirstKey = keyOnly(storeFirstKey);<a name="line.933"></a> |
| <span class="sourceLineNo">934</span> currentRegionBoundariesInformation.storesLastKey = keyOnly(storeLastKey);<a name="line.934"></a> |
| <span class="sourceLineNo">935</span> if (currentRegionBoundariesInformation.metaFirstKey.length == 0)<a name="line.935"></a> |
| <span class="sourceLineNo">936</span> currentRegionBoundariesInformation.metaFirstKey = null;<a name="line.936"></a> |
| <span class="sourceLineNo">937</span> if (currentRegionBoundariesInformation.metaLastKey.length == 0)<a name="line.937"></a> |
| <span class="sourceLineNo">938</span> currentRegionBoundariesInformation.metaLastKey = null;<a name="line.938"></a> |
| <span class="sourceLineNo">939</span><a name="line.939"></a> |
| <span class="sourceLineNo">940</span> // For a region to be correct, we need the META start key to be smaller or equal to the<a name="line.940"></a> |
| <span class="sourceLineNo">941</span> // smallest start key from all the stores, and the start key from the next META entry to<a name="line.941"></a> |
| <span class="sourceLineNo">942</span> // be bigger than the last key from all the current stores. First region start key is null;<a name="line.942"></a> |
| <span class="sourceLineNo">943</span> // Last region end key is null; some regions can be empty and not have any store.<a name="line.943"></a> |
| <span class="sourceLineNo">944</span><a name="line.944"></a> |
| <span class="sourceLineNo">945</span> boolean valid = true;<a name="line.945"></a> |
| <span class="sourceLineNo">946</span> // Checking start key.<a name="line.946"></a> |
| <span class="sourceLineNo">947</span> if ((currentRegionBoundariesInformation.storesFirstKey != null)<a name="line.947"></a> |
| <span class="sourceLineNo">948</span> && (currentRegionBoundariesInformation.metaFirstKey != null)) {<a name="line.948"></a> |
| <span class="sourceLineNo">949</span> valid = valid<a name="line.949"></a> |
| <span class="sourceLineNo">950</span> && comparator.compare(currentRegionBoundariesInformation.storesFirstKey,<a name="line.950"></a> |
| <span class="sourceLineNo">951</span> currentRegionBoundariesInformation.metaFirstKey) >= 0;<a name="line.951"></a> |
| <span class="sourceLineNo">952</span> }<a name="line.952"></a> |
| <span class="sourceLineNo">953</span> // Checking stop key.<a name="line.953"></a> |
| <span class="sourceLineNo">954</span> if ((currentRegionBoundariesInformation.storesLastKey != null)<a name="line.954"></a> |
| <span class="sourceLineNo">955</span> && (currentRegionBoundariesInformation.metaLastKey != null)) {<a name="line.955"></a> |
| <span class="sourceLineNo">956</span> valid = valid<a name="line.956"></a> |
| <span class="sourceLineNo">957</span> && comparator.compare(currentRegionBoundariesInformation.storesLastKey,<a name="line.957"></a> |
| <span class="sourceLineNo">958</span> currentRegionBoundariesInformation.metaLastKey) < 0;<a name="line.958"></a> |
| <span class="sourceLineNo">959</span> }<a name="line.959"></a> |
| <span class="sourceLineNo">960</span> if (!valid) {<a name="line.960"></a> |
| <span class="sourceLineNo">961</span> errors.reportError(ERROR_CODE.BOUNDARIES_ERROR, "Found issues with regions boundaries",<a name="line.961"></a> |
| <span class="sourceLineNo">962</span> tablesInfo.get(regionInfo.getTable()));<a name="line.962"></a> |
| <span class="sourceLineNo">963</span> LOG.warn("Region's boundaries not aligned between stores and META for:");<a name="line.963"></a> |
| <span class="sourceLineNo">964</span> LOG.warn(Objects.toString(currentRegionBoundariesInformation));<a name="line.964"></a> |
| <span class="sourceLineNo">965</span> }<a name="line.965"></a> |
| <span class="sourceLineNo">966</span> }<a name="line.966"></a> |
| <span class="sourceLineNo">967</span> } catch (IOException e) {<a name="line.967"></a> |
| <span class="sourceLineNo">968</span> LOG.error(e.toString(), e);<a name="line.968"></a> |
| <span class="sourceLineNo">969</span> }<a name="line.969"></a> |
| <span class="sourceLineNo">970</span> }<a name="line.970"></a> |
| <span class="sourceLineNo">971</span><a name="line.971"></a> |
| <span class="sourceLineNo">972</span> /**<a name="line.972"></a> |
| <span class="sourceLineNo">973</span> * Iterates through the list of all orphan/invalid regiondirs.<a name="line.973"></a> |
| <span class="sourceLineNo">974</span> */<a name="line.974"></a> |
| <span class="sourceLineNo">975</span> private void adoptHdfsOrphans(Collection<HbckRegionInfo> orphanHdfsDirs) throws IOException {<a name="line.975"></a> |
| <span class="sourceLineNo">976</span> for (HbckRegionInfo hi : orphanHdfsDirs) {<a name="line.976"></a> |
| <span class="sourceLineNo">977</span> LOG.info("Attempting to handle orphan hdfs dir: " + hi.getHdfsRegionDir());<a name="line.977"></a> |
| <span class="sourceLineNo">978</span> adoptHdfsOrphan(hi);<a name="line.978"></a> |
| <span class="sourceLineNo">979</span> }<a name="line.979"></a> |
| <span class="sourceLineNo">980</span> }<a name="line.980"></a> |
| <span class="sourceLineNo">981</span><a name="line.981"></a> |
| <span class="sourceLineNo">982</span> /**<a name="line.982"></a> |
| <span class="sourceLineNo">983</span> * Orphaned regions are regions without a .regioninfo file in them. We "adopt"<a name="line.983"></a> |
| <span class="sourceLineNo">984</span> * these orphans by creating a new region, and moving the column families,<a name="line.984"></a> |
| <span class="sourceLineNo">985</span> * recovered edits, WALs, into the new region dir. We determine the region<a name="line.985"></a> |
| <span class="sourceLineNo">986</span> * startkey and endkeys by looking at all of the hfiles inside the column<a name="line.986"></a> |
| <span class="sourceLineNo">987</span> * families to identify the min and max keys. The resulting region will<a name="line.987"></a> |
| <span class="sourceLineNo">988</span> * likely violate table integrity but will be dealt with by merging<a name="line.988"></a> |
| <span class="sourceLineNo">989</span> * overlapping regions.<a name="line.989"></a> |
| <span class="sourceLineNo">990</span> */<a name="line.990"></a> |
| <span class="sourceLineNo">991</span> @SuppressWarnings("deprecation")<a name="line.991"></a> |
| <span class="sourceLineNo">992</span> private void adoptHdfsOrphan(HbckRegionInfo hi) throws IOException {<a name="line.992"></a> |
| <span class="sourceLineNo">993</span> Path p = hi.getHdfsRegionDir();<a name="line.993"></a> |
| <span class="sourceLineNo">994</span> FileSystem fs = p.getFileSystem(getConf());<a name="line.994"></a> |
| <span class="sourceLineNo">995</span> FileStatus[] dirs = fs.listStatus(p);<a name="line.995"></a> |
| <span class="sourceLineNo">996</span> if (dirs == null) {<a name="line.996"></a> |
| <span class="sourceLineNo">997</span> LOG.warn("Attempt to adopt orphan hdfs region skipped because no files present in " +<a name="line.997"></a> |
| <span class="sourceLineNo">998</span> p + ". This dir could probably be deleted.");<a name="line.998"></a> |
| <span class="sourceLineNo">999</span> return ;<a name="line.999"></a> |
| <span class="sourceLineNo">1000</span> }<a name="line.1000"></a> |
| <span class="sourceLineNo">1001</span><a name="line.1001"></a> |
| <span class="sourceLineNo">1002</span> TableName tableName = hi.getTableName();<a name="line.1002"></a> |
| <span class="sourceLineNo">1003</span> HbckTableInfo tableInfo = tablesInfo.get(tableName);<a name="line.1003"></a> |
| <span class="sourceLineNo">1004</span> Preconditions.checkNotNull(tableInfo, "Table '" + tableName + "' not present!");<a name="line.1004"></a> |
| <span class="sourceLineNo">1005</span> TableDescriptor template = tableInfo.getTableDescriptor();<a name="line.1005"></a> |
| <span class="sourceLineNo">1006</span><a name="line.1006"></a> |
| <span class="sourceLineNo">1007</span> // find min and max key values<a name="line.1007"></a> |
| <span class="sourceLineNo">1008</span> Pair<byte[],byte[]> orphanRegionRange = null;<a name="line.1008"></a> |
| <span class="sourceLineNo">1009</span> for (FileStatus cf : dirs) {<a name="line.1009"></a> |
| <span class="sourceLineNo">1010</span> String cfName= cf.getPath().getName();<a name="line.1010"></a> |
| <span class="sourceLineNo">1011</span> // TODO Figure out what the special dirs are<a name="line.1011"></a> |
| <span class="sourceLineNo">1012</span> if (cfName.startsWith(".") || cfName.equals(HConstants.SPLIT_LOGDIR_NAME)) continue;<a name="line.1012"></a> |
| <span class="sourceLineNo">1013</span><a name="line.1013"></a> |
| <span class="sourceLineNo">1014</span> FileStatus[] hfiles = fs.listStatus(cf.getPath());<a name="line.1014"></a> |
| <span class="sourceLineNo">1015</span> for (FileStatus hfile : hfiles) {<a name="line.1015"></a> |
| <span class="sourceLineNo">1016</span> byte[] start, end;<a name="line.1016"></a> |
| <span class="sourceLineNo">1017</span> HFile.Reader hf = null;<a name="line.1017"></a> |
| <span class="sourceLineNo">1018</span> try {<a name="line.1018"></a> |
| <span class="sourceLineNo">1019</span> hf = HFile.createReader(fs, hfile.getPath(), CacheConfig.DISABLED, true, getConf());<a name="line.1019"></a> |
| <span class="sourceLineNo">1020</span> Optional<Cell> startKv = hf.getFirstKey();<a name="line.1020"></a> |
| <span class="sourceLineNo">1021</span> start = CellUtil.cloneRow(startKv.get());<a name="line.1021"></a> |
| <span class="sourceLineNo">1022</span> Optional<Cell> endKv = hf.getLastKey();<a name="line.1022"></a> |
| <span class="sourceLineNo">1023</span> end = CellUtil.cloneRow(endKv.get());<a name="line.1023"></a> |
| <span class="sourceLineNo">1024</span> } catch (IOException ioe) {<a name="line.1024"></a> |
| <span class="sourceLineNo">1025</span> LOG.warn("Problem reading orphan file " + hfile + ", skipping");<a name="line.1025"></a> |
| <span class="sourceLineNo">1026</span> continue;<a name="line.1026"></a> |
| <span class="sourceLineNo">1027</span> } catch (NullPointerException ioe) {<a name="line.1027"></a> |
| <span class="sourceLineNo">1028</span> LOG.warn("Orphan file " + hfile + " is possibly corrupted HFile, skipping");<a name="line.1028"></a> |
| <span class="sourceLineNo">1029</span> continue;<a name="line.1029"></a> |
| <span class="sourceLineNo">1030</span> } finally {<a name="line.1030"></a> |
| <span class="sourceLineNo">1031</span> if (hf != null) {<a name="line.1031"></a> |
| <span class="sourceLineNo">1032</span> hf.close();<a name="line.1032"></a> |
| <span class="sourceLineNo">1033</span> }<a name="line.1033"></a> |
| <span class="sourceLineNo">1034</span> }<a name="line.1034"></a> |
| <span class="sourceLineNo">1035</span><a name="line.1035"></a> |
| <span class="sourceLineNo">1036</span> // expand the range to include the range of all hfiles<a name="line.1036"></a> |
| <span class="sourceLineNo">1037</span> if (orphanRegionRange == null) {<a name="line.1037"></a> |
| <span class="sourceLineNo">1038</span> // first range<a name="line.1038"></a> |
| <span class="sourceLineNo">1039</span> orphanRegionRange = new Pair<>(start, end);<a name="line.1039"></a> |
| <span class="sourceLineNo">1040</span> } else {<a name="line.1040"></a> |
| <span class="sourceLineNo">1041</span> // TODO add test<a name="line.1041"></a> |
| <span class="sourceLineNo">1042</span><a name="line.1042"></a> |
| <span class="sourceLineNo">1043</span> // expand range only if the hfile is wider.<a name="line.1043"></a> |
| <span class="sourceLineNo">1044</span> if (Bytes.compareTo(orphanRegionRange.getFirst(), start) > 0) {<a name="line.1044"></a> |
| <span class="sourceLineNo">1045</span> orphanRegionRange.setFirst(start);<a name="line.1045"></a> |
| <span class="sourceLineNo">1046</span> }<a name="line.1046"></a> |
| <span class="sourceLineNo">1047</span> if (Bytes.compareTo(orphanRegionRange.getSecond(), end) < 0 ) {<a name="line.1047"></a> |
| <span class="sourceLineNo">1048</span> orphanRegionRange.setSecond(end);<a name="line.1048"></a> |
| <span class="sourceLineNo">1049</span> }<a name="line.1049"></a> |
| <span class="sourceLineNo">1050</span> }<a name="line.1050"></a> |
| <span class="sourceLineNo">1051</span> }<a name="line.1051"></a> |
| <span class="sourceLineNo">1052</span> }<a name="line.1052"></a> |
| <span class="sourceLineNo">1053</span> if (orphanRegionRange == null) {<a name="line.1053"></a> |
| <span class="sourceLineNo">1054</span> LOG.warn("No data in dir " + p + ", sidelining data");<a name="line.1054"></a> |
| <span class="sourceLineNo">1055</span> fixes++;<a name="line.1055"></a> |
| <span class="sourceLineNo">1056</span> sidelineRegionDir(fs, hi);<a name="line.1056"></a> |
| <span class="sourceLineNo">1057</span> return;<a name="line.1057"></a> |
| <span class="sourceLineNo">1058</span> }<a name="line.1058"></a> |
| <span class="sourceLineNo">1059</span> LOG.info("Min max keys are : [" + Bytes.toString(orphanRegionRange.getFirst()) + ", " +<a name="line.1059"></a> |
| <span class="sourceLineNo">1060</span> Bytes.toString(orphanRegionRange.getSecond()) + ")");<a name="line.1060"></a> |
| <span class="sourceLineNo">1061</span><a name="line.1061"></a> |
| <span class="sourceLineNo">1062</span> // create new region on hdfs. move data into place.<a name="line.1062"></a> |
| <span class="sourceLineNo">1063</span> RegionInfo regionInfo = RegionInfoBuilder.newBuilder(template.getTableName())<a name="line.1063"></a> |
| <span class="sourceLineNo">1064</span> .setStartKey(orphanRegionRange.getFirst())<a name="line.1064"></a> |
| <span class="sourceLineNo">1065</span> .setEndKey(Bytes.add(orphanRegionRange.getSecond(), new byte[1]))<a name="line.1065"></a> |
| <span class="sourceLineNo">1066</span> .build();<a name="line.1066"></a> |
| <span class="sourceLineNo">1067</span> LOG.info("Creating new region : " + regionInfo);<a name="line.1067"></a> |
| <span class="sourceLineNo">1068</span> HRegion region = HBaseFsckRepair.createHDFSRegionDir(getConf(), regionInfo, template);<a name="line.1068"></a> |
| <span class="sourceLineNo">1069</span> Path target = region.getRegionFileSystem().getRegionDir();<a name="line.1069"></a> |
| <span class="sourceLineNo">1070</span><a name="line.1070"></a> |
| <span class="sourceLineNo">1071</span> // rename all the data to new region<a name="line.1071"></a> |
| <span class="sourceLineNo">1072</span> mergeRegionDirs(target, hi);<a name="line.1072"></a> |
| <span class="sourceLineNo">1073</span> fixes++;<a name="line.1073"></a> |
| <span class="sourceLineNo">1074</span> }<a name="line.1074"></a> |
| <span class="sourceLineNo">1075</span><a name="line.1075"></a> |
| <span class="sourceLineNo">1076</span> /**<a name="line.1076"></a> |
| <span class="sourceLineNo">1077</span> * This method determines if there are table integrity errors in HDFS. If<a name="line.1077"></a> |
| <span class="sourceLineNo">1078</span> * there are errors and the appropriate "fix" options are enabled, the method<a name="line.1078"></a> |
| <span class="sourceLineNo">1079</span> * will first correct orphan regions making them into legit regiondirs, and<a name="line.1079"></a> |
| <span class="sourceLineNo">1080</span> * then reload to merge potentially overlapping regions.<a name="line.1080"></a> |
| <span class="sourceLineNo">1081</span> *<a name="line.1081"></a> |
| <span class="sourceLineNo">1082</span> * @return number of table integrity errors found<a name="line.1082"></a> |
| <span class="sourceLineNo">1083</span> */<a name="line.1083"></a> |
| <span class="sourceLineNo">1084</span> private int restoreHdfsIntegrity() throws IOException, InterruptedException {<a name="line.1084"></a> |
| <span class="sourceLineNo">1085</span> // Determine what's on HDFS<a name="line.1085"></a> |
| <span class="sourceLineNo">1086</span> LOG.info("Loading HBase regioninfo from HDFS...");<a name="line.1086"></a> |
| <span class="sourceLineNo">1087</span> loadHdfsRegionDirs(); // populating regioninfo table.<a name="line.1087"></a> |
| <span class="sourceLineNo">1088</span><a name="line.1088"></a> |
| <span class="sourceLineNo">1089</span> int errs = errors.getErrorList().size();<a name="line.1089"></a> |
| <span class="sourceLineNo">1090</span> // First time just get suggestions.<a name="line.1090"></a> |
| <span class="sourceLineNo">1091</span> tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.<a name="line.1091"></a> |
| <span class="sourceLineNo">1092</span> checkHdfsIntegrity(false, false);<a name="line.1092"></a> |
| <span class="sourceLineNo">1093</span><a name="line.1093"></a> |
| <span class="sourceLineNo">1094</span> if (errors.getErrorList().size() == errs) {<a name="line.1094"></a> |
| <span class="sourceLineNo">1095</span> LOG.info("No integrity errors. We are done with this phase. Glorious.");<a name="line.1095"></a> |
| <span class="sourceLineNo">1096</span> return 0;<a name="line.1096"></a> |
| <span class="sourceLineNo">1097</span> }<a name="line.1097"></a> |
| <span class="sourceLineNo">1098</span><a name="line.1098"></a> |
| <span class="sourceLineNo">1099</span> if (shouldFixHdfsOrphans() && orphanHdfsDirs.size() > 0) {<a name="line.1099"></a> |
| <span class="sourceLineNo">1100</span> adoptHdfsOrphans(orphanHdfsDirs);<a name="line.1100"></a> |
| <span class="sourceLineNo">1101</span> // TODO optimize by incrementally adding instead of reloading.<a name="line.1101"></a> |
| <span class="sourceLineNo">1102</span> }<a name="line.1102"></a> |
| <span class="sourceLineNo">1103</span><a name="line.1103"></a> |
| <span class="sourceLineNo">1104</span> // Make sure there are no holes now.<a name="line.1104"></a> |
| <span class="sourceLineNo">1105</span> if (shouldFixHdfsHoles()) {<a name="line.1105"></a> |
| <span class="sourceLineNo">1106</span> clearState(); // this also resets # fixes.<a name="line.1106"></a> |
| <span class="sourceLineNo">1107</span> loadHdfsRegionDirs();<a name="line.1107"></a> |
| <span class="sourceLineNo">1108</span> tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.<a name="line.1108"></a> |
| <span class="sourceLineNo">1109</span> tablesInfo = checkHdfsIntegrity(shouldFixHdfsHoles(), false);<a name="line.1109"></a> |
| <span class="sourceLineNo">1110</span> }<a name="line.1110"></a> |
| <span class="sourceLineNo">1111</span><a name="line.1111"></a> |
| <span class="sourceLineNo">1112</span> // Now we fix overlaps<a name="line.1112"></a> |
| <span class="sourceLineNo">1113</span> if (shouldFixHdfsOverlaps()) {<a name="line.1113"></a> |
| <span class="sourceLineNo">1114</span> // second pass we fix overlaps.<a name="line.1114"></a> |
| <span class="sourceLineNo">1115</span> clearState(); // this also resets # fixes.<a name="line.1115"></a> |
| <span class="sourceLineNo">1116</span> loadHdfsRegionDirs();<a name="line.1116"></a> |
| <span class="sourceLineNo">1117</span> tablesInfo = loadHdfsRegionInfos(); // update tableInfos based on region info in fs.<a name="line.1117"></a> |
| <span class="sourceLineNo">1118</span> tablesInfo = checkHdfsIntegrity(false, shouldFixHdfsOverlaps());<a name="line.1118"></a> |
| <span class="sourceLineNo">1119</span> }<a name="line.1119"></a> |
| <span class="sourceLineNo">1120</span><a name="line.1120"></a> |
| <span class="sourceLineNo">1121</span> return errors.getErrorList().size();<a name="line.1121"></a> |
| <span class="sourceLineNo">1122</span> }<a name="line.1122"></a> |
| <span class="sourceLineNo">1123</span><a name="line.1123"></a> |
| <span class="sourceLineNo">1124</span> /**<a name="line.1124"></a> |
| <span class="sourceLineNo">1125</span> * Scan all the store file names to find any lingering reference files,<a name="line.1125"></a> |
| <span class="sourceLineNo">1126</span> * which refer to some none-exiting files. If "fix" option is enabled,<a name="line.1126"></a> |
| <span class="sourceLineNo">1127</span> * any lingering reference file will be sidelined if found.<a name="line.1127"></a> |
| <span class="sourceLineNo">1128</span> * <p><a name="line.1128"></a> |
| <span class="sourceLineNo">1129</span> * Lingering reference file prevents a region from opening. It has to<a name="line.1129"></a> |
| <span class="sourceLineNo">1130</span> * be fixed before a cluster can start properly.<a name="line.1130"></a> |
| <span class="sourceLineNo">1131</span> */<a name="line.1131"></a> |
| <span class="sourceLineNo">1132</span> private void offlineReferenceFileRepair() throws IOException, InterruptedException {<a name="line.1132"></a> |
| <span class="sourceLineNo">1133</span> clearState();<a name="line.1133"></a> |
| <span class="sourceLineNo">1134</span> Configuration conf = getConf();<a name="line.1134"></a> |
| <span class="sourceLineNo">1135</span> Path hbaseRoot = CommonFSUtils.getRootDir(conf);<a name="line.1135"></a> |
| <span class="sourceLineNo">1136</span> FileSystem fs = hbaseRoot.getFileSystem(conf);<a name="line.1136"></a> |
| <span class="sourceLineNo">1137</span> LOG.info("Computing mapping of all store files");<a name="line.1137"></a> |
| <span class="sourceLineNo">1138</span> Map<String, Path> allFiles = FSUtils.getTableStoreFilePathMap(fs, hbaseRoot,<a name="line.1138"></a> |
| <span class="sourceLineNo">1139</span> new FSUtils.ReferenceFileFilter(fs), executor, errors);<a name="line.1139"></a> |
| <span class="sourceLineNo">1140</span> errors.print("");<a name="line.1140"></a> |
| <span class="sourceLineNo">1141</span> LOG.info("Validating mapping using HDFS state");<a name="line.1141"></a> |
| <span class="sourceLineNo">1142</span> for (Path path: allFiles.values()) {<a name="line.1142"></a> |
| <span class="sourceLineNo">1143</span> Path referredToFile = StoreFileInfo.getReferredToFile(path);<a name="line.1143"></a> |
| <span class="sourceLineNo">1144</span> if (fs.exists(referredToFile)) continue; // good, expected<a name="line.1144"></a> |
| <span class="sourceLineNo">1145</span><a name="line.1145"></a> |
| <span class="sourceLineNo">1146</span> // Found a lingering reference file<a name="line.1146"></a> |
| <span class="sourceLineNo">1147</span> errors.reportError(ERROR_CODE.LINGERING_REFERENCE_HFILE,<a name="line.1147"></a> |
| <span class="sourceLineNo">1148</span> "Found lingering reference file " + path);<a name="line.1148"></a> |
| <span class="sourceLineNo">1149</span> if (!shouldFixReferenceFiles()) continue;<a name="line.1149"></a> |
| <span class="sourceLineNo">1150</span><a name="line.1150"></a> |
| <span class="sourceLineNo">1151</span> // Now, trying to fix it since requested<a name="line.1151"></a> |
| <span class="sourceLineNo">1152</span> boolean success = false;<a name="line.1152"></a> |
| <span class="sourceLineNo">1153</span> String pathStr = path.toString();<a name="line.1153"></a> |
| <span class="sourceLineNo">1154</span><a name="line.1154"></a> |
| <span class="sourceLineNo">1155</span> // A reference file path should be like<a name="line.1155"></a> |
| <span class="sourceLineNo">1156</span> // ${hbase.rootdir}/data/namespace/table_name/region_id/family_name/referred_file.region_name<a name="line.1156"></a> |
| <span class="sourceLineNo">1157</span> // Up 5 directories to get the root folder.<a name="line.1157"></a> |
| <span class="sourceLineNo">1158</span> // So the file will be sidelined to a similar folder structure.<a name="line.1158"></a> |
| <span class="sourceLineNo">1159</span> int index = pathStr.lastIndexOf(Path.SEPARATOR_CHAR);<a name="line.1159"></a> |
| <span class="sourceLineNo">1160</span> for (int i = 0; index > 0 && i < 5; i++) {<a name="line.1160"></a> |
| <span class="sourceLineNo">1161</span> index = pathStr.lastIndexOf(Path.SEPARATOR_CHAR, index - 1);<a name="line.1161"></a> |
| <span class="sourceLineNo">1162</span> }<a name="line.1162"></a> |
| <span class="sourceLineNo">1163</span> if (index > 0) {<a name="line.1163"></a> |
| <span class="sourceLineNo">1164</span> Path rootDir = getSidelineDir();<a name="line.1164"></a> |
| <span class="sourceLineNo">1165</span> Path dst = new Path(rootDir, pathStr.substring(index + 1));<a name="line.1165"></a> |
| <span class="sourceLineNo">1166</span> fs.mkdirs(dst.getParent());<a name="line.1166"></a> |
| <span class="sourceLineNo">1167</span> LOG.info("Trying to sideline reference file "<a name="line.1167"></a> |
| <span class="sourceLineNo">1168</span> + path + " to " + dst);<a name="line.1168"></a> |
| <span class="sourceLineNo">1169</span> setShouldRerun();<a name="line.1169"></a> |
| <span class="sourceLineNo">1170</span><a name="line.1170"></a> |
| <span class="sourceLineNo">1171</span> success = fs.rename(path, dst);<a name="line.1171"></a> |
| <span class="sourceLineNo">1172</span> debugLsr(dst);<a name="line.1172"></a> |
| <span class="sourceLineNo">1173</span><a name="line.1173"></a> |
| <span class="sourceLineNo">1174</span> }<a name="line.1174"></a> |
| <span class="sourceLineNo">1175</span> if (!success) {<a name="line.1175"></a> |
| <span class="sourceLineNo">1176</span> LOG.error("Failed to sideline reference file " + path);<a name="line.1176"></a> |
| <span class="sourceLineNo">1177</span> }<a name="line.1177"></a> |
| <span class="sourceLineNo">1178</span> }<a name="line.1178"></a> |
| <span class="sourceLineNo">1179</span> }<a name="line.1179"></a> |
| <span class="sourceLineNo">1180</span><a name="line.1180"></a> |
| <span class="sourceLineNo">1181</span> /**<a name="line.1181"></a> |
| <span class="sourceLineNo">1182</span> * Scan all the store file names to find any lingering HFileLink files,<a name="line.1182"></a> |
| <span class="sourceLineNo">1183</span> * which refer to some none-exiting files. If "fix" option is enabled,<a name="line.1183"></a> |
| <span class="sourceLineNo">1184</span> * any lingering HFileLink file will be sidelined if found.<a name="line.1184"></a> |
| <span class="sourceLineNo">1185</span> */<a name="line.1185"></a> |
| <span class="sourceLineNo">1186</span> private void offlineHLinkFileRepair() throws IOException, InterruptedException {<a name="line.1186"></a> |
| <span class="sourceLineNo">1187</span> Configuration conf = getConf();<a name="line.1187"></a> |
| <span class="sourceLineNo">1188</span> Path hbaseRoot = CommonFSUtils.getRootDir(conf);<a name="line.1188"></a> |
| <span class="sourceLineNo">1189</span> FileSystem fs = hbaseRoot.getFileSystem(conf);<a name="line.1189"></a> |
| <span class="sourceLineNo">1190</span> LOG.info("Computing mapping of all link files");<a name="line.1190"></a> |
| <span class="sourceLineNo">1191</span> Map<String, Path> allFiles = FSUtils<a name="line.1191"></a> |
| <span class="sourceLineNo">1192</span> .getTableStoreFilePathMap(fs, hbaseRoot, new FSUtils.HFileLinkFilter(), executor, errors);<a name="line.1192"></a> |
| <span class="sourceLineNo">1193</span> errors.print("");<a name="line.1193"></a> |
| <span class="sourceLineNo">1194</span><a name="line.1194"></a> |
| <span class="sourceLineNo">1195</span> LOG.info("Validating mapping using HDFS state");<a name="line.1195"></a> |
| <span class="sourceLineNo">1196</span> for (Path path : allFiles.values()) {<a name="line.1196"></a> |
| <span class="sourceLineNo">1197</span> // building HFileLink object to gather locations<a name="line.1197"></a> |
| <span class="sourceLineNo">1198</span> HFileLink actualLink = HFileLink.buildFromHFileLinkPattern(conf, path);<a name="line.1198"></a> |
| <span class="sourceLineNo">1199</span> if (actualLink.exists(fs)) continue; // good, expected<a name="line.1199"></a> |
| <span class="sourceLineNo">1200</span><a name="line.1200"></a> |
| <span class="sourceLineNo">1201</span> // Found a lingering HFileLink<a name="line.1201"></a> |
| <span class="sourceLineNo">1202</span> errors.reportError(ERROR_CODE.LINGERING_HFILELINK, "Found lingering HFileLink " + path);<a name="line.1202"></a> |
| <span class="sourceLineNo">1203</span> if (!shouldFixHFileLinks()) continue;<a name="line.1203"></a> |
| <span class="sourceLineNo">1204</span><a name="line.1204"></a> |
| <span class="sourceLineNo">1205</span> // Now, trying to fix it since requested<a name="line.1205"></a> |
| <span class="sourceLineNo">1206</span> setShouldRerun();<a name="line.1206"></a> |
| <span class="sourceLineNo">1207</span><a name="line.1207"></a> |
| <span class="sourceLineNo">1208</span> // An HFileLink path should be like<a name="line.1208"></a> |
| <span class="sourceLineNo">1209</span> // ${hbase.rootdir}/data/namespace/table_name/region_id/family_name/linkedtable=linkedregionname-linkedhfilename<a name="line.1209"></a> |
| <span class="sourceLineNo">1210</span> // sidelineing will happen in the ${hbase.rootdir}/${sidelinedir} directory with the same folder structure.<a name="line.1210"></a> |
| <span class="sourceLineNo">1211</span> boolean success = sidelineFile(fs, hbaseRoot, path);<a name="line.1211"></a> |
| <span class="sourceLineNo">1212</span><a name="line.1212"></a> |
| <span class="sourceLineNo">1213</span> if (!success) {<a name="line.1213"></a> |
| <span class="sourceLineNo">1214</span> LOG.error("Failed to sideline HFileLink file " + path);<a name="line.1214"></a> |
| <span class="sourceLineNo">1215</span> }<a name="line.1215"></a> |
| <span class="sourceLineNo">1216</span><a name="line.1216"></a> |
| <span class="sourceLineNo">1217</span> // An HFileLink backreference path should be like<a name="line.1217"></a> |
| <span class="sourceLineNo">1218</span> // ${hbase.rootdir}/archive/data/namespace/table_name/region_id/family_name/.links-linkedhfilename<a name="line.1218"></a> |
| <span class="sourceLineNo">1219</span> // sidelineing will happen in the ${hbase.rootdir}/${sidelinedir} directory with the same folder structure.<a name="line.1219"></a> |
| <span class="sourceLineNo">1220</span> Path backRefPath = FileLink.getBackReferencesDir(HFileArchiveUtil<a name="line.1220"></a> |
| <span class="sourceLineNo">1221</span> .getStoreArchivePath(conf, HFileLink.getReferencedTableName(path.getName().toString()),<a name="line.1221"></a> |
| <span class="sourceLineNo">1222</span> HFileLink.getReferencedRegionName(path.getName().toString()),<a name="line.1222"></a> |
| <span class="sourceLineNo">1223</span> path.getParent().getName()),<a name="line.1223"></a> |
| <span class="sourceLineNo">1224</span> HFileLink.getReferencedHFileName(path.getName().toString()));<a name="line.1224"></a> |
| <span class="sourceLineNo">1225</span> success = sidelineFile(fs, hbaseRoot, backRefPath);<a name="line.1225"></a> |
| <span class="sourceLineNo">1226</span><a name="line.1226"></a> |
| <span class="sourceLineNo">1227</span> if (!success) {<a name="line.1227"></a> |
| <span class="sourceLineNo">1228</span> LOG.error("Failed to sideline HFileLink backreference file " + path);<a name="line.1228"></a> |
| <span class="sourceLineNo">1229</span> }<a name="line.1229"></a> |
| <span class="sourceLineNo">1230</span> }<a name="line.1230"></a> |
| <span class="sourceLineNo">1231</span> }<a name="line.1231"></a> |
| <span class="sourceLineNo">1232</span><a name="line.1232"></a> |
| <span class="sourceLineNo">1233</span> private boolean sidelineFile(FileSystem fs, Path hbaseRoot, Path path) throws IOException {<a name="line.1233"></a> |
| <span class="sourceLineNo">1234</span> URI uri = hbaseRoot.toUri().relativize(path.toUri());<a name="line.1234"></a> |
| <span class="sourceLineNo">1235</span> if (uri.isAbsolute()) return false;<a name="line.1235"></a> |
| <span class="sourceLineNo">1236</span> String relativePath = uri.getPath();<a name="line.1236"></a> |
| <span class="sourceLineNo">1237</span> Path rootDir = getSidelineDir();<a name="line.1237"></a> |
| <span class="sourceLineNo">1238</span> Path dst = new Path(rootDir, relativePath);<a name="line.1238"></a> |
| <span class="sourceLineNo">1239</span> boolean pathCreated = fs.mkdirs(dst.getParent());<a name="line.1239"></a> |
| <span class="sourceLineNo">1240</span> if (!pathCreated) {<a name="line.1240"></a> |
| <span class="sourceLineNo">1241</span> LOG.error("Failed to create path: " + dst.getParent());<a name="line.1241"></a> |
| <span class="sourceLineNo">1242</span> return false;<a name="line.1242"></a> |
| <span class="sourceLineNo">1243</span> }<a name="line.1243"></a> |
| <span class="sourceLineNo">1244</span> LOG.info("Trying to sideline file " + path + " to " + dst);<a name="line.1244"></a> |
| <span class="sourceLineNo">1245</span> return fs.rename(path, dst);<a name="line.1245"></a> |
| <span class="sourceLineNo">1246</span> }<a name="line.1246"></a> |
| <span class="sourceLineNo">1247</span><a name="line.1247"></a> |
| <span class="sourceLineNo">1248</span> /**<a name="line.1248"></a> |
| <span class="sourceLineNo">1249</span> * TODO -- need to add tests for this.<a name="line.1249"></a> |
| <span class="sourceLineNo">1250</span> */<a name="line.1250"></a> |
| <span class="sourceLineNo">1251</span> private void reportEmptyMetaCells() {<a name="line.1251"></a> |
| <span class="sourceLineNo">1252</span> errors.print("Number of empty REGIONINFO_QUALIFIER rows in hbase:meta: " +<a name="line.1252"></a> |
| <span class="sourceLineNo">1253</span> emptyRegionInfoQualifiers.size());<a name="line.1253"></a> |
| <span class="sourceLineNo">1254</span> if (details) {<a name="line.1254"></a> |
| <span class="sourceLineNo">1255</span> for (Result r: emptyRegionInfoQualifiers) {<a name="line.1255"></a> |
| <span class="sourceLineNo">1256</span> errors.print(" " + r);<a name="line.1256"></a> |
| <span class="sourceLineNo">1257</span> }<a name="line.1257"></a> |
| <span class="sourceLineNo">1258</span> }<a name="line.1258"></a> |
| <span class="sourceLineNo">1259</span> }<a name="line.1259"></a> |
| <span class="sourceLineNo">1260</span><a name="line.1260"></a> |
| <span class="sourceLineNo">1261</span> /**<a name="line.1261"></a> |
| <span class="sourceLineNo">1262</span> * TODO -- need to add tests for this.<a name="line.1262"></a> |
| <span class="sourceLineNo">1263</span> */<a name="line.1263"></a> |
| <span class="sourceLineNo">1264</span> private void reportTablesInFlux() {<a name="line.1264"></a> |
| <span class="sourceLineNo">1265</span> AtomicInteger numSkipped = new AtomicInteger(0);<a name="line.1265"></a> |
| <span class="sourceLineNo">1266</span> TableDescriptor[] allTables = getTables(numSkipped);<a name="line.1266"></a> |
| <span class="sourceLineNo">1267</span> errors.print("Number of Tables: " + allTables.length);<a name="line.1267"></a> |
| <span class="sourceLineNo">1268</span> if (details) {<a name="line.1268"></a> |
| <span class="sourceLineNo">1269</span> if (numSkipped.get() > 0) {<a name="line.1269"></a> |
| <span class="sourceLineNo">1270</span> errors.detail("Number of Tables in flux: " + numSkipped.get());<a name="line.1270"></a> |
| <span class="sourceLineNo">1271</span> }<a name="line.1271"></a> |
| <span class="sourceLineNo">1272</span> for (TableDescriptor td : allTables) {<a name="line.1272"></a> |
| <span class="sourceLineNo">1273</span> errors.detail(" Table: " + td.getTableName() + "\t" +<a name="line.1273"></a> |
| <span class="sourceLineNo">1274</span> (td.isReadOnly() ? "ro" : "rw") + "\t" +<a name="line.1274"></a> |
| <span class="sourceLineNo">1275</span> (td.isMetaRegion() ? "META" : " ") + "\t" +<a name="line.1275"></a> |
| <span class="sourceLineNo">1276</span> " families: " + td.getColumnFamilyCount());<a name="line.1276"></a> |
| <span class="sourceLineNo">1277</span> }<a name="line.1277"></a> |
| <span class="sourceLineNo">1278</span> }<a name="line.1278"></a> |
| <span class="sourceLineNo">1279</span> }<a name="line.1279"></a> |
| <span class="sourceLineNo">1280</span><a name="line.1280"></a> |
| <span class="sourceLineNo">1281</span> public HbckErrorReporter getErrors() {<a name="line.1281"></a> |
| <span class="sourceLineNo">1282</span> return errors;<a name="line.1282"></a> |
| <span class="sourceLineNo">1283</span> }<a name="line.1283"></a> |
| <span class="sourceLineNo">1284</span><a name="line.1284"></a> |
| <span class="sourceLineNo">1285</span> /**<a name="line.1285"></a> |
| <span class="sourceLineNo">1286</span> * Populate hbi's from regionInfos loaded from file system.<a name="line.1286"></a> |
| <span class="sourceLineNo">1287</span> */<a name="line.1287"></a> |
| <span class="sourceLineNo">1288</span> private SortedMap<TableName, HbckTableInfo> loadHdfsRegionInfos()<a name="line.1288"></a> |
| <span class="sourceLineNo">1289</span> throws IOException, InterruptedException {<a name="line.1289"></a> |
| <span class="sourceLineNo">1290</span> tablesInfo.clear(); // regenerating the data<a name="line.1290"></a> |
| <span class="sourceLineNo">1291</span> // generate region split structure<a name="line.1291"></a> |
| <span class="sourceLineNo">1292</span> Collection<HbckRegionInfo> hbckRegionInfos = regionInfoMap.values();<a name="line.1292"></a> |
| <span class="sourceLineNo">1293</span><a name="line.1293"></a> |
| <span class="sourceLineNo">1294</span> // Parallelized read of .regioninfo files.<a name="line.1294"></a> |
| <span class="sourceLineNo">1295</span> List<WorkItemHdfsRegionInfo> hbis = new ArrayList<>(hbckRegionInfos.size());<a name="line.1295"></a> |
| <span class="sourceLineNo">1296</span> List<Future<Void>> hbiFutures;<a name="line.1296"></a> |
| <span class="sourceLineNo">1297</span><a name="line.1297"></a> |
| <span class="sourceLineNo">1298</span> for (HbckRegionInfo hbi : hbckRegionInfos) {<a name="line.1298"></a> |
| <span class="sourceLineNo">1299</span> WorkItemHdfsRegionInfo work = new WorkItemHdfsRegionInfo(hbi, this, errors);<a name="line.1299"></a> |
| <span class="sourceLineNo">1300</span> hbis.add(work);<a name="line.1300"></a> |
| <span class="sourceLineNo">1301</span> }<a name="line.1301"></a> |
| <span class="sourceLineNo">1302</span><a name="line.1302"></a> |
| <span class="sourceLineNo">1303</span> // Submit and wait for completion<a name="line.1303"></a> |
| <span class="sourceLineNo">1304</span> hbiFutures = executor.invokeAll(hbis);<a name="line.1304"></a> |
| <span class="sourceLineNo">1305</span><a name="line.1305"></a> |
| <span class="sourceLineNo">1306</span> for(int i=0; i<hbiFutures.size(); i++) {<a name="line.1306"></a> |
| <span class="sourceLineNo">1307</span> WorkItemHdfsRegionInfo work = hbis.get(i);<a name="line.1307"></a> |
| <span class="sourceLineNo">1308</span> Future<Void> f = hbiFutures.get(i);<a name="line.1308"></a> |
| <span class="sourceLineNo">1309</span> try {<a name="line.1309"></a> |
| <span class="sourceLineNo">1310</span> f.get();<a name="line.1310"></a> |
| <span class="sourceLineNo">1311</span> } catch(ExecutionException e) {<a name="line.1311"></a> |
| <span class="sourceLineNo">1312</span> LOG.warn("Failed to read .regioninfo file for region " +<a name="line.1312"></a> |
| <span class="sourceLineNo">1313</span> work.hbi.getRegionNameAsString(), e.getCause());<a name="line.1313"></a> |
| <span class="sourceLineNo">1314</span> }<a name="line.1314"></a> |
| <span class="sourceLineNo">1315</span> }<a name="line.1315"></a> |
| <span class="sourceLineNo">1316</span><a name="line.1316"></a> |
| <span class="sourceLineNo">1317</span> Path hbaseRoot = CommonFSUtils.getRootDir(getConf());<a name="line.1317"></a> |
| <span class="sourceLineNo">1318</span> FileSystem fs = hbaseRoot.getFileSystem(getConf());<a name="line.1318"></a> |
| <span class="sourceLineNo">1319</span> // serialized table info gathering.<a name="line.1319"></a> |
| <span class="sourceLineNo">1320</span> for (HbckRegionInfo hbi: hbckRegionInfos) {<a name="line.1320"></a> |
| <span class="sourceLineNo">1321</span><a name="line.1321"></a> |
| <span class="sourceLineNo">1322</span> if (hbi.getHdfsHRI() == null) {<a name="line.1322"></a> |
| <span class="sourceLineNo">1323</span> // was an orphan<a name="line.1323"></a> |
| <span class="sourceLineNo">1324</span> continue;<a name="line.1324"></a> |
| <span class="sourceLineNo">1325</span> }<a name="line.1325"></a> |
| <span class="sourceLineNo">1326</span><a name="line.1326"></a> |
| <span class="sourceLineNo">1327</span><a name="line.1327"></a> |
| <span class="sourceLineNo">1328</span> // get table name from hdfs, populate various HBaseFsck tables.<a name="line.1328"></a> |
| <span class="sourceLineNo">1329</span> TableName tableName = hbi.getTableName();<a name="line.1329"></a> |
| <span class="sourceLineNo">1330</span> if (tableName == null) {<a name="line.1330"></a> |
| <span class="sourceLineNo">1331</span> // There was an entry in hbase:meta not in the HDFS?<a name="line.1331"></a> |
| <span class="sourceLineNo">1332</span> LOG.warn("tableName was null for: " + hbi);<a name="line.1332"></a> |
| <span class="sourceLineNo">1333</span> continue;<a name="line.1333"></a> |
| <span class="sourceLineNo">1334</span> }<a name="line.1334"></a> |
| <span class="sourceLineNo">1335</span><a name="line.1335"></a> |
| <span class="sourceLineNo">1336</span> HbckTableInfo modTInfo = tablesInfo.get(tableName);<a name="line.1336"></a> |
| <span class="sourceLineNo">1337</span> if (modTInfo == null) {<a name="line.1337"></a> |
| <span class="sourceLineNo">1338</span> // only executed once per table.<a name="line.1338"></a> |
| <span class="sourceLineNo">1339</span> modTInfo = new HbckTableInfo(tableName, this);<a name="line.1339"></a> |
| <span class="sourceLineNo">1340</span> tablesInfo.put(tableName, modTInfo);<a name="line.1340"></a> |
| <span class="sourceLineNo">1341</span> try {<a name="line.1341"></a> |
| <span class="sourceLineNo">1342</span> TableDescriptor htd =<a name="line.1342"></a> |
| <span class="sourceLineNo">1343</span> FSTableDescriptors.getTableDescriptorFromFs(fs, hbaseRoot, tableName);<a name="line.1343"></a> |
| <span class="sourceLineNo">1344</span> modTInfo.htds.add(htd);<a name="line.1344"></a> |
| <span class="sourceLineNo">1345</span> } catch (IOException ioe) {<a name="line.1345"></a> |
| <span class="sourceLineNo">1346</span> if (!orphanTableDirs.containsKey(tableName)) {<a name="line.1346"></a> |
| <span class="sourceLineNo">1347</span> LOG.warn("Unable to read .tableinfo from " + hbaseRoot, ioe);<a name="line.1347"></a> |
| <span class="sourceLineNo">1348</span> //should only report once for each table<a name="line.1348"></a> |
| <span class="sourceLineNo">1349</span> errors.reportError(ERROR_CODE.NO_TABLEINFO_FILE,<a name="line.1349"></a> |
| <span class="sourceLineNo">1350</span> "Unable to read .tableinfo from " + hbaseRoot + "/" + tableName);<a name="line.1350"></a> |
| <span class="sourceLineNo">1351</span> Set<String> columns = new HashSet<>();<a name="line.1351"></a> |
| <span class="sourceLineNo">1352</span> orphanTableDirs.put(tableName, getColumnFamilyList(columns, hbi));<a name="line.1352"></a> |
| <span class="sourceLineNo">1353</span> }<a name="line.1353"></a> |
| <span class="sourceLineNo">1354</span> }<a name="line.1354"></a> |
| <span class="sourceLineNo">1355</span> }<a name="line.1355"></a> |
| <span class="sourceLineNo">1356</span> if (!hbi.isSkipChecks()) {<a name="line.1356"></a> |
| <span class="sourceLineNo">1357</span> modTInfo.addRegionInfo(hbi);<a name="line.1357"></a> |
| <span class="sourceLineNo">1358</span> }<a name="line.1358"></a> |
| <span class="sourceLineNo">1359</span> }<a name="line.1359"></a> |
| <span class="sourceLineNo">1360</span><a name="line.1360"></a> |
| <span class="sourceLineNo">1361</span> loadTableInfosForTablesWithNoRegion();<a name="line.1361"></a> |
| <span class="sourceLineNo">1362</span> errors.print("");<a name="line.1362"></a> |
| <span class="sourceLineNo">1363</span><a name="line.1363"></a> |
| <span class="sourceLineNo">1364</span> return tablesInfo;<a name="line.1364"></a> |
| <span class="sourceLineNo">1365</span> }<a name="line.1365"></a> |
| <span class="sourceLineNo">1366</span><a name="line.1366"></a> |
| <span class="sourceLineNo">1367</span> /**<a name="line.1367"></a> |
| <span class="sourceLineNo">1368</span> * To get the column family list according to the column family dirs<a name="line.1368"></a> |
| <span class="sourceLineNo">1369</span> * @param columns<a name="line.1369"></a> |
| <span class="sourceLineNo">1370</span> * @param hbi<a name="line.1370"></a> |
| <span class="sourceLineNo">1371</span> * @return a set of column families<a name="line.1371"></a> |
| <span class="sourceLineNo">1372</span> * @throws IOException<a name="line.1372"></a> |
| <span class="sourceLineNo">1373</span> */<a name="line.1373"></a> |
| <span class="sourceLineNo">1374</span> private Set<String> getColumnFamilyList(Set<String> columns, HbckRegionInfo hbi)<a name="line.1374"></a> |
| <span class="sourceLineNo">1375</span> throws IOException {<a name="line.1375"></a> |
| <span class="sourceLineNo">1376</span> Path regionDir = hbi.getHdfsRegionDir();<a name="line.1376"></a> |
| <span class="sourceLineNo">1377</span> FileSystem fs = regionDir.getFileSystem(getConf());<a name="line.1377"></a> |
| <span class="sourceLineNo">1378</span> FileStatus[] subDirs = fs.listStatus(regionDir, new FSUtils.FamilyDirFilter(fs));<a name="line.1378"></a> |
| <span class="sourceLineNo">1379</span> for (FileStatus subdir : subDirs) {<a name="line.1379"></a> |
| <span class="sourceLineNo">1380</span> String columnfamily = subdir.getPath().getName();<a name="line.1380"></a> |
| <span class="sourceLineNo">1381</span> columns.add(columnfamily);<a name="line.1381"></a> |
| <span class="sourceLineNo">1382</span> }<a name="line.1382"></a> |
| <span class="sourceLineNo">1383</span> return columns;<a name="line.1383"></a> |
| <span class="sourceLineNo">1384</span> }<a name="line.1384"></a> |
| <span class="sourceLineNo">1385</span><a name="line.1385"></a> |
| <span class="sourceLineNo">1386</span> /**<a name="line.1386"></a> |
| <span class="sourceLineNo">1387</span> * To fabricate a .tableinfo file with following contents<br><a name="line.1387"></a> |
| <span class="sourceLineNo">1388</span> * 1. the correct tablename <br><a name="line.1388"></a> |
| <span class="sourceLineNo">1389</span> * 2. the correct colfamily list<br><a name="line.1389"></a> |
| <span class="sourceLineNo">1390</span> * 3. the default properties for both {@link TableDescriptor} and {@link ColumnFamilyDescriptor}<br><a name="line.1390"></a> |
| <span class="sourceLineNo">1391</span> * @throws IOException<a name="line.1391"></a> |
| <span class="sourceLineNo">1392</span> */<a name="line.1392"></a> |
| <span class="sourceLineNo">1393</span> private boolean fabricateTableInfo(FSTableDescriptors fstd, TableName tableName,<a name="line.1393"></a> |
| <span class="sourceLineNo">1394</span> Set<String> columns) throws IOException {<a name="line.1394"></a> |
| <span class="sourceLineNo">1395</span> if (columns ==null || columns.isEmpty()) return false;<a name="line.1395"></a> |
| <span class="sourceLineNo">1396</span> TableDescriptorBuilder builder = TableDescriptorBuilder.newBuilder(tableName);<a name="line.1396"></a> |
| <span class="sourceLineNo">1397</span> for (String columnfamimly : columns) {<a name="line.1397"></a> |
| <span class="sourceLineNo">1398</span> builder.setColumnFamily(ColumnFamilyDescriptorBuilder.of(columnfamimly));<a name="line.1398"></a> |
| <span class="sourceLineNo">1399</span> }<a name="line.1399"></a> |
| <span class="sourceLineNo">1400</span> fstd.createTableDescriptor(builder.build(), true);<a name="line.1400"></a> |
| <span class="sourceLineNo">1401</span> return true;<a name="line.1401"></a> |
| <span class="sourceLineNo">1402</span> }<a name="line.1402"></a> |
| <span class="sourceLineNo">1403</span><a name="line.1403"></a> |
| <span class="sourceLineNo">1404</span> /**<a name="line.1404"></a> |
| <span class="sourceLineNo">1405</span> * To fix the empty REGIONINFO_QUALIFIER rows from hbase:meta <br><a name="line.1405"></a> |
| <span class="sourceLineNo">1406</span> * @throws IOException<a name="line.1406"></a> |
| <span class="sourceLineNo">1407</span> */<a name="line.1407"></a> |
| <span class="sourceLineNo">1408</span> public void fixEmptyMetaCells() throws IOException {<a name="line.1408"></a> |
| <span class="sourceLineNo">1409</span> if (shouldFixEmptyMetaCells() && !emptyRegionInfoQualifiers.isEmpty()) {<a name="line.1409"></a> |
| <span class="sourceLineNo">1410</span> LOG.info("Trying to fix empty REGIONINFO_QUALIFIER hbase:meta rows.");<a name="line.1410"></a> |
| <span class="sourceLineNo">1411</span> for (Result region : emptyRegionInfoQualifiers) {<a name="line.1411"></a> |
| <span class="sourceLineNo">1412</span> deleteMetaRegion(region.getRow());<a name="line.1412"></a> |
| <span class="sourceLineNo">1413</span> errors.getErrorList().remove(ERROR_CODE.EMPTY_META_CELL);<a name="line.1413"></a> |
| <span class="sourceLineNo">1414</span> }<a name="line.1414"></a> |
| <span class="sourceLineNo">1415</span> emptyRegionInfoQualifiers.clear();<a name="line.1415"></a> |
| <span class="sourceLineNo">1416</span> }<a name="line.1416"></a> |
| <span class="sourceLineNo">1417</span> }<a name="line.1417"></a> |
| <span class="sourceLineNo">1418</span><a name="line.1418"></a> |
| <span class="sourceLineNo">1419</span> /**<a name="line.1419"></a> |
| <span class="sourceLineNo">1420</span> * To fix orphan table by creating a .tableinfo file under tableDir <br><a name="line.1420"></a> |
| <span class="sourceLineNo">1421</span> * 1. if TableInfo is cached, to recover the .tableinfo accordingly <br><a name="line.1421"></a> |
| <span class="sourceLineNo">1422</span> * 2. else create a default .tableinfo file with following items<br><a name="line.1422"></a> |
| <span class="sourceLineNo">1423</span> * &nbsp;2.1 the correct tablename <br><a name="line.1423"></a> |
| <span class="sourceLineNo">1424</span> * &nbsp;2.2 the correct colfamily list<br><a name="line.1424"></a> |
| <span class="sourceLineNo">1425</span> * &nbsp;2.3 the default properties for both {@link TableDescriptor} and {@link ColumnFamilyDescriptor}<br><a name="line.1425"></a> |
| <span class="sourceLineNo">1426</span> * @throws IOException<a name="line.1426"></a> |
| <span class="sourceLineNo">1427</span> */<a name="line.1427"></a> |
| <span class="sourceLineNo">1428</span> public void fixOrphanTables() throws IOException {<a name="line.1428"></a> |
| <span class="sourceLineNo">1429</span> if (shouldFixTableOrphans() && !orphanTableDirs.isEmpty()) {<a name="line.1429"></a> |
| <span class="sourceLineNo">1430</span><a name="line.1430"></a> |
| <span class="sourceLineNo">1431</span> List<TableName> tmpList = new ArrayList<>(orphanTableDirs.keySet().size());<a name="line.1431"></a> |
| <span class="sourceLineNo">1432</span> tmpList.addAll(orphanTableDirs.keySet());<a name="line.1432"></a> |
| <span class="sourceLineNo">1433</span> TableDescriptor[] htds = getTableDescriptors(tmpList);<a name="line.1433"></a> |
| <span class="sourceLineNo">1434</span> Iterator<Entry<TableName, Set<String>>> iter =<a name="line.1434"></a> |
| <span class="sourceLineNo">1435</span> orphanTableDirs.entrySet().iterator();<a name="line.1435"></a> |
| <span class="sourceLineNo">1436</span> int j = 0;<a name="line.1436"></a> |
| <span class="sourceLineNo">1437</span> int numFailedCase = 0;<a name="line.1437"></a> |
| <span class="sourceLineNo">1438</span> FSTableDescriptors fstd = new FSTableDescriptors(getConf());<a name="line.1438"></a> |
| <span class="sourceLineNo">1439</span> while (iter.hasNext()) {<a name="line.1439"></a> |
| <span class="sourceLineNo">1440</span> Entry<TableName, Set<String>> entry =<a name="line.1440"></a> |
| <span class="sourceLineNo">1441</span> iter.next();<a name="line.1441"></a> |
| <span class="sourceLineNo">1442</span> TableName tableName = entry.getKey();<a name="line.1442"></a> |
| <span class="sourceLineNo">1443</span> LOG.info("Trying to fix orphan table error: " + tableName);<a name="line.1443"></a> |
| <span class="sourceLineNo">1444</span> if (j < htds.length) {<a name="line.1444"></a> |
| <span class="sourceLineNo">1445</span> if (tableName.equals(htds[j].getTableName())) {<a name="line.1445"></a> |
| <span class="sourceLineNo">1446</span> TableDescriptor htd = htds[j];<a name="line.1446"></a> |
| <span class="sourceLineNo">1447</span> LOG.info("fixing orphan table: " + tableName + " from cache");<a name="line.1447"></a> |
| <span class="sourceLineNo">1448</span> fstd.createTableDescriptor(htd, true);<a name="line.1448"></a> |
| <span class="sourceLineNo">1449</span> j++;<a name="line.1449"></a> |
| <span class="sourceLineNo">1450</span> iter.remove();<a name="line.1450"></a> |
| <span class="sourceLineNo">1451</span> }<a name="line.1451"></a> |
| <span class="sourceLineNo">1452</span> } else {<a name="line.1452"></a> |
| <span class="sourceLineNo">1453</span> if (fabricateTableInfo(fstd, tableName, entry.getValue())) {<a name="line.1453"></a> |
| <span class="sourceLineNo">1454</span> LOG.warn("fixing orphan table: " + tableName + " with a default .tableinfo file");<a name="line.1454"></a> |
| <span class="sourceLineNo">1455</span> LOG.warn("Strongly recommend to modify the TableDescriptor if necessary for: " + tableName);<a name="line.1455"></a> |
| <span class="sourceLineNo">1456</span> iter.remove();<a name="line.1456"></a> |
| <span class="sourceLineNo">1457</span> } else {<a name="line.1457"></a> |
| <span class="sourceLineNo">1458</span> LOG.error("Unable to create default .tableinfo for " + tableName + " while missing column family information");<a name="line.1458"></a> |
| <span class="sourceLineNo">1459</span> numFailedCase++;<a name="line.1459"></a> |
| <span class="sourceLineNo">1460</span> }<a name="line.1460"></a> |
| <span class="sourceLineNo">1461</span> }<a name="line.1461"></a> |
| <span class="sourceLineNo">1462</span> fixes++;<a name="line.1462"></a> |
| <span class="sourceLineNo">1463</span> }<a name="line.1463"></a> |
| <span class="sourceLineNo">1464</span><a name="line.1464"></a> |
| <span class="sourceLineNo">1465</span> if (orphanTableDirs.isEmpty()) {<a name="line.1465"></a> |
| <span class="sourceLineNo">1466</span> // all orphanTableDirs are luckily recovered<a name="line.1466"></a> |
| <span class="sourceLineNo">1467</span> // re-run doFsck after recovering the .tableinfo file<a name="line.1467"></a> |
| <span class="sourceLineNo">1468</span> setShouldRerun();<a name="line.1468"></a> |
| <span class="sourceLineNo">1469</span> LOG.warn("Strongly recommend to re-run manually hfsck after all orphanTableDirs being fixed");<a name="line.1469"></a> |
| <span class="sourceLineNo">1470</span> } else if (numFailedCase > 0) {<a name="line.1470"></a> |
| <span class="sourceLineNo">1471</span> LOG.error("Failed to fix " + numFailedCase<a name="line.1471"></a> |
| <span class="sourceLineNo">1472</span> + " OrphanTables with default .tableinfo files");<a name="line.1472"></a> |
| <span class="sourceLineNo">1473</span> }<a name="line.1473"></a> |
| <span class="sourceLineNo">1474</span><a name="line.1474"></a> |
| <span class="sourceLineNo">1475</span> }<a name="line.1475"></a> |
| <span class="sourceLineNo">1476</span> //cleanup the list<a name="line.1476"></a> |
| <span class="sourceLineNo">1477</span> orphanTableDirs.clear();<a name="line.1477"></a> |
| <span class="sourceLineNo">1478</span><a name="line.1478"></a> |
| <span class="sourceLineNo">1479</span> }<a name="line.1479"></a> |
| <span class="sourceLineNo">1480</span><a name="line.1480"></a> |
| <span class="sourceLineNo">1481</span> /**<a name="line.1481"></a> |
| <span class="sourceLineNo">1482</span> * Log an appropriate message about whether or not overlapping merges are computed in parallel.<a name="line.1482"></a> |
| <span class="sourceLineNo">1483</span> */<a name="line.1483"></a> |
| <span class="sourceLineNo">1484</span> private void logParallelMerge() {<a name="line.1484"></a> |
| <span class="sourceLineNo">1485</span> if (getConf().getBoolean("hbasefsck.overlap.merge.parallel", true)) {<a name="line.1485"></a> |
| <span class="sourceLineNo">1486</span> LOG.info("Handling overlap merges in parallel. set hbasefsck.overlap.merge.parallel to" +<a name="line.1486"></a> |
| <span class="sourceLineNo">1487</span> " false to run serially.");<a name="line.1487"></a> |
| <span class="sourceLineNo">1488</span> } else {<a name="line.1488"></a> |
| <span class="sourceLineNo">1489</span> LOG.info("Handling overlap merges serially. set hbasefsck.overlap.merge.parallel to" +<a name="line.1489"></a> |
| <span class="sourceLineNo">1490</span> " true to run in parallel.");<a name="line.1490"></a> |
| <span class="sourceLineNo">1491</span> }<a name="line.1491"></a> |
| <span class="sourceLineNo">1492</span> }<a name="line.1492"></a> |
| <span class="sourceLineNo">1493</span><a name="line.1493"></a> |
| <span class="sourceLineNo">1494</span> private SortedMap<TableName, HbckTableInfo> checkHdfsIntegrity(boolean fixHoles,<a name="line.1494"></a> |
| <span class="sourceLineNo">1495</span> boolean fixOverlaps) throws IOException {<a name="line.1495"></a> |
| <span class="sourceLineNo">1496</span> LOG.info("Checking HBase region split map from HDFS data...");<a name="line.1496"></a> |
| <span class="sourceLineNo">1497</span> logParallelMerge();<a name="line.1497"></a> |
| <span class="sourceLineNo">1498</span> for (HbckTableInfo tInfo : tablesInfo.values()) {<a name="line.1498"></a> |
| <span class="sourceLineNo">1499</span> TableIntegrityErrorHandler handler;<a name="line.1499"></a> |
| <span class="sourceLineNo">1500</span> if (fixHoles || fixOverlaps) {<a name="line.1500"></a> |
| <span class="sourceLineNo">1501</span> handler = tInfo.new HDFSIntegrityFixer(tInfo, errors, getConf(),<a name="line.1501"></a> |
| <span class="sourceLineNo">1502</span> fixHoles, fixOverlaps);<a name="line.1502"></a> |
| <span class="sourceLineNo">1503</span> } else {<a name="line.1503"></a> |
| <span class="sourceLineNo">1504</span> handler = tInfo.new IntegrityFixSuggester(tInfo, errors);<a name="line.1504"></a> |
| <span class="sourceLineNo">1505</span> }<a name="line.1505"></a> |
| <span class="sourceLineNo">1506</span> if (!tInfo.checkRegionChain(handler)) {<a name="line.1506"></a> |
| <span class="sourceLineNo">1507</span> // should dump info as well.<a name="line.1507"></a> |
| <span class="sourceLineNo">1508</span> errors.report("Found inconsistency in table " + tInfo.getName());<a name="line.1508"></a> |
| <span class="sourceLineNo">1509</span> }<a name="line.1509"></a> |
| <span class="sourceLineNo">1510</span> }<a name="line.1510"></a> |
| <span class="sourceLineNo">1511</span> return tablesInfo;<a name="line.1511"></a> |
| <span class="sourceLineNo">1512</span> }<a name="line.1512"></a> |
| <span class="sourceLineNo">1513</span><a name="line.1513"></a> |
| <span class="sourceLineNo">1514</span> Path getSidelineDir() throws IOException {<a name="line.1514"></a> |
| <span class="sourceLineNo">1515</span> if (sidelineDir == null) {<a name="line.1515"></a> |
| <span class="sourceLineNo">1516</span> Path hbaseDir = CommonFSUtils.getRootDir(getConf());<a name="line.1516"></a> |
| <span class="sourceLineNo">1517</span> Path hbckDir = new Path(hbaseDir, HConstants.HBCK_SIDELINEDIR_NAME);<a name="line.1517"></a> |
| <span class="sourceLineNo">1518</span> sidelineDir = new Path(hbckDir, hbaseDir.getName() + "-"<a name="line.1518"></a> |
| <span class="sourceLineNo">1519</span> + startMillis);<a name="line.1519"></a> |
| <span class="sourceLineNo">1520</span> }<a name="line.1520"></a> |
| <span class="sourceLineNo">1521</span> return sidelineDir;<a name="line.1521"></a> |
| <span class="sourceLineNo">1522</span> }<a name="line.1522"></a> |
| <span class="sourceLineNo">1523</span><a name="line.1523"></a> |
| <span class="sourceLineNo">1524</span> /**<a name="line.1524"></a> |
| <span class="sourceLineNo">1525</span> * Sideline a region dir (instead of deleting it)<a name="line.1525"></a> |
| <span class="sourceLineNo">1526</span> */<a name="line.1526"></a> |
| <span class="sourceLineNo">1527</span> Path sidelineRegionDir(FileSystem fs, HbckRegionInfo hi) throws IOException {<a name="line.1527"></a> |
| <span class="sourceLineNo">1528</span> return sidelineRegionDir(fs, null, hi);<a name="line.1528"></a> |
| <span class="sourceLineNo">1529</span> }<a name="line.1529"></a> |
| <span class="sourceLineNo">1530</span><a name="line.1530"></a> |
| <span class="sourceLineNo">1531</span> /**<a name="line.1531"></a> |
| <span class="sourceLineNo">1532</span> * Sideline a region dir (instead of deleting it)<a name="line.1532"></a> |
| <span class="sourceLineNo">1533</span> *<a name="line.1533"></a> |
| <span class="sourceLineNo">1534</span> * @param parentDir if specified, the region will be sidelined to folder like<a name="line.1534"></a> |
| <span class="sourceLineNo">1535</span> * {@literal .../parentDir/<table name>/<region name>}. The purpose is to group together<a name="line.1535"></a> |
| <span class="sourceLineNo">1536</span> * similar regions sidelined, for example, those regions should be bulk loaded back later<a name="line.1536"></a> |
| <span class="sourceLineNo">1537</span> * on. If NULL, it is ignored.<a name="line.1537"></a> |
| <span class="sourceLineNo">1538</span> */<a name="line.1538"></a> |
| <span class="sourceLineNo">1539</span> Path sidelineRegionDir(FileSystem fs,<a name="line.1539"></a> |
| <span class="sourceLineNo">1540</span> String parentDir, HbckRegionInfo hi) throws IOException {<a name="line.1540"></a> |
| <span class="sourceLineNo">1541</span> TableName tableName = hi.getTableName();<a name="line.1541"></a> |
| <span class="sourceLineNo">1542</span> Path regionDir = hi.getHdfsRegionDir();<a name="line.1542"></a> |
| <span class="sourceLineNo">1543</span><a name="line.1543"></a> |
| <span class="sourceLineNo">1544</span> if (!fs.exists(regionDir)) {<a name="line.1544"></a> |
| <span class="sourceLineNo">1545</span> LOG.warn("No previous " + regionDir + " exists. Continuing.");<a name="line.1545"></a> |
| <span class="sourceLineNo">1546</span> return null;<a name="line.1546"></a> |
| <span class="sourceLineNo">1547</span> }<a name="line.1547"></a> |
| <span class="sourceLineNo">1548</span><a name="line.1548"></a> |
| <span class="sourceLineNo">1549</span> Path rootDir = getSidelineDir();<a name="line.1549"></a> |
| <span class="sourceLineNo">1550</span> if (parentDir != null) {<a name="line.1550"></a> |
| <span class="sourceLineNo">1551</span> rootDir = new Path(rootDir, parentDir);<a name="line.1551"></a> |
| <span class="sourceLineNo">1552</span> }<a name="line.1552"></a> |
| <span class="sourceLineNo">1553</span> Path sidelineTableDir= CommonFSUtils.getTableDir(rootDir, tableName);<a name="line.1553"></a> |
| <span class="sourceLineNo">1554</span> Path sidelineRegionDir = new Path(sidelineTableDir, regionDir.getName());<a name="line.1554"></a> |
| <span class="sourceLineNo">1555</span> fs.mkdirs(sidelineRegionDir);<a name="line.1555"></a> |
| <span class="sourceLineNo">1556</span> boolean success = false;<a name="line.1556"></a> |
| <span class="sourceLineNo">1557</span> FileStatus[] cfs = fs.listStatus(regionDir);<a name="line.1557"></a> |
| <span class="sourceLineNo">1558</span> if (cfs == null) {<a name="line.1558"></a> |
| <span class="sourceLineNo">1559</span> LOG.info("Region dir is empty: " + regionDir);<a name="line.1559"></a> |
| <span class="sourceLineNo">1560</span> } else {<a name="line.1560"></a> |
| <span class="sourceLineNo">1561</span> for (FileStatus cf : cfs) {<a name="line.1561"></a> |
| <span class="sourceLineNo">1562</span> Path src = cf.getPath();<a name="line.1562"></a> |
| <span class="sourceLineNo">1563</span> Path dst = new Path(sidelineRegionDir, src.getName());<a name="line.1563"></a> |
| <span class="sourceLineNo">1564</span> if (fs.isFile(src)) {<a name="line.1564"></a> |
| <span class="sourceLineNo">1565</span> // simple file<a name="line.1565"></a> |
| <span class="sourceLineNo">1566</span> success = fs.rename(src, dst);<a name="line.1566"></a> |
| <span class="sourceLineNo">1567</span> if (!success) {<a name="line.1567"></a> |
| <span class="sourceLineNo">1568</span> String msg = "Unable to rename file " + src + " to " + dst;<a name="line.1568"></a> |
| <span class="sourceLineNo">1569</span> LOG.error(msg);<a name="line.1569"></a> |
| <span class="sourceLineNo">1570</span> throw new IOException(msg);<a name="line.1570"></a> |
| <span class="sourceLineNo">1571</span> }<a name="line.1571"></a> |
| <span class="sourceLineNo">1572</span> continue;<a name="line.1572"></a> |
| <span class="sourceLineNo">1573</span> }<a name="line.1573"></a> |
| <span class="sourceLineNo">1574</span><a name="line.1574"></a> |
| <span class="sourceLineNo">1575</span> // is a directory.<a name="line.1575"></a> |
| <span class="sourceLineNo">1576</span> fs.mkdirs(dst);<a name="line.1576"></a> |
| <span class="sourceLineNo">1577</span><a name="line.1577"></a> |
| <span class="sourceLineNo">1578</span> LOG.info("Sidelining files from " + src + " into containing region " + dst);<a name="line.1578"></a> |
| <span class="sourceLineNo">1579</span> // FileSystem.rename is inconsistent with directories -- if the<a name="line.1579"></a> |
| <span class="sourceLineNo">1580</span> // dst (foo/a) exists and is a dir, and the src (foo/b) is a dir,<a name="line.1580"></a> |
| <span class="sourceLineNo">1581</span> // it moves the src into the dst dir resulting in (foo/a/b). If<a name="line.1581"></a> |
| <span class="sourceLineNo">1582</span> // the dst does not exist, and the src a dir, src becomes dst. (foo/b)<a name="line.1582"></a> |
| <span class="sourceLineNo">1583</span> FileStatus[] hfiles = fs.listStatus(src);<a name="line.1583"></a> |
| <span class="sourceLineNo">1584</span> if (hfiles != null && hfiles.length > 0) {<a name="line.1584"></a> |
| <span class="sourceLineNo">1585</span> for (FileStatus hfile : hfiles) {<a name="line.1585"></a> |
| <span class="sourceLineNo">1586</span> success = fs.rename(hfile.getPath(), dst);<a name="line.1586"></a> |
| <span class="sourceLineNo">1587</span> if (!success) {<a name="line.1587"></a> |
| <span class="sourceLineNo">1588</span> String msg = "Unable to rename file " + src + " to " + dst;<a name="line.1588"></a> |
| <span class="sourceLineNo">1589</span> LOG.error(msg);<a name="line.1589"></a> |
| <span class="sourceLineNo">1590</span> throw new IOException(msg);<a name="line.1590"></a> |
| <span class="sourceLineNo">1591</span> }<a name="line.1591"></a> |
| <span class="sourceLineNo">1592</span> }<a name="line.1592"></a> |
| <span class="sourceLineNo">1593</span> }<a name="line.1593"></a> |
| <span class="sourceLineNo">1594</span> LOG.debug("Sideline directory contents:");<a name="line.1594"></a> |
| <span class="sourceLineNo">1595</span> debugLsr(sidelineRegionDir);<a name="line.1595"></a> |
| <span class="sourceLineNo">1596</span> }<a name="line.1596"></a> |
| <span class="sourceLineNo">1597</span> }<a name="line.1597"></a> |
| <span class="sourceLineNo">1598</span><a name="line.1598"></a> |
| <span class="sourceLineNo">1599</span> LOG.info("Removing old region dir: " + regionDir);<a name="line.1599"></a> |
| <span class="sourceLineNo">1600</span> success = fs.delete(regionDir, true);<a name="line.1600"></a> |
| <span class="sourceLineNo">1601</span> if (!success) {<a name="line.1601"></a> |
| <span class="sourceLineNo">1602</span> String msg = "Unable to delete dir " + regionDir;<a name="line.1602"></a> |
| <span class="sourceLineNo">1603</span> LOG.error(msg);<a name="line.1603"></a> |
| <span class="sourceLineNo">1604</span> throw new IOException(msg);<a name="line.1604"></a> |
| <span class="sourceLineNo">1605</span> }<a name="line.1605"></a> |
| <span class="sourceLineNo">1606</span> return sidelineRegionDir;<a name="line.1606"></a> |
| <span class="sourceLineNo">1607</span> }<a name="line.1607"></a> |
| <span class="sourceLineNo">1608</span><a name="line.1608"></a> |
| <span class="sourceLineNo">1609</span> /**<a name="line.1609"></a> |
| <span class="sourceLineNo">1610</span> * Load the list of disabled tables in ZK into local set.<a name="line.1610"></a> |
| <span class="sourceLineNo">1611</span> * @throws ZooKeeperConnectionException<a name="line.1611"></a> |
| <span class="sourceLineNo">1612</span> * @throws IOException<a name="line.1612"></a> |
| <span class="sourceLineNo">1613</span> */<a name="line.1613"></a> |
| <span class="sourceLineNo">1614</span> private void loadTableStates()<a name="line.1614"></a> |
| <span class="sourceLineNo">1615</span> throws IOException {<a name="line.1615"></a> |
| <span class="sourceLineNo">1616</span> tableStates = MetaTableAccessor.getTableStates(connection);<a name="line.1616"></a> |
| <span class="sourceLineNo">1617</span> // Add hbase:meta so this tool keeps working. In hbase2, meta is always enabled though it<a name="line.1617"></a> |
| <span class="sourceLineNo">1618</span> // has no entry in the table states. HBCK doesn't work right w/ hbase2 but just do this in<a name="line.1618"></a> |
| <span class="sourceLineNo">1619</span> // meantime.<a name="line.1619"></a> |
| <span class="sourceLineNo">1620</span> this.tableStates.put(TableName.META_TABLE_NAME,<a name="line.1620"></a> |
| <span class="sourceLineNo">1621</span> new TableState(TableName.META_TABLE_NAME, TableState.State.ENABLED));<a name="line.1621"></a> |
| <span class="sourceLineNo">1622</span> }<a name="line.1622"></a> |
| <span class="sourceLineNo">1623</span><a name="line.1623"></a> |
| <span class="sourceLineNo">1624</span> /**<a name="line.1624"></a> |
| <span class="sourceLineNo">1625</span> * Check if the specified region's table is disabled.<a name="line.1625"></a> |
| <span class="sourceLineNo">1626</span> * @param tableName table to check status of<a name="line.1626"></a> |
| <span class="sourceLineNo">1627</span> */<a name="line.1627"></a> |
| <span class="sourceLineNo">1628</span> boolean isTableDisabled(TableName tableName) {<a name="line.1628"></a> |
| <span class="sourceLineNo">1629</span> return tableStates.containsKey(tableName)<a name="line.1629"></a> |
| <span class="sourceLineNo">1630</span> && tableStates.get(tableName)<a name="line.1630"></a> |
| <span class="sourceLineNo">1631</span> .inStates(TableState.State.DISABLED, TableState.State.DISABLING);<a name="line.1631"></a> |
| <span class="sourceLineNo">1632</span> }<a name="line.1632"></a> |
| <span class="sourceLineNo">1633</span><a name="line.1633"></a> |
| <span class="sourceLineNo">1634</span> /**<a name="line.1634"></a> |
| <span class="sourceLineNo">1635</span> * Scan HDFS for all regions, recording their information into<a name="line.1635"></a> |
| <span class="sourceLineNo">1636</span> * regionInfoMap<a name="line.1636"></a> |
| <span class="sourceLineNo">1637</span> */<a name="line.1637"></a> |
| <span class="sourceLineNo">1638</span> public void loadHdfsRegionDirs() throws IOException, InterruptedException {<a name="line.1638"></a> |
| <span class="sourceLineNo">1639</span> Path rootDir = CommonFSUtils.getRootDir(getConf());<a name="line.1639"></a> |
| <span class="sourceLineNo">1640</span> FileSystem fs = rootDir.getFileSystem(getConf());<a name="line.1640"></a> |
| <span class="sourceLineNo">1641</span><a name="line.1641"></a> |
| <span class="sourceLineNo">1642</span> // list all tables from HDFS<a name="line.1642"></a> |
| <span class="sourceLineNo">1643</span> List<FileStatus> tableDirs = Lists.newArrayList();<a name="line.1643"></a> |
| <span class="sourceLineNo">1644</span><a name="line.1644"></a> |
| <span class="sourceLineNo">1645</span> boolean foundVersionFile = fs.exists(new Path(rootDir, HConstants.VERSION_FILE_NAME));<a name="line.1645"></a> |
| <span class="sourceLineNo">1646</span><a name="line.1646"></a> |
| <span class="sourceLineNo">1647</span> List<Path> paths = FSUtils.getTableDirs(fs, rootDir);<a name="line.1647"></a> |
| <span class="sourceLineNo">1648</span> for (Path path : paths) {<a name="line.1648"></a> |
| <span class="sourceLineNo">1649</span> TableName tableName = CommonFSUtils.getTableName(path);<a name="line.1649"></a> |
| <span class="sourceLineNo">1650</span> if ((!checkMetaOnly &&<a name="line.1650"></a> |
| <span class="sourceLineNo">1651</span> isTableIncluded(tableName)) ||<a name="line.1651"></a> |
| <span class="sourceLineNo">1652</span> tableName.equals(TableName.META_TABLE_NAME)) {<a name="line.1652"></a> |
| <span class="sourceLineNo">1653</span> tableDirs.add(fs.getFileStatus(path));<a name="line.1653"></a> |
| <span class="sourceLineNo">1654</span> }<a name="line.1654"></a> |
| <span class="sourceLineNo">1655</span> }<a name="line.1655"></a> |
| <span class="sourceLineNo">1656</span><a name="line.1656"></a> |
| <span class="sourceLineNo">1657</span> // verify that version file exists<a name="line.1657"></a> |
| <span class="sourceLineNo">1658</span> if (!foundVersionFile) {<a name="line.1658"></a> |
| <span class="sourceLineNo">1659</span> errors.reportError(ERROR_CODE.NO_VERSION_FILE,<a name="line.1659"></a> |
| <span class="sourceLineNo">1660</span> "Version file does not exist in root dir " + rootDir);<a name="line.1660"></a> |
| <span class="sourceLineNo">1661</span> if (shouldFixVersionFile()) {<a name="line.1661"></a> |
| <span class="sourceLineNo">1662</span> LOG.info("Trying to create a new " + HConstants.VERSION_FILE_NAME<a name="line.1662"></a> |
| <span class="sourceLineNo">1663</span> + " file.");<a name="line.1663"></a> |
| <span class="sourceLineNo">1664</span> setShouldRerun();<a name="line.1664"></a> |
| <span class="sourceLineNo">1665</span> FSUtils.setVersion(fs, rootDir, getConf().getInt(<a name="line.1665"></a> |
| <span class="sourceLineNo">1666</span> HConstants.THREAD_WAKE_FREQUENCY, 10 * 1000), getConf().getInt(<a name="line.1666"></a> |
| <span class="sourceLineNo">1667</span> HConstants.VERSION_FILE_WRITE_ATTEMPTS,<a name="line.1667"></a> |
| <span class="sourceLineNo">1668</span> HConstants.DEFAULT_VERSION_FILE_WRITE_ATTEMPTS));<a name="line.1668"></a> |
| <span class="sourceLineNo">1669</span> }<a name="line.1669"></a> |
| <span class="sourceLineNo">1670</span> }<a name="line.1670"></a> |
| <span class="sourceLineNo">1671</span><a name="line.1671"></a> |
| <span class="sourceLineNo">1672</span> // Avoid multithreading at table-level because already multithreaded internally at<a name="line.1672"></a> |
| <span class="sourceLineNo">1673</span> // region-level. Additionally multithreading at table-level can lead to deadlock<a name="line.1673"></a> |
| <span class="sourceLineNo">1674</span> // if there are many tables in the cluster. Since there are a limited # of threads<a name="line.1674"></a> |
| <span class="sourceLineNo">1675</span> // in the executor's thread pool and if we multithread at the table-level by putting<a name="line.1675"></a> |
| <span class="sourceLineNo">1676</span> // WorkItemHdfsDir callables into the executor, then we will have some threads in the<a name="line.1676"></a> |
| <span class="sourceLineNo">1677</span> // executor tied up solely in waiting for the tables' region-level calls to complete.<a name="line.1677"></a> |
| <span class="sourceLineNo">1678</span> // If there are enough tables then there will be no actual threads in the pool left<a name="line.1678"></a> |
| <span class="sourceLineNo">1679</span> // for the region-level callables to be serviced.<a name="line.1679"></a> |
| <span class="sourceLineNo">1680</span> for (FileStatus tableDir : tableDirs) {<a name="line.1680"></a> |
| <span class="sourceLineNo">1681</span> LOG.debug("Loading region dirs from " +tableDir.getPath());<a name="line.1681"></a> |
| <span class="sourceLineNo">1682</span> WorkItemHdfsDir item = new WorkItemHdfsDir(fs, errors, tableDir);<a name="line.1682"></a> |
| <span class="sourceLineNo">1683</span> try {<a name="line.1683"></a> |
| <span class="sourceLineNo">1684</span> item.call();<a name="line.1684"></a> |
| <span class="sourceLineNo">1685</span> } catch (ExecutionException e) {<a name="line.1685"></a> |
| <span class="sourceLineNo">1686</span> LOG.warn("Could not completely load table dir " +<a name="line.1686"></a> |
| <span class="sourceLineNo">1687</span> tableDir.getPath(), e.getCause());<a name="line.1687"></a> |
| <span class="sourceLineNo">1688</span> }<a name="line.1688"></a> |
| <span class="sourceLineNo">1689</span> }<a name="line.1689"></a> |
| <span class="sourceLineNo">1690</span> errors.print("");<a name="line.1690"></a> |
| <span class="sourceLineNo">1691</span> }<a name="line.1691"></a> |
| <span class="sourceLineNo">1692</span><a name="line.1692"></a> |
| <span class="sourceLineNo">1693</span> /**<a name="line.1693"></a> |
| <span class="sourceLineNo">1694</span> * Record the location of the hbase:meta region as found in ZooKeeper.<a name="line.1694"></a> |
| <span class="sourceLineNo">1695</span> */<a name="line.1695"></a> |
| <span class="sourceLineNo">1696</span> private boolean recordMetaRegion() throws IOException {<a name="line.1696"></a> |
| <span class="sourceLineNo">1697</span> List<HRegionLocation> locs;<a name="line.1697"></a> |
| <span class="sourceLineNo">1698</span> try (RegionLocator locator = connection.getRegionLocator(TableName.META_TABLE_NAME)) {<a name="line.1698"></a> |
| <span class="sourceLineNo">1699</span> locs = locator.getRegionLocations(HConstants.EMPTY_START_ROW, true);<a name="line.1699"></a> |
| <span class="sourceLineNo">1700</span> }<a name="line.1700"></a> |
| <span class="sourceLineNo">1701</span> if (locs == null || locs.isEmpty()) {<a name="line.1701"></a> |
| <span class="sourceLineNo">1702</span> errors.reportError(ERROR_CODE.NULL_META_REGION, "META region was not found in ZooKeeper");<a name="line.1702"></a> |
| <span class="sourceLineNo">1703</span> return false;<a name="line.1703"></a> |
| <span class="sourceLineNo">1704</span> }<a name="line.1704"></a> |
| <span class="sourceLineNo">1705</span> for (HRegionLocation metaLocation : locs) {<a name="line.1705"></a> |
| <span class="sourceLineNo">1706</span> // Check if Meta region is valid and existing<a name="line.1706"></a> |
| <span class="sourceLineNo">1707</span> if (metaLocation == null) {<a name="line.1707"></a> |
| <span class="sourceLineNo">1708</span> errors.reportError(ERROR_CODE.NULL_META_REGION, "META region location is null");<a name="line.1708"></a> |
| <span class="sourceLineNo">1709</span> return false;<a name="line.1709"></a> |
| <span class="sourceLineNo">1710</span> }<a name="line.1710"></a> |
| <span class="sourceLineNo">1711</span> if (metaLocation.getRegion() == null) {<a name="line.1711"></a> |
| <span class="sourceLineNo">1712</span> errors.reportError(ERROR_CODE.NULL_META_REGION, "META location regionInfo is null");<a name="line.1712"></a> |
| <span class="sourceLineNo">1713</span> return false;<a name="line.1713"></a> |
| <span class="sourceLineNo">1714</span> }<a name="line.1714"></a> |
| <span class="sourceLineNo">1715</span> if (metaLocation.getHostname() == null) {<a name="line.1715"></a> |
| <span class="sourceLineNo">1716</span> errors.reportError(ERROR_CODE.NULL_META_REGION, "META location hostName is null");<a name="line.1716"></a> |
| <span class="sourceLineNo">1717</span> return false;<a name="line.1717"></a> |
| <span class="sourceLineNo">1718</span> }<a name="line.1718"></a> |
| <span class="sourceLineNo">1719</span> ServerName sn = metaLocation.getServerName();<a name="line.1719"></a> |
| <span class="sourceLineNo">1720</span> HbckRegionInfo.MetaEntry m = new HbckRegionInfo.MetaEntry(metaLocation.getRegion(), sn,<a name="line.1720"></a> |
| <span class="sourceLineNo">1721</span> EnvironmentEdgeManager.currentTime());<a name="line.1721"></a> |
| <span class="sourceLineNo">1722</span> HbckRegionInfo hbckRegionInfo = regionInfoMap.get(metaLocation.getRegion().getEncodedName());<a name="line.1722"></a> |
| <span class="sourceLineNo">1723</span> if (hbckRegionInfo == null) {<a name="line.1723"></a> |
| <span class="sourceLineNo">1724</span> regionInfoMap.put(metaLocation.getRegion().getEncodedName(), new HbckRegionInfo(m));<a name="line.1724"></a> |
| <span class="sourceLineNo">1725</span> } else {<a name="line.1725"></a> |
| <span class="sourceLineNo">1726</span> hbckRegionInfo.setMetaEntry(m);<a name="line.1726"></a> |
| <span class="sourceLineNo">1727</span> }<a name="line.1727"></a> |
| <span class="sourceLineNo">1728</span> }<a name="line.1728"></a> |
| <span class="sourceLineNo">1729</span> return true;<a name="line.1729"></a> |
| <span class="sourceLineNo">1730</span> }<a name="line.1730"></a> |
| <span class="sourceLineNo">1731</span><a name="line.1731"></a> |
| <span class="sourceLineNo">1732</span> private ZKWatcher createZooKeeperWatcher() throws IOException {<a name="line.1732"></a> |
| <span class="sourceLineNo">1733</span> return new ZKWatcher(getConf(), "hbase Fsck", new Abortable() {<a name="line.1733"></a> |
| <span class="sourceLineNo">1734</span> @Override<a name="line.1734"></a> |
| <span class="sourceLineNo">1735</span> public void abort(String why, Throwable e) {<a name="line.1735"></a> |
| <span class="sourceLineNo">1736</span> LOG.error(why, e);<a name="line.1736"></a> |
| <span class="sourceLineNo">1737</span> System.exit(1);<a name="line.1737"></a> |
| <span class="sourceLineNo">1738</span> }<a name="line.1738"></a> |
| <span class="sourceLineNo">1739</span><a name="line.1739"></a> |
| <span class="sourceLineNo">1740</span> @Override<a name="line.1740"></a> |
| <span class="sourceLineNo">1741</span> public boolean isAborted() {<a name="line.1741"></a> |
| <span class="sourceLineNo">1742</span> return false;<a name="line.1742"></a> |
| <span class="sourceLineNo">1743</span> }<a name="line.1743"></a> |
| <span class="sourceLineNo">1744</span><a name="line.1744"></a> |
| <span class="sourceLineNo">1745</span> });<a name="line.1745"></a> |
| <span class="sourceLineNo">1746</span> }<a name="line.1746"></a> |
| <span class="sourceLineNo">1747</span><a name="line.1747"></a> |
| <span class="sourceLineNo">1748</span> /**<a name="line.1748"></a> |
| <span class="sourceLineNo">1749</span> * Contacts each regionserver and fetches metadata about regions.<a name="line.1749"></a> |
| <span class="sourceLineNo">1750</span> * @param regionServerList - the list of region servers to connect to<a name="line.1750"></a> |
| <span class="sourceLineNo">1751</span> * @throws IOException if a remote or network exception occurs<a name="line.1751"></a> |
| <span class="sourceLineNo">1752</span> */<a name="line.1752"></a> |
| <span class="sourceLineNo">1753</span> void processRegionServers(Collection<ServerName> regionServerList)<a name="line.1753"></a> |
| <span class="sourceLineNo">1754</span> throws IOException, InterruptedException {<a name="line.1754"></a> |
| <span class="sourceLineNo">1755</span><a name="line.1755"></a> |
| <span class="sourceLineNo">1756</span> List<WorkItemRegion> workItems = new ArrayList<>(regionServerList.size());<a name="line.1756"></a> |
| <span class="sourceLineNo">1757</span> List<Future<Void>> workFutures;<a name="line.1757"></a> |
| <span class="sourceLineNo">1758</span><a name="line.1758"></a> |
| <span class="sourceLineNo">1759</span> // loop to contact each region server in parallel<a name="line.1759"></a> |
| <span class="sourceLineNo">1760</span> for (ServerName rsinfo: regionServerList) {<a name="line.1760"></a> |
| <span class="sourceLineNo">1761</span> workItems.add(new WorkItemRegion(this, rsinfo, errors, connection));<a name="line.1761"></a> |
| <span class="sourceLineNo">1762</span> }<a name="line.1762"></a> |
| <span class="sourceLineNo">1763</span><a name="line.1763"></a> |
| <span class="sourceLineNo">1764</span> workFutures = executor.invokeAll(workItems);<a name="line.1764"></a> |
| <span class="sourceLineNo">1765</span><a name="line.1765"></a> |
| <span class="sourceLineNo">1766</span> for(int i=0; i<workFutures.size(); i++) {<a name="line.1766"></a> |
| <span class="sourceLineNo">1767</span> WorkItemRegion item = workItems.get(i);<a name="line.1767"></a> |
| <span class="sourceLineNo">1768</span> Future<Void> f = workFutures.get(i);<a name="line.1768"></a> |
| <span class="sourceLineNo">1769</span> try {<a name="line.1769"></a> |
| <span class="sourceLineNo">1770</span> f.get();<a name="line.1770"></a> |
| <span class="sourceLineNo">1771</span> } catch(ExecutionException e) {<a name="line.1771"></a> |
| <span class="sourceLineNo">1772</span> LOG.warn("Could not process regionserver {}", item.rsinfo.getAddress(),<a name="line.1772"></a> |
| <span class="sourceLineNo">1773</span> e.getCause());<a name="line.1773"></a> |
| <span class="sourceLineNo">1774</span> }<a name="line.1774"></a> |
| <span class="sourceLineNo">1775</span> }<a name="line.1775"></a> |
| <span class="sourceLineNo">1776</span> }<a name="line.1776"></a> |
| <span class="sourceLineNo">1777</span><a name="line.1777"></a> |
| <span class="sourceLineNo">1778</span> /**<a name="line.1778"></a> |
| <span class="sourceLineNo">1779</span> * Check consistency of all regions that have been found in previous phases.<a name="line.1779"></a> |
| <span class="sourceLineNo">1780</span> */<a name="line.1780"></a> |
| <span class="sourceLineNo">1781</span> private void checkAndFixConsistency()<a name="line.1781"></a> |
| <span class="sourceLineNo">1782</span> throws IOException, KeeperException, InterruptedException {<a name="line.1782"></a> |
| <span class="sourceLineNo">1783</span> // Divide the checks in two phases. One for default/primary replicas and another<a name="line.1783"></a> |
| <span class="sourceLineNo">1784</span> // for the non-primary ones. Keeps code cleaner this way.<a name="line.1784"></a> |
| <span class="sourceLineNo">1785</span><a name="line.1785"></a> |
| <span class="sourceLineNo">1786</span> List<CheckRegionConsistencyWorkItem> workItems = new ArrayList<>(regionInfoMap.size());<a name="line.1786"></a> |
| <span class="sourceLineNo">1787</span> for (java.util.Map.Entry<String, HbckRegionInfo> e: regionInfoMap.entrySet()) {<a name="line.1787"></a> |
| <span class="sourceLineNo">1788</span> if (e.getValue().getReplicaId() == RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.1788"></a> |
| <span class="sourceLineNo">1789</span> workItems.add(new CheckRegionConsistencyWorkItem(e.getKey(), e.getValue()));<a name="line.1789"></a> |
| <span class="sourceLineNo">1790</span> }<a name="line.1790"></a> |
| <span class="sourceLineNo">1791</span> }<a name="line.1791"></a> |
| <span class="sourceLineNo">1792</span> checkRegionConsistencyConcurrently(workItems);<a name="line.1792"></a> |
| <span class="sourceLineNo">1793</span><a name="line.1793"></a> |
| <span class="sourceLineNo">1794</span> boolean prevHdfsCheck = shouldCheckHdfs();<a name="line.1794"></a> |
| <span class="sourceLineNo">1795</span> setCheckHdfs(false); //replicas don't have any hdfs data<a name="line.1795"></a> |
| <span class="sourceLineNo">1796</span> // Run a pass over the replicas and fix any assignment issues that exist on the currently<a name="line.1796"></a> |
| <span class="sourceLineNo">1797</span> // deployed/undeployed replicas.<a name="line.1797"></a> |
| <span class="sourceLineNo">1798</span> List<CheckRegionConsistencyWorkItem> replicaWorkItems = new ArrayList<>(regionInfoMap.size());<a name="line.1798"></a> |
| <span class="sourceLineNo">1799</span> for (java.util.Map.Entry<String, HbckRegionInfo> e: regionInfoMap.entrySet()) {<a name="line.1799"></a> |
| <span class="sourceLineNo">1800</span> if (e.getValue().getReplicaId() != RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.1800"></a> |
| <span class="sourceLineNo">1801</span> replicaWorkItems.add(new CheckRegionConsistencyWorkItem(e.getKey(), e.getValue()));<a name="line.1801"></a> |
| <span class="sourceLineNo">1802</span> }<a name="line.1802"></a> |
| <span class="sourceLineNo">1803</span> }<a name="line.1803"></a> |
| <span class="sourceLineNo">1804</span> checkRegionConsistencyConcurrently(replicaWorkItems);<a name="line.1804"></a> |
| <span class="sourceLineNo">1805</span> setCheckHdfs(prevHdfsCheck);<a name="line.1805"></a> |
| <span class="sourceLineNo">1806</span><a name="line.1806"></a> |
| <span class="sourceLineNo">1807</span> // If some regions is skipped during checkRegionConsistencyConcurrently() phase, we might<a name="line.1807"></a> |
| <span class="sourceLineNo">1808</span> // not get accurate state of the hbase if continuing. The config here allows users to tune<a name="line.1808"></a> |
| <span class="sourceLineNo">1809</span> // the tolerance of number of skipped region.<a name="line.1809"></a> |
| <span class="sourceLineNo">1810</span> // TODO: evaluate the consequence to continue the hbck operation without config.<a name="line.1810"></a> |
| <span class="sourceLineNo">1811</span> int terminateThreshold = getConf().getInt("hbase.hbck.skipped.regions.limit", 0);<a name="line.1811"></a> |
| <span class="sourceLineNo">1812</span> int numOfSkippedRegions = skippedRegions.size();<a name="line.1812"></a> |
| <span class="sourceLineNo">1813</span> if (numOfSkippedRegions > 0 && numOfSkippedRegions > terminateThreshold) {<a name="line.1813"></a> |
| <span class="sourceLineNo">1814</span> throw new IOException(numOfSkippedRegions<a name="line.1814"></a> |
| <span class="sourceLineNo">1815</span> + " region(s) could not be checked or repaired. See logs for detail.");<a name="line.1815"></a> |
| <span class="sourceLineNo">1816</span> }<a name="line.1816"></a> |
| <span class="sourceLineNo">1817</span><a name="line.1817"></a> |
| <span class="sourceLineNo">1818</span> if (shouldCheckHdfs()) {<a name="line.1818"></a> |
| <span class="sourceLineNo">1819</span> checkAndFixTableStates();<a name="line.1819"></a> |
| <span class="sourceLineNo">1820</span> }<a name="line.1820"></a> |
| <span class="sourceLineNo">1821</span> }<a name="line.1821"></a> |
| <span class="sourceLineNo">1822</span><a name="line.1822"></a> |
| <span class="sourceLineNo">1823</span> /**<a name="line.1823"></a> |
| <span class="sourceLineNo">1824</span> * Check consistency of all regions using mulitple threads concurrently.<a name="line.1824"></a> |
| <span class="sourceLineNo">1825</span> */<a name="line.1825"></a> |
| <span class="sourceLineNo">1826</span> private void checkRegionConsistencyConcurrently(<a name="line.1826"></a> |
| <span class="sourceLineNo">1827</span> final List<CheckRegionConsistencyWorkItem> workItems)<a name="line.1827"></a> |
| <span class="sourceLineNo">1828</span> throws IOException, KeeperException, InterruptedException {<a name="line.1828"></a> |
| <span class="sourceLineNo">1829</span> if (workItems.isEmpty()) {<a name="line.1829"></a> |
| <span class="sourceLineNo">1830</span> return; // nothing to check<a name="line.1830"></a> |
| <span class="sourceLineNo">1831</span> }<a name="line.1831"></a> |
| <span class="sourceLineNo">1832</span><a name="line.1832"></a> |
| <span class="sourceLineNo">1833</span> List<Future<Void>> workFutures = executor.invokeAll(workItems);<a name="line.1833"></a> |
| <span class="sourceLineNo">1834</span> for(Future<Void> f: workFutures) {<a name="line.1834"></a> |
| <span class="sourceLineNo">1835</span> try {<a name="line.1835"></a> |
| <span class="sourceLineNo">1836</span> f.get();<a name="line.1836"></a> |
| <span class="sourceLineNo">1837</span> } catch(ExecutionException e1) {<a name="line.1837"></a> |
| <span class="sourceLineNo">1838</span> LOG.warn("Could not check region consistency " , e1.getCause());<a name="line.1838"></a> |
| <span class="sourceLineNo">1839</span> if (e1.getCause() instanceof IOException) {<a name="line.1839"></a> |
| <span class="sourceLineNo">1840</span> throw (IOException)e1.getCause();<a name="line.1840"></a> |
| <span class="sourceLineNo">1841</span> } else if (e1.getCause() instanceof KeeperException) {<a name="line.1841"></a> |
| <span class="sourceLineNo">1842</span> throw (KeeperException)e1.getCause();<a name="line.1842"></a> |
| <span class="sourceLineNo">1843</span> } else if (e1.getCause() instanceof InterruptedException) {<a name="line.1843"></a> |
| <span class="sourceLineNo">1844</span> throw (InterruptedException)e1.getCause();<a name="line.1844"></a> |
| <span class="sourceLineNo">1845</span> } else {<a name="line.1845"></a> |
| <span class="sourceLineNo">1846</span> throw new IOException(e1.getCause());<a name="line.1846"></a> |
| <span class="sourceLineNo">1847</span> }<a name="line.1847"></a> |
| <span class="sourceLineNo">1848</span> }<a name="line.1848"></a> |
| <span class="sourceLineNo">1849</span> }<a name="line.1849"></a> |
| <span class="sourceLineNo">1850</span> }<a name="line.1850"></a> |
| <span class="sourceLineNo">1851</span><a name="line.1851"></a> |
| <span class="sourceLineNo">1852</span> class CheckRegionConsistencyWorkItem implements Callable<Void> {<a name="line.1852"></a> |
| <span class="sourceLineNo">1853</span> private final String key;<a name="line.1853"></a> |
| <span class="sourceLineNo">1854</span> private final HbckRegionInfo hbi;<a name="line.1854"></a> |
| <span class="sourceLineNo">1855</span><a name="line.1855"></a> |
| <span class="sourceLineNo">1856</span> CheckRegionConsistencyWorkItem(String key, HbckRegionInfo hbi) {<a name="line.1856"></a> |
| <span class="sourceLineNo">1857</span> this.key = key;<a name="line.1857"></a> |
| <span class="sourceLineNo">1858</span> this.hbi = hbi;<a name="line.1858"></a> |
| <span class="sourceLineNo">1859</span> }<a name="line.1859"></a> |
| <span class="sourceLineNo">1860</span><a name="line.1860"></a> |
| <span class="sourceLineNo">1861</span> @Override<a name="line.1861"></a> |
| <span class="sourceLineNo">1862</span> public synchronized Void call() throws Exception {<a name="line.1862"></a> |
| <span class="sourceLineNo">1863</span> try {<a name="line.1863"></a> |
| <span class="sourceLineNo">1864</span> checkRegionConsistency(key, hbi);<a name="line.1864"></a> |
| <span class="sourceLineNo">1865</span> } catch (Exception e) {<a name="line.1865"></a> |
| <span class="sourceLineNo">1866</span> // If the region is non-META region, skip this region and send warning/error message; if<a name="line.1866"></a> |
| <span class="sourceLineNo">1867</span> // the region is META region, we should not continue.<a name="line.1867"></a> |
| <span class="sourceLineNo">1868</span> LOG.warn("Unable to complete check or repair the region '" + hbi.getRegionNameAsString()<a name="line.1868"></a> |
| <span class="sourceLineNo">1869</span> + "'.", e);<a name="line.1869"></a> |
| <span class="sourceLineNo">1870</span> if (hbi.getHdfsHRI().isMetaRegion()) {<a name="line.1870"></a> |
| <span class="sourceLineNo">1871</span> throw e;<a name="line.1871"></a> |
| <span class="sourceLineNo">1872</span> }<a name="line.1872"></a> |
| <span class="sourceLineNo">1873</span> LOG.warn("Skip region '" + hbi.getRegionNameAsString() + "'");<a name="line.1873"></a> |
| <span class="sourceLineNo">1874</span> addSkippedRegion(hbi);<a name="line.1874"></a> |
| <span class="sourceLineNo">1875</span> }<a name="line.1875"></a> |
| <span class="sourceLineNo">1876</span> return null;<a name="line.1876"></a> |
| <span class="sourceLineNo">1877</span> }<a name="line.1877"></a> |
| <span class="sourceLineNo">1878</span> }<a name="line.1878"></a> |
| <span class="sourceLineNo">1879</span><a name="line.1879"></a> |
| <span class="sourceLineNo">1880</span> private void addSkippedRegion(final HbckRegionInfo hbi) {<a name="line.1880"></a> |
| <span class="sourceLineNo">1881</span> Set<String> skippedRegionNames = skippedRegions.get(hbi.getTableName());<a name="line.1881"></a> |
| <span class="sourceLineNo">1882</span> if (skippedRegionNames == null) {<a name="line.1882"></a> |
| <span class="sourceLineNo">1883</span> skippedRegionNames = new HashSet<>();<a name="line.1883"></a> |
| <span class="sourceLineNo">1884</span> }<a name="line.1884"></a> |
| <span class="sourceLineNo">1885</span> skippedRegionNames.add(hbi.getRegionNameAsString());<a name="line.1885"></a> |
| <span class="sourceLineNo">1886</span> skippedRegions.put(hbi.getTableName(), skippedRegionNames);<a name="line.1886"></a> |
| <span class="sourceLineNo">1887</span> }<a name="line.1887"></a> |
| <span class="sourceLineNo">1888</span><a name="line.1888"></a> |
| <span class="sourceLineNo">1889</span> /**<a name="line.1889"></a> |
| <span class="sourceLineNo">1890</span> * Check and fix table states, assumes full info available:<a name="line.1890"></a> |
| <span class="sourceLineNo">1891</span> * - tableInfos<a name="line.1891"></a> |
| <span class="sourceLineNo">1892</span> * - empty tables loaded<a name="line.1892"></a> |
| <span class="sourceLineNo">1893</span> */<a name="line.1893"></a> |
| <span class="sourceLineNo">1894</span> private void checkAndFixTableStates() throws IOException {<a name="line.1894"></a> |
| <span class="sourceLineNo">1895</span> // first check dangling states<a name="line.1895"></a> |
| <span class="sourceLineNo">1896</span> for (Entry<TableName, TableState> entry : tableStates.entrySet()) {<a name="line.1896"></a> |
| <span class="sourceLineNo">1897</span> TableName tableName = entry.getKey();<a name="line.1897"></a> |
| <span class="sourceLineNo">1898</span> TableState tableState = entry.getValue();<a name="line.1898"></a> |
| <span class="sourceLineNo">1899</span> HbckTableInfo tableInfo = tablesInfo.get(tableName);<a name="line.1899"></a> |
| <span class="sourceLineNo">1900</span> if (isTableIncluded(tableName)<a name="line.1900"></a> |
| <span class="sourceLineNo">1901</span> && !tableName.isSystemTable()<a name="line.1901"></a> |
| <span class="sourceLineNo">1902</span> && tableInfo == null) {<a name="line.1902"></a> |
| <span class="sourceLineNo">1903</span> if (fixMeta) {<a name="line.1903"></a> |
| <span class="sourceLineNo">1904</span> MetaTableAccessor.deleteTableState(connection, tableName);<a name="line.1904"></a> |
| <span class="sourceLineNo">1905</span> TableState state = MetaTableAccessor.getTableState(connection, tableName);<a name="line.1905"></a> |
| <span class="sourceLineNo">1906</span> if (state != null) {<a name="line.1906"></a> |
| <span class="sourceLineNo">1907</span> errors.reportError(ERROR_CODE.ORPHAN_TABLE_STATE,<a name="line.1907"></a> |
| <span class="sourceLineNo">1908</span> tableName + " unable to delete dangling table state " + tableState);<a name="line.1908"></a> |
| <span class="sourceLineNo">1909</span> }<a name="line.1909"></a> |
| <span class="sourceLineNo">1910</span> } else if (!checkMetaOnly) {<a name="line.1910"></a> |
| <span class="sourceLineNo">1911</span> // dangling table state in meta if checkMetaOnly is false. If checkMetaOnly is<a name="line.1911"></a> |
| <span class="sourceLineNo">1912</span> // true, tableInfo will be null as tablesInfo are not polulated for all tables from hdfs<a name="line.1912"></a> |
| <span class="sourceLineNo">1913</span> errors.reportError(ERROR_CODE.ORPHAN_TABLE_STATE,<a name="line.1913"></a> |
| <span class="sourceLineNo">1914</span> tableName + " has dangling table state " + tableState);<a name="line.1914"></a> |
| <span class="sourceLineNo">1915</span> }<a name="line.1915"></a> |
| <span class="sourceLineNo">1916</span> }<a name="line.1916"></a> |
| <span class="sourceLineNo">1917</span> }<a name="line.1917"></a> |
| <span class="sourceLineNo">1918</span> // check that all tables have states<a name="line.1918"></a> |
| <span class="sourceLineNo">1919</span> for (TableName tableName : tablesInfo.keySet()) {<a name="line.1919"></a> |
| <span class="sourceLineNo">1920</span> if (isTableIncluded(tableName) && !tableStates.containsKey(tableName)) {<a name="line.1920"></a> |
| <span class="sourceLineNo">1921</span> if (fixMeta) {<a name="line.1921"></a> |
| <span class="sourceLineNo">1922</span> MetaTableAccessor.updateTableState(connection, tableName, TableState.State.ENABLED);<a name="line.1922"></a> |
| <span class="sourceLineNo">1923</span> TableState newState = MetaTableAccessor.getTableState(connection, tableName);<a name="line.1923"></a> |
| <span class="sourceLineNo">1924</span> if (newState == null) {<a name="line.1924"></a> |
| <span class="sourceLineNo">1925</span> errors.reportError(ERROR_CODE.NO_TABLE_STATE,<a name="line.1925"></a> |
| <span class="sourceLineNo">1926</span> "Unable to change state for table " + tableName + " in meta ");<a name="line.1926"></a> |
| <span class="sourceLineNo">1927</span> }<a name="line.1927"></a> |
| <span class="sourceLineNo">1928</span> } else {<a name="line.1928"></a> |
| <span class="sourceLineNo">1929</span> errors.reportError(ERROR_CODE.NO_TABLE_STATE,<a name="line.1929"></a> |
| <span class="sourceLineNo">1930</span> tableName + " has no state in meta ");<a name="line.1930"></a> |
| <span class="sourceLineNo">1931</span> }<a name="line.1931"></a> |
| <span class="sourceLineNo">1932</span> }<a name="line.1932"></a> |
| <span class="sourceLineNo">1933</span> }<a name="line.1933"></a> |
| <span class="sourceLineNo">1934</span> }<a name="line.1934"></a> |
| <span class="sourceLineNo">1935</span><a name="line.1935"></a> |
| <span class="sourceLineNo">1936</span> private void preCheckPermission() throws IOException {<a name="line.1936"></a> |
| <span class="sourceLineNo">1937</span> if (shouldIgnorePreCheckPermission()) {<a name="line.1937"></a> |
| <span class="sourceLineNo">1938</span> return;<a name="line.1938"></a> |
| <span class="sourceLineNo">1939</span> }<a name="line.1939"></a> |
| <span class="sourceLineNo">1940</span><a name="line.1940"></a> |
| <span class="sourceLineNo">1941</span> Path hbaseDir = CommonFSUtils.getRootDir(getConf());<a name="line.1941"></a> |
| <span class="sourceLineNo">1942</span> FileSystem fs = hbaseDir.getFileSystem(getConf());<a name="line.1942"></a> |
| <span class="sourceLineNo">1943</span> UserProvider userProvider = UserProvider.instantiate(getConf());<a name="line.1943"></a> |
| <span class="sourceLineNo">1944</span> UserGroupInformation ugi = userProvider.getCurrent().getUGI();<a name="line.1944"></a> |
| <span class="sourceLineNo">1945</span> FileStatus[] files = fs.listStatus(hbaseDir);<a name="line.1945"></a> |
| <span class="sourceLineNo">1946</span> for (FileStatus file : files) {<a name="line.1946"></a> |
| <span class="sourceLineNo">1947</span> try {<a name="line.1947"></a> |
| <span class="sourceLineNo">1948</span> fs.access(file.getPath(), FsAction.WRITE);<a name="line.1948"></a> |
| <span class="sourceLineNo">1949</span> } catch (AccessControlException ace) {<a name="line.1949"></a> |
| <span class="sourceLineNo">1950</span> LOG.warn("Got AccessDeniedException when preCheckPermission ", ace);<a name="line.1950"></a> |
| <span class="sourceLineNo">1951</span> errors.reportError(ERROR_CODE.WRONG_USAGE, "Current user " + ugi.getUserName()<a name="line.1951"></a> |
| <span class="sourceLineNo">1952</span> + " does not have write perms to " + file.getPath()<a name="line.1952"></a> |
| <span class="sourceLineNo">1953</span> + ". Please rerun hbck as hdfs user " + file.getOwner());<a name="line.1953"></a> |
| <span class="sourceLineNo">1954</span> throw ace;<a name="line.1954"></a> |
| <span class="sourceLineNo">1955</span> }<a name="line.1955"></a> |
| <span class="sourceLineNo">1956</span> }<a name="line.1956"></a> |
| <span class="sourceLineNo">1957</span> }<a name="line.1957"></a> |
| <span class="sourceLineNo">1958</span><a name="line.1958"></a> |
| <span class="sourceLineNo">1959</span> /**<a name="line.1959"></a> |
| <span class="sourceLineNo">1960</span> * Deletes region from meta table<a name="line.1960"></a> |
| <span class="sourceLineNo">1961</span> */<a name="line.1961"></a> |
| <span class="sourceLineNo">1962</span> private void deleteMetaRegion(HbckRegionInfo hi) throws IOException {<a name="line.1962"></a> |
| <span class="sourceLineNo">1963</span> deleteMetaRegion(hi.getMetaEntry().getRegionInfo().getRegionName());<a name="line.1963"></a> |
| <span class="sourceLineNo">1964</span> }<a name="line.1964"></a> |
| <span class="sourceLineNo">1965</span><a name="line.1965"></a> |
| <span class="sourceLineNo">1966</span> /**<a name="line.1966"></a> |
| <span class="sourceLineNo">1967</span> * Deletes region from meta table<a name="line.1967"></a> |
| <span class="sourceLineNo">1968</span> */<a name="line.1968"></a> |
| <span class="sourceLineNo">1969</span> private void deleteMetaRegion(byte[] metaKey) throws IOException {<a name="line.1969"></a> |
| <span class="sourceLineNo">1970</span> Delete d = new Delete(metaKey);<a name="line.1970"></a> |
| <span class="sourceLineNo">1971</span> meta.delete(d);<a name="line.1971"></a> |
| <span class="sourceLineNo">1972</span> LOG.info("Deleted " + Bytes.toString(metaKey) + " from META" );<a name="line.1972"></a> |
| <span class="sourceLineNo">1973</span> }<a name="line.1973"></a> |
| <span class="sourceLineNo">1974</span><a name="line.1974"></a> |
| <span class="sourceLineNo">1975</span> /**<a name="line.1975"></a> |
| <span class="sourceLineNo">1976</span> * Reset the split parent region info in meta table<a name="line.1976"></a> |
| <span class="sourceLineNo">1977</span> */<a name="line.1977"></a> |
| <span class="sourceLineNo">1978</span> private void resetSplitParent(HbckRegionInfo hi) throws IOException {<a name="line.1978"></a> |
| <span class="sourceLineNo">1979</span> RowMutations mutations = new RowMutations(hi.getMetaEntry().getRegionInfo().getRegionName());<a name="line.1979"></a> |
| <span class="sourceLineNo">1980</span> Delete d = new Delete(hi.getMetaEntry().getRegionInfo().getRegionName());<a name="line.1980"></a> |
| <span class="sourceLineNo">1981</span> d.addColumn(HConstants.CATALOG_FAMILY, HConstants.SPLITA_QUALIFIER);<a name="line.1981"></a> |
| <span class="sourceLineNo">1982</span> d.addColumn(HConstants.CATALOG_FAMILY, HConstants.SPLITB_QUALIFIER);<a name="line.1982"></a> |
| <span class="sourceLineNo">1983</span> mutations.add(d);<a name="line.1983"></a> |
| <span class="sourceLineNo">1984</span><a name="line.1984"></a> |
| <span class="sourceLineNo">1985</span> RegionInfo hri = RegionInfoBuilder.newBuilder(hi.getMetaEntry().getRegionInfo())<a name="line.1985"></a> |
| <span class="sourceLineNo">1986</span> .setOffline(false).setSplit(false).build();<a name="line.1986"></a> |
| <span class="sourceLineNo">1987</span> Put p = MetaTableAccessor.makePutFromRegionInfo(hri, EnvironmentEdgeManager.currentTime());<a name="line.1987"></a> |
| <span class="sourceLineNo">1988</span> mutations.add(p);<a name="line.1988"></a> |
| <span class="sourceLineNo">1989</span><a name="line.1989"></a> |
| <span class="sourceLineNo">1990</span> meta.mutateRow(mutations);<a name="line.1990"></a> |
| <span class="sourceLineNo">1991</span> LOG.info("Reset split parent " + hi.getMetaEntry().getRegionInfo().getRegionNameAsString() +<a name="line.1991"></a> |
| <span class="sourceLineNo">1992</span> " in META");<a name="line.1992"></a> |
| <span class="sourceLineNo">1993</span> }<a name="line.1993"></a> |
| <span class="sourceLineNo">1994</span><a name="line.1994"></a> |
| <span class="sourceLineNo">1995</span> /**<a name="line.1995"></a> |
| <span class="sourceLineNo">1996</span> * This backwards-compatibility wrapper for permanently offlining a region<a name="line.1996"></a> |
| <span class="sourceLineNo">1997</span> * that should not be alive. If the region server does not support the<a name="line.1997"></a> |
| <span class="sourceLineNo">1998</span> * "offline" method, it will use the closest unassign method instead. This<a name="line.1998"></a> |
| <span class="sourceLineNo">1999</span> * will basically work until one attempts to disable or delete the affected<a name="line.1999"></a> |
| <span class="sourceLineNo">2000</span> * table. The problem has to do with in-memory only master state, so<a name="line.2000"></a> |
| <span class="sourceLineNo">2001</span> * restarting the HMaster or failing over to another should fix this.<a name="line.2001"></a> |
| <span class="sourceLineNo">2002</span> */<a name="line.2002"></a> |
| <span class="sourceLineNo">2003</span> void offline(byte[] regionName) throws IOException {<a name="line.2003"></a> |
| <span class="sourceLineNo">2004</span> String regionString = Bytes.toStringBinary(regionName);<a name="line.2004"></a> |
| <span class="sourceLineNo">2005</span> if (!rsSupportsOffline) {<a name="line.2005"></a> |
| <span class="sourceLineNo">2006</span> LOG.warn(<a name="line.2006"></a> |
| <span class="sourceLineNo">2007</span> "Using unassign region " + regionString + " instead of using offline method, you should" +<a name="line.2007"></a> |
| <span class="sourceLineNo">2008</span> " restart HMaster after these repairs");<a name="line.2008"></a> |
| <span class="sourceLineNo">2009</span> admin.unassign(regionName, true);<a name="line.2009"></a> |
| <span class="sourceLineNo">2010</span> return;<a name="line.2010"></a> |
| <span class="sourceLineNo">2011</span> }<a name="line.2011"></a> |
| <span class="sourceLineNo">2012</span><a name="line.2012"></a> |
| <span class="sourceLineNo">2013</span> // first time we assume the rs's supports #offline.<a name="line.2013"></a> |
| <span class="sourceLineNo">2014</span> try {<a name="line.2014"></a> |
| <span class="sourceLineNo">2015</span> LOG.info("Offlining region " + regionString);<a name="line.2015"></a> |
| <span class="sourceLineNo">2016</span> admin.offline(regionName);<a name="line.2016"></a> |
| <span class="sourceLineNo">2017</span> } catch (IOException ioe) {<a name="line.2017"></a> |
| <span class="sourceLineNo">2018</span> String notFoundMsg = "java.lang.NoSuchMethodException: " +<a name="line.2018"></a> |
| <span class="sourceLineNo">2019</span> "org.apache.hadoop.hbase.master.HMaster.offline([B)";<a name="line.2019"></a> |
| <span class="sourceLineNo">2020</span> if (ioe.getMessage().contains(notFoundMsg)) {<a name="line.2020"></a> |
| <span class="sourceLineNo">2021</span> LOG.warn("Using unassign region " + regionString +<a name="line.2021"></a> |
| <span class="sourceLineNo">2022</span> " instead of using offline method, you should" +<a name="line.2022"></a> |
| <span class="sourceLineNo">2023</span> " restart HMaster after these repairs");<a name="line.2023"></a> |
| <span class="sourceLineNo">2024</span> rsSupportsOffline = false; // in the future just use unassign<a name="line.2024"></a> |
| <span class="sourceLineNo">2025</span> admin.unassign(regionName, true);<a name="line.2025"></a> |
| <span class="sourceLineNo">2026</span> return;<a name="line.2026"></a> |
| <span class="sourceLineNo">2027</span> }<a name="line.2027"></a> |
| <span class="sourceLineNo">2028</span> throw ioe;<a name="line.2028"></a> |
| <span class="sourceLineNo">2029</span> }<a name="line.2029"></a> |
| <span class="sourceLineNo">2030</span> }<a name="line.2030"></a> |
| <span class="sourceLineNo">2031</span><a name="line.2031"></a> |
| <span class="sourceLineNo">2032</span> /**<a name="line.2032"></a> |
| <span class="sourceLineNo">2033</span> * Attempts to undeploy a region from a region server based in information in<a name="line.2033"></a> |
| <span class="sourceLineNo">2034</span> * META. Any operations that modify the file system should make sure that<a name="line.2034"></a> |
| <span class="sourceLineNo">2035</span> * its corresponding region is not deployed to prevent data races.<a name="line.2035"></a> |
| <span class="sourceLineNo">2036</span> *<a name="line.2036"></a> |
| <span class="sourceLineNo">2037</span> * A separate call is required to update the master in-memory region state<a name="line.2037"></a> |
| <span class="sourceLineNo">2038</span> * kept in the AssignementManager. Because disable uses this state instead of<a name="line.2038"></a> |
| <span class="sourceLineNo">2039</span> * that found in META, we can't seem to cleanly disable/delete tables that<a name="line.2039"></a> |
| <span class="sourceLineNo">2040</span> * have been hbck fixed. When used on a version of HBase that does not have<a name="line.2040"></a> |
| <span class="sourceLineNo">2041</span> * the offline ipc call exposed on the master (&lt;0.90.5, &lt;0.92.0) a master<a name="line.2041"></a> |
| <span class="sourceLineNo">2042</span> * restart or failover may be required.<a name="line.2042"></a> |
| <span class="sourceLineNo">2043</span> */<a name="line.2043"></a> |
| <span class="sourceLineNo">2044</span> void closeRegion(HbckRegionInfo hi) throws IOException, InterruptedException {<a name="line.2044"></a> |
| <span class="sourceLineNo">2045</span> if (hi.getMetaEntry() == null && hi.getHdfsEntry() == null) {<a name="line.2045"></a> |
| <span class="sourceLineNo">2046</span> undeployRegions(hi);<a name="line.2046"></a> |
| <span class="sourceLineNo">2047</span> return;<a name="line.2047"></a> |
| <span class="sourceLineNo">2048</span> }<a name="line.2048"></a> |
| <span class="sourceLineNo">2049</span><a name="line.2049"></a> |
| <span class="sourceLineNo">2050</span> // get assignment info and hregioninfo from meta.<a name="line.2050"></a> |
| <span class="sourceLineNo">2051</span> Get get = new Get(hi.getRegionName());<a name="line.2051"></a> |
| <span class="sourceLineNo">2052</span> get.addColumn(HConstants.CATALOG_FAMILY, HConstants.REGIONINFO_QUALIFIER);<a name="line.2052"></a> |
| <span class="sourceLineNo">2053</span> get.addColumn(HConstants.CATALOG_FAMILY, HConstants.SERVER_QUALIFIER);<a name="line.2053"></a> |
| <span class="sourceLineNo">2054</span> get.addColumn(HConstants.CATALOG_FAMILY, HConstants.STARTCODE_QUALIFIER);<a name="line.2054"></a> |
| <span class="sourceLineNo">2055</span> // also get the locations of the replicas to close if the primary region is being closed<a name="line.2055"></a> |
| <span class="sourceLineNo">2056</span> if (hi.getReplicaId() == RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.2056"></a> |
| <span class="sourceLineNo">2057</span> int numReplicas = admin.getDescriptor(hi.getTableName()).getRegionReplication();<a name="line.2057"></a> |
| <span class="sourceLineNo">2058</span> for (int i = 0; i < numReplicas; i++) {<a name="line.2058"></a> |
| <span class="sourceLineNo">2059</span> get.addColumn(HConstants.CATALOG_FAMILY, CatalogFamilyFormat.getServerColumn(i));<a name="line.2059"></a> |
| <span class="sourceLineNo">2060</span> get.addColumn(HConstants.CATALOG_FAMILY, CatalogFamilyFormat.getStartCodeColumn(i));<a name="line.2060"></a> |
| <span class="sourceLineNo">2061</span> }<a name="line.2061"></a> |
| <span class="sourceLineNo">2062</span> }<a name="line.2062"></a> |
| <span class="sourceLineNo">2063</span> Result r = meta.get(get);<a name="line.2063"></a> |
| <span class="sourceLineNo">2064</span> RegionLocations rl = CatalogFamilyFormat.getRegionLocations(r);<a name="line.2064"></a> |
| <span class="sourceLineNo">2065</span> if (rl == null) {<a name="line.2065"></a> |
| <span class="sourceLineNo">2066</span> LOG.warn("Unable to close region " + hi.getRegionNameAsString() +<a name="line.2066"></a> |
| <span class="sourceLineNo">2067</span> " since meta does not have handle to reach it");<a name="line.2067"></a> |
| <span class="sourceLineNo">2068</span> return;<a name="line.2068"></a> |
| <span class="sourceLineNo">2069</span> }<a name="line.2069"></a> |
| <span class="sourceLineNo">2070</span> for (HRegionLocation h : rl.getRegionLocations()) {<a name="line.2070"></a> |
| <span class="sourceLineNo">2071</span> ServerName serverName = h.getServerName();<a name="line.2071"></a> |
| <span class="sourceLineNo">2072</span> if (serverName == null) {<a name="line.2072"></a> |
| <span class="sourceLineNo">2073</span> errors.reportError("Unable to close region "<a name="line.2073"></a> |
| <span class="sourceLineNo">2074</span> + hi.getRegionNameAsString() + " because meta does not "<a name="line.2074"></a> |
| <span class="sourceLineNo">2075</span> + "have handle to reach it.");<a name="line.2075"></a> |
| <span class="sourceLineNo">2076</span> continue;<a name="line.2076"></a> |
| <span class="sourceLineNo">2077</span> }<a name="line.2077"></a> |
| <span class="sourceLineNo">2078</span> RegionInfo hri = h.getRegion();<a name="line.2078"></a> |
| <span class="sourceLineNo">2079</span> if (hri == null) {<a name="line.2079"></a> |
| <span class="sourceLineNo">2080</span> LOG.warn("Unable to close region " + hi.getRegionNameAsString()<a name="line.2080"></a> |
| <span class="sourceLineNo">2081</span> + " because hbase:meta had invalid or missing "<a name="line.2081"></a> |
| <span class="sourceLineNo">2082</span> + HConstants.CATALOG_FAMILY_STR + ":"<a name="line.2082"></a> |
| <span class="sourceLineNo">2083</span> + Bytes.toString(HConstants.REGIONINFO_QUALIFIER)<a name="line.2083"></a> |
| <span class="sourceLineNo">2084</span> + " qualifier value.");<a name="line.2084"></a> |
| <span class="sourceLineNo">2085</span> continue;<a name="line.2085"></a> |
| <span class="sourceLineNo">2086</span> }<a name="line.2086"></a> |
| <span class="sourceLineNo">2087</span> // close the region -- close files and remove assignment<a name="line.2087"></a> |
| <span class="sourceLineNo">2088</span> HBaseFsckRepair.closeRegionSilentlyAndWait(connection, serverName, hri);<a name="line.2088"></a> |
| <span class="sourceLineNo">2089</span> }<a name="line.2089"></a> |
| <span class="sourceLineNo">2090</span> }<a name="line.2090"></a> |
| <span class="sourceLineNo">2091</span><a name="line.2091"></a> |
| <span class="sourceLineNo">2092</span> private void undeployRegions(HbckRegionInfo hi) throws IOException, InterruptedException {<a name="line.2092"></a> |
| <span class="sourceLineNo">2093</span> undeployRegionsForHbi(hi);<a name="line.2093"></a> |
| <span class="sourceLineNo">2094</span> // undeploy replicas of the region (but only if the method is invoked for the primary)<a name="line.2094"></a> |
| <span class="sourceLineNo">2095</span> if (hi.getReplicaId() != RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.2095"></a> |
| <span class="sourceLineNo">2096</span> return;<a name="line.2096"></a> |
| <span class="sourceLineNo">2097</span> }<a name="line.2097"></a> |
| <span class="sourceLineNo">2098</span> int numReplicas = admin.getDescriptor(hi.getTableName()).getRegionReplication();<a name="line.2098"></a> |
| <span class="sourceLineNo">2099</span> for (int i = 1; i < numReplicas; i++) {<a name="line.2099"></a> |
| <span class="sourceLineNo">2100</span> if (hi.getPrimaryHRIForDeployedReplica() == null) continue;<a name="line.2100"></a> |
| <span class="sourceLineNo">2101</span> RegionInfo hri = RegionReplicaUtil.getRegionInfoForReplica(<a name="line.2101"></a> |
| <span class="sourceLineNo">2102</span> hi.getPrimaryHRIForDeployedReplica(), i);<a name="line.2102"></a> |
| <span class="sourceLineNo">2103</span> HbckRegionInfo h = regionInfoMap.get(hri.getEncodedName());<a name="line.2103"></a> |
| <span class="sourceLineNo">2104</span> if (h != null) {<a name="line.2104"></a> |
| <span class="sourceLineNo">2105</span> undeployRegionsForHbi(h);<a name="line.2105"></a> |
| <span class="sourceLineNo">2106</span> //set skip checks; we undeployed it, and we don't want to evaluate this anymore<a name="line.2106"></a> |
| <span class="sourceLineNo">2107</span> //in consistency checks<a name="line.2107"></a> |
| <span class="sourceLineNo">2108</span> h.setSkipChecks(true);<a name="line.2108"></a> |
| <span class="sourceLineNo">2109</span> }<a name="line.2109"></a> |
| <span class="sourceLineNo">2110</span> }<a name="line.2110"></a> |
| <span class="sourceLineNo">2111</span> }<a name="line.2111"></a> |
| <span class="sourceLineNo">2112</span><a name="line.2112"></a> |
| <span class="sourceLineNo">2113</span> private void undeployRegionsForHbi(HbckRegionInfo hi) throws IOException, InterruptedException {<a name="line.2113"></a> |
| <span class="sourceLineNo">2114</span> for (HbckRegionInfo.OnlineEntry rse : hi.getOnlineEntries()) {<a name="line.2114"></a> |
| <span class="sourceLineNo">2115</span> LOG.debug("Undeploy region " + rse.getRegionInfo() + " from " + rse.getServerName());<a name="line.2115"></a> |
| <span class="sourceLineNo">2116</span> try {<a name="line.2116"></a> |
| <span class="sourceLineNo">2117</span> HBaseFsckRepair<a name="line.2117"></a> |
| <span class="sourceLineNo">2118</span> .closeRegionSilentlyAndWait(connection, rse.getServerName(), rse.getRegionInfo());<a name="line.2118"></a> |
| <span class="sourceLineNo">2119</span> offline(rse.getRegionInfo().getRegionName());<a name="line.2119"></a> |
| <span class="sourceLineNo">2120</span> } catch (IOException ioe) {<a name="line.2120"></a> |
| <span class="sourceLineNo">2121</span> LOG.warn("Got exception when attempting to offline region "<a name="line.2121"></a> |
| <span class="sourceLineNo">2122</span> + Bytes.toString(rse.getRegionInfo().getRegionName()), ioe);<a name="line.2122"></a> |
| <span class="sourceLineNo">2123</span> }<a name="line.2123"></a> |
| <span class="sourceLineNo">2124</span> }<a name="line.2124"></a> |
| <span class="sourceLineNo">2125</span> }<a name="line.2125"></a> |
| <span class="sourceLineNo">2126</span><a name="line.2126"></a> |
| <span class="sourceLineNo">2127</span> private void tryAssignmentRepair(HbckRegionInfo hbi, String msg) throws IOException,<a name="line.2127"></a> |
| <span class="sourceLineNo">2128</span> KeeperException, InterruptedException {<a name="line.2128"></a> |
| <span class="sourceLineNo">2129</span> // If we are trying to fix the errors<a name="line.2129"></a> |
| <span class="sourceLineNo">2130</span> if (shouldFixAssignments()) {<a name="line.2130"></a> |
| <span class="sourceLineNo">2131</span> errors.print(msg);<a name="line.2131"></a> |
| <span class="sourceLineNo">2132</span> undeployRegions(hbi);<a name="line.2132"></a> |
| <span class="sourceLineNo">2133</span> setShouldRerun();<a name="line.2133"></a> |
| <span class="sourceLineNo">2134</span> RegionInfo hri = hbi.getHdfsHRI();<a name="line.2134"></a> |
| <span class="sourceLineNo">2135</span> if (hri == null) {<a name="line.2135"></a> |
| <span class="sourceLineNo">2136</span> hri = hbi.getMetaEntry().getRegionInfo();<a name="line.2136"></a> |
| <span class="sourceLineNo">2137</span> }<a name="line.2137"></a> |
| <span class="sourceLineNo">2138</span> HBaseFsckRepair.fixUnassigned(admin, hri);<a name="line.2138"></a> |
| <span class="sourceLineNo">2139</span> HBaseFsckRepair.waitUntilAssigned(admin, hri);<a name="line.2139"></a> |
| <span class="sourceLineNo">2140</span><a name="line.2140"></a> |
| <span class="sourceLineNo">2141</span> // also assign replicas if needed (do it only when this call operates on a primary replica)<a name="line.2141"></a> |
| <span class="sourceLineNo">2142</span> if (hbi.getReplicaId() != RegionInfo.DEFAULT_REPLICA_ID) return;<a name="line.2142"></a> |
| <span class="sourceLineNo">2143</span> int replicationCount = admin.getDescriptor(hri.getTable()).getRegionReplication();<a name="line.2143"></a> |
| <span class="sourceLineNo">2144</span> for (int i = 1; i < replicationCount; i++) {<a name="line.2144"></a> |
| <span class="sourceLineNo">2145</span> hri = RegionReplicaUtil.getRegionInfoForReplica(hri, i);<a name="line.2145"></a> |
| <span class="sourceLineNo">2146</span> HbckRegionInfo h = regionInfoMap.get(hri.getEncodedName());<a name="line.2146"></a> |
| <span class="sourceLineNo">2147</span> if (h != null) {<a name="line.2147"></a> |
| <span class="sourceLineNo">2148</span> undeployRegions(h);<a name="line.2148"></a> |
| <span class="sourceLineNo">2149</span> //set skip checks; we undeploy & deploy it; we don't want to evaluate this hbi anymore<a name="line.2149"></a> |
| <span class="sourceLineNo">2150</span> //in consistency checks<a name="line.2150"></a> |
| <span class="sourceLineNo">2151</span> h.setSkipChecks(true);<a name="line.2151"></a> |
| <span class="sourceLineNo">2152</span> }<a name="line.2152"></a> |
| <span class="sourceLineNo">2153</span> HBaseFsckRepair.fixUnassigned(admin, hri);<a name="line.2153"></a> |
| <span class="sourceLineNo">2154</span> HBaseFsckRepair.waitUntilAssigned(admin, hri);<a name="line.2154"></a> |
| <span class="sourceLineNo">2155</span> }<a name="line.2155"></a> |
| <span class="sourceLineNo">2156</span><a name="line.2156"></a> |
| <span class="sourceLineNo">2157</span> }<a name="line.2157"></a> |
| <span class="sourceLineNo">2158</span> }<a name="line.2158"></a> |
| <span class="sourceLineNo">2159</span><a name="line.2159"></a> |
| <span class="sourceLineNo">2160</span> /**<a name="line.2160"></a> |
| <span class="sourceLineNo">2161</span> * Check a single region for consistency and correct deployment.<a name="line.2161"></a> |
| <span class="sourceLineNo">2162</span> */<a name="line.2162"></a> |
| <span class="sourceLineNo">2163</span> private void checkRegionConsistency(final String key, final HbckRegionInfo hbi)<a name="line.2163"></a> |
| <span class="sourceLineNo">2164</span> throws IOException, KeeperException, InterruptedException {<a name="line.2164"></a> |
| <span class="sourceLineNo">2165</span><a name="line.2165"></a> |
| <span class="sourceLineNo">2166</span> if (hbi.isSkipChecks()) return;<a name="line.2166"></a> |
| <span class="sourceLineNo">2167</span> String descriptiveName = hbi.toString();<a name="line.2167"></a> |
| <span class="sourceLineNo">2168</span> boolean inMeta = hbi.getMetaEntry() != null;<a name="line.2168"></a> |
| <span class="sourceLineNo">2169</span> // In case not checking HDFS, assume the region is on HDFS<a name="line.2169"></a> |
| <span class="sourceLineNo">2170</span> boolean inHdfs = !shouldCheckHdfs() || hbi.getHdfsRegionDir() != null;<a name="line.2170"></a> |
| <span class="sourceLineNo">2171</span> boolean hasMetaAssignment = inMeta && hbi.getMetaEntry().regionServer != null;<a name="line.2171"></a> |
| <span class="sourceLineNo">2172</span> boolean isDeployed = !hbi.getDeployedOn().isEmpty();<a name="line.2172"></a> |
| <span class="sourceLineNo">2173</span> boolean isMultiplyDeployed = hbi.getDeployedOn().size() > 1;<a name="line.2173"></a> |
| <span class="sourceLineNo">2174</span> boolean deploymentMatchesMeta = hasMetaAssignment && isDeployed && !isMultiplyDeployed &&<a name="line.2174"></a> |
| <span class="sourceLineNo">2175</span> hbi.getMetaEntry().regionServer.equals(hbi.getDeployedOn().get(0));<a name="line.2175"></a> |
| <span class="sourceLineNo">2176</span> boolean splitParent = inMeta && hbi.getMetaEntry().getRegionInfo().isSplit() &&<a name="line.2176"></a> |
| <span class="sourceLineNo">2177</span> hbi.getMetaEntry().getRegionInfo().isOffline();<a name="line.2177"></a> |
| <span class="sourceLineNo">2178</span> boolean shouldBeDeployed =<a name="line.2178"></a> |
| <span class="sourceLineNo">2179</span> inMeta && !isTableDisabled(hbi.getMetaEntry().getRegionInfo().getTable());<a name="line.2179"></a> |
| <span class="sourceLineNo">2180</span> boolean recentlyModified = inHdfs &&<a name="line.2180"></a> |
| <span class="sourceLineNo">2181</span> hbi.getModTime() + timelag > EnvironmentEdgeManager.currentTime();<a name="line.2181"></a> |
| <span class="sourceLineNo">2182</span><a name="line.2182"></a> |
| <span class="sourceLineNo">2183</span> // ========== First the healthy cases =============<a name="line.2183"></a> |
| <span class="sourceLineNo">2184</span> if (hbi.containsOnlyHdfsEdits()) {<a name="line.2184"></a> |
| <span class="sourceLineNo">2185</span> return;<a name="line.2185"></a> |
| <span class="sourceLineNo">2186</span> }<a name="line.2186"></a> |
| <span class="sourceLineNo">2187</span> if (inMeta && inHdfs && isDeployed && deploymentMatchesMeta && shouldBeDeployed) {<a name="line.2187"></a> |
| <span class="sourceLineNo">2188</span> return;<a name="line.2188"></a> |
| <span class="sourceLineNo">2189</span> } else if (inMeta && inHdfs && !shouldBeDeployed && !isDeployed) {<a name="line.2189"></a> |
| <span class="sourceLineNo">2190</span> LOG.info("Region " + descriptiveName + " is in META, and in a disabled " +<a name="line.2190"></a> |
| <span class="sourceLineNo">2191</span> "tabled that is not deployed");<a name="line.2191"></a> |
| <span class="sourceLineNo">2192</span> return;<a name="line.2192"></a> |
| <span class="sourceLineNo">2193</span> } else if (recentlyModified) {<a name="line.2193"></a> |
| <span class="sourceLineNo">2194</span> LOG.warn("Region " + descriptiveName + " was recently modified -- skipping");<a name="line.2194"></a> |
| <span class="sourceLineNo">2195</span> return;<a name="line.2195"></a> |
| <span class="sourceLineNo">2196</span> }<a name="line.2196"></a> |
| <span class="sourceLineNo">2197</span> // ========== Cases where the region is not in hbase:meta =============<a name="line.2197"></a> |
| <span class="sourceLineNo">2198</span> else if (!inMeta && !inHdfs && !isDeployed) {<a name="line.2198"></a> |
| <span class="sourceLineNo">2199</span> // We shouldn't have record of this region at all then!<a name="line.2199"></a> |
| <span class="sourceLineNo">2200</span> assert false : "Entry for region with no data";<a name="line.2200"></a> |
| <span class="sourceLineNo">2201</span> } else if (!inMeta && !inHdfs && isDeployed) {<a name="line.2201"></a> |
| <span class="sourceLineNo">2202</span> errors.reportError(ERROR_CODE.NOT_IN_META_HDFS, "Region "<a name="line.2202"></a> |
| <span class="sourceLineNo">2203</span> + descriptiveName + ", key=" + key + ", not on HDFS or in hbase:meta but " +<a name="line.2203"></a> |
| <span class="sourceLineNo">2204</span> "deployed on " + Joiner.on(", ").join(hbi.getDeployedOn()));<a name="line.2204"></a> |
| <span class="sourceLineNo">2205</span> if (shouldFixAssignments()) {<a name="line.2205"></a> |
| <span class="sourceLineNo">2206</span> undeployRegions(hbi);<a name="line.2206"></a> |
| <span class="sourceLineNo">2207</span> }<a name="line.2207"></a> |
| <span class="sourceLineNo">2208</span><a name="line.2208"></a> |
| <span class="sourceLineNo">2209</span> } else if (!inMeta && inHdfs && !isDeployed) {<a name="line.2209"></a> |
| <span class="sourceLineNo">2210</span> if (hbi.isMerged()) {<a name="line.2210"></a> |
| <span class="sourceLineNo">2211</span> // This region has already been merged, the remaining hdfs file will be<a name="line.2211"></a> |
| <span class="sourceLineNo">2212</span> // cleaned by CatalogJanitor later<a name="line.2212"></a> |
| <span class="sourceLineNo">2213</span> hbi.setSkipChecks(true);<a name="line.2213"></a> |
| <span class="sourceLineNo">2214</span> LOG.info("Region " + descriptiveName<a name="line.2214"></a> |
| <span class="sourceLineNo">2215</span> + " got merge recently, its file(s) will be cleaned by CatalogJanitor later");<a name="line.2215"></a> |
| <span class="sourceLineNo">2216</span> return;<a name="line.2216"></a> |
| <span class="sourceLineNo">2217</span> }<a name="line.2217"></a> |
| <span class="sourceLineNo">2218</span> errors.reportError(ERROR_CODE.NOT_IN_META_OR_DEPLOYED, "Region "<a name="line.2218"></a> |
| <span class="sourceLineNo">2219</span> + descriptiveName + " on HDFS, but not listed in hbase:meta " +<a name="line.2219"></a> |
| <span class="sourceLineNo">2220</span> "or deployed on any region server");<a name="line.2220"></a> |
| <span class="sourceLineNo">2221</span> // restore region consistency of an adopted orphan<a name="line.2221"></a> |
| <span class="sourceLineNo">2222</span> if (shouldFixMeta()) {<a name="line.2222"></a> |
| <span class="sourceLineNo">2223</span> if (!hbi.isHdfsRegioninfoPresent()) {<a name="line.2223"></a> |
| <span class="sourceLineNo">2224</span> LOG.error("Region " + hbi.getHdfsHRI() + " could have been repaired"<a name="line.2224"></a> |
| <span class="sourceLineNo">2225</span> + " in table integrity repair phase if -fixHdfsOrphans was" +<a name="line.2225"></a> |
| <span class="sourceLineNo">2226</span> " used.");<a name="line.2226"></a> |
| <span class="sourceLineNo">2227</span> return;<a name="line.2227"></a> |
| <span class="sourceLineNo">2228</span> }<a name="line.2228"></a> |
| <span class="sourceLineNo">2229</span><a name="line.2229"></a> |
| <span class="sourceLineNo">2230</span> RegionInfo hri = hbi.getHdfsHRI();<a name="line.2230"></a> |
| <span class="sourceLineNo">2231</span> HbckTableInfo tableInfo = tablesInfo.get(hri.getTable());<a name="line.2231"></a> |
| <span class="sourceLineNo">2232</span><a name="line.2232"></a> |
| <span class="sourceLineNo">2233</span> for (RegionInfo region : tableInfo.getRegionsFromMeta(this.regionInfoMap)) {<a name="line.2233"></a> |
| <span class="sourceLineNo">2234</span> if (Bytes.compareTo(region.getStartKey(), hri.getStartKey()) <= 0<a name="line.2234"></a> |
| <span class="sourceLineNo">2235</span> && (region.getEndKey().length == 0 || Bytes.compareTo(region.getEndKey(),<a name="line.2235"></a> |
| <span class="sourceLineNo">2236</span> hri.getEndKey()) >= 0)<a name="line.2236"></a> |
| <span class="sourceLineNo">2237</span> && Bytes.compareTo(region.getStartKey(), hri.getEndKey()) <= 0) {<a name="line.2237"></a> |
| <span class="sourceLineNo">2238</span> if(region.isSplit() || region.isOffline()) continue;<a name="line.2238"></a> |
| <span class="sourceLineNo">2239</span> Path regionDir = hbi.getHdfsRegionDir();<a name="line.2239"></a> |
| <span class="sourceLineNo">2240</span> FileSystem fs = regionDir.getFileSystem(getConf());<a name="line.2240"></a> |
| <span class="sourceLineNo">2241</span> List<Path> familyDirs = FSUtils.getFamilyDirs(fs, regionDir);<a name="line.2241"></a> |
| <span class="sourceLineNo">2242</span> for (Path familyDir : familyDirs) {<a name="line.2242"></a> |
| <span class="sourceLineNo">2243</span> List<Path> referenceFilePaths = FSUtils.getReferenceFilePaths(fs, familyDir);<a name="line.2243"></a> |
| <span class="sourceLineNo">2244</span> for (Path referenceFilePath : referenceFilePaths) {<a name="line.2244"></a> |
| <span class="sourceLineNo">2245</span> Path parentRegionDir =<a name="line.2245"></a> |
| <span class="sourceLineNo">2246</span> StoreFileInfo.getReferredToFile(referenceFilePath).getParent().getParent();<a name="line.2246"></a> |
| <span class="sourceLineNo">2247</span> if (parentRegionDir.toString().endsWith(region.getEncodedName())) {<a name="line.2247"></a> |
| <span class="sourceLineNo">2248</span> LOG.warn(hri + " start and stop keys are in the range of " + region<a name="line.2248"></a> |
| <span class="sourceLineNo">2249</span> + ". The region might not be cleaned up from hdfs when region " + region<a name="line.2249"></a> |
| <span class="sourceLineNo">2250</span> + " split failed. Hence deleting from hdfs.");<a name="line.2250"></a> |
| <span class="sourceLineNo">2251</span> HRegionFileSystem.deleteRegionFromFileSystem(getConf(), fs,<a name="line.2251"></a> |
| <span class="sourceLineNo">2252</span> regionDir.getParent(), hri);<a name="line.2252"></a> |
| <span class="sourceLineNo">2253</span> return;<a name="line.2253"></a> |
| <span class="sourceLineNo">2254</span> }<a name="line.2254"></a> |
| <span class="sourceLineNo">2255</span> }<a name="line.2255"></a> |
| <span class="sourceLineNo">2256</span> }<a name="line.2256"></a> |
| <span class="sourceLineNo">2257</span> }<a name="line.2257"></a> |
| <span class="sourceLineNo">2258</span> }<a name="line.2258"></a> |
| <span class="sourceLineNo">2259</span> LOG.info("Patching hbase:meta with .regioninfo: " + hbi.getHdfsHRI());<a name="line.2259"></a> |
| <span class="sourceLineNo">2260</span> int numReplicas = admin.getDescriptor(hbi.getTableName()).getRegionReplication();<a name="line.2260"></a> |
| <span class="sourceLineNo">2261</span> HBaseFsckRepair.fixMetaHoleOnlineAndAddReplicas(getConf(), hbi.getHdfsHRI(),<a name="line.2261"></a> |
| <span class="sourceLineNo">2262</span> admin.getClusterMetrics(EnumSet.of(Option.LIVE_SERVERS))<a name="line.2262"></a> |
| <span class="sourceLineNo">2263</span> .getLiveServerMetrics().keySet(), numReplicas);<a name="line.2263"></a> |
| <span class="sourceLineNo">2264</span><a name="line.2264"></a> |
| <span class="sourceLineNo">2265</span> tryAssignmentRepair(hbi, "Trying to reassign region...");<a name="line.2265"></a> |
| <span class="sourceLineNo">2266</span> }<a name="line.2266"></a> |
| <span class="sourceLineNo">2267</span><a name="line.2267"></a> |
| <span class="sourceLineNo">2268</span> } else if (!inMeta && inHdfs && isDeployed) {<a name="line.2268"></a> |
| <span class="sourceLineNo">2269</span> errors.reportError(ERROR_CODE.NOT_IN_META, "Region " + descriptiveName<a name="line.2269"></a> |
| <span class="sourceLineNo">2270</span> + " not in META, but deployed on " + Joiner.on(", ").join(hbi.getDeployedOn()));<a name="line.2270"></a> |
| <span class="sourceLineNo">2271</span> debugLsr(hbi.getHdfsRegionDir());<a name="line.2271"></a> |
| <span class="sourceLineNo">2272</span> if (hbi.getReplicaId() != RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.2272"></a> |
| <span class="sourceLineNo">2273</span> // for replicas, this means that we should undeploy the region (we would have<a name="line.2273"></a> |
| <span class="sourceLineNo">2274</span> // gone over the primaries and fixed meta holes in first phase under<a name="line.2274"></a> |
| <span class="sourceLineNo">2275</span> // checkAndFixConsistency; we shouldn't get the condition !inMeta at<a name="line.2275"></a> |
| <span class="sourceLineNo">2276</span> // this stage unless unwanted replica)<a name="line.2276"></a> |
| <span class="sourceLineNo">2277</span> if (shouldFixAssignments()) {<a name="line.2277"></a> |
| <span class="sourceLineNo">2278</span> undeployRegionsForHbi(hbi);<a name="line.2278"></a> |
| <span class="sourceLineNo">2279</span> }<a name="line.2279"></a> |
| <span class="sourceLineNo">2280</span> }<a name="line.2280"></a> |
| <span class="sourceLineNo">2281</span> if (shouldFixMeta() && hbi.getReplicaId() == RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.2281"></a> |
| <span class="sourceLineNo">2282</span> if (!hbi.isHdfsRegioninfoPresent()) {<a name="line.2282"></a> |
| <span class="sourceLineNo">2283</span> LOG.error("This should have been repaired in table integrity repair phase");<a name="line.2283"></a> |
| <span class="sourceLineNo">2284</span> return;<a name="line.2284"></a> |
| <span class="sourceLineNo">2285</span> }<a name="line.2285"></a> |
| <span class="sourceLineNo">2286</span><a name="line.2286"></a> |
| <span class="sourceLineNo">2287</span> LOG.info("Patching hbase:meta with with .regioninfo: " + hbi.getHdfsHRI());<a name="line.2287"></a> |
| <span class="sourceLineNo">2288</span> int numReplicas = admin.getDescriptor(hbi.getTableName()).getRegionReplication();<a name="line.2288"></a> |
| <span class="sourceLineNo">2289</span> HBaseFsckRepair.fixMetaHoleOnlineAndAddReplicas(getConf(), hbi.getHdfsHRI(),<a name="line.2289"></a> |
| <span class="sourceLineNo">2290</span> admin.getClusterMetrics(EnumSet.of(Option.LIVE_SERVERS))<a name="line.2290"></a> |
| <span class="sourceLineNo">2291</span> .getLiveServerMetrics().keySet(), numReplicas);<a name="line.2291"></a> |
| <span class="sourceLineNo">2292</span> tryAssignmentRepair(hbi, "Trying to fix unassigned region...");<a name="line.2292"></a> |
| <span class="sourceLineNo">2293</span> }<a name="line.2293"></a> |
| <span class="sourceLineNo">2294</span><a name="line.2294"></a> |
| <span class="sourceLineNo">2295</span> // ========== Cases where the region is in hbase:meta =============<a name="line.2295"></a> |
| <span class="sourceLineNo">2296</span> } else if (inMeta && inHdfs && !isDeployed && splitParent) {<a name="line.2296"></a> |
| <span class="sourceLineNo">2297</span> // check whether this is an actual error, or just transient state where parent<a name="line.2297"></a> |
| <span class="sourceLineNo">2298</span> // is not cleaned<a name="line.2298"></a> |
| <span class="sourceLineNo">2299</span> if (hbi.getMetaEntry().splitA != null && hbi.getMetaEntry().splitB != null) {<a name="line.2299"></a> |
| <span class="sourceLineNo">2300</span> // check that split daughters are there<a name="line.2300"></a> |
| <span class="sourceLineNo">2301</span> HbckRegionInfo infoA = this.regionInfoMap.get(hbi.getMetaEntry().splitA.getEncodedName());<a name="line.2301"></a> |
| <span class="sourceLineNo">2302</span> HbckRegionInfo infoB = this.regionInfoMap.get(hbi.getMetaEntry().splitB.getEncodedName());<a name="line.2302"></a> |
| <span class="sourceLineNo">2303</span> if (infoA != null && infoB != null) {<a name="line.2303"></a> |
| <span class="sourceLineNo">2304</span> // we already processed or will process daughters. Move on, nothing to see here.<a name="line.2304"></a> |
| <span class="sourceLineNo">2305</span> hbi.setSkipChecks(true);<a name="line.2305"></a> |
| <span class="sourceLineNo">2306</span> return;<a name="line.2306"></a> |
| <span class="sourceLineNo">2307</span> }<a name="line.2307"></a> |
| <span class="sourceLineNo">2308</span> }<a name="line.2308"></a> |
| <span class="sourceLineNo">2309</span><a name="line.2309"></a> |
| <span class="sourceLineNo">2310</span> // For Replica region, we need to do a similar check. If replica is not split successfully,<a name="line.2310"></a> |
| <span class="sourceLineNo">2311</span> // error is going to be reported against primary daughter region.<a name="line.2311"></a> |
| <span class="sourceLineNo">2312</span> if (hbi.getReplicaId() != RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.2312"></a> |
| <span class="sourceLineNo">2313</span> LOG.info("Region " + descriptiveName + " is a split parent in META, in HDFS, "<a name="line.2313"></a> |
| <span class="sourceLineNo">2314</span> + "and not deployed on any region server. This may be transient.");<a name="line.2314"></a> |
| <span class="sourceLineNo">2315</span> hbi.setSkipChecks(true);<a name="line.2315"></a> |
| <span class="sourceLineNo">2316</span> return;<a name="line.2316"></a> |
| <span class="sourceLineNo">2317</span> }<a name="line.2317"></a> |
| <span class="sourceLineNo">2318</span><a name="line.2318"></a> |
| <span class="sourceLineNo">2319</span> errors.reportError(ERROR_CODE.LINGERING_SPLIT_PARENT, "Region "<a name="line.2319"></a> |
| <span class="sourceLineNo">2320</span> + descriptiveName + " is a split parent in META, in HDFS, "<a name="line.2320"></a> |
| <span class="sourceLineNo">2321</span> + "and not deployed on any region server. This could be transient, "<a name="line.2321"></a> |
| <span class="sourceLineNo">2322</span> + "consider to run the catalog janitor first!");<a name="line.2322"></a> |
| <span class="sourceLineNo">2323</span> if (shouldFixSplitParents()) {<a name="line.2323"></a> |
| <span class="sourceLineNo">2324</span> setShouldRerun();<a name="line.2324"></a> |
| <span class="sourceLineNo">2325</span> resetSplitParent(hbi);<a name="line.2325"></a> |
| <span class="sourceLineNo">2326</span> }<a name="line.2326"></a> |
| <span class="sourceLineNo">2327</span> } else if (inMeta && !inHdfs && !isDeployed) {<a name="line.2327"></a> |
| <span class="sourceLineNo">2328</span> errors.reportError(ERROR_CODE.NOT_IN_HDFS_OR_DEPLOYED, "Region "<a name="line.2328"></a> |
| <span class="sourceLineNo">2329</span> + descriptiveName + " found in META, but not in HDFS "<a name="line.2329"></a> |
| <span class="sourceLineNo">2330</span> + "or deployed on any region server.");<a name="line.2330"></a> |
| <span class="sourceLineNo">2331</span> if (shouldFixMeta()) {<a name="line.2331"></a> |
| <span class="sourceLineNo">2332</span> deleteMetaRegion(hbi);<a name="line.2332"></a> |
| <span class="sourceLineNo">2333</span> }<a name="line.2333"></a> |
| <span class="sourceLineNo">2334</span> } else if (inMeta && !inHdfs && isDeployed) {<a name="line.2334"></a> |
| <span class="sourceLineNo">2335</span> errors.reportError(ERROR_CODE.NOT_IN_HDFS, "Region " + descriptiveName<a name="line.2335"></a> |
| <span class="sourceLineNo">2336</span> + " found in META, but not in HDFS, " +<a name="line.2336"></a> |
| <span class="sourceLineNo">2337</span> "and deployed on " + Joiner.on(", ").join(hbi.getDeployedOn()));<a name="line.2337"></a> |
| <span class="sourceLineNo">2338</span> // We treat HDFS as ground truth. Any information in meta is transient<a name="line.2338"></a> |
| <span class="sourceLineNo">2339</span> // and equivalent data can be regenerated. So, lets unassign and remove<a name="line.2339"></a> |
| <span class="sourceLineNo">2340</span> // these problems from META.<a name="line.2340"></a> |
| <span class="sourceLineNo">2341</span> if (shouldFixAssignments()) {<a name="line.2341"></a> |
| <span class="sourceLineNo">2342</span> errors.print("Trying to fix unassigned region...");<a name="line.2342"></a> |
| <span class="sourceLineNo">2343</span> undeployRegions(hbi);<a name="line.2343"></a> |
| <span class="sourceLineNo">2344</span> }<a name="line.2344"></a> |
| <span class="sourceLineNo">2345</span> if (shouldFixMeta()) {<a name="line.2345"></a> |
| <span class="sourceLineNo">2346</span> // wait for it to complete<a name="line.2346"></a> |
| <span class="sourceLineNo">2347</span> deleteMetaRegion(hbi);<a name="line.2347"></a> |
| <span class="sourceLineNo">2348</span> }<a name="line.2348"></a> |
| <span class="sourceLineNo">2349</span> } else if (inMeta && inHdfs && !isDeployed && shouldBeDeployed) {<a name="line.2349"></a> |
| <span class="sourceLineNo">2350</span> errors.reportError(ERROR_CODE.NOT_DEPLOYED, "Region " + descriptiveName<a name="line.2350"></a> |
| <span class="sourceLineNo">2351</span> + " not deployed on any region server.");<a name="line.2351"></a> |
| <span class="sourceLineNo">2352</span> tryAssignmentRepair(hbi, "Trying to fix unassigned region...");<a name="line.2352"></a> |
| <span class="sourceLineNo">2353</span> } else if (inMeta && inHdfs && isDeployed && !shouldBeDeployed) {<a name="line.2353"></a> |
| <span class="sourceLineNo">2354</span> errors.reportError(ERROR_CODE.SHOULD_NOT_BE_DEPLOYED,<a name="line.2354"></a> |
| <span class="sourceLineNo">2355</span> "Region " + descriptiveName + " should not be deployed according " +<a name="line.2355"></a> |
| <span class="sourceLineNo">2356</span> "to META, but is deployed on " + Joiner.on(", ").join(hbi.getDeployedOn()));<a name="line.2356"></a> |
| <span class="sourceLineNo">2357</span> if (shouldFixAssignments()) {<a name="line.2357"></a> |
| <span class="sourceLineNo">2358</span> errors.print("Trying to close the region " + descriptiveName);<a name="line.2358"></a> |
| <span class="sourceLineNo">2359</span> setShouldRerun();<a name="line.2359"></a> |
| <span class="sourceLineNo">2360</span> HBaseFsckRepair.fixMultiAssignment(connection, hbi.getMetaEntry().getRegionInfo(),<a name="line.2360"></a> |
| <span class="sourceLineNo">2361</span> hbi.getDeployedOn());<a name="line.2361"></a> |
| <span class="sourceLineNo">2362</span> }<a name="line.2362"></a> |
| <span class="sourceLineNo">2363</span> } else if (inMeta && inHdfs && isMultiplyDeployed) {<a name="line.2363"></a> |
| <span class="sourceLineNo">2364</span> errors.reportError(ERROR_CODE.MULTI_DEPLOYED,<a name="line.2364"></a> |
| <span class="sourceLineNo">2365</span> "Region " + descriptiveName + " is listed in hbase:meta on region server " +<a name="line.2365"></a> |
| <span class="sourceLineNo">2366</span> hbi.getMetaEntry().regionServer + " but is multiply assigned to region servers " +<a name="line.2366"></a> |
| <span class="sourceLineNo">2367</span> Joiner.on(", ").join(hbi.getDeployedOn()));<a name="line.2367"></a> |
| <span class="sourceLineNo">2368</span> // If we are trying to fix the errors<a name="line.2368"></a> |
| <span class="sourceLineNo">2369</span> if (shouldFixAssignments()) {<a name="line.2369"></a> |
| <span class="sourceLineNo">2370</span> errors.print("Trying to fix assignment error...");<a name="line.2370"></a> |
| <span class="sourceLineNo">2371</span> setShouldRerun();<a name="line.2371"></a> |
| <span class="sourceLineNo">2372</span> HBaseFsckRepair.fixMultiAssignment(connection, hbi.getMetaEntry().getRegionInfo(),<a name="line.2372"></a> |
| <span class="sourceLineNo">2373</span> hbi.getDeployedOn());<a name="line.2373"></a> |
| <span class="sourceLineNo">2374</span> }<a name="line.2374"></a> |
| <span class="sourceLineNo">2375</span> } else if (inMeta && inHdfs && isDeployed && !deploymentMatchesMeta) {<a name="line.2375"></a> |
| <span class="sourceLineNo">2376</span> errors.reportError(ERROR_CODE.SERVER_DOES_NOT_MATCH_META, "Region "<a name="line.2376"></a> |
| <span class="sourceLineNo">2377</span> + descriptiveName + " listed in hbase:meta on region server " +<a name="line.2377"></a> |
| <span class="sourceLineNo">2378</span> hbi.getMetaEntry().regionServer + " but found on region server " +<a name="line.2378"></a> |
| <span class="sourceLineNo">2379</span> hbi.getDeployedOn().get(0));<a name="line.2379"></a> |
| <span class="sourceLineNo">2380</span> // If we are trying to fix the errors<a name="line.2380"></a> |
| <span class="sourceLineNo">2381</span> if (shouldFixAssignments()) {<a name="line.2381"></a> |
| <span class="sourceLineNo">2382</span> errors.print("Trying to fix assignment error...");<a name="line.2382"></a> |
| <span class="sourceLineNo">2383</span> setShouldRerun();<a name="line.2383"></a> |
| <span class="sourceLineNo">2384</span> HBaseFsckRepair.fixMultiAssignment(connection, hbi.getMetaEntry().getRegionInfo(),<a name="line.2384"></a> |
| <span class="sourceLineNo">2385</span> hbi.getDeployedOn());<a name="line.2385"></a> |
| <span class="sourceLineNo">2386</span> HBaseFsckRepair.waitUntilAssigned(admin, hbi.getHdfsHRI());<a name="line.2386"></a> |
| <span class="sourceLineNo">2387</span> }<a name="line.2387"></a> |
| <span class="sourceLineNo">2388</span> } else {<a name="line.2388"></a> |
| <span class="sourceLineNo">2389</span> errors.reportError(ERROR_CODE.UNKNOWN, "Region " + descriptiveName +<a name="line.2389"></a> |
| <span class="sourceLineNo">2390</span> " is in an unforeseen state:" +<a name="line.2390"></a> |
| <span class="sourceLineNo">2391</span> " inMeta=" + inMeta +<a name="line.2391"></a> |
| <span class="sourceLineNo">2392</span> " inHdfs=" + inHdfs +<a name="line.2392"></a> |
| <span class="sourceLineNo">2393</span> " isDeployed=" + isDeployed +<a name="line.2393"></a> |
| <span class="sourceLineNo">2394</span> " isMultiplyDeployed=" + isMultiplyDeployed +<a name="line.2394"></a> |
| <span class="sourceLineNo">2395</span> " deploymentMatchesMeta=" + deploymentMatchesMeta +<a name="line.2395"></a> |
| <span class="sourceLineNo">2396</span> " shouldBeDeployed=" + shouldBeDeployed);<a name="line.2396"></a> |
| <span class="sourceLineNo">2397</span> }<a name="line.2397"></a> |
| <span class="sourceLineNo">2398</span> }<a name="line.2398"></a> |
| <span class="sourceLineNo">2399</span><a name="line.2399"></a> |
| <span class="sourceLineNo">2400</span> /**<a name="line.2400"></a> |
| <span class="sourceLineNo">2401</span> * Checks tables integrity. Goes over all regions and scans the tables.<a name="line.2401"></a> |
| <span class="sourceLineNo">2402</span> * Collects all the pieces for each table and checks if there are missing,<a name="line.2402"></a> |
| <span class="sourceLineNo">2403</span> * repeated or overlapping ones.<a name="line.2403"></a> |
| <span class="sourceLineNo">2404</span> * @throws IOException<a name="line.2404"></a> |
| <span class="sourceLineNo">2405</span> */<a name="line.2405"></a> |
| <span class="sourceLineNo">2406</span> SortedMap<TableName, HbckTableInfo> checkIntegrity() throws IOException {<a name="line.2406"></a> |
| <span class="sourceLineNo">2407</span> tablesInfo = new TreeMap<>();<a name="line.2407"></a> |
| <span class="sourceLineNo">2408</span> LOG.debug("There are " + regionInfoMap.size() + " region info entries");<a name="line.2408"></a> |
| <span class="sourceLineNo">2409</span> for (HbckRegionInfo hbi : regionInfoMap.values()) {<a name="line.2409"></a> |
| <span class="sourceLineNo">2410</span> // Check only valid, working regions<a name="line.2410"></a> |
| <span class="sourceLineNo">2411</span> if (hbi.getMetaEntry() == null) {<a name="line.2411"></a> |
| <span class="sourceLineNo">2412</span> // this assumes that consistency check has run loadMetaEntry<a name="line.2412"></a> |
| <span class="sourceLineNo">2413</span> Path p = hbi.getHdfsRegionDir();<a name="line.2413"></a> |
| <span class="sourceLineNo">2414</span> if (p == null) {<a name="line.2414"></a> |
| <span class="sourceLineNo">2415</span> errors.report("No regioninfo in Meta or HDFS. " + hbi);<a name="line.2415"></a> |
| <span class="sourceLineNo">2416</span> }<a name="line.2416"></a> |
| <span class="sourceLineNo">2417</span><a name="line.2417"></a> |
| <span class="sourceLineNo">2418</span> // TODO test.<a name="line.2418"></a> |
| <span class="sourceLineNo">2419</span> continue;<a name="line.2419"></a> |
| <span class="sourceLineNo">2420</span> }<a name="line.2420"></a> |
| <span class="sourceLineNo">2421</span> if (hbi.getMetaEntry().regionServer == null) {<a name="line.2421"></a> |
| <span class="sourceLineNo">2422</span> errors.detail("Skipping region because no region server: " + hbi);<a name="line.2422"></a> |
| <span class="sourceLineNo">2423</span> continue;<a name="line.2423"></a> |
| <span class="sourceLineNo">2424</span> }<a name="line.2424"></a> |
| <span class="sourceLineNo">2425</span> if (hbi.getMetaEntry().getRegionInfo().isOffline()) {<a name="line.2425"></a> |
| <span class="sourceLineNo">2426</span> errors.detail("Skipping region because it is offline: " + hbi);<a name="line.2426"></a> |
| <span class="sourceLineNo">2427</span> continue;<a name="line.2427"></a> |
| <span class="sourceLineNo">2428</span> }<a name="line.2428"></a> |
| <span class="sourceLineNo">2429</span> if (hbi.containsOnlyHdfsEdits()) {<a name="line.2429"></a> |
| <span class="sourceLineNo">2430</span> errors.detail("Skipping region because it only contains edits" + hbi);<a name="line.2430"></a> |
| <span class="sourceLineNo">2431</span> continue;<a name="line.2431"></a> |
| <span class="sourceLineNo">2432</span> }<a name="line.2432"></a> |
| <span class="sourceLineNo">2433</span><a name="line.2433"></a> |
| <span class="sourceLineNo">2434</span> // Missing regionDir or over-deployment is checked elsewhere. Include<a name="line.2434"></a> |
| <span class="sourceLineNo">2435</span> // these cases in modTInfo, so we can evaluate those regions as part of<a name="line.2435"></a> |
| <span class="sourceLineNo">2436</span> // the region chain in META<a name="line.2436"></a> |
| <span class="sourceLineNo">2437</span> //if (hbi.foundRegionDir == null) continue;<a name="line.2437"></a> |
| <span class="sourceLineNo">2438</span> //if (hbi.deployedOn.size() != 1) continue;<a name="line.2438"></a> |
| <span class="sourceLineNo">2439</span> if (hbi.getDeployedOn().isEmpty()) {<a name="line.2439"></a> |
| <span class="sourceLineNo">2440</span> continue;<a name="line.2440"></a> |
| <span class="sourceLineNo">2441</span> }<a name="line.2441"></a> |
| <span class="sourceLineNo">2442</span><a name="line.2442"></a> |
| <span class="sourceLineNo">2443</span> // We should be safe here<a name="line.2443"></a> |
| <span class="sourceLineNo">2444</span> TableName tableName = hbi.getMetaEntry().getRegionInfo().getTable();<a name="line.2444"></a> |
| <span class="sourceLineNo">2445</span> HbckTableInfo modTInfo = tablesInfo.get(tableName);<a name="line.2445"></a> |
| <span class="sourceLineNo">2446</span> if (modTInfo == null) {<a name="line.2446"></a> |
| <span class="sourceLineNo">2447</span> modTInfo = new HbckTableInfo(tableName, this);<a name="line.2447"></a> |
| <span class="sourceLineNo">2448</span> }<a name="line.2448"></a> |
| <span class="sourceLineNo">2449</span> for (ServerName server : hbi.getDeployedOn()) {<a name="line.2449"></a> |
| <span class="sourceLineNo">2450</span> modTInfo.addServer(server);<a name="line.2450"></a> |
| <span class="sourceLineNo">2451</span> }<a name="line.2451"></a> |
| <span class="sourceLineNo">2452</span><a name="line.2452"></a> |
| <span class="sourceLineNo">2453</span> if (!hbi.isSkipChecks()) {<a name="line.2453"></a> |
| <span class="sourceLineNo">2454</span> modTInfo.addRegionInfo(hbi);<a name="line.2454"></a> |
| <span class="sourceLineNo">2455</span> }<a name="line.2455"></a> |
| <span class="sourceLineNo">2456</span><a name="line.2456"></a> |
| <span class="sourceLineNo">2457</span> tablesInfo.put(tableName, modTInfo);<a name="line.2457"></a> |
| <span class="sourceLineNo">2458</span> }<a name="line.2458"></a> |
| <span class="sourceLineNo">2459</span><a name="line.2459"></a> |
| <span class="sourceLineNo">2460</span> loadTableInfosForTablesWithNoRegion();<a name="line.2460"></a> |
| <span class="sourceLineNo">2461</span><a name="line.2461"></a> |
| <span class="sourceLineNo">2462</span> logParallelMerge();<a name="line.2462"></a> |
| <span class="sourceLineNo">2463</span> for (HbckTableInfo tInfo : tablesInfo.values()) {<a name="line.2463"></a> |
| <span class="sourceLineNo">2464</span> TableIntegrityErrorHandler handler = tInfo.new IntegrityFixSuggester(tInfo, errors);<a name="line.2464"></a> |
| <span class="sourceLineNo">2465</span> if (!tInfo.checkRegionChain(handler)) {<a name="line.2465"></a> |
| <span class="sourceLineNo">2466</span> errors.report("Found inconsistency in table " + tInfo.getName());<a name="line.2466"></a> |
| <span class="sourceLineNo">2467</span> }<a name="line.2467"></a> |
| <span class="sourceLineNo">2468</span> }<a name="line.2468"></a> |
| <span class="sourceLineNo">2469</span> return tablesInfo;<a name="line.2469"></a> |
| <span class="sourceLineNo">2470</span> }<a name="line.2470"></a> |
| <span class="sourceLineNo">2471</span><a name="line.2471"></a> |
| <span class="sourceLineNo">2472</span> /** Loads table info's for tables that may not have been included, since there are no<a name="line.2472"></a> |
| <span class="sourceLineNo">2473</span> * regions reported for the table, but table dir is there in hdfs<a name="line.2473"></a> |
| <span class="sourceLineNo">2474</span> */<a name="line.2474"></a> |
| <span class="sourceLineNo">2475</span> private void loadTableInfosForTablesWithNoRegion() throws IOException {<a name="line.2475"></a> |
| <span class="sourceLineNo">2476</span> Map<String, TableDescriptor> allTables = new FSTableDescriptors(getConf()).getAll();<a name="line.2476"></a> |
| <span class="sourceLineNo">2477</span> for (TableDescriptor htd : allTables.values()) {<a name="line.2477"></a> |
| <span class="sourceLineNo">2478</span> if (checkMetaOnly && !htd.isMetaTable()) {<a name="line.2478"></a> |
| <span class="sourceLineNo">2479</span> continue;<a name="line.2479"></a> |
| <span class="sourceLineNo">2480</span> }<a name="line.2480"></a> |
| <span class="sourceLineNo">2481</span><a name="line.2481"></a> |
| <span class="sourceLineNo">2482</span> TableName tableName = htd.getTableName();<a name="line.2482"></a> |
| <span class="sourceLineNo">2483</span> if (isTableIncluded(tableName) && !tablesInfo.containsKey(tableName)) {<a name="line.2483"></a> |
| <span class="sourceLineNo">2484</span> HbckTableInfo tableInfo = new HbckTableInfo(tableName, this);<a name="line.2484"></a> |
| <span class="sourceLineNo">2485</span> tableInfo.htds.add(htd);<a name="line.2485"></a> |
| <span class="sourceLineNo">2486</span> tablesInfo.put(htd.getTableName(), tableInfo);<a name="line.2486"></a> |
| <span class="sourceLineNo">2487</span> }<a name="line.2487"></a> |
| <span class="sourceLineNo">2488</span> }<a name="line.2488"></a> |
| <span class="sourceLineNo">2489</span> }<a name="line.2489"></a> |
| <span class="sourceLineNo">2490</span><a name="line.2490"></a> |
| <span class="sourceLineNo">2491</span> /**<a name="line.2491"></a> |
| <span class="sourceLineNo">2492</span> * Merge hdfs data by moving from contained HbckRegionInfo into targetRegionDir.<a name="line.2492"></a> |
| <span class="sourceLineNo">2493</span> * @return number of file move fixes done to merge regions.<a name="line.2493"></a> |
| <span class="sourceLineNo">2494</span> */<a name="line.2494"></a> |
| <span class="sourceLineNo">2495</span> public int mergeRegionDirs(Path targetRegionDir, HbckRegionInfo contained) throws IOException {<a name="line.2495"></a> |
| <span class="sourceLineNo">2496</span> int fileMoves = 0;<a name="line.2496"></a> |
| <span class="sourceLineNo">2497</span> String thread = Thread.currentThread().getName();<a name="line.2497"></a> |
| <span class="sourceLineNo">2498</span> LOG.debug("[" + thread + "] Contained region dir after close and pause");<a name="line.2498"></a> |
| <span class="sourceLineNo">2499</span> debugLsr(contained.getHdfsRegionDir());<a name="line.2499"></a> |
| <span class="sourceLineNo">2500</span><a name="line.2500"></a> |
| <span class="sourceLineNo">2501</span> // rename the contained into the container.<a name="line.2501"></a> |
| <span class="sourceLineNo">2502</span> FileSystem fs = targetRegionDir.getFileSystem(getConf());<a name="line.2502"></a> |
| <span class="sourceLineNo">2503</span> FileStatus[] dirs = null;<a name="line.2503"></a> |
| <span class="sourceLineNo">2504</span> try {<a name="line.2504"></a> |
| <span class="sourceLineNo">2505</span> dirs = fs.listStatus(contained.getHdfsRegionDir());<a name="line.2505"></a> |
| <span class="sourceLineNo">2506</span> } catch (FileNotFoundException fnfe) {<a name="line.2506"></a> |
| <span class="sourceLineNo">2507</span> // region we are attempting to merge in is not present! Since this is a merge, there is<a name="line.2507"></a> |
| <span class="sourceLineNo">2508</span> // no harm skipping this region if it does not exist.<a name="line.2508"></a> |
| <span class="sourceLineNo">2509</span> if (!fs.exists(contained.getHdfsRegionDir())) {<a name="line.2509"></a> |
| <span class="sourceLineNo">2510</span> LOG.warn("[" + thread + "] HDFS region dir " + contained.getHdfsRegionDir()<a name="line.2510"></a> |
| <span class="sourceLineNo">2511</span> + " is missing. Assuming already sidelined or moved.");<a name="line.2511"></a> |
| <span class="sourceLineNo">2512</span> } else {<a name="line.2512"></a> |
| <span class="sourceLineNo">2513</span> sidelineRegionDir(fs, contained);<a name="line.2513"></a> |
| <span class="sourceLineNo">2514</span> }<a name="line.2514"></a> |
| <span class="sourceLineNo">2515</span> return fileMoves;<a name="line.2515"></a> |
| <span class="sourceLineNo">2516</span> }<a name="line.2516"></a> |
| <span class="sourceLineNo">2517</span><a name="line.2517"></a> |
| <span class="sourceLineNo">2518</span> if (dirs == null) {<a name="line.2518"></a> |
| <span class="sourceLineNo">2519</span> if (!fs.exists(contained.getHdfsRegionDir())) {<a name="line.2519"></a> |
| <span class="sourceLineNo">2520</span> LOG.warn("[" + thread + "] HDFS region dir " + contained.getHdfsRegionDir()<a name="line.2520"></a> |
| <span class="sourceLineNo">2521</span> + " already sidelined.");<a name="line.2521"></a> |
| <span class="sourceLineNo">2522</span> } else {<a name="line.2522"></a> |
| <span class="sourceLineNo">2523</span> sidelineRegionDir(fs, contained);<a name="line.2523"></a> |
| <span class="sourceLineNo">2524</span> }<a name="line.2524"></a> |
| <span class="sourceLineNo">2525</span> return fileMoves;<a name="line.2525"></a> |
| <span class="sourceLineNo">2526</span> }<a name="line.2526"></a> |
| <span class="sourceLineNo">2527</span><a name="line.2527"></a> |
| <span class="sourceLineNo">2528</span> for (FileStatus cf : dirs) {<a name="line.2528"></a> |
| <span class="sourceLineNo">2529</span> Path src = cf.getPath();<a name="line.2529"></a> |
| <span class="sourceLineNo">2530</span> Path dst = new Path(targetRegionDir, src.getName());<a name="line.2530"></a> |
| <span class="sourceLineNo">2531</span><a name="line.2531"></a> |
| <span class="sourceLineNo">2532</span> if (src.getName().equals(HRegionFileSystem.REGION_INFO_FILE)) {<a name="line.2532"></a> |
| <span class="sourceLineNo">2533</span> // do not copy the old .regioninfo file.<a name="line.2533"></a> |
| <span class="sourceLineNo">2534</span> continue;<a name="line.2534"></a> |
| <span class="sourceLineNo">2535</span> }<a name="line.2535"></a> |
| <span class="sourceLineNo">2536</span><a name="line.2536"></a> |
| <span class="sourceLineNo">2537</span> if (src.getName().equals(HConstants.HREGION_OLDLOGDIR_NAME)) {<a name="line.2537"></a> |
| <span class="sourceLineNo">2538</span> // do not copy the .oldlogs files<a name="line.2538"></a> |
| <span class="sourceLineNo">2539</span> continue;<a name="line.2539"></a> |
| <span class="sourceLineNo">2540</span> }<a name="line.2540"></a> |
| <span class="sourceLineNo">2541</span><a name="line.2541"></a> |
| <span class="sourceLineNo">2542</span> LOG.info("[" + thread + "] Moving files from " + src + " into containing region " + dst);<a name="line.2542"></a> |
| <span class="sourceLineNo">2543</span> // FileSystem.rename is inconsistent with directories -- if the<a name="line.2543"></a> |
| <span class="sourceLineNo">2544</span> // dst (foo/a) exists and is a dir, and the src (foo/b) is a dir,<a name="line.2544"></a> |
| <span class="sourceLineNo">2545</span> // it moves the src into the dst dir resulting in (foo/a/b). If<a name="line.2545"></a> |
| <span class="sourceLineNo">2546</span> // the dst does not exist, and the src a dir, src becomes dst. (foo/b)<a name="line.2546"></a> |
| <span class="sourceLineNo">2547</span> for (FileStatus hfile : fs.listStatus(src)) {<a name="line.2547"></a> |
| <span class="sourceLineNo">2548</span> boolean success = fs.rename(hfile.getPath(), dst);<a name="line.2548"></a> |
| <span class="sourceLineNo">2549</span> if (success) {<a name="line.2549"></a> |
| <span class="sourceLineNo">2550</span> fileMoves++;<a name="line.2550"></a> |
| <span class="sourceLineNo">2551</span> }<a name="line.2551"></a> |
| <span class="sourceLineNo">2552</span> }<a name="line.2552"></a> |
| <span class="sourceLineNo">2553</span> LOG.debug("[" + thread + "] Sideline directory contents:");<a name="line.2553"></a> |
| <span class="sourceLineNo">2554</span> debugLsr(targetRegionDir);<a name="line.2554"></a> |
| <span class="sourceLineNo">2555</span> }<a name="line.2555"></a> |
| <span class="sourceLineNo">2556</span><a name="line.2556"></a> |
| <span class="sourceLineNo">2557</span> // if all success.<a name="line.2557"></a> |
| <span class="sourceLineNo">2558</span> sidelineRegionDir(fs, contained);<a name="line.2558"></a> |
| <span class="sourceLineNo">2559</span> LOG.info("[" + thread + "] Sidelined region dir "+ contained.getHdfsRegionDir() + " into " +<a name="line.2559"></a> |
| <span class="sourceLineNo">2560</span> getSidelineDir());<a name="line.2560"></a> |
| <span class="sourceLineNo">2561</span> debugLsr(contained.getHdfsRegionDir());<a name="line.2561"></a> |
| <span class="sourceLineNo">2562</span><a name="line.2562"></a> |
| <span class="sourceLineNo">2563</span> return fileMoves;<a name="line.2563"></a> |
| <span class="sourceLineNo">2564</span> }<a name="line.2564"></a> |
| <span class="sourceLineNo">2565</span><a name="line.2565"></a> |
| <span class="sourceLineNo">2566</span><a name="line.2566"></a> |
| <span class="sourceLineNo">2567</span> static class WorkItemOverlapMerge implements Callable<Void> {<a name="line.2567"></a> |
| <span class="sourceLineNo">2568</span> private TableIntegrityErrorHandler handler;<a name="line.2568"></a> |
| <span class="sourceLineNo">2569</span> Collection<HbckRegionInfo> overlapgroup;<a name="line.2569"></a> |
| <span class="sourceLineNo">2570</span><a name="line.2570"></a> |
| <span class="sourceLineNo">2571</span> WorkItemOverlapMerge(Collection<HbckRegionInfo> overlapgroup,<a name="line.2571"></a> |
| <span class="sourceLineNo">2572</span> TableIntegrityErrorHandler handler) {<a name="line.2572"></a> |
| <span class="sourceLineNo">2573</span> this.handler = handler;<a name="line.2573"></a> |
| <span class="sourceLineNo">2574</span> this.overlapgroup = overlapgroup;<a name="line.2574"></a> |
| <span class="sourceLineNo">2575</span> }<a name="line.2575"></a> |
| <span class="sourceLineNo">2576</span><a name="line.2576"></a> |
| <span class="sourceLineNo">2577</span> @Override<a name="line.2577"></a> |
| <span class="sourceLineNo">2578</span> public Void call() throws Exception {<a name="line.2578"></a> |
| <span class="sourceLineNo">2579</span> handler.handleOverlapGroup(overlapgroup);<a name="line.2579"></a> |
| <span class="sourceLineNo">2580</span> return null;<a name="line.2580"></a> |
| <span class="sourceLineNo">2581</span> }<a name="line.2581"></a> |
| <span class="sourceLineNo">2582</span> }<a name="line.2582"></a> |
| <span class="sourceLineNo">2583</span><a name="line.2583"></a> |
| <span class="sourceLineNo">2584</span> /**<a name="line.2584"></a> |
| <span class="sourceLineNo">2585</span> * Return a list of user-space table names whose metadata have not been<a name="line.2585"></a> |
| <span class="sourceLineNo">2586</span> * modified in the last few milliseconds specified by timelag<a name="line.2586"></a> |
| <span class="sourceLineNo">2587</span> * if any of the REGIONINFO_QUALIFIER, SERVER_QUALIFIER, STARTCODE_QUALIFIER,<a name="line.2587"></a> |
| <span class="sourceLineNo">2588</span> * SPLITA_QUALIFIER, SPLITB_QUALIFIER have not changed in the last<a name="line.2588"></a> |
| <span class="sourceLineNo">2589</span> * milliseconds specified by timelag, then the table is a candidate to be returned.<a name="line.2589"></a> |
| <span class="sourceLineNo">2590</span> * @return tables that have not been modified recently<a name="line.2590"></a> |
| <span class="sourceLineNo">2591</span> * @throws IOException if an error is encountered<a name="line.2591"></a> |
| <span class="sourceLineNo">2592</span> */<a name="line.2592"></a> |
| <span class="sourceLineNo">2593</span> TableDescriptor[] getTables(AtomicInteger numSkipped) {<a name="line.2593"></a> |
| <span class="sourceLineNo">2594</span> List<TableName> tableNames = new ArrayList<>();<a name="line.2594"></a> |
| <span class="sourceLineNo">2595</span> long now = EnvironmentEdgeManager.currentTime();<a name="line.2595"></a> |
| <span class="sourceLineNo">2596</span><a name="line.2596"></a> |
| <span class="sourceLineNo">2597</span> for (HbckRegionInfo hbi : regionInfoMap.values()) {<a name="line.2597"></a> |
| <span class="sourceLineNo">2598</span> HbckRegionInfo.MetaEntry info = hbi.getMetaEntry();<a name="line.2598"></a> |
| <span class="sourceLineNo">2599</span><a name="line.2599"></a> |
| <span class="sourceLineNo">2600</span> // if the start key is zero, then we have found the first region of a table.<a name="line.2600"></a> |
| <span class="sourceLineNo">2601</span> // pick only those tables that were not modified in the last few milliseconds.<a name="line.2601"></a> |
| <span class="sourceLineNo">2602</span> if (info != null && info.getRegionInfo().getStartKey().length == 0 &&<a name="line.2602"></a> |
| <span class="sourceLineNo">2603</span> !info.getRegionInfo().isMetaRegion()) {<a name="line.2603"></a> |
| <span class="sourceLineNo">2604</span> if (info.modTime + timelag < now) {<a name="line.2604"></a> |
| <span class="sourceLineNo">2605</span> tableNames.add(info.getRegionInfo().getTable());<a name="line.2605"></a> |
| <span class="sourceLineNo">2606</span> } else {<a name="line.2606"></a> |
| <span class="sourceLineNo">2607</span> numSkipped.incrementAndGet(); // one more in-flux table<a name="line.2607"></a> |
| <span class="sourceLineNo">2608</span> }<a name="line.2608"></a> |
| <span class="sourceLineNo">2609</span> }<a name="line.2609"></a> |
| <span class="sourceLineNo">2610</span> }<a name="line.2610"></a> |
| <span class="sourceLineNo">2611</span> return getTableDescriptors(tableNames);<a name="line.2611"></a> |
| <span class="sourceLineNo">2612</span> }<a name="line.2612"></a> |
| <span class="sourceLineNo">2613</span><a name="line.2613"></a> |
| <span class="sourceLineNo">2614</span> TableDescriptor[] getTableDescriptors(List<TableName> tableNames) {<a name="line.2614"></a> |
| <span class="sourceLineNo">2615</span> LOG.info("getTableDescriptors == tableNames => " + tableNames);<a name="line.2615"></a> |
| <span class="sourceLineNo">2616</span> try (Connection conn = ConnectionFactory.createConnection(getConf());<a name="line.2616"></a> |
| <span class="sourceLineNo">2617</span> Admin admin = conn.getAdmin()) {<a name="line.2617"></a> |
| <span class="sourceLineNo">2618</span> List<TableDescriptor> tds = admin.listTableDescriptors(tableNames);<a name="line.2618"></a> |
| <span class="sourceLineNo">2619</span> return tds.toArray(new TableDescriptor[tds.size()]);<a name="line.2619"></a> |
| <span class="sourceLineNo">2620</span> } catch (IOException e) {<a name="line.2620"></a> |
| <span class="sourceLineNo">2621</span> LOG.debug("Exception getting table descriptors", e);<a name="line.2621"></a> |
| <span class="sourceLineNo">2622</span> }<a name="line.2622"></a> |
| <span class="sourceLineNo">2623</span> return new TableDescriptor[0];<a name="line.2623"></a> |
| <span class="sourceLineNo">2624</span> }<a name="line.2624"></a> |
| <span class="sourceLineNo">2625</span><a name="line.2625"></a> |
| <span class="sourceLineNo">2626</span> /**<a name="line.2626"></a> |
| <span class="sourceLineNo">2627</span> * Gets the entry in regionInfo corresponding to the the given encoded<a name="line.2627"></a> |
| <span class="sourceLineNo">2628</span> * region name. If the region has not been seen yet, a new entry is added<a name="line.2628"></a> |
| <span class="sourceLineNo">2629</span> * and returned.<a name="line.2629"></a> |
| <span class="sourceLineNo">2630</span> */<a name="line.2630"></a> |
| <span class="sourceLineNo">2631</span> private synchronized HbckRegionInfo getOrCreateInfo(String name) {<a name="line.2631"></a> |
| <span class="sourceLineNo">2632</span> HbckRegionInfo hbi = regionInfoMap.get(name);<a name="line.2632"></a> |
| <span class="sourceLineNo">2633</span> if (hbi == null) {<a name="line.2633"></a> |
| <span class="sourceLineNo">2634</span> hbi = new HbckRegionInfo(null);<a name="line.2634"></a> |
| <span class="sourceLineNo">2635</span> regionInfoMap.put(name, hbi);<a name="line.2635"></a> |
| <span class="sourceLineNo">2636</span> }<a name="line.2636"></a> |
| <span class="sourceLineNo">2637</span> return hbi;<a name="line.2637"></a> |
| <span class="sourceLineNo">2638</span> }<a name="line.2638"></a> |
| <span class="sourceLineNo">2639</span><a name="line.2639"></a> |
| <span class="sourceLineNo">2640</span> private void checkAndFixReplication() throws ReplicationException {<a name="line.2640"></a> |
| <span class="sourceLineNo">2641</span> ReplicationChecker checker = new ReplicationChecker(getConf(), zkw, errors);<a name="line.2641"></a> |
| <span class="sourceLineNo">2642</span> checker.checkUnDeletedQueues();<a name="line.2642"></a> |
| <span class="sourceLineNo">2643</span><a name="line.2643"></a> |
| <span class="sourceLineNo">2644</span> if (checker.hasUnDeletedQueues() && this.fixReplication) {<a name="line.2644"></a> |
| <span class="sourceLineNo">2645</span> checker.fixUnDeletedQueues();<a name="line.2645"></a> |
| <span class="sourceLineNo">2646</span> setShouldRerun();<a name="line.2646"></a> |
| <span class="sourceLineNo">2647</span> }<a name="line.2647"></a> |
| <span class="sourceLineNo">2648</span> }<a name="line.2648"></a> |
| <span class="sourceLineNo">2649</span><a name="line.2649"></a> |
| <span class="sourceLineNo">2650</span> /**<a name="line.2650"></a> |
| <span class="sourceLineNo">2651</span> * Check values in regionInfo for hbase:meta<a name="line.2651"></a> |
| <span class="sourceLineNo">2652</span> * Check if zero or more than one regions with hbase:meta are found.<a name="line.2652"></a> |
| <span class="sourceLineNo">2653</span> * If there are inconsistencies (i.e. zero or more than one regions<a name="line.2653"></a> |
| <span class="sourceLineNo">2654</span> * pretend to be holding the hbase:meta) try to fix that and report an error.<a name="line.2654"></a> |
| <span class="sourceLineNo">2655</span> * @throws IOException from HBaseFsckRepair functions<a name="line.2655"></a> |
| <span class="sourceLineNo">2656</span> * @throws KeeperException<a name="line.2656"></a> |
| <span class="sourceLineNo">2657</span> * @throws InterruptedException<a name="line.2657"></a> |
| <span class="sourceLineNo">2658</span> */<a name="line.2658"></a> |
| <span class="sourceLineNo">2659</span> boolean checkMetaRegion() throws IOException, KeeperException, InterruptedException {<a name="line.2659"></a> |
| <span class="sourceLineNo">2660</span> Map<Integer, HbckRegionInfo> metaRegions = new HashMap<>();<a name="line.2660"></a> |
| <span class="sourceLineNo">2661</span> for (HbckRegionInfo value : regionInfoMap.values()) {<a name="line.2661"></a> |
| <span class="sourceLineNo">2662</span> if (value.getMetaEntry() != null && value.getMetaEntry().getRegionInfo().isMetaRegion()) {<a name="line.2662"></a> |
| <span class="sourceLineNo">2663</span> metaRegions.put(value.getReplicaId(), value);<a name="line.2663"></a> |
| <span class="sourceLineNo">2664</span> }<a name="line.2664"></a> |
| <span class="sourceLineNo">2665</span> }<a name="line.2665"></a> |
| <span class="sourceLineNo">2666</span> int metaReplication = admin.getDescriptor(TableName.META_TABLE_NAME)<a name="line.2666"></a> |
| <span class="sourceLineNo">2667</span> .getRegionReplication();<a name="line.2667"></a> |
| <span class="sourceLineNo">2668</span> boolean noProblem = true;<a name="line.2668"></a> |
| <span class="sourceLineNo">2669</span> // There will be always entries in regionInfoMap corresponding to hbase:meta & its replicas<a name="line.2669"></a> |
| <span class="sourceLineNo">2670</span> // Check the deployed servers. It should be exactly one server for each replica.<a name="line.2670"></a> |
| <span class="sourceLineNo">2671</span> for (int i = 0; i < metaReplication; i++) {<a name="line.2671"></a> |
| <span class="sourceLineNo">2672</span> HbckRegionInfo metaHbckRegionInfo = metaRegions.remove(i);<a name="line.2672"></a> |
| <span class="sourceLineNo">2673</span> List<ServerName> servers = new ArrayList<>();<a name="line.2673"></a> |
| <span class="sourceLineNo">2674</span> if (metaHbckRegionInfo != null) {<a name="line.2674"></a> |
| <span class="sourceLineNo">2675</span> servers = metaHbckRegionInfo.getDeployedOn();<a name="line.2675"></a> |
| <span class="sourceLineNo">2676</span> }<a name="line.2676"></a> |
| <span class="sourceLineNo">2677</span> if (servers.size() != 1) {<a name="line.2677"></a> |
| <span class="sourceLineNo">2678</span> noProblem = false;<a name="line.2678"></a> |
| <span class="sourceLineNo">2679</span> if (servers.isEmpty()) {<a name="line.2679"></a> |
| <span class="sourceLineNo">2680</span> assignMetaReplica(i);<a name="line.2680"></a> |
| <span class="sourceLineNo">2681</span> } else if (servers.size() > 1) {<a name="line.2681"></a> |
| <span class="sourceLineNo">2682</span> errors<a name="line.2682"></a> |
| <span class="sourceLineNo">2683</span> .reportError(ERROR_CODE.MULTI_META_REGION, "hbase:meta, replicaId " +<a name="line.2683"></a> |
| <span class="sourceLineNo">2684</span> metaHbckRegionInfo.getReplicaId() + " is found on more than one region.");<a name="line.2684"></a> |
| <span class="sourceLineNo">2685</span> if (shouldFixAssignments()) {<a name="line.2685"></a> |
| <span class="sourceLineNo">2686</span> errors.print("Trying to fix a problem with hbase:meta, replicaId " +<a name="line.2686"></a> |
| <span class="sourceLineNo">2687</span> metaHbckRegionInfo.getReplicaId() + "..");<a name="line.2687"></a> |
| <span class="sourceLineNo">2688</span> setShouldRerun();<a name="line.2688"></a> |
| <span class="sourceLineNo">2689</span> // try fix it (treat is a dupe assignment)<a name="line.2689"></a> |
| <span class="sourceLineNo">2690</span> HBaseFsckRepair.fixMultiAssignment(connection,<a name="line.2690"></a> |
| <span class="sourceLineNo">2691</span> metaHbckRegionInfo.getMetaEntry().getRegionInfo(), servers);<a name="line.2691"></a> |
| <span class="sourceLineNo">2692</span> }<a name="line.2692"></a> |
| <span class="sourceLineNo">2693</span> }<a name="line.2693"></a> |
| <span class="sourceLineNo">2694</span> }<a name="line.2694"></a> |
| <span class="sourceLineNo">2695</span> }<a name="line.2695"></a> |
| <span class="sourceLineNo">2696</span> // unassign whatever is remaining in metaRegions. They are excess replicas.<a name="line.2696"></a> |
| <span class="sourceLineNo">2697</span> for (Map.Entry<Integer, HbckRegionInfo> entry : metaRegions.entrySet()) {<a name="line.2697"></a> |
| <span class="sourceLineNo">2698</span> noProblem = false;<a name="line.2698"></a> |
| <span class="sourceLineNo">2699</span> errors.reportError(ERROR_CODE.SHOULD_NOT_BE_DEPLOYED,<a name="line.2699"></a> |
| <span class="sourceLineNo">2700</span> "hbase:meta replicas are deployed in excess. Configured " + metaReplication +<a name="line.2700"></a> |
| <span class="sourceLineNo">2701</span> ", deployed " + metaRegions.size());<a name="line.2701"></a> |
| <span class="sourceLineNo">2702</span> if (shouldFixAssignments()) {<a name="line.2702"></a> |
| <span class="sourceLineNo">2703</span> errors.print("Trying to undeploy excess replica, replicaId: " + entry.getKey() +<a name="line.2703"></a> |
| <span class="sourceLineNo">2704</span> " of hbase:meta..");<a name="line.2704"></a> |
| <span class="sourceLineNo">2705</span> setShouldRerun();<a name="line.2705"></a> |
| <span class="sourceLineNo">2706</span> unassignMetaReplica(entry.getValue());<a name="line.2706"></a> |
| <span class="sourceLineNo">2707</span> }<a name="line.2707"></a> |
| <span class="sourceLineNo">2708</span> }<a name="line.2708"></a> |
| <span class="sourceLineNo">2709</span> // if noProblem is false, rerun hbck with hopefully fixed META<a name="line.2709"></a> |
| <span class="sourceLineNo">2710</span> // if noProblem is true, no errors, so continue normally<a name="line.2710"></a> |
| <span class="sourceLineNo">2711</span> return noProblem;<a name="line.2711"></a> |
| <span class="sourceLineNo">2712</span> }<a name="line.2712"></a> |
| <span class="sourceLineNo">2713</span><a name="line.2713"></a> |
| <span class="sourceLineNo">2714</span> private void unassignMetaReplica(HbckRegionInfo hi)<a name="line.2714"></a> |
| <span class="sourceLineNo">2715</span> throws IOException, InterruptedException, KeeperException {<a name="line.2715"></a> |
| <span class="sourceLineNo">2716</span> undeployRegions(hi);<a name="line.2716"></a> |
| <span class="sourceLineNo">2717</span> ZKUtil.deleteNode(zkw,<a name="line.2717"></a> |
| <span class="sourceLineNo">2718</span> zkw.getZNodePaths().getZNodeForReplica(hi.getMetaEntry().getRegionInfo().getReplicaId()));<a name="line.2718"></a> |
| <span class="sourceLineNo">2719</span> }<a name="line.2719"></a> |
| <span class="sourceLineNo">2720</span><a name="line.2720"></a> |
| <span class="sourceLineNo">2721</span> private void assignMetaReplica(int replicaId)<a name="line.2721"></a> |
| <span class="sourceLineNo">2722</span> throws IOException, KeeperException, InterruptedException {<a name="line.2722"></a> |
| <span class="sourceLineNo">2723</span> errors.reportError(ERROR_CODE.NO_META_REGION, "hbase:meta, replicaId " +<a name="line.2723"></a> |
| <span class="sourceLineNo">2724</span> replicaId +" is not found on any region.");<a name="line.2724"></a> |
| <span class="sourceLineNo">2725</span> if (shouldFixAssignments()) {<a name="line.2725"></a> |
| <span class="sourceLineNo">2726</span> errors.print("Trying to fix a problem with hbase:meta..");<a name="line.2726"></a> |
| <span class="sourceLineNo">2727</span> setShouldRerun();<a name="line.2727"></a> |
| <span class="sourceLineNo">2728</span> // try to fix it (treat it as unassigned region)<a name="line.2728"></a> |
| <span class="sourceLineNo">2729</span> RegionInfo h = RegionReplicaUtil.getRegionInfoForReplica(<a name="line.2729"></a> |
| <span class="sourceLineNo">2730</span> RegionInfoBuilder.FIRST_META_REGIONINFO, replicaId);<a name="line.2730"></a> |
| <span class="sourceLineNo">2731</span> HBaseFsckRepair.fixUnassigned(admin, h);<a name="line.2731"></a> |
| <span class="sourceLineNo">2732</span> HBaseFsckRepair.waitUntilAssigned(admin, h);<a name="line.2732"></a> |
| <span class="sourceLineNo">2733</span> }<a name="line.2733"></a> |
| <span class="sourceLineNo">2734</span> }<a name="line.2734"></a> |
| <span class="sourceLineNo">2735</span><a name="line.2735"></a> |
| <span class="sourceLineNo">2736</span> /**<a name="line.2736"></a> |
| <span class="sourceLineNo">2737</span> * Scan hbase:meta, adding all regions found to the regionInfo map.<a name="line.2737"></a> |
| <span class="sourceLineNo">2738</span> * @throws IOException if an error is encountered<a name="line.2738"></a> |
| <span class="sourceLineNo">2739</span> */<a name="line.2739"></a> |
| <span class="sourceLineNo">2740</span> boolean loadMetaEntries() throws IOException {<a name="line.2740"></a> |
| <span class="sourceLineNo">2741</span> ClientMetaTableAccessor.Visitor visitor = new ClientMetaTableAccessor.Visitor() {<a name="line.2741"></a> |
| <span class="sourceLineNo">2742</span> int countRecord = 1;<a name="line.2742"></a> |
| <span class="sourceLineNo">2743</span><a name="line.2743"></a> |
| <span class="sourceLineNo">2744</span> // comparator to sort KeyValues with latest modtime<a name="line.2744"></a> |
| <span class="sourceLineNo">2745</span> final Comparator<Cell> comp = new Comparator<Cell>() {<a name="line.2745"></a> |
| <span class="sourceLineNo">2746</span> @Override<a name="line.2746"></a> |
| <span class="sourceLineNo">2747</span> public int compare(Cell k1, Cell k2) {<a name="line.2747"></a> |
| <span class="sourceLineNo">2748</span> return Long.compare(k1.getTimestamp(), k2.getTimestamp());<a name="line.2748"></a> |
| <span class="sourceLineNo">2749</span> }<a name="line.2749"></a> |
| <span class="sourceLineNo">2750</span> };<a name="line.2750"></a> |
| <span class="sourceLineNo">2751</span><a name="line.2751"></a> |
| <span class="sourceLineNo">2752</span> @Override<a name="line.2752"></a> |
| <span class="sourceLineNo">2753</span> public boolean visit(Result result) throws IOException {<a name="line.2753"></a> |
| <span class="sourceLineNo">2754</span> try {<a name="line.2754"></a> |
| <span class="sourceLineNo">2755</span><a name="line.2755"></a> |
| <span class="sourceLineNo">2756</span> // record the latest modification of this META record<a name="line.2756"></a> |
| <span class="sourceLineNo">2757</span> long ts = Collections.max(result.listCells(), comp).getTimestamp();<a name="line.2757"></a> |
| <span class="sourceLineNo">2758</span> RegionLocations rl = CatalogFamilyFormat.getRegionLocations(result);<a name="line.2758"></a> |
| <span class="sourceLineNo">2759</span> if (rl == null) {<a name="line.2759"></a> |
| <span class="sourceLineNo">2760</span> emptyRegionInfoQualifiers.add(result);<a name="line.2760"></a> |
| <span class="sourceLineNo">2761</span> errors.reportError(ERROR_CODE.EMPTY_META_CELL,<a name="line.2761"></a> |
| <span class="sourceLineNo">2762</span> "Empty REGIONINFO_QUALIFIER found in hbase:meta");<a name="line.2762"></a> |
| <span class="sourceLineNo">2763</span> return true;<a name="line.2763"></a> |
| <span class="sourceLineNo">2764</span> }<a name="line.2764"></a> |
| <span class="sourceLineNo">2765</span> ServerName sn = null;<a name="line.2765"></a> |
| <span class="sourceLineNo">2766</span> if (rl.getRegionLocation(RegionInfo.DEFAULT_REPLICA_ID) == null ||<a name="line.2766"></a> |
| <span class="sourceLineNo">2767</span> rl.getRegionLocation(RegionInfo.DEFAULT_REPLICA_ID).getRegion() == null) {<a name="line.2767"></a> |
| <span class="sourceLineNo">2768</span> emptyRegionInfoQualifiers.add(result);<a name="line.2768"></a> |
| <span class="sourceLineNo">2769</span> errors.reportError(ERROR_CODE.EMPTY_META_CELL,<a name="line.2769"></a> |
| <span class="sourceLineNo">2770</span> "Empty REGIONINFO_QUALIFIER found in hbase:meta");<a name="line.2770"></a> |
| <span class="sourceLineNo">2771</span> return true;<a name="line.2771"></a> |
| <span class="sourceLineNo">2772</span> }<a name="line.2772"></a> |
| <span class="sourceLineNo">2773</span> RegionInfo hri = rl.getRegionLocation(RegionInfo.DEFAULT_REPLICA_ID).getRegion();<a name="line.2773"></a> |
| <span class="sourceLineNo">2774</span> if (!(isTableIncluded(hri.getTable())<a name="line.2774"></a> |
| <span class="sourceLineNo">2775</span> || hri.isMetaRegion())) {<a name="line.2775"></a> |
| <span class="sourceLineNo">2776</span> return true;<a name="line.2776"></a> |
| <span class="sourceLineNo">2777</span> }<a name="line.2777"></a> |
| <span class="sourceLineNo">2778</span> PairOfSameType<RegionInfo> daughters = MetaTableAccessor.getDaughterRegions(result);<a name="line.2778"></a> |
| <span class="sourceLineNo">2779</span> for (HRegionLocation h : rl.getRegionLocations()) {<a name="line.2779"></a> |
| <span class="sourceLineNo">2780</span> if (h == null || h.getRegion() == null) {<a name="line.2780"></a> |
| <span class="sourceLineNo">2781</span> continue;<a name="line.2781"></a> |
| <span class="sourceLineNo">2782</span> }<a name="line.2782"></a> |
| <span class="sourceLineNo">2783</span> sn = h.getServerName();<a name="line.2783"></a> |
| <span class="sourceLineNo">2784</span> hri = h.getRegion();<a name="line.2784"></a> |
| <span class="sourceLineNo">2785</span><a name="line.2785"></a> |
| <span class="sourceLineNo">2786</span> HbckRegionInfo.MetaEntry m = null;<a name="line.2786"></a> |
| <span class="sourceLineNo">2787</span> if (hri.getReplicaId() == RegionInfo.DEFAULT_REPLICA_ID) {<a name="line.2787"></a> |
| <span class="sourceLineNo">2788</span> m = new HbckRegionInfo.MetaEntry(hri, sn, ts, daughters.getFirst(),<a name="line.2788"></a> |
| <span class="sourceLineNo">2789</span> daughters.getSecond());<a name="line.2789"></a> |
| <span class="sourceLineNo">2790</span> } else {<a name="line.2790"></a> |
| <span class="sourceLineNo">2791</span> m = new HbckRegionInfo.MetaEntry(hri, sn, ts, null, null);<a name="line.2791"></a> |
| <span class="sourceLineNo">2792</span> }<a name="line.2792"></a> |
| <span class="sourceLineNo">2793</span> HbckRegionInfo previous = regionInfoMap.get(hri.getEncodedName());<a name="line.2793"></a> |
| <span class="sourceLineNo">2794</span> if (previous == null) {<a name="line.2794"></a> |
| <span class="sourceLineNo">2795</span> regionInfoMap.put(hri.getEncodedName(), new HbckRegionInfo(m));<a name="line.2795"></a> |
| <span class="sourceLineNo">2796</span> } else if (previous.getMetaEntry() == null) {<a name="line.2796"></a> |
| <span class="sourceLineNo">2797</span> previous.setMetaEntry(m);<a name="line.2797"></a> |
| <span class="sourceLineNo">2798</span> } else {<a name="line.2798"></a> |
| <span class="sourceLineNo">2799</span> throw new IOException("Two entries in hbase:meta are same " + previous);<a name="line.2799"></a> |
| <span class="sourceLineNo">2800</span> }<a name="line.2800"></a> |
| <span class="sourceLineNo">2801</span> }<a name="line.2801"></a> |
| <span class="sourceLineNo">2802</span> List<RegionInfo> mergeParents = CatalogFamilyFormat.getMergeRegions(result.rawCells());<a name="line.2802"></a> |
| <span class="sourceLineNo">2803</span> if (mergeParents != null) {<a name="line.2803"></a> |
| <span class="sourceLineNo">2804</span> for (RegionInfo mergeRegion : mergeParents) {<a name="line.2804"></a> |
| <span class="sourceLineNo">2805</span> if (mergeRegion != null) {<a name="line.2805"></a> |
| <span class="sourceLineNo">2806</span> // This region is already being merged<a name="line.2806"></a> |
| <span class="sourceLineNo">2807</span> HbckRegionInfo hbInfo = getOrCreateInfo(mergeRegion.getEncodedName());<a name="line.2807"></a> |
| <span class="sourceLineNo">2808</span> hbInfo.setMerged(true);<a name="line.2808"></a> |
| <span class="sourceLineNo">2809</span> }<a name="line.2809"></a> |
| <span class="sourceLineNo">2810</span> }<a name="line.2810"></a> |
| <span class="sourceLineNo">2811</span> }<a name="line.2811"></a> |
| <span class="sourceLineNo">2812</span><a name="line.2812"></a> |
| <span class="sourceLineNo">2813</span> // show proof of progress to the user, once for every 100 records.<a name="line.2813"></a> |
| <span class="sourceLineNo">2814</span> if (countRecord % 100 == 0) {<a name="line.2814"></a> |
| <span class="sourceLineNo">2815</span> errors.progress();<a name="line.2815"></a> |
| <span class="sourceLineNo">2816</span> }<a name="line.2816"></a> |
| <span class="sourceLineNo">2817</span> countRecord++;<a name="line.2817"></a> |
| <span class="sourceLineNo">2818</span> return true;<a name="line.2818"></a> |
| <span class="sourceLineNo">2819</span> } catch (RuntimeException e) {<a name="line.2819"></a> |
| <span class="sourceLineNo">2820</span> LOG.error("Result=" + result);<a name="line.2820"></a> |
| <span class="sourceLineNo">2821</span> throw e;<a name="line.2821"></a> |
| <span class="sourceLineNo">2822</span> }<a name="line.2822"></a> |
| <span class="sourceLineNo">2823</span> }<a name="line.2823"></a> |
| <span class="sourceLineNo">2824</span> };<a name="line.2824"></a> |
| <span class="sourceLineNo">2825</span> if (!checkMetaOnly) {<a name="line.2825"></a> |
| <span class="sourceLineNo">2826</span> // Scan hbase:meta to pick up user regions<a name="line.2826"></a> |
| <span class="sourceLineNo">2827</span> MetaTableAccessor.fullScanRegions(connection, visitor);<a name="line.2827"></a> |
| <span class="sourceLineNo">2828</span> }<a name="line.2828"></a> |
| <span class="sourceLineNo">2829</span><a name="line.2829"></a> |
| <span class="sourceLineNo">2830</span> errors.print("");<a name="line.2830"></a> |
| <span class="sourceLineNo">2831</span> return true;<a name="line.2831"></a> |
| <span class="sourceLineNo">2832</span> }<a name="line.2832"></a> |
| <span class="sourceLineNo">2833</span><a name="line.2833"></a> |
| <span class="sourceLineNo">2834</span> /**<a name="line.2834"></a> |
| <span class="sourceLineNo">2835</span> * Prints summary of all tables found on the system.<a name="line.2835"></a> |
| <span class="sourceLineNo">2836</span> */<a name="line.2836"></a> |
| <span class="sourceLineNo">2837</span> private void printTableSummary(SortedMap<TableName, HbckTableInfo> tablesInfo) {<a name="line.2837"></a> |
| <span class="sourceLineNo">2838</span> StringBuilder sb = new StringBuilder();<a name="line.2838"></a> |
| <span class="sourceLineNo">2839</span> int numOfSkippedRegions;<a name="line.2839"></a> |
| <span class="sourceLineNo">2840</span> errors.print("Summary:");<a name="line.2840"></a> |
| <span class="sourceLineNo">2841</span> for (HbckTableInfo tInfo : tablesInfo.values()) {<a name="line.2841"></a> |
| <span class="sourceLineNo">2842</span> numOfSkippedRegions = (skippedRegions.containsKey(tInfo.getName())) ?<a name="line.2842"></a> |
| <span class="sourceLineNo">2843</span> skippedRegions.get(tInfo.getName()).size() : 0;<a name="line.2843"></a> |
| <span class="sourceLineNo">2844</span><a name="line.2844"></a> |
| <span class="sourceLineNo">2845</span> if (errors.tableHasErrors(tInfo)) {<a name="line.2845"></a> |
| <span class="sourceLineNo">2846</span> errors.print("Table " + tInfo.getName() + " is inconsistent.");<a name="line.2846"></a> |
| <span class="sourceLineNo">2847</span> } else if (numOfSkippedRegions > 0){<a name="line.2847"></a> |
| <span class="sourceLineNo">2848</span> errors.print("Table " + tInfo.getName() + " is okay (with "<a name="line.2848"></a> |
| <span class="sourceLineNo">2849</span> + numOfSkippedRegions + " skipped regions).");<a name="line.2849"></a> |
| <span class="sourceLineNo">2850</span> }<a name="line.2850"></a> |
| <span class="sourceLineNo">2851</span> else {<a name="line.2851"></a> |
| <span class="sourceLineNo">2852</span> errors.print("Table " + tInfo.getName() + " is okay.");<a name="line.2852"></a> |
| <span class="sourceLineNo">2853</span> }<a name="line.2853"></a> |
| <span class="sourceLineNo">2854</span> errors.print(" Number of regions: " + tInfo.getNumRegions());<a name="line.2854"></a> |
| <span class="sourceLineNo">2855</span> if (numOfSkippedRegions > 0) {<a name="line.2855"></a> |
| <span class="sourceLineNo">2856</span> Set<String> skippedRegionStrings = skippedRegions.get(tInfo.getName());<a name="line.2856"></a> |
| <span class="sourceLineNo">2857</span> System.out.println(" Number of skipped regions: " + numOfSkippedRegions);<a name="line.2857"></a> |
| <span class="sourceLineNo">2858</span> System.out.println(" List of skipped regions:");<a name="line.2858"></a> |
| <span class="sourceLineNo">2859</span> for(String sr : skippedRegionStrings) {<a name="line.2859"></a> |
| <span class="sourceLineNo">2860</span> System.out.println(" " + sr);<a name="line.2860"></a> |
| <span class="sourceLineNo">2861</span> }<a name="line.2861"></a> |
| <span class="sourceLineNo">2862</span> }<a name="line.2862"></a> |
| <span class="sourceLineNo">2863</span> sb.setLength(0); // clear out existing buffer, if any.<a name="line.2863"></a> |
| <span class="sourceLineNo">2864</span> sb.append(" Deployed on: ");<a name="line.2864"></a> |
| <span class="sourceLineNo">2865</span> for (ServerName server : tInfo.deployedOn) {<a name="line.2865"></a> |
| <span class="sourceLineNo">2866</span> sb.append(" " + server.toString());<a name="line.2866"></a> |
| <span class="sourceLineNo">2867</span> }<a name="line.2867"></a> |
| <span class="sourceLineNo">2868</span> errors.print(sb.toString());<a name="line.2868"></a> |
| <span class="sourceLineNo">2869</span> }<a name="line.2869"></a> |
| <span class="sourceLineNo">2870</span> }<a name="line.2870"></a> |
| <span class="sourceLineNo">2871</span><a name="line.2871"></a> |
| <span class="sourceLineNo">2872</span> static HbckErrorReporter getErrorReporter(final Configuration conf)<a name="line.2872"></a> |
| <span class="sourceLineNo">2873</span> throws ClassNotFoundException {<a name="line.2873"></a> |
| <span class="sourceLineNo">2874</span> Class<? extends HbckErrorReporter> reporter =<a name="line.2874"></a> |
| <span class="sourceLineNo">2875</span> conf.getClass("hbasefsck.errorreporter", PrintingErrorReporter.class,<a name="line.2875"></a> |
| <span class="sourceLineNo">2876</span> HbckErrorReporter.class);<a name="line.2876"></a> |
| <span class="sourceLineNo">2877</span> return ReflectionUtils.newInstance(reporter, conf);<a name="line.2877"></a> |
| <span class="sourceLineNo">2878</span> }<a name="line.2878"></a> |
| <span class="sourceLineNo">2879</span><a name="line.2879"></a> |
| <span class="sourceLineNo">2880</span> static class PrintingErrorReporter implements HbckErrorReporter {<a name="line.2880"></a> |
| <span class="sourceLineNo">2881</span> public int errorCount = 0;<a name="line.2881"></a> |
| <span class="sourceLineNo">2882</span> private int showProgress;<a name="line.2882"></a> |
| <span class="sourceLineNo">2883</span> // How frequently calls to progress() will create output<a name="line.2883"></a> |
| <span class="sourceLineNo">2884</span> private static final int progressThreshold = 100;<a name="line.2884"></a> |
| <span class="sourceLineNo">2885</span><a name="line.2885"></a> |
| <span class="sourceLineNo">2886</span> Set<HbckTableInfo> errorTables = new HashSet<>();<a name="line.2886"></a> |
| <span class="sourceLineNo">2887</span><a name="line.2887"></a> |
| <span class="sourceLineNo">2888</span> // for use by unit tests to verify which errors were discovered<a name="line.2888"></a> |
| <span class="sourceLineNo">2889</span> private ArrayList<ERROR_CODE> errorList = new ArrayList<>();<a name="line.2889"></a> |
| <span class="sourceLineNo">2890</span><a name="line.2890"></a> |
| <span class="sourceLineNo">2891</span> @Override<a name="line.2891"></a> |
| <span class="sourceLineNo">2892</span> public void clear() {<a name="line.2892"></a> |
| <span class="sourceLineNo">2893</span> errorTables.clear();<a name="line.2893"></a> |
| <span class="sourceLineNo">2894</span> errorList.clear();<a name="line.2894"></a> |
| <span class="sourceLineNo">2895</span> errorCount = 0;<a name="line.2895"></a> |
| <span class="sourceLineNo">2896</span> }<a name="line.2896"></a> |
| <span class="sourceLineNo">2897</span><a name="line.2897"></a> |
| <span class="sourceLineNo">2898</span> @Override<a name="line.2898"></a> |
| <span class="sourceLineNo">2899</span> public synchronized void reportError(ERROR_CODE errorCode, String message) {<a name="line.2899"></a> |
| <span class="sourceLineNo">2900</span> if (errorCode == ERROR_CODE.WRONG_USAGE) {<a name="line.2900"></a> |
| <span class="sourceLineNo">2901</span> System.err.println(message);<a name="line.2901"></a> |
| <span class="sourceLineNo">2902</span> return;<a name="line.2902"></a> |
| <span class="sourceLineNo">2903</span> }<a name="line.2903"></a> |
| <span class="sourceLineNo">2904</span><a name="line.2904"></a> |
| <span class="sourceLineNo">2905</span> errorList.add(errorCode);<a name="line.2905"></a> |
| <span class="sourceLineNo">2906</span> if (!summary) {<a name="line.2906"></a> |
| <span class="sourceLineNo">2907</span> System.out.println("ERROR: " + message);<a name="line.2907"></a> |
| <span class="sourceLineNo">2908</span> }<a name="line.2908"></a> |
| <span class="sourceLineNo">2909</span> errorCount++;<a name="line.2909"></a> |
| <span class="sourceLineNo">2910</span> showProgress = 0;<a name="line.2910"></a> |
| <span class="sourceLineNo">2911</span> }<a name="line.2911"></a> |
| <span class="sourceLineNo">2912</span><a name="line.2912"></a> |
| <span class="sourceLineNo">2913</span> @Override<a name="line.2913"></a> |
| <span class="sourceLineNo">2914</span> public synchronized void reportError(ERROR_CODE errorCode, String message,<a name="line.2914"></a> |
| <span class="sourceLineNo">2915</span> HbckTableInfo table) {<a name="line.2915"></a> |
| <span class="sourceLineNo">2916</span> errorTables.add(table);<a name="line.2916"></a> |
| <span class="sourceLineNo">2917</span> reportError(errorCode, message);<a name="line.2917"></a> |
| <span class="sourceLineNo">2918</span> }<a name="line.2918"></a> |
| <span class="sourceLineNo">2919</span><a name="line.2919"></a> |
| <span class="sourceLineNo">2920</span> @Override<a name="line.2920"></a> |
| <span class="sourceLineNo">2921</span> public synchronized void reportError(ERROR_CODE errorCode, String message, HbckTableInfo table,<a name="line.2921"></a> |
| <span class="sourceLineNo">2922</span> HbckRegionInfo info) {<a name="line.2922"></a> |
| <span class="sourceLineNo">2923</span> errorTables.add(table);<a name="line.2923"></a> |
| <span class="sourceLineNo">2924</span> String reference = "(region " + info.getRegionNameAsString() + ")";<a name="line.2924"></a> |
| <span class="sourceLineNo">2925</span> reportError(errorCode, reference + " " + message);<a name="line.2925"></a> |
| <span class="sourceLineNo">2926</span> }<a name="line.2926"></a> |
| <span class="sourceLineNo">2927</span><a name="line.2927"></a> |
| <span class="sourceLineNo">2928</span> @Override<a name="line.2928"></a> |
| <span class="sourceLineNo">2929</span> public synchronized void reportError(ERROR_CODE errorCode, String message, HbckTableInfo table,<a name="line.2929"></a> |
| <span class="sourceLineNo">2930</span> HbckRegionInfo info1, HbckRegionInfo info2) {<a name="line.2930"></a> |
| <span class="sourceLineNo">2931</span> errorTables.add(table);<a name="line.2931"></a> |
| <span class="sourceLineNo">2932</span> String reference = "(regions " + info1.getRegionNameAsString()<a name="line.2932"></a> |
| <span class="sourceLineNo">2933</span> + " and " + info2.getRegionNameAsString() + ")";<a name="line.2933"></a> |
| <span class="sourceLineNo">2934</span> reportError(errorCode, reference + " " + message);<a name="line.2934"></a> |
| <span class="sourceLineNo">2935</span> }<a name="line.2935"></a> |
| <span class="sourceLineNo">2936</span><a name="line.2936"></a> |
| <span class="sourceLineNo">2937</span> @Override<a name="line.2937"></a> |
| <span class="sourceLineNo">2938</span> public synchronized void reportError(String message) {<a name="line.2938"></a> |
| <span class="sourceLineNo">2939</span> reportError(ERROR_CODE.UNKNOWN, message);<a name="line.2939"></a> |
| <span class="sourceLineNo">2940</span> }<a name="line.2940"></a> |
| <span class="sourceLineNo">2941</span><a name="line.2941"></a> |
| <span class="sourceLineNo">2942</span> /**<a name="line.2942"></a> |
| <span class="sourceLineNo">2943</span> * Report error information, but do not increment the error count. Intended for cases<a name="line.2943"></a> |
| <span class="sourceLineNo">2944</span> * where the actual error would have been reported previously.<a name="line.2944"></a> |
| <span class="sourceLineNo">2945</span> * @param message<a name="line.2945"></a> |
| <span class="sourceLineNo">2946</span> */<a name="line.2946"></a> |
| <span class="sourceLineNo">2947</span> @Override<a name="line.2947"></a> |
| <span class="sourceLineNo">2948</span> public synchronized void report(String message) {<a name="line.2948"></a> |
| <span class="sourceLineNo">2949</span> if (! summary) {<a name="line.2949"></a> |
| <span class="sourceLineNo">2950</span> System.out.println("ERROR: " + message);<a name="line.2950"></a> |
| <span class="sourceLineNo">2951</span> }<a name="line.2951"></a> |
| <span class="sourceLineNo">2952</span> showProgress = 0;<a name="line.2952"></a> |
| <span class="sourceLineNo">2953</span> }<a name="line.2953"></a> |
| <span class="sourceLineNo">2954</span><a name="line.2954"></a> |
| <span class="sourceLineNo">2955</span> @Override<a name="line.2955"></a> |
| <span class="sourceLineNo">2956</span> public synchronized int summarize() {<a name="line.2956"></a> |
| <span class="sourceLineNo">2957</span> System.out.println(Integer.toString(errorCount) +<a name="line.2957"></a> |
| <span class="sourceLineNo">2958</span> " inconsistencies detected.");<a name="line.2958"></a> |
| <span class="sourceLineNo">2959</span> if (errorCount == 0) {<a name="line.2959"></a> |
| <span class="sourceLineNo">2960</span> System.out.println("Status: OK");<a name="line.2960"></a> |
| <span class="sourceLineNo">2961</span> return 0;<a name="line.2961"></a> |
| <span class="sourceLineNo">2962</span> } else {<a name="line.2962"></a> |
| <span class="sourceLineNo">2963</span> System.out.println("Status: INCONSISTENT");<a name="line.2963"></a> |
| <span class="sourceLineNo">2964</span> return -1;<a name="line.2964"></a> |
| <span class="sourceLineNo">2965</span> }<a name="line.2965"></a> |
| <span class="sourceLineNo">2966</span> }<a name="line.2966"></a> |
| <span class="sourceLineNo">2967</span><a name="line.2967"></a> |
| <span class="sourceLineNo">2968</span> @Override<a name="line.2968"></a> |
| <span class="sourceLineNo">2969</span> public ArrayList<ERROR_CODE> getErrorList() {<a name="line.2969"></a> |
| <span class="sourceLineNo">2970</span> return errorList;<a name="line.2970"></a> |
| <span class="sourceLineNo">2971</span> }<a name="line.2971"></a> |
| <span class="sourceLineNo">2972</span><a name="line.2972"></a> |
| <span class="sourceLineNo">2973</span> @Override<a name="line.2973"></a> |
| <span class="sourceLineNo">2974</span> public synchronized void print(String message) {<a name="line.2974"></a> |
| <span class="sourceLineNo">2975</span> if (!summary) {<a name="line.2975"></a> |
| <span class="sourceLineNo">2976</span> System.out.println(message);<a name="line.2976"></a> |
| <span class="sourceLineNo">2977</span> }<a name="line.2977"></a> |
| <span class="sourceLineNo">2978</span> }<a name="line.2978"></a> |
| <span class="sourceLineNo">2979</span><a name="line.2979"></a> |
| <span class="sourceLineNo">2980</span> @Override<a name="line.2980"></a> |
| <span class="sourceLineNo">2981</span> public boolean tableHasErrors(HbckTableInfo table) {<a name="line.2981"></a> |
| <span class="sourceLineNo">2982</span> return errorTables.contains(table);<a name="line.2982"></a> |
| <span class="sourceLineNo">2983</span> }<a name="line.2983"></a> |
| <span class="sourceLineNo">2984</span><a name="line.2984"></a> |
| <span class="sourceLineNo">2985</span> @Override<a name="line.2985"></a> |
| <span class="sourceLineNo">2986</span> public void resetErrors() {<a name="line.2986"></a> |
| <span class="sourceLineNo">2987</span> errorCount = 0;<a name="line.2987"></a> |
| <span class="sourceLineNo">2988</span> }<a name="line.2988"></a> |
| <span class="sourceLineNo">2989</span><a name="line.2989"></a> |
| <span class="sourceLineNo">2990</span> @Override<a name="line.2990"></a> |
| <span class="sourceLineNo">2991</span> public synchronized void detail(String message) {<a name="line.2991"></a> |
| <span class="sourceLineNo">2992</span> if (details) {<a name="line.2992"></a> |
| <span class="sourceLineNo">2993</span> System.out.println(message);<a name="line.2993"></a> |
| <span class="sourceLineNo">2994</span> }<a name="line.2994"></a> |
| <span class="sourceLineNo">2995</span> showProgress = 0;<a name="line.2995"></a> |
| <span class="sourceLineNo">2996</span> }<a name="line.2996"></a> |
| <span class="sourceLineNo">2997</span><a name="line.2997"></a> |
| <span class="sourceLineNo">2998</span> @Override<a name="line.2998"></a> |
| <span class="sourceLineNo">2999</span> public synchronized void progress() {<a name="line.2999"></a> |
| <span class="sourceLineNo">3000</span> if (showProgress++ == progressThreshold) {<a name="line.3000"></a> |
| <span class="sourceLineNo">3001</span> if (!summary) {<a name="line.3001"></a> |
| <span class="sourceLineNo">3002</span> System.out.print(".");<a name="line.3002"></a> |
| <span class="sourceLineNo">3003</span> }<a name="line.3003"></a> |
| <span class="sourceLineNo">3004</span> showProgress = 0;<a name="line.3004"></a> |
| <span class="sourceLineNo">3005</span> }<a name="line.3005"></a> |
| <span class="sourceLineNo">3006</span> }<a name="line.3006"></a> |
| <span class="sourceLineNo">3007</span> }<a name="line.3007"></a> |
| <span class="sourceLineNo">3008</span><a name="line.3008"></a> |
| <span class="sourceLineNo">3009</span> /**<a name="line.3009"></a> |
| <span class="sourceLineNo">3010</span> * Contact a region server and get all information from it<a name="line.3010"></a> |
| <span class="sourceLineNo">3011</span> */<a name="line.3011"></a> |
| <span class="sourceLineNo">3012</span> static class WorkItemRegion implements Callable<Void> {<a name="line.3012"></a> |
| <span class="sourceLineNo">3013</span> private final HBaseFsck hbck;<a name="line.3013"></a> |
| <span class="sourceLineNo">3014</span> private final ServerName rsinfo;<a name="line.3014"></a> |
| <span class="sourceLineNo">3015</span> private final HbckErrorReporter errors;<a name="line.3015"></a> |
| <span class="sourceLineNo">3016</span> private final Connection connection;<a name="line.3016"></a> |
| <span class="sourceLineNo">3017</span><a name="line.3017"></a> |
| <span class="sourceLineNo">3018</span> WorkItemRegion(HBaseFsck hbck, ServerName info, HbckErrorReporter errors,<a name="line.3018"></a> |
| <span class="sourceLineNo">3019</span> Connection connection) {<a name="line.3019"></a> |
| <span class="sourceLineNo">3020</span> this.hbck = hbck;<a name="line.3020"></a> |
| <span class="sourceLineNo">3021</span> this.rsinfo = info;<a name="line.3021"></a> |
| <span class="sourceLineNo">3022</span> this.errors = errors;<a name="line.3022"></a> |
| <span class="sourceLineNo">3023</span> this.connection = connection;<a name="line.3023"></a> |
| <span class="sourceLineNo">3024</span> }<a name="line.3024"></a> |
| <span class="sourceLineNo">3025</span><a name="line.3025"></a> |
| <span class="sourceLineNo">3026</span> @Override<a name="line.3026"></a> |
| <span class="sourceLineNo">3027</span> public synchronized Void call() throws IOException {<a name="line.3027"></a> |
| <span class="sourceLineNo">3028</span> errors.progress();<a name="line.3028"></a> |
| <span class="sourceLineNo">3029</span> try {<a name="line.3029"></a> |
| <span class="sourceLineNo">3030</span> // list all online regions from this region server<a name="line.3030"></a> |
| <span class="sourceLineNo">3031</span> List<RegionInfo> regions = connection.getAdmin().getRegions(rsinfo);<a name="line.3031"></a> |
| <span class="sourceLineNo">3032</span> regions = filterRegions(regions);<a name="line.3032"></a> |
| <span class="sourceLineNo">3033</span><a name="line.3033"></a> |
| <span class="sourceLineNo">3034</span> if (details) {<a name="line.3034"></a> |
| <span class="sourceLineNo">3035</span> errors.detail(<a name="line.3035"></a> |
| <span class="sourceLineNo">3036</span> "RegionServer: " + rsinfo.getServerName() + " number of regions: " + regions.size());<a name="line.3036"></a> |
| <span class="sourceLineNo">3037</span> for (RegionInfo rinfo : regions) {<a name="line.3037"></a> |
| <span class="sourceLineNo">3038</span> errors.detail(" " + rinfo.getRegionNameAsString() + " id: " + rinfo.getRegionId() +<a name="line.3038"></a> |
| <span class="sourceLineNo">3039</span> " encoded_name: " + rinfo.getEncodedName() + " start: " +<a name="line.3039"></a> |
| <span class="sourceLineNo">3040</span> Bytes.toStringBinary(rinfo.getStartKey()) + " end: " +<a name="line.3040"></a> |
| <span class="sourceLineNo">3041</span> Bytes.toStringBinary(rinfo.getEndKey()));<a name="line.3041"></a> |
| <span class="sourceLineNo">3042</span> }<a name="line.3042"></a> |
| <span class="sourceLineNo">3043</span> }<a name="line.3043"></a> |
| <span class="sourceLineNo">3044</span><a name="line.3044"></a> |
| <span class="sourceLineNo">3045</span> // check to see if the existence of this region matches the region in META<a name="line.3045"></a> |
| <span class="sourceLineNo">3046</span> for (RegionInfo r : regions) {<a name="line.3046"></a> |
| <span class="sourceLineNo">3047</span> HbckRegionInfo hbi = hbck.getOrCreateInfo(r.getEncodedName());<a name="line.3047"></a> |
| <span class="sourceLineNo">3048</span> hbi.addServer(r, rsinfo);<a name="line.3048"></a> |
| <span class="sourceLineNo">3049</span> }<a name="line.3049"></a> |
| <span class="sourceLineNo">3050</span> } catch (IOException e) { // unable to connect to the region server.<a name="line.3050"></a> |
| <span class="sourceLineNo">3051</span> errors.reportError(ERROR_CODE.RS_CONNECT_FAILURE,<a name="line.3051"></a> |
| <span class="sourceLineNo">3052</span> "RegionServer: " + rsinfo.getServerName() + " Unable to fetch region information. " + e);<a name="line.3052"></a> |
| <span class="sourceLineNo">3053</span> throw e;<a name="line.3053"></a> |
| <span class="sourceLineNo">3054</span> }<a name="line.3054"></a> |
| <span class="sourceLineNo">3055</span> return null;<a name="line.3055"></a> |
| <span class="sourceLineNo">3056</span> }<a name="line.3056"></a> |
| <span class="sourceLineNo">3057</span><a name="line.3057"></a> |
| <span class="sourceLineNo">3058</span> private List<RegionInfo> filterRegions(List<RegionInfo> regions) {<a name="line.3058"></a> |
| <span class="sourceLineNo">3059</span> List<RegionInfo> ret = Lists.newArrayList();<a name="line.3059"></a> |
| <span class="sourceLineNo">3060</span> for (RegionInfo hri : regions) {<a name="line.3060"></a> |
| <span class="sourceLineNo">3061</span> if (hri.isMetaRegion() || (!hbck.checkMetaOnly<a name="line.3061"></a> |
| <span class="sourceLineNo">3062</span> && hbck.isTableIncluded(hri.getTable()))) {<a name="line.3062"></a> |
| <span class="sourceLineNo">3063</span> ret.add(hri);<a name="line.3063"></a> |
| <span class="sourceLineNo">3064</span> }<a name="line.3064"></a> |
| <span class="sourceLineNo">3065</span> }<a name="line.3065"></a> |
| <span class="sourceLineNo">3066</span> return ret;<a name="line.3066"></a> |
| <span class="sourceLineNo">3067</span> }<a name="line.3067"></a> |
| <span class="sourceLineNo">3068</span> }<a name="line.3068"></a> |
| <span class="sourceLineNo">3069</span><a name="line.3069"></a> |
| <span class="sourceLineNo">3070</span> /**<a name="line.3070"></a> |
| <span class="sourceLineNo">3071</span> * Contact hdfs and get all information about specified table directory into<a name="line.3071"></a> |
| <span class="sourceLineNo">3072</span> * regioninfo list.<a name="line.3072"></a> |
| <span class="sourceLineNo">3073</span> */<a name="line.3073"></a> |
| <span class="sourceLineNo">3074</span> class WorkItemHdfsDir implements Callable<Void> {<a name="line.3074"></a> |
| <span class="sourceLineNo">3075</span> private FileStatus tableDir;<a name="line.3075"></a> |
| <span class="sourceLineNo">3076</span> private HbckErrorReporter errors;<a name="line.3076"></a> |
| <span class="sourceLineNo">3077</span> private FileSystem fs;<a name="line.3077"></a> |
| <span class="sourceLineNo">3078</span><a name="line.3078"></a> |
| <span class="sourceLineNo">3079</span> WorkItemHdfsDir(FileSystem fs, HbckErrorReporter errors, FileStatus status) {<a name="line.3079"></a> |
| <span class="sourceLineNo">3080</span> this.fs = fs;<a name="line.3080"></a> |
| <span class="sourceLineNo">3081</span> this.tableDir = status;<a name="line.3081"></a> |
| <span class="sourceLineNo">3082</span> this.errors = errors;<a name="line.3082"></a> |
| <span class="sourceLineNo">3083</span> }<a name="line.3083"></a> |
| <span class="sourceLineNo">3084</span><a name="line.3084"></a> |
| <span class="sourceLineNo">3085</span> @Override<a name="line.3085"></a> |
| <span class="sourceLineNo">3086</span> public synchronized Void call() throws InterruptedException, ExecutionException {<a name="line.3086"></a> |
| <span class="sourceLineNo">3087</span> final Vector<Exception> exceptions = new Vector<>();<a name="line.3087"></a> |
| <span class="sourceLineNo">3088</span><a name="line.3088"></a> |
| <span class="sourceLineNo">3089</span> try {<a name="line.3089"></a> |
| <span class="sourceLineNo">3090</span> final FileStatus[] regionDirs = fs.listStatus(tableDir.getPath());<a name="line.3090"></a> |
| <span class="sourceLineNo">3091</span> final List<Future<?>> futures = new ArrayList<>(regionDirs.length);<a name="line.3091"></a> |
| <span class="sourceLineNo">3092</span><a name="line.3092"></a> |
| <span class="sourceLineNo">3093</span> for (final FileStatus regionDir : regionDirs) {<a name="line.3093"></a> |
| <span class="sourceLineNo">3094</span> errors.progress();<a name="line.3094"></a> |
| <span class="sourceLineNo">3095</span> final String encodedName = regionDir.getPath().getName();<a name="line.3095"></a> |
| <span class="sourceLineNo">3096</span> // ignore directories that aren't hexadecimal<a name="line.3096"></a> |
| <span class="sourceLineNo">3097</span> if (!encodedName.toLowerCase(Locale.ROOT).matches("[0-9a-f]+")) {<a name="line.3097"></a> |
| <span class="sourceLineNo">3098</span> continue;<a name="line.3098"></a> |
| <span class="sourceLineNo">3099</span> }<a name="line.3099"></a> |
| <span class="sourceLineNo">3100</span><a name="line.3100"></a> |
| <span class="sourceLineNo">3101</span> if (!exceptions.isEmpty()) {<a name="line.3101"></a> |
| <span class="sourceLineNo">3102</span> break;<a name="line.3102"></a> |
| <span class="sourceLineNo">3103</span> }<a name="line.3103"></a> |
| <span class="sourceLineNo">3104</span><a name="line.3104"></a> |
| <span class="sourceLineNo">3105</span> futures.add(executor.submit(new Runnable() {<a name="line.3105"></a> |
| <span class="sourceLineNo">3106</span> @Override<a name="line.3106"></a> |
| <span class="sourceLineNo">3107</span> public void run() {<a name="line.3107"></a> |
| <span class="sourceLineNo">3108</span> try {<a name="line.3108"></a> |
| <span class="sourceLineNo">3109</span> LOG.debug("Loading region info from hdfs:"+ regionDir.getPath());<a name="line.3109"></a> |
| <span class="sourceLineNo">3110</span><a name="line.3110"></a> |
| <span class="sourceLineNo">3111</span> Path regioninfoFile = new Path(regionDir.getPath(), HRegionFileSystem.REGION_INFO_FILE);<a name="line.3111"></a> |
| <span class="sourceLineNo">3112</span> boolean regioninfoFileExists = fs.exists(regioninfoFile);<a name="line.3112"></a> |
| <span class="sourceLineNo">3113</span><a name="line.3113"></a> |
| <span class="sourceLineNo">3114</span> if (!regioninfoFileExists) {<a name="line.3114"></a> |
| <span class="sourceLineNo">3115</span> // As tables become larger it is more and more likely that by the time you<a name="line.3115"></a> |
| <span class="sourceLineNo">3116</span> // reach a given region that it will be gone due to region splits/merges.<a name="line.3116"></a> |
| <span class="sourceLineNo">3117</span> if (!fs.exists(regionDir.getPath())) {<a name="line.3117"></a> |
| <span class="sourceLineNo">3118</span> LOG.warn("By the time we tried to process this region dir it was already gone: "<a name="line.3118"></a> |
| <span class="sourceLineNo">3119</span> + regionDir.getPath());<a name="line.3119"></a> |
| <span class="sourceLineNo">3120</span> return;<a name="line.3120"></a> |
| <span class="sourceLineNo">3121</span> }<a name="line.3121"></a> |
| <span class="sourceLineNo">3122</span> }<a name="line.3122"></a> |
| <span class="sourceLineNo">3123</span><a name="line.3123"></a> |
| <span class="sourceLineNo">3124</span> HbckRegionInfo hbi = HBaseFsck.this.getOrCreateInfo(encodedName);<a name="line.3124"></a> |
| <span class="sourceLineNo">3125</span> HbckRegionInfo.HdfsEntry he = new HbckRegionInfo.HdfsEntry();<a name="line.3125"></a> |
| <span class="sourceLineNo">3126</span> synchronized (hbi) {<a name="line.3126"></a> |
| <span class="sourceLineNo">3127</span> if (hbi.getHdfsRegionDir() != null) {<a name="line.3127"></a> |
| <span class="sourceLineNo">3128</span> errors.print("Directory " + encodedName + " duplicate??" +<a name="line.3128"></a> |
| <span class="sourceLineNo">3129</span> hbi.getHdfsRegionDir());<a name="line.3129"></a> |
| <span class="sourceLineNo">3130</span> }<a name="line.3130"></a> |
| <span class="sourceLineNo">3131</span><a name="line.3131"></a> |
| <span class="sourceLineNo">3132</span> he.regionDir = regionDir.getPath();<a name="line.3132"></a> |
| <span class="sourceLineNo">3133</span> he.regionDirModTime = regionDir.getModificationTime();<a name="line.3133"></a> |
| <span class="sourceLineNo">3134</span> he.hdfsRegioninfoFilePresent = regioninfoFileExists;<a name="line.3134"></a> |
| <span class="sourceLineNo">3135</span> // we add to orphan list when we attempt to read .regioninfo<a name="line.3135"></a> |
| <span class="sourceLineNo">3136</span><a name="line.3136"></a> |
| <span class="sourceLineNo">3137</span> // Set a flag if this region contains only edits<a name="line.3137"></a> |
| <span class="sourceLineNo">3138</span> // This is special case if a region is left after split<a name="line.3138"></a> |
| <span class="sourceLineNo">3139</span> he.hdfsOnlyEdits = true;<a name="line.3139"></a> |
| <span class="sourceLineNo">3140</span> FileStatus[] subDirs = fs.listStatus(regionDir.getPath());<a name="line.3140"></a> |
| <span class="sourceLineNo">3141</span> Path ePath = WALSplitUtil.getRegionDirRecoveredEditsDir(regionDir.getPath());<a name="line.3141"></a> |
| <span class="sourceLineNo">3142</span> for (FileStatus subDir : subDirs) {<a name="line.3142"></a> |
| <span class="sourceLineNo">3143</span> errors.progress();<a name="line.3143"></a> |
| <span class="sourceLineNo">3144</span> String sdName = subDir.getPath().getName();<a name="line.3144"></a> |
| <span class="sourceLineNo">3145</span> if (!sdName.startsWith(".") && !sdName.equals(ePath.getName())) {<a name="line.3145"></a> |
| <span class="sourceLineNo">3146</span> he.hdfsOnlyEdits = false;<a name="line.3146"></a> |
| <span class="sourceLineNo">3147</span> break;<a name="line.3147"></a> |
| <span class="sourceLineNo">3148</span> }<a name="line.3148"></a> |
| <span class="sourceLineNo">3149</span> }<a name="line.3149"></a> |
| <span class="sourceLineNo">3150</span> hbi.setHdfsEntry(he);<a name="line.3150"></a> |
| <span class="sourceLineNo">3151</span> }<a name="line.3151"></a> |
| <span class="sourceLineNo">3152</span> } catch (Exception e) {<a name="line.3152"></a> |
| <span class="sourceLineNo">3153</span> LOG.error("Could not load region dir", e);<a name="line.3153"></a> |
| <span class="sourceLineNo">3154</span> exceptions.add(e);<a name="line.3154"></a> |
| <span class="sourceLineNo">3155</span> }<a name="line.3155"></a> |
| <span class="sourceLineNo">3156</span> }<a name="line.3156"></a> |
| <span class="sourceLineNo">3157</span> }));<a name="line.3157"></a> |
| <span class="sourceLineNo">3158</span> }<a name="line.3158"></a> |
| <span class="sourceLineNo">3159</span><a name="line.3159"></a> |
| <span class="sourceLineNo">3160</span> // Ensure all pending tasks are complete (or that we run into an exception)<a name="line.3160"></a> |
| <span class="sourceLineNo">3161</span> for (Future<?> f : futures) {<a name="line.3161"></a> |
| <span class="sourceLineNo">3162</span> if (!exceptions.isEmpty()) {<a name="line.3162"></a> |
| <span class="sourceLineNo">3163</span> break;<a name="line.3163"></a> |
| <span class="sourceLineNo">3164</span> }<a name="line.3164"></a> |
| <span class="sourceLineNo">3165</span> try {<a name="line.3165"></a> |
| <span class="sourceLineNo">3166</span> f.get();<a name="line.3166"></a> |
| <span class="sourceLineNo">3167</span> } catch (ExecutionException e) {<a name="line.3167"></a> |
| <span class="sourceLineNo">3168</span> LOG.error("Unexpected exec exception! Should've been caught already. (Bug?)", e);<a name="line.3168"></a> |
| <span class="sourceLineNo">3169</span> // Shouldn't happen, we already logged/caught any exceptions in the Runnable<a name="line.3169"></a> |
| <span class="sourceLineNo">3170</span> }<a name="line.3170"></a> |
| <span class="sourceLineNo">3171</span> }<a name="line.3171"></a> |
| <span class="sourceLineNo">3172</span> } catch (IOException e) {<a name="line.3172"></a> |
| <span class="sourceLineNo">3173</span> LOG.error("Cannot execute WorkItemHdfsDir for " + tableDir, e);<a name="line.3173"></a> |
| <span class="sourceLineNo">3174</span> exceptions.add(e);<a name="line.3174"></a> |
| <span class="sourceLineNo">3175</span> } finally {<a name="line.3175"></a> |
| <span class="sourceLineNo">3176</span> if (!exceptions.isEmpty()) {<a name="line.3176"></a> |
| <span class="sourceLineNo">3177</span> errors.reportError(ERROR_CODE.RS_CONNECT_FAILURE, "Table Directory: "<a name="line.3177"></a> |
| <span class="sourceLineNo">3178</span> + tableDir.getPath().getName()<a name="line.3178"></a> |
| <span class="sourceLineNo">3179</span> + " Unable to fetch all HDFS region information. ");<a name="line.3179"></a> |
| <span class="sourceLineNo">3180</span> // Just throw the first exception as an indication something bad happened<a name="line.3180"></a> |
| <span class="sourceLineNo">3181</span> // Don't need to propagate all the exceptions, we already logged them all anyway<a name="line.3181"></a> |
| <span class="sourceLineNo">3182</span> throw new ExecutionException("First exception in WorkItemHdfsDir", exceptions.firstElement());<a name="line.3182"></a> |
| <span class="sourceLineNo">3183</span> }<a name="line.3183"></a> |
| <span class="sourceLineNo">3184</span> }<a name="line.3184"></a> |
| <span class="sourceLineNo">3185</span> return null;<a name="line.3185"></a> |
| <span class="sourceLineNo">3186</span> }<a name="line.3186"></a> |
| <span class="sourceLineNo">3187</span> }<a name="line.3187"></a> |
| <span class="sourceLineNo">3188</span><a name="line.3188"></a> |
| <span class="sourceLineNo">3189</span> /**<a name="line.3189"></a> |
| <span class="sourceLineNo">3190</span> * Contact hdfs and get all information about specified table directory into<a name="line.3190"></a> |
| <span class="sourceLineNo">3191</span> * regioninfo list.<a name="line.3191"></a> |
| <span class="sourceLineNo">3192</span> */<a name="line.3192"></a> |
| <span class="sourceLineNo">3193</span> static class WorkItemHdfsRegionInfo implements Callable<Void> {<a name="line.3193"></a> |
| <span class="sourceLineNo">3194</span> private HbckRegionInfo hbi;<a name="line.3194"></a> |
| <span class="sourceLineNo">3195</span> private HBaseFsck hbck;<a name="line.3195"></a> |
| <span class="sourceLineNo">3196</span> private HbckErrorReporter errors;<a name="line.3196"></a> |
| <span class="sourceLineNo">3197</span><a name="line.3197"></a> |
| <span class="sourceLineNo">3198</span> WorkItemHdfsRegionInfo(HbckRegionInfo hbi, HBaseFsck hbck, HbckErrorReporter errors) {<a name="line.3198"></a> |
| <span class="sourceLineNo">3199</span> this.hbi = hbi;<a name="line.3199"></a> |
| <span class="sourceLineNo">3200</span> this.hbck = hbck;<a name="line.3200"></a> |
| <span class="sourceLineNo">3201</span> this.errors = errors;<a name="line.3201"></a> |
| <span class="sourceLineNo">3202</span> }<a name="line.3202"></a> |
| <span class="sourceLineNo">3203</span><a name="line.3203"></a> |
| <span class="sourceLineNo">3204</span> @Override<a name="line.3204"></a> |
| <span class="sourceLineNo">3205</span> public synchronized Void call() throws IOException {<a name="line.3205"></a> |
| <span class="sourceLineNo">3206</span> // only load entries that haven't been loaded yet.<a name="line.3206"></a> |
| <span class="sourceLineNo">3207</span> if (hbi.getHdfsHRI() == null) {<a name="line.3207"></a> |
| <span class="sourceLineNo">3208</span> try {<a name="line.3208"></a> |
| <span class="sourceLineNo">3209</span> errors.progress();<a name="line.3209"></a> |
| <span class="sourceLineNo">3210</span> hbi.loadHdfsRegioninfo(hbck.getConf());<a name="line.3210"></a> |
| <span class="sourceLineNo">3211</span> } catch (IOException ioe) {<a name="line.3211"></a> |
| <span class="sourceLineNo">3212</span> String msg = "Orphan region in HDFS: Unable to load .regioninfo from table "<a name="line.3212"></a> |
| <span class="sourceLineNo">3213</span> + hbi.getTableName() + " in hdfs dir "<a name="line.3213"></a> |
| <span class="sourceLineNo">3214</span> + hbi.getHdfsRegionDir()<a name="line.3214"></a> |
| <span class="sourceLineNo">3215</span> + "! It may be an invalid format or version file. Treating as "<a name="line.3215"></a> |
| <span class="sourceLineNo">3216</span> + "an orphaned regiondir.";<a name="line.3216"></a> |
| <span class="sourceLineNo">3217</span> errors.reportError(ERROR_CODE.ORPHAN_HDFS_REGION, msg);<a name="line.3217"></a> |
| <span class="sourceLineNo">3218</span> try {<a name="line.3218"></a> |
| <span class="sourceLineNo">3219</span> hbck.debugLsr(hbi.getHdfsRegionDir());<a name="line.3219"></a> |
| <span class="sourceLineNo">3220</span> } catch (IOException ioe2) {<a name="line.3220"></a> |
| <span class="sourceLineNo">3221</span> LOG.error("Unable to read directory " + hbi.getHdfsRegionDir(), ioe2);<a name="line.3221"></a> |
| <span class="sourceLineNo">3222</span> throw ioe2;<a name="line.3222"></a> |
| <span class="sourceLineNo">3223</span> }<a name="line.3223"></a> |
| <span class="sourceLineNo">3224</span> hbck.orphanHdfsDirs.add(hbi);<a name="line.3224"></a> |
| <span class="sourceLineNo">3225</span> throw ioe;<a name="line.3225"></a> |
| <span class="sourceLineNo">3226</span> }<a name="line.3226"></a> |
| <span class="sourceLineNo">3227</span> }<a name="line.3227"></a> |
| <span class="sourceLineNo">3228</span> return null;<a name="line.3228"></a> |
| <span class="sourceLineNo">3229</span> }<a name="line.3229"></a> |
| <span class="sourceLineNo">3230</span> }<a name="line.3230"></a> |
| <span class="sourceLineNo">3231</span><a name="line.3231"></a> |
| <span class="sourceLineNo">3232</span> /**<a name="line.3232"></a> |
| <span class="sourceLineNo">3233</span> * Display the full report from fsck. This displays all live and dead region<a name="line.3233"></a> |
| <span class="sourceLineNo">3234</span> * servers, and all known regions.<a name="line.3234"></a> |
| <span class="sourceLineNo">3235</span> */<a name="line.3235"></a> |
| <span class="sourceLineNo">3236</span> public static void setDisplayFullReport() {<a name="line.3236"></a> |
| <span class="sourceLineNo">3237</span> details = true;<a name="line.3237"></a> |
| <span class="sourceLineNo">3238</span> }<a name="line.3238"></a> |
| <span class="sourceLineNo">3239</span><a name="line.3239"></a> |
| <span class="sourceLineNo">3240</span> public static boolean shouldDisplayFullReport() {<a name="line.3240"></a> |
| <span class="sourceLineNo">3241</span> return details;<a name="line.3241"></a> |
| <span class="sourceLineNo">3242</span> }<a name="line.3242"></a> |
| <span class="sourceLineNo">3243</span><a name="line.3243"></a> |
| <span class="sourceLineNo">3244</span> /**<a name="line.3244"></a> |
| <span class="sourceLineNo">3245</span> * Set exclusive mode.<a name="line.3245"></a> |
| <span class="sourceLineNo">3246</span> */<a name="line.3246"></a> |
| <span class="sourceLineNo">3247</span> public static void setForceExclusive() {<a name="line.3247"></a> |
| <span class="sourceLineNo">3248</span> forceExclusive = true;<a name="line.3248"></a> |
| <span class="sourceLineNo">3249</span> }<a name="line.3249"></a> |
| <span class="sourceLineNo">3250</span><a name="line.3250"></a> |
| <span class="sourceLineNo">3251</span> /**<a name="line.3251"></a> |
| <span class="sourceLineNo">3252</span> * Only one instance of hbck can modify HBase at a time.<a name="line.3252"></a> |
| <span class="sourceLineNo">3253</span> */<a name="line.3253"></a> |
| <span class="sourceLineNo">3254</span> public boolean isExclusive() {<a name="line.3254"></a> |
| <span class="sourceLineNo">3255</span> return fixAny || forceExclusive;<a name="line.3255"></a> |
| <span class="sourceLineNo">3256</span> }<a name="line.3256"></a> |
| <span class="sourceLineNo">3257</span><a name="line.3257"></a> |
| <span class="sourceLineNo">3258</span> /**<a name="line.3258"></a> |
| <span class="sourceLineNo">3259</span> * Set summary mode.<a name="line.3259"></a> |
| <span class="sourceLineNo">3260</span> * Print only summary of the tables and status (OK or INCONSISTENT)<a name="line.3260"></a> |
| <span class="sourceLineNo">3261</span> */<a name="line.3261"></a> |
| <span class="sourceLineNo">3262</span> static void setSummary() {<a name="line.3262"></a> |
| <span class="sourceLineNo">3263</span> summary = true;<a name="line.3263"></a> |
| <span class="sourceLineNo">3264</span> }<a name="line.3264"></a> |
| <span class="sourceLineNo">3265</span><a name="line.3265"></a> |
| <span class="sourceLineNo">3266</span> /**<a name="line.3266"></a> |
| <span class="sourceLineNo">3267</span> * Set hbase:meta check mode.<a name="line.3267"></a> |
| <span class="sourceLineNo">3268</span> * Print only info about hbase:meta table deployment/state<a name="line.3268"></a> |
| <span class="sourceLineNo">3269</span> */<a name="line.3269"></a> |
| <span class="sourceLineNo">3270</span> void setCheckMetaOnly() {<a name="line.3270"></a> |
| <span class="sourceLineNo">3271</span> checkMetaOnly = true;<a name="line.3271"></a> |
| <span class="sourceLineNo">3272</span> }<a name="line.3272"></a> |
| <span class="sourceLineNo">3273</span><a name="line.3273"></a> |
| <span class="sourceLineNo">3274</span> /**<a name="line.3274"></a> |
| <span class="sourceLineNo">3275</span> * Set region boundaries check mode.<a name="line.3275"></a> |
| <span class="sourceLineNo">3276</span> */<a name="line.3276"></a> |
| <span class="sourceLineNo">3277</span> void setRegionBoundariesCheck() {<a name="line.3277"></a> |
| <span class="sourceLineNo">3278</span> checkRegionBoundaries = true;<a name="line.3278"></a> |
| <span class="sourceLineNo">3279</span> }<a name="line.3279"></a> |
| <span class="sourceLineNo">3280</span><a name="line.3280"></a> |
| <span class="sourceLineNo">3281</span> /**<a name="line.3281"></a> |
| <span class="sourceLineNo">3282</span> * Set replication fix mode.<a name="line.3282"></a> |
| <span class="sourceLineNo">3283</span> */<a name="line.3283"></a> |
| <span class="sourceLineNo">3284</span> public void setFixReplication(boolean shouldFix) {<a name="line.3284"></a> |
| <span class="sourceLineNo">3285</span> fixReplication = shouldFix;<a name="line.3285"></a> |
| <span class="sourceLineNo">3286</span> fixAny |= shouldFix;<a name="line.3286"></a> |
| <span class="sourceLineNo">3287</span> }<a name="line.3287"></a> |
| <span class="sourceLineNo">3288</span><a name="line.3288"></a> |
| <span class="sourceLineNo">3289</span> public void setCleanReplicationBarrier(boolean shouldClean) {<a name="line.3289"></a> |
| <span class="sourceLineNo">3290</span> cleanReplicationBarrier = shouldClean;<a name="line.3290"></a> |
| <span class="sourceLineNo">3291</span> }<a name="line.3291"></a> |
| <span class="sourceLineNo">3292</span><a name="line.3292"></a> |
| <span class="sourceLineNo">3293</span> /**<a name="line.3293"></a> |
| <span class="sourceLineNo">3294</span> * Check if we should rerun fsck again. This checks if we've tried to<a name="line.3294"></a> |
| <span class="sourceLineNo">3295</span> * fix something and we should rerun fsck tool again.<a name="line.3295"></a> |
| <span class="sourceLineNo">3296</span> * Display the full report from fsck. This displays all live and dead<a name="line.3296"></a> |
| <span class="sourceLineNo">3297</span> * region servers, and all known regions.<a name="line.3297"></a> |
| <span class="sourceLineNo">3298</span> */<a name="line.3298"></a> |
| <span class="sourceLineNo">3299</span> void setShouldRerun() {<a name="line.3299"></a> |
| <span class="sourceLineNo">3300</span> rerun = true;<a name="line.3300"></a> |
| <span class="sourceLineNo">3301</span> }<a name="line.3301"></a> |
| <span class="sourceLineNo">3302</span><a name="line.3302"></a> |
| <span class="sourceLineNo">3303</span> public boolean shouldRerun() {<a name="line.3303"></a> |
| <span class="sourceLineNo">3304</span> return rerun;<a name="line.3304"></a> |
| <span class="sourceLineNo">3305</span> }<a name="line.3305"></a> |
| <span class="sourceLineNo">3306</span><a name="line.3306"></a> |
| <span class="sourceLineNo">3307</span> /**<a name="line.3307"></a> |
| <span class="sourceLineNo">3308</span> * Fix inconsistencies found by fsck. This should try to fix errors (if any)<a name="line.3308"></a> |
| <span class="sourceLineNo">3309</span> * found by fsck utility.<a name="line.3309"></a> |
| <span class="sourceLineNo">3310</span> */<a name="line.3310"></a> |
| <span class="sourceLineNo">3311</span> public void setFixAssignments(boolean shouldFix) {<a name="line.3311"></a> |
| <span class="sourceLineNo">3312</span> fixAssignments = shouldFix;<a name="line.3312"></a> |
| <span class="sourceLineNo">3313</span> fixAny |= shouldFix;<a name="line.3313"></a> |
| <span class="sourceLineNo">3314</span> }<a name="line.3314"></a> |
| <span class="sourceLineNo">3315</span><a name="line.3315"></a> |
| <span class="sourceLineNo">3316</span> boolean shouldFixAssignments() {<a name="line.3316"></a> |
| <span class="sourceLineNo">3317</span> return fixAssignments;<a name="line.3317"></a> |
| <span class="sourceLineNo">3318</span> }<a name="line.3318"></a> |
| <span class="sourceLineNo">3319</span><a name="line.3319"></a> |
| <span class="sourceLineNo">3320</span> public void setFixMeta(boolean shouldFix) {<a name="line.3320"></a> |
| <span class="sourceLineNo">3321</span> fixMeta = shouldFix;<a name="line.3321"></a> |
| <span class="sourceLineNo">3322</span> fixAny |= shouldFix;<a name="line.3322"></a> |
| <span class="sourceLineNo">3323</span> }<a name="line.3323"></a> |
| <span class="sourceLineNo">3324</span><a name="line.3324"></a> |
| <span class="sourceLineNo">3325</span> boolean shouldFixMeta() {<a name="line.3325"></a> |
| <span class="sourceLineNo">3326</span> return fixMeta;<a name="line.3326"></a> |
| <span class="sourceLineNo">3327</span> }<a name="line.3327"></a> |
| <span class="sourceLineNo">3328</span><a name="line.3328"></a> |
| <span class="sourceLineNo">3329</span> public void setFixEmptyMetaCells(boolean shouldFix) {<a name="line.3329"></a> |
| <span class="sourceLineNo">3330</span> fixEmptyMetaCells = shouldFix;<a name="line.3330"></a> |
| <span class="sourceLineNo">3331</span> fixAny |= shouldFix;<a name="line.3331"></a> |
| <span class="sourceLineNo">3332</span> }<a name="line.3332"></a> |
| <span class="sourceLineNo">3333</span><a name="line.3333"></a> |
| <span class="sourceLineNo">3334</span> boolean shouldFixEmptyMetaCells() {<a name="line.3334"></a> |
| <span class="sourceLineNo">3335</span> return fixEmptyMetaCells;<a name="line.3335"></a> |
| <span class="sourceLineNo">3336</span> }<a name="line.3336"></a> |
| <span class="sourceLineNo">3337</span><a name="line.3337"></a> |
| <span class="sourceLineNo">3338</span> public void setCheckHdfs(boolean checking) {<a name="line.3338"></a> |
| <span class="sourceLineNo">3339</span> checkHdfs = checking;<a name="line.3339"></a> |
| <span class="sourceLineNo">3340</span> }<a name="line.3340"></a> |
| <span class="sourceLineNo">3341</span><a name="line.3341"></a> |
| <span class="sourceLineNo">3342</span> boolean shouldCheckHdfs() {<a name="line.3342"></a> |
| <span class="sourceLineNo">3343</span> return checkHdfs;<a name="line.3343"></a> |
| <span class="sourceLineNo">3344</span> }<a name="line.3344"></a> |
| <span class="sourceLineNo">3345</span><a name="line.3345"></a> |
| <span class="sourceLineNo">3346</span> public void setFixHdfsHoles(boolean shouldFix) {<a name="line.3346"></a> |
| <span class="sourceLineNo">3347</span> fixHdfsHoles = shouldFix;<a name="line.3347"></a> |
| <span class="sourceLineNo">3348</span> fixAny |= shouldFix;<a name="line.3348"></a> |
| <span class="sourceLineNo">3349</span> }<a name="line.3349"></a> |
| <span class="sourceLineNo">3350</span><a name="line.3350"></a> |
| <span class="sourceLineNo">3351</span> boolean shouldFixHdfsHoles() {<a name="line.3351"></a> |
| <span class="sourceLineNo">3352</span> return fixHdfsHoles;<a name="line.3352"></a> |
| <span class="sourceLineNo">3353</span> }<a name="line.3353"></a> |
| <span class="sourceLineNo">3354</span><a name="line.3354"></a> |
| <span class="sourceLineNo">3355</span> public void setFixTableOrphans(boolean shouldFix) {<a name="line.3355"></a> |
| <span class="sourceLineNo">3356</span> fixTableOrphans = shouldFix;<a name="line.3356"></a> |
| <span class="sourceLineNo">3357</span> fixAny |= shouldFix;<a name="line.3357"></a> |
| <span class="sourceLineNo">3358</span> }<a name="line.3358"></a> |
| <span class="sourceLineNo">3359</span><a name="line.3359"></a> |
| <span class="sourceLineNo">3360</span> boolean shouldFixTableOrphans() {<a name="line.3360"></a> |
| <span class="sourceLineNo">3361</span> return fixTableOrphans;<a name="line.3361"></a> |
| <span class="sourceLineNo">3362</span> }<a name="line.3362"></a> |
| <span class="sourceLineNo">3363</span><a name="line.3363"></a> |
| <span class="sourceLineNo">3364</span> public void setFixHdfsOverlaps(boolean shouldFix) {<a name="line.3364"></a> |
| <span class="sourceLineNo">3365</span> fixHdfsOverlaps = shouldFix;<a name="line.3365"></a> |
| <span class="sourceLineNo">3366</span> fixAny |= shouldFix;<a name="line.3366"></a> |
| <span class="sourceLineNo">3367</span> }<a name="line.3367"></a> |
| <span class="sourceLineNo">3368</span><a name="line.3368"></a> |
| <span class="sourceLineNo">3369</span> boolean shouldFixHdfsOverlaps() {<a name="line.3369"></a> |
| <span class="sourceLineNo">3370</span> return fixHdfsOverlaps;<a name="line.3370"></a> |
| <span class="sourceLineNo">3371</span> }<a name="line.3371"></a> |
| <span class="sourceLineNo">3372</span><a name="line.3372"></a> |
| <span class="sourceLineNo">3373</span> public void setFixHdfsOrphans(boolean shouldFix) {<a name="line.3373"></a> |
| <span class="sourceLineNo">3374</span> fixHdfsOrphans = shouldFix;<a name="line.3374"></a> |
| <span class="sourceLineNo">3375</span> fixAny |= shouldFix;<a name="line.3375"></a> |
| <span class="sourceLineNo">3376</span> }<a name="line.3376"></a> |
| <span class="sourceLineNo">3377</span><a name="line.3377"></a> |
| <span class="sourceLineNo">3378</span> boolean shouldFixHdfsOrphans() {<a name="line.3378"></a> |
| <span class="sourceLineNo">3379</span> return fixHdfsOrphans;<a name="line.3379"></a> |
| <span class="sourceLineNo">3380</span> }<a name="line.3380"></a> |
| <span class="sourceLineNo">3381</span><a name="line.3381"></a> |
| <span class="sourceLineNo">3382</span> public void setFixVersionFile(boolean shouldFix) {<a name="line.3382"></a> |
| <span class="sourceLineNo">3383</span> fixVersionFile = shouldFix;<a name="line.3383"></a> |
| <span class="sourceLineNo">3384</span> fixAny |= shouldFix;<a name="line.3384"></a> |
| <span class="sourceLineNo">3385</span> }<a name="line.3385"></a> |
| <span class="sourceLineNo">3386</span><a name="line.3386"></a> |
| <span class="sourceLineNo">3387</span> public boolean shouldFixVersionFile() {<a name="line.3387"></a> |
| <span class="sourceLineNo">3388</span> return fixVersionFile;<a name="line.3388"></a> |
| <span class="sourceLineNo">3389</span> }<a name="line.3389"></a> |
| <span class="sourceLineNo">3390</span><a name="line.3390"></a> |
| <span class="sourceLineNo">3391</span> public void setSidelineBigOverlaps(boolean sbo) {<a name="line.3391"></a> |
| <span class="sourceLineNo">3392</span> this.sidelineBigOverlaps = sbo;<a name="line.3392"></a> |
| <span class="sourceLineNo">3393</span> }<a name="line.3393"></a> |
| <span class="sourceLineNo">3394</span><a name="line.3394"></a> |
| <span class="sourceLineNo">3395</span> public boolean shouldSidelineBigOverlaps() {<a name="line.3395"></a> |
| <span class="sourceLineNo">3396</span> return sidelineBigOverlaps;<a name="line.3396"></a> |
| <span class="sourceLineNo">3397</span> }<a name="line.3397"></a> |
| <span class="sourceLineNo">3398</span><a name="line.3398"></a> |
| <span class="sourceLineNo">3399</span> public void setFixSplitParents(boolean shouldFix) {<a name="line.3399"></a> |
| <span class="sourceLineNo">3400</span> fixSplitParents = shouldFix;<a name="line.3400"></a> |
| <span class="sourceLineNo">3401</span> fixAny |= shouldFix;<a name="line.3401"></a> |
| <span class="sourceLineNo">3402</span> }<a name="line.3402"></a> |
| <span class="sourceLineNo">3403</span><a name="line.3403"></a> |
| <span class="sourceLineNo">3404</span> public void setRemoveParents(boolean shouldFix) {<a name="line.3404"></a> |
| <span class="sourceLineNo">3405</span> removeParents = shouldFix;<a name="line.3405"></a> |
| <span class="sourceLineNo">3406</span> fixAny |= shouldFix;<a name="line.3406"></a> |
| <span class="sourceLineNo">3407</span> }<a name="line.3407"></a> |
| <span class="sourceLineNo">3408</span><a name="line.3408"></a> |
| <span class="sourceLineNo">3409</span> boolean shouldFixSplitParents() {<a name="line.3409"></a> |
| <span class="sourceLineNo">3410</span> return fixSplitParents;<a name="line.3410"></a> |
| <span class="sourceLineNo">3411</span> }<a name="line.3411"></a> |
| <span class="sourceLineNo">3412</span><a name="line.3412"></a> |
| <span class="sourceLineNo">3413</span> boolean shouldRemoveParents() {<a name="line.3413"></a> |
| <span class="sourceLineNo">3414</span> return removeParents;<a name="line.3414"></a> |
| <span class="sourceLineNo">3415</span> }<a name="line.3415"></a> |
| <span class="sourceLineNo">3416</span><a name="line.3416"></a> |
| <span class="sourceLineNo">3417</span> public void setFixReferenceFiles(boolean shouldFix) {<a name="line.3417"></a> |
| <span class="sourceLineNo">3418</span> fixReferenceFiles = shouldFix;<a name="line.3418"></a> |
| <span class="sourceLineNo">3419</span> fixAny |= shouldFix;<a name="line.3419"></a> |
| <span class="sourceLineNo">3420</span> }<a name="line.3420"></a> |
| <span class="sourceLineNo">3421</span><a name="line.3421"></a> |
| <span class="sourceLineNo">3422</span> boolean shouldFixReferenceFiles() {<a name="line.3422"></a> |
| <span class="sourceLineNo">3423</span> return fixReferenceFiles;<a name="line.3423"></a> |
| <span class="sourceLineNo">3424</span> }<a name="line.3424"></a> |
| <span class="sourceLineNo">3425</span><a name="line.3425"></a> |
| <span class="sourceLineNo">3426</span> public void setFixHFileLinks(boolean shouldFix) {<a name="line.3426"></a> |
| <span class="sourceLineNo">3427</span> fixHFileLinks = shouldFix;<a name="line.3427"></a> |
| <span class="sourceLineNo">3428</span> fixAny |= shouldFix;<a name="line.3428"></a> |
| <span class="sourceLineNo">3429</span> }<a name="line.3429"></a> |
| <span class="sourceLineNo">3430</span><a name="line.3430"></a> |
| <span class="sourceLineNo">3431</span> boolean shouldFixHFileLinks() {<a name="line.3431"></a> |
| <span class="sourceLineNo">3432</span> return fixHFileLinks;<a name="line.3432"></a> |
| <span class="sourceLineNo">3433</span> }<a name="line.3433"></a> |
| <span class="sourceLineNo">3434</span><a name="line.3434"></a> |
| <span class="sourceLineNo">3435</span> public boolean shouldIgnorePreCheckPermission() {<a name="line.3435"></a> |
| <span class="sourceLineNo">3436</span> return !fixAny || ignorePreCheckPermission;<a name="line.3436"></a> |
| <span class="sourceLineNo">3437</span> }<a name="line.3437"></a> |
| <span class="sourceLineNo">3438</span><a name="line.3438"></a> |
| <span class="sourceLineNo">3439</span> public void setIgnorePreCheckPermission(boolean ignorePreCheckPermission) {<a name="line.3439"></a> |
| <span class="sourceLineNo">3440</span> this.ignorePreCheckPermission = ignorePreCheckPermission;<a name="line.3440"></a> |
| <span class="sourceLineNo">3441</span> }<a name="line.3441"></a> |
| <span class="sourceLineNo">3442</span><a name="line.3442"></a> |
| <span class="sourceLineNo">3443</span> /**<a name="line.3443"></a> |
| <span class="sourceLineNo">3444</span> * @param mm maximum number of regions to merge into a single region.<a name="line.3444"></a> |
| <span class="sourceLineNo">3445</span> */<a name="line.3445"></a> |
| <span class="sourceLineNo">3446</span> public void setMaxMerge(int mm) {<a name="line.3446"></a> |
| <span class="sourceLineNo">3447</span> this.maxMerge = mm;<a name="line.3447"></a> |
| <span class="sourceLineNo">3448</span> }<a name="line.3448"></a> |
| <span class="sourceLineNo">3449</span><a name="line.3449"></a> |
| <span class="sourceLineNo">3450</span> public int getMaxMerge() {<a name="line.3450"></a> |
| <span class="sourceLineNo">3451</span> return maxMerge;<a name="line.3451"></a> |
| <span class="sourceLineNo">3452</span> }<a name="line.3452"></a> |
| <span class="sourceLineNo">3453</span><a name="line.3453"></a> |
| <span class="sourceLineNo">3454</span> public void setMaxOverlapsToSideline(int mo) {<a name="line.3454"></a> |
| <span class="sourceLineNo">3455</span> this.maxOverlapsToSideline = mo;<a name="line.3455"></a> |
| <span class="sourceLineNo">3456</span> }<a name="line.3456"></a> |
| <span class="sourceLineNo">3457</span><a name="line.3457"></a> |
| <span class="sourceLineNo">3458</span> public int getMaxOverlapsToSideline() {<a name="line.3458"></a> |
| <span class="sourceLineNo">3459</span> return maxOverlapsToSideline;<a name="line.3459"></a> |
| <span class="sourceLineNo">3460</span> }<a name="line.3460"></a> |
| <span class="sourceLineNo">3461</span><a name="line.3461"></a> |
| <span class="sourceLineNo">3462</span> /**<a name="line.3462"></a> |
| <span class="sourceLineNo">3463</span> * Only check/fix tables specified by the list,<a name="line.3463"></a> |
| <span class="sourceLineNo">3464</span> * Empty list means all tables are included.<a name="line.3464"></a> |
| <span class="sourceLineNo">3465</span> */<a name="line.3465"></a> |
| <span class="sourceLineNo">3466</span> boolean isTableIncluded(TableName table) {<a name="line.3466"></a> |
| <span class="sourceLineNo">3467</span> return (tablesIncluded.isEmpty()) || tablesIncluded.contains(table);<a name="line.3467"></a> |
| <span class="sourceLineNo">3468</span> }<a name="line.3468"></a> |
| <span class="sourceLineNo">3469</span><a name="line.3469"></a> |
| <span class="sourceLineNo">3470</span> public void includeTable(TableName table) {<a name="line.3470"></a> |
| <span class="sourceLineNo">3471</span> tablesIncluded.add(table);<a name="line.3471"></a> |
| <span class="sourceLineNo">3472</span> }<a name="line.3472"></a> |
| <span class="sourceLineNo">3473</span><a name="line.3473"></a> |
| <span class="sourceLineNo">3474</span> Set<TableName> getIncludedTables() {<a name="line.3474"></a> |
| <span class="sourceLineNo">3475</span> return new HashSet<>(tablesIncluded);<a name="line.3475"></a> |
| <span class="sourceLineNo">3476</span> }<a name="line.3476"></a> |
| <span class="sourceLineNo">3477</span><a name="line.3477"></a> |
| <span class="sourceLineNo">3478</span> /**<a name="line.3478"></a> |
| <span class="sourceLineNo">3479</span> * We are interested in only those tables that have not changed their state in<a name="line.3479"></a> |
| <span class="sourceLineNo">3480</span> * hbase:meta during the last few seconds specified by hbase.admin.fsck.timelag<a name="line.3480"></a> |
| <span class="sourceLineNo">3481</span> * @param seconds - the time in seconds<a name="line.3481"></a> |
| <span class="sourceLineNo">3482</span> */<a name="line.3482"></a> |
| <span class="sourceLineNo">3483</span> public void setTimeLag(long seconds) {<a name="line.3483"></a> |
| <span class="sourceLineNo">3484</span> timelag = seconds * 1000; // convert to milliseconds<a name="line.3484"></a> |
| <span class="sourceLineNo">3485</span> }<a name="line.3485"></a> |
| <span class="sourceLineNo">3486</span><a name="line.3486"></a> |
| <span class="sourceLineNo">3487</span> /**<a name="line.3487"></a> |
| <span class="sourceLineNo">3488</span> *<a name="line.3488"></a> |
| <span class="sourceLineNo">3489</span> * @param sidelineDir - HDFS path to sideline data<a name="line.3489"></a> |
| <span class="sourceLineNo">3490</span> */<a name="line.3490"></a> |
| <span class="sourceLineNo">3491</span> public void setSidelineDir(String sidelineDir) {<a name="line.3491"></a> |
| <span class="sourceLineNo">3492</span> this.sidelineDir = new Path(sidelineDir);<a name="line.3492"></a> |
| <span class="sourceLineNo">3493</span> }<a name="line.3493"></a> |
| <span class="sourceLineNo">3494</span><a name="line.3494"></a> |
| <span class="sourceLineNo">3495</span> protected HFileCorruptionChecker createHFileCorruptionChecker(boolean sidelineCorruptHFiles) throws IOException {<a name="line.3495"></a> |
| <span class="sourceLineNo">3496</span> return new HFileCorruptionChecker(getConf(), executor, sidelineCorruptHFiles);<a name="line.3496"></a> |
| <span class="sourceLineNo">3497</span> }<a name="line.3497"></a> |
| <span class="sourceLineNo">3498</span><a name="line.3498"></a> |
| <span class="sourceLineNo">3499</span> public HFileCorruptionChecker getHFilecorruptionChecker() {<a name="line.3499"></a> |
| <span class="sourceLineNo">3500</span> return hfcc;<a name="line.3500"></a> |
| <span class="sourceLineNo">3501</span> }<a name="line.3501"></a> |
| <span class="sourceLineNo">3502</span><a name="line.3502"></a> |
| <span class="sourceLineNo">3503</span> public void setHFileCorruptionChecker(HFileCorruptionChecker hfcc) {<a name="line.3503"></a> |
| <span class="sourceLineNo">3504</span> this.hfcc = hfcc;<a name="line.3504"></a> |
| <span class="sourceLineNo">3505</span> }<a name="line.3505"></a> |
| <span class="sourceLineNo">3506</span><a name="line.3506"></a> |
| <span class="sourceLineNo">3507</span> public void setRetCode(int code) {<a name="line.3507"></a> |
| <span class="sourceLineNo">3508</span> this.retcode = code;<a name="line.3508"></a> |
| <span class="sourceLineNo">3509</span> }<a name="line.3509"></a> |
| <span class="sourceLineNo">3510</span><a name="line.3510"></a> |
| <span class="sourceLineNo">3511</span> public int getRetCode() {<a name="line.3511"></a> |
| <span class="sourceLineNo">3512</span> return retcode;<a name="line.3512"></a> |
| <span class="sourceLineNo">3513</span> }<a name="line.3513"></a> |
| <span class="sourceLineNo">3514</span><a name="line.3514"></a> |
| <span class="sourceLineNo">3515</span> protected HBaseFsck printUsageAndExit() {<a name="line.3515"></a> |
| <span class="sourceLineNo">3516</span> StringWriter sw = new StringWriter(2048);<a name="line.3516"></a> |
| <span class="sourceLineNo">3517</span> PrintWriter out = new PrintWriter(sw);<a name="line.3517"></a> |
| <span class="sourceLineNo">3518</span> out.println("");<a name="line.3518"></a> |
| <span class="sourceLineNo">3519</span> out.println("-----------------------------------------------------------------------");<a name="line.3519"></a> |
| <span class="sourceLineNo">3520</span> out.println("NOTE: As of HBase version 2.0, the hbck tool is significantly changed.");<a name="line.3520"></a> |
| <span class="sourceLineNo">3521</span> out.println("In general, all Read-Only options are supported and can be be used");<a name="line.3521"></a> |
| <span class="sourceLineNo">3522</span> out.println("safely. Most -fix/ -repair options are NOT supported. Please see usage");<a name="line.3522"></a> |
| <span class="sourceLineNo">3523</span> out.println("below for details on which options are not supported.");<a name="line.3523"></a> |
| <span class="sourceLineNo">3524</span> out.println("-----------------------------------------------------------------------");<a name="line.3524"></a> |
| <span class="sourceLineNo">3525</span> out.println("");<a name="line.3525"></a> |
| <span class="sourceLineNo">3526</span> out.println("Usage: fsck [opts] {only tables}");<a name="line.3526"></a> |
| <span class="sourceLineNo">3527</span> out.println(" where [opts] are:");<a name="line.3527"></a> |
| <span class="sourceLineNo">3528</span> out.println(" -help Display help options (this)");<a name="line.3528"></a> |
| <span class="sourceLineNo">3529</span> out.println(" -details Display full report of all regions.");<a name="line.3529"></a> |
| <span class="sourceLineNo">3530</span> out.println(" -timelag <timeInSeconds> Process only regions that " +<a name="line.3530"></a> |
| <span class="sourceLineNo">3531</span> " have not experienced any metadata updates in the last " +<a name="line.3531"></a> |
| <span class="sourceLineNo">3532</span> " <timeInSeconds> seconds.");<a name="line.3532"></a> |
| <span class="sourceLineNo">3533</span> out.println(" -sleepBeforeRerun <timeInSeconds> Sleep this many seconds" +<a name="line.3533"></a> |
| <span class="sourceLineNo">3534</span> " before checking if the fix worked if run with -fix");<a name="line.3534"></a> |
| <span class="sourceLineNo">3535</span> out.println(" -summary Print only summary of the tables and status.");<a name="line.3535"></a> |
| <span class="sourceLineNo">3536</span> out.println(" -metaonly Only check the state of the hbase:meta table.");<a name="line.3536"></a> |
| <span class="sourceLineNo">3537</span> out.println(" -sidelineDir <hdfs://> HDFS path to backup existing meta.");<a name="line.3537"></a> |
| <span class="sourceLineNo">3538</span> out.println(" -boundaries Verify that regions boundaries are the same between META and store files.");<a name="line.3538"></a> |
| <span class="sourceLineNo">3539</span> out.println(" -exclusive Abort if another hbck is exclusive or fixing.");<a name="line.3539"></a> |
| <span class="sourceLineNo">3540</span><a name="line.3540"></a> |
| <span class="sourceLineNo">3541</span> out.println("");<a name="line.3541"></a> |
| <span class="sourceLineNo">3542</span> out.println(" Datafile Repair options: (expert features, use with caution!)");<a name="line.3542"></a> |
| <span class="sourceLineNo">3543</span> out.println(" -checkCorruptHFiles Check all Hfiles by opening them to make sure they are valid");<a name="line.3543"></a> |
| <span class="sourceLineNo">3544</span> out.println(" -sidelineCorruptHFiles Quarantine corrupted HFiles. implies -checkCorruptHFiles");<a name="line.3544"></a> |
| <span class="sourceLineNo">3545</span><a name="line.3545"></a> |
| <span class="sourceLineNo">3546</span> out.println("");<a name="line.3546"></a> |
| <span class="sourceLineNo">3547</span> out.println(" Replication options");<a name="line.3547"></a> |
| <span class="sourceLineNo">3548</span> out.println(" -fixReplication Deletes replication queues for removed peers");<a name="line.3548"></a> |
| <span class="sourceLineNo">3549</span><a name="line.3549"></a> |
| <span class="sourceLineNo">3550</span> out.println("");<a name="line.3550"></a> |
| <span class="sourceLineNo">3551</span> out.println(" Metadata Repair options supported as of version 2.0: (expert features, use with caution!)");<a name="line.3551"></a> |
| <span class="sourceLineNo">3552</span> out.println(" -fixVersionFile Try to fix missing hbase.version file in hdfs.");<a name="line.3552"></a> |
| <span class="sourceLineNo">3553</span> out.println(" -fixReferenceFiles Try to offline lingering reference store files");<a name="line.3553"></a> |
| <span class="sourceLineNo">3554</span> out.println(" -fixHFileLinks Try to offline lingering HFileLinks");<a name="line.3554"></a> |
| <span class="sourceLineNo">3555</span> out.println(" -noHdfsChecking Don't load/check region info from HDFS."<a name="line.3555"></a> |
| <span class="sourceLineNo">3556</span> + " Assumes hbase:meta region info is good. Won't check/fix any HDFS issue, e.g. hole, orphan, or overlap");<a name="line.3556"></a> |
| <span class="sourceLineNo">3557</span> out.println(" -ignorePreCheckPermission ignore filesystem permission pre-check");<a name="line.3557"></a> |
| <span class="sourceLineNo">3558</span><a name="line.3558"></a> |
| <span class="sourceLineNo">3559</span> out.println("");<a name="line.3559"></a> |
| <span class="sourceLineNo">3560</span> out.println("NOTE: Following options are NOT supported as of HBase version 2.0+.");<a name="line.3560"></a> |
| <span class="sourceLineNo">3561</span> out.println("");<a name="line.3561"></a> |
| <span class="sourceLineNo">3562</span> out.println(" UNSUPPORTED Metadata Repair options: (expert features, use with caution!)");<a name="line.3562"></a> |
| <span class="sourceLineNo">3563</span> out.println(" -fix Try to fix region assignments. This is for backwards compatiblity");<a name="line.3563"></a> |
| <span class="sourceLineNo">3564</span> out.println(" -fixAssignments Try to fix region assignments. Replaces the old -fix");<a name="line.3564"></a> |
| <span class="sourceLineNo">3565</span> out.println(" -fixMeta Try to fix meta problems. This assumes HDFS region info is good.");<a name="line.3565"></a> |
| <span class="sourceLineNo">3566</span> out.println(" -fixHdfsHoles Try to fix region holes in hdfs.");<a name="line.3566"></a> |
| <span class="sourceLineNo">3567</span> out.println(" -fixHdfsOrphans Try to fix region dirs with no .regioninfo file in hdfs");<a name="line.3567"></a> |
| <span class="sourceLineNo">3568</span> out.println(" -fixTableOrphans Try to fix table dirs with no .tableinfo file in hdfs (online mode only)");<a name="line.3568"></a> |
| <span class="sourceLineNo">3569</span> out.println(" -fixHdfsOverlaps Try to fix region overlaps in hdfs.");<a name="line.3569"></a> |
| <span class="sourceLineNo">3570</span> out.println(" -maxMerge <n> When fixing region overlaps, allow at most <n> regions to merge. (n=" + DEFAULT_MAX_MERGE +" by default)");<a name="line.3570"></a> |
| <span class="sourceLineNo">3571</span> out.println(" -sidelineBigOverlaps When fixing region overlaps, allow to sideline big overlaps");<a name="line.3571"></a> |
| <span class="sourceLineNo">3572</span> out.println(" -maxOverlapsToSideline <n> When fixing region overlaps, allow at most <n> regions to sideline per group. (n=" + DEFAULT_OVERLAPS_TO_SIDELINE +" by default)");<a name="line.3572"></a> |
| <span class="sourceLineNo">3573</span> out.println(" -fixSplitParents Try to force offline split parents to be online.");<a name="line.3573"></a> |
| <span class="sourceLineNo">3574</span> out.println(" -removeParents Try to offline and sideline lingering parents and keep daughter regions.");<a name="line.3574"></a> |
| <span class="sourceLineNo">3575</span> out.println(" -fixEmptyMetaCells Try to fix hbase:meta entries not referencing any region"<a name="line.3575"></a> |
| <span class="sourceLineNo">3576</span> + " (empty REGIONINFO_QUALIFIER rows)");<a name="line.3576"></a> |
| <span class="sourceLineNo">3577</span><a name="line.3577"></a> |
| <span class="sourceLineNo">3578</span> out.println("");<a name="line.3578"></a> |
| <span class="sourceLineNo">3579</span> out.println(" UNSUPPORTED Metadata Repair shortcuts");<a name="line.3579"></a> |
| <span class="sourceLineNo">3580</span> out.println(" -repair Shortcut for -fixAssignments -fixMeta -fixHdfsHoles " +<a name="line.3580"></a> |
| <span class="sourceLineNo">3581</span> "-fixHdfsOrphans -fixHdfsOverlaps -fixVersionFile -sidelineBigOverlaps -fixReferenceFiles" +<a name="line.3581"></a> |
| <span class="sourceLineNo">3582</span> "-fixHFileLinks");<a name="line.3582"></a> |
| <span class="sourceLineNo">3583</span> out.println(" -repairHoles Shortcut for -fixAssignments -fixMeta -fixHdfsHoles");<a name="line.3583"></a> |
| <span class="sourceLineNo">3584</span> out.println("");<a name="line.3584"></a> |
| <span class="sourceLineNo">3585</span> out.println(" Replication options");<a name="line.3585"></a> |
| <span class="sourceLineNo">3586</span> out.println(" -fixReplication Deletes replication queues for removed peers");<a name="line.3586"></a> |
| <span class="sourceLineNo">3587</span> out.println(" -cleanReplicationBrarier [tableName] clean the replication barriers " +<a name="line.3587"></a> |
| <span class="sourceLineNo">3588</span> "of a specified table, tableName is required");<a name="line.3588"></a> |
| <span class="sourceLineNo">3589</span> out.flush();<a name="line.3589"></a> |
| <span class="sourceLineNo">3590</span> errors.reportError(ERROR_CODE.WRONG_USAGE, sw.toString());<a name="line.3590"></a> |
| <span class="sourceLineNo">3591</span><a name="line.3591"></a> |
| <span class="sourceLineNo">3592</span> setRetCode(-2);<a name="line.3592"></a> |
| <span class="sourceLineNo">3593</span> return this;<a name="line.3593"></a> |
| <span class="sourceLineNo">3594</span> }<a name="line.3594"></a> |
| <span class="sourceLineNo">3595</span><a name="line.3595"></a> |
| <span class="sourceLineNo">3596</span> /**<a name="line.3596"></a> |
| <span class="sourceLineNo">3597</span> * Main program<a name="line.3597"></a> |
| <span class="sourceLineNo">3598</span> *<a name="line.3598"></a> |
| <span class="sourceLineNo">3599</span> * @param args<a name="line.3599"></a> |
| <span class="sourceLineNo">3600</span> * @throws Exception<a name="line.3600"></a> |
| <span class="sourceLineNo">3601</span> */<a name="line.3601"></a> |
| <span class="sourceLineNo">3602</span> public static void main(String[] args) throws Exception {<a name="line.3602"></a> |
| <span class="sourceLineNo">3603</span> // create a fsck object<a name="line.3603"></a> |
| <span class="sourceLineNo">3604</span> Configuration conf = HBaseConfiguration.create();<a name="line.3604"></a> |
| <span class="sourceLineNo">3605</span> Path hbasedir = CommonFSUtils.getRootDir(conf);<a name="line.3605"></a> |
| <span class="sourceLineNo">3606</span> URI defaultFs = hbasedir.getFileSystem(conf).getUri();<a name="line.3606"></a> |
| <span class="sourceLineNo">3607</span> CommonFSUtils.setFsDefault(conf, new Path(defaultFs));<a name="line.3607"></a> |
| <span class="sourceLineNo">3608</span> int ret = ToolRunner.run(new HBaseFsckTool(conf), args);<a name="line.3608"></a> |
| <span class="sourceLineNo">3609</span> System.exit(ret);<a name="line.3609"></a> |
| <span class="sourceLineNo">3610</span> }<a name="line.3610"></a> |
| <span class="sourceLineNo">3611</span><a name="line.3611"></a> |
| <span class="sourceLineNo">3612</span> /**<a name="line.3612"></a> |
| <span class="sourceLineNo">3613</span> * This is a Tool wrapper that gathers -Dxxx=yyy configuration settings from the command line.<a name="line.3613"></a> |
| <span class="sourceLineNo">3614</span> */<a name="line.3614"></a> |
| <span class="sourceLineNo">3615</span> static class HBaseFsckTool extends Configured implements Tool {<a name="line.3615"></a> |
| <span class="sourceLineNo">3616</span> HBaseFsckTool(Configuration conf) { super(conf); }<a name="line.3616"></a> |
| <span class="sourceLineNo">3617</span> @Override<a name="line.3617"></a> |
| <span class="sourceLineNo">3618</span> public int run(String[] args) throws Exception {<a name="line.3618"></a> |
| <span class="sourceLineNo">3619</span> HBaseFsck hbck = new HBaseFsck(getConf());<a name="line.3619"></a> |
| <span class="sourceLineNo">3620</span> hbck.exec(hbck.executor, args);<a name="line.3620"></a> |
| <span class="sourceLineNo">3621</span> hbck.close();<a name="line.3621"></a> |
| <span class="sourceLineNo">3622</span> return hbck.getRetCode();<a name="line.3622"></a> |
| <span class="sourceLineNo">3623</span> }<a name="line.3623"></a> |
| <span class="sourceLineNo">3624</span> }<a name="line.3624"></a> |
| <span class="sourceLineNo">3625</span><a name="line.3625"></a> |
| <span class="sourceLineNo">3626</span> public HBaseFsck exec(ExecutorService exec, String[] args)<a name="line.3626"></a> |
| <span class="sourceLineNo">3627</span> throws KeeperException, IOException, InterruptedException, ReplicationException {<a name="line.3627"></a> |
| <span class="sourceLineNo">3628</span> long sleepBeforeRerun = DEFAULT_SLEEP_BEFORE_RERUN;<a name="line.3628"></a> |
| <span class="sourceLineNo">3629</span><a name="line.3629"></a> |
| <span class="sourceLineNo">3630</span> boolean checkCorruptHFiles = false;<a name="line.3630"></a> |
| <span class="sourceLineNo">3631</span> boolean sidelineCorruptHFiles = false;<a name="line.3631"></a> |
| <span class="sourceLineNo">3632</span><a name="line.3632"></a> |
| <span class="sourceLineNo">3633</span> // Process command-line args.<a name="line.3633"></a> |
| <span class="sourceLineNo">3634</span> for (int i = 0; i < args.length; i++) {<a name="line.3634"></a> |
| <span class="sourceLineNo">3635</span> String cmd = args[i];<a name="line.3635"></a> |
| <span class="sourceLineNo">3636</span> if (cmd.equals("-help") || cmd.equals("-h")) {<a name="line.3636"></a> |
| <span class="sourceLineNo">3637</span> return printUsageAndExit();<a name="line.3637"></a> |
| <span class="sourceLineNo">3638</span> } else if (cmd.equals("-details")) {<a name="line.3638"></a> |
| <span class="sourceLineNo">3639</span> setDisplayFullReport();<a name="line.3639"></a> |
| <span class="sourceLineNo">3640</span> } else if (cmd.equals("-exclusive")) {<a name="line.3640"></a> |
| <span class="sourceLineNo">3641</span> setForceExclusive();<a name="line.3641"></a> |
| <span class="sourceLineNo">3642</span> } else if (cmd.equals("-timelag")) {<a name="line.3642"></a> |
| <span class="sourceLineNo">3643</span> if (i == args.length - 1) {<a name="line.3643"></a> |
| <span class="sourceLineNo">3644</span> errors.reportError(ERROR_CODE.WRONG_USAGE, "HBaseFsck: -timelag needs a value.");<a name="line.3644"></a> |
| <span class="sourceLineNo">3645</span> return printUsageAndExit();<a name="line.3645"></a> |
| <span class="sourceLineNo">3646</span> }<a name="line.3646"></a> |
| <span class="sourceLineNo">3647</span> try {<a name="line.3647"></a> |
| <span class="sourceLineNo">3648</span> long timelag = Long.parseLong(args[++i]);<a name="line.3648"></a> |
| <span class="sourceLineNo">3649</span> setTimeLag(timelag);<a name="line.3649"></a> |
| <span class="sourceLineNo">3650</span> } catch (NumberFormatException e) {<a name="line.3650"></a> |
| <span class="sourceLineNo">3651</span> errors.reportError(ERROR_CODE.WRONG_USAGE, "-timelag needs a numeric value.");<a name="line.3651"></a> |
| <span class="sourceLineNo">3652</span> return printUsageAndExit();<a name="line.3652"></a> |
| <span class="sourceLineNo">3653</span> }<a name="line.3653"></a> |
| <span class="sourceLineNo">3654</span> } else if (cmd.equals("-sleepBeforeRerun")) {<a name="line.3654"></a> |
| <span class="sourceLineNo">3655</span> if (i == args.length - 1) {<a name="line.3655"></a> |
| <span class="sourceLineNo">3656</span> errors.reportError(ERROR_CODE.WRONG_USAGE,<a name="line.3656"></a> |
| <span class="sourceLineNo">3657</span> "HBaseFsck: -sleepBeforeRerun needs a value.");<a name="line.3657"></a> |
| <span class="sourceLineNo">3658</span> return printUsageAndExit();<a name="line.3658"></a> |
| <span class="sourceLineNo">3659</span> }<a name="line.3659"></a> |
| <span class="sourceLineNo">3660</span> try {<a name="line.3660"></a> |
| <span class="sourceLineNo">3661</span> sleepBeforeRerun = Long.parseLong(args[++i]);<a name="line.3661"></a> |
| <span class="sourceLineNo">3662</span> } catch (NumberFormatException e) {<a name="line.3662"></a> |
| <span class="sourceLineNo">3663</span> errors.reportError(ERROR_CODE.WRONG_USAGE, "-sleepBeforeRerun needs a numeric value.");<a name="line.3663"></a> |
| <span class="sourceLineNo">3664</span> return printUsageAndExit();<a name="line.3664"></a> |
| <span class="sourceLineNo">3665</span> }<a name="line.3665"></a> |
| <span class="sourceLineNo">3666</span> } else if (cmd.equals("-sidelineDir")) {<a name="line.3666"></a> |
| <span class="sourceLineNo">3667</span> if (i == args.length - 1) {<a name="line.3667"></a> |
| <span class="sourceLineNo">3668</span> errors.reportError(ERROR_CODE.WRONG_USAGE, "HBaseFsck: -sidelineDir needs a value.");<a name="line.3668"></a> |
| <span class="sourceLineNo">3669</span> return printUsageAndExit();<a name="line.3669"></a> |
| <span class="sourceLineNo">3670</span> }<a name="line.3670"></a> |
| <span class="sourceLineNo">3671</span> setSidelineDir(args[++i]);<a name="line.3671"></a> |
| <span class="sourceLineNo">3672</span> } else if (cmd.equals("-fix")) {<a name="line.3672"></a> |
| <span class="sourceLineNo">3673</span> errors.reportError(ERROR_CODE.WRONG_USAGE,<a name="line.3673"></a> |
| <span class="sourceLineNo">3674</span> "This option is deprecated, please use -fixAssignments instead.");<a name="line.3674"></a> |
| <span class="sourceLineNo">3675</span> setFixAssignments(true);<a name="line.3675"></a> |
| <span class="sourceLineNo">3676</span> } else if (cmd.equals("-fixAssignments")) {<a name="line.3676"></a> |
| <span class="sourceLineNo">3677</span> setFixAssignments(true);<a name="line.3677"></a> |
| <span class="sourceLineNo">3678</span> } else if (cmd.equals("-fixMeta")) {<a name="line.3678"></a> |
| <span class="sourceLineNo">3679</span> setFixMeta(true);<a name="line.3679"></a> |
| <span class="sourceLineNo">3680</span> } else if (cmd.equals("-noHdfsChecking")) {<a name="line.3680"></a> |
| <span class="sourceLineNo">3681</span> setCheckHdfs(false);<a name="line.3681"></a> |
| <span class="sourceLineNo">3682</span> } else if (cmd.equals("-fixHdfsHoles")) {<a name="line.3682"></a> |
| <span class="sourceLineNo">3683</span> setFixHdfsHoles(true);<a name="line.3683"></a> |
| <span class="sourceLineNo">3684</span> } else if (cmd.equals("-fixHdfsOrphans")) {<a name="line.3684"></a> |
| <span class="sourceLineNo">3685</span> setFixHdfsOrphans(true);<a name="line.3685"></a> |
| <span class="sourceLineNo">3686</span> } else if (cmd.equals("-fixTableOrphans")) {<a name="line.3686"></a> |
| <span class="sourceLineNo">3687</span> setFixTableOrphans(true);<a name="line.3687"></a> |
| <span class="sourceLineNo">3688</span> } else if (cmd.equals("-fixHdfsOverlaps")) {<a name="line.3688"></a> |
| <span class="sourceLineNo">3689</span> setFixHdfsOverlaps(true);<a name="line.3689"></a> |
| <span class="sourceLineNo">3690</span> } else if (cmd.equals("-fixVersionFile")) {<a name="line.3690"></a> |
| <span class="sourceLineNo">3691</span> setFixVersionFile(true);<a name="line.3691"></a> |
| <span class="sourceLineNo">3692</span> } else if (cmd.equals("-sidelineBigOverlaps")) {<a name="line.3692"></a> |
| <span class="sourceLineNo">3693</span> setSidelineBigOverlaps(true);<a name="line.3693"></a> |
| <span class="sourceLineNo">3694</span> } else if (cmd.equals("-fixSplitParents")) {<a name="line.3694"></a> |
| <span class="sourceLineNo">3695</span> setFixSplitParents(true);<a name="line.3695"></a> |
| <span class="sourceLineNo">3696</span> } else if (cmd.equals("-removeParents")) {<a name="line.3696"></a> |
| <span class="sourceLineNo">3697</span> setRemoveParents(true);<a name="line.3697"></a> |
| <span class="sourceLineNo">3698</span> } else if (cmd.equals("-ignorePreCheckPermission")) {<a name="line.3698"></a> |
| <span class="sourceLineNo">3699</span> setIgnorePreCheckPermission(true);<a name="line.3699"></a> |
| <span class="sourceLineNo">3700</span> } else if (cmd.equals("-checkCorruptHFiles")) {<a name="line.3700"></a> |
| <span class="sourceLineNo">3701</span> checkCorruptHFiles = true;<a name="line.3701"></a> |
| <span class="sourceLineNo">3702</span> } else if (cmd.equals("-sidelineCorruptHFiles")) {<a name="line.3702"></a> |
| <span class="sourceLineNo">3703</span> sidelineCorruptHFiles = true;<a name="line.3703"></a> |
| <span class="sourceLineNo">3704</span> } else if (cmd.equals("-fixReferenceFiles")) {<a name="line.3704"></a> |
| <span class="sourceLineNo">3705</span> setFixReferenceFiles(true);<a name="line.3705"></a> |
| <span class="sourceLineNo">3706</span> } else if (cmd.equals("-fixHFileLinks")) {<a name="line.3706"></a> |
| <span class="sourceLineNo">3707</span> setFixHFileLinks(true);<a name="line.3707"></a> |
| <span class="sourceLineNo">3708</span> } else if (cmd.equals("-fixEmptyMetaCells")) {<a name="line.3708"></a> |
| <span class="sourceLineNo">3709</span> setFixEmptyMetaCells(true);<a name="line.3709"></a> |
| <span class="sourceLineNo">3710</span> } else if (cmd.equals("-repair")) {<a name="line.3710"></a> |
| <span class="sourceLineNo">3711</span> // this attempts to merge overlapping hdfs regions, needs testing<a name="line.3711"></a> |
| <span class="sourceLineNo">3712</span> // under load<a name="line.3712"></a> |
| <span class="sourceLineNo">3713</span> setFixHdfsHoles(true);<a name="line.3713"></a> |
| <span class="sourceLineNo">3714</span> setFixHdfsOrphans(true);<a name="line.3714"></a> |
| <span class="sourceLineNo">3715</span> setFixMeta(true);<a name="line.3715"></a> |
| <span class="sourceLineNo">3716</span> setFixAssignments(true);<a name="line.3716"></a> |
| <span class="sourceLineNo">3717</span> setFixHdfsOverlaps(true);<a name="line.3717"></a> |
| <span class="sourceLineNo">3718</span> setFixVersionFile(true);<a name="line.3718"></a> |
| <span class="sourceLineNo">3719</span> setSidelineBigOverlaps(true);<a name="line.3719"></a> |
| <span class="sourceLineNo">3720</span> setFixSplitParents(false);<a name="line.3720"></a> |
| <span class="sourceLineNo">3721</span> setCheckHdfs(true);<a name="line.3721"></a> |
| <span class="sourceLineNo">3722</span> setFixReferenceFiles(true);<a name="line.3722"></a> |
| <span class="sourceLineNo">3723</span> setFixHFileLinks(true);<a name="line.3723"></a> |
| <span class="sourceLineNo">3724</span> } else if (cmd.equals("-repairHoles")) {<a name="line.3724"></a> |
| <span class="sourceLineNo">3725</span> // this will make all missing hdfs regions available but may lose data<a name="line.3725"></a> |
| <span class="sourceLineNo">3726</span> setFixHdfsHoles(true);<a name="line.3726"></a> |
| <span class="sourceLineNo">3727</span> setFixHdfsOrphans(false);<a name="line.3727"></a> |
| <span class="sourceLineNo">3728</span> setFixMeta(true);<a name="line.3728"></a> |
| <span class="sourceLineNo">3729</span> setFixAssignments(true);<a name="line.3729"></a> |
| <span class="sourceLineNo">3730</span> setFixHdfsOverlaps(false);<a name="line.3730"></a> |
| <span class="sourceLineNo">3731</span> setSidelineBigOverlaps(false);<a name="line.3731"></a> |
| <span class="sourceLineNo">3732</span> setFixSplitParents(false);<a name="line.3732"></a> |
| <span class="sourceLineNo">3733</span> setCheckHdfs(true);<a name="line.3733"></a> |
| <span class="sourceLineNo">3734</span> } else if (cmd.equals("-maxOverlapsToSideline")) {<a name="line.3734"></a> |
| <span class="sourceLineNo">3735</span> if (i == args.length - 1) {<a name="line.3735"></a> |
| <span class="sourceLineNo">3736</span> errors.reportError(ERROR_CODE.WRONG_USAGE,<a name="line.3736"></a> |
| <span class="sourceLineNo">3737</span> "-maxOverlapsToSideline needs a numeric value argument.");<a name="line.3737"></a> |
| <span class="sourceLineNo">3738</span> return printUsageAndExit();<a name="line.3738"></a> |
| <span class="sourceLineNo">3739</span> }<a name="line.3739"></a> |
| <span class="sourceLineNo">3740</span> try {<a name="line.3740"></a> |
| <span class="sourceLineNo">3741</span> int maxOverlapsToSideline = Integer.parseInt(args[++i]);<a name="line.3741"></a> |
| <span class="sourceLineNo">3742</span> setMaxOverlapsToSideline(maxOverlapsToSideline);<a name="line.3742"></a> |
| <span class="sourceLineNo">3743</span> } catch (NumberFormatException e) {<a name="line.3743"></a> |
| <span class="sourceLineNo">3744</span> errors.reportError(ERROR_CODE.WRONG_USAGE,<a name="line.3744"></a> |
| <span class="sourceLineNo">3745</span> "-maxOverlapsToSideline needs a numeric value argument.");<a name="line.3745"></a> |
| <span class="sourceLineNo">3746</span> return printUsageAndExit();<a name="line.3746"></a> |
| <span class="sourceLineNo">3747</span> }<a name="line.3747"></a> |
| <span class="sourceLineNo">3748</span> } else if (cmd.equals("-maxMerge")) {<a name="line.3748"></a> |
| <span class="sourceLineNo">3749</span> if (i == args.length - 1) {<a name="line.3749"></a> |
| <span class="sourceLineNo">3750</span> errors.reportError(ERROR_CODE.WRONG_USAGE,<a name="line.3750"></a> |
| <span class="sourceLineNo">3751</span> "-maxMerge needs a numeric value argument.");<a name="line.3751"></a> |
| <span class="sourceLineNo">3752</span> return printUsageAndExit();<a name="line.3752"></a> |
| <span class="sourceLineNo">3753</span> }<a name="line.3753"></a> |
| <span class="sourceLineNo">3754</span> try {<a name="line.3754"></a> |
| <span class="sourceLineNo">3755</span> int maxMerge = Integer.parseInt(args[++i]);<a name="line.3755"></a> |
| <span class="sourceLineNo">3756</span> setMaxMerge(maxMerge);<a name="line.3756"></a> |
| <span class="sourceLineNo">3757</span> } catch (NumberFormatException e) {<a name="line.3757"></a> |
| <span class="sourceLineNo">3758</span> errors.reportError(ERROR_CODE.WRONG_USAGE,<a name="line.3758"></a> |
| <span class="sourceLineNo">3759</span> "-maxMerge needs a numeric value argument.");<a name="line.3759"></a> |
| <span class="sourceLineNo">3760</span> return printUsageAndExit();<a name="line.3760"></a> |
| <span class="sourceLineNo">3761</span> }<a name="line.3761"></a> |
| <span class="sourceLineNo">3762</span> } else if (cmd.equals("-summary")) {<a name="line.3762"></a> |
| <span class="sourceLineNo">3763</span> setSummary();<a name="line.3763"></a> |
| <span class="sourceLineNo">3764</span> } else if (cmd.equals("-metaonly")) {<a name="line.3764"></a> |
| <span class="sourceLineNo">3765</span> setCheckMetaOnly();<a name="line.3765"></a> |
| <span class="sourceLineNo">3766</span> } else if (cmd.equals("-boundaries")) {<a name="line.3766"></a> |
| <span class="sourceLineNo">3767</span> setRegionBoundariesCheck();<a name="line.3767"></a> |
| <span class="sourceLineNo">3768</span> } else if (cmd.equals("-fixReplication")) {<a name="line.3768"></a> |
| <span class="sourceLineNo">3769</span> setFixReplication(true);<a name="line.3769"></a> |
| <span class="sourceLineNo">3770</span> } else if (cmd.equals("-cleanReplicationBarrier")) {<a name="line.3770"></a> |
| <span class="sourceLineNo">3771</span> setCleanReplicationBarrier(true);<a name="line.3771"></a> |
| <span class="sourceLineNo">3772</span> if(args[++i].startsWith("-")){<a name="line.3772"></a> |
| <span class="sourceLineNo">3773</span> printUsageAndExit();<a name="line.3773"></a> |
| <span class="sourceLineNo">3774</span> }<a name="line.3774"></a> |
| <span class="sourceLineNo">3775</span> setCleanReplicationBarrierTable(args[i]);<a name="line.3775"></a> |
| <span class="sourceLineNo">3776</span> } else if (cmd.startsWith("-")) {<a name="line.3776"></a> |
| <span class="sourceLineNo">3777</span> errors.reportError(ERROR_CODE.WRONG_USAGE, "Unrecognized option:" + cmd);<a name="line.3777"></a> |
| <span class="sourceLineNo">3778</span> return printUsageAndExit();<a name="line.3778"></a> |
| <span class="sourceLineNo">3779</span> } else {<a name="line.3779"></a> |
| <span class="sourceLineNo">3780</span> includeTable(TableName.valueOf(cmd));<a name="line.3780"></a> |
| <span class="sourceLineNo">3781</span> errors.print("Allow checking/fixes for table: " + cmd);<a name="line.3781"></a> |
| <span class="sourceLineNo">3782</span> }<a name="line.3782"></a> |
| <span class="sourceLineNo">3783</span> }<a name="line.3783"></a> |
| <span class="sourceLineNo">3784</span><a name="line.3784"></a> |
| <span class="sourceLineNo">3785</span> errors.print("HBaseFsck command line options: " + StringUtils.join(args, " "));<a name="line.3785"></a> |
| <span class="sourceLineNo">3786</span><a name="line.3786"></a> |
| <span class="sourceLineNo">3787</span> // pre-check current user has FS write permission or not<a name="line.3787"></a> |
| <span class="sourceLineNo">3788</span> try {<a name="line.3788"></a> |
| <span class="sourceLineNo">3789</span> preCheckPermission();<a name="line.3789"></a> |
| <span class="sourceLineNo">3790</span> } catch (IOException ioe) {<a name="line.3790"></a> |
| <span class="sourceLineNo">3791</span> Runtime.getRuntime().exit(-1);<a name="line.3791"></a> |
| <span class="sourceLineNo">3792</span> }<a name="line.3792"></a> |
| <span class="sourceLineNo">3793</span><a name="line.3793"></a> |
| <span class="sourceLineNo">3794</span> // do the real work of hbck<a name="line.3794"></a> |
| <span class="sourceLineNo">3795</span> connect();<a name="line.3795"></a> |
| <span class="sourceLineNo">3796</span><a name="line.3796"></a> |
| <span class="sourceLineNo">3797</span> // after connecting to server above, we have server version<a name="line.3797"></a> |
| <span class="sourceLineNo">3798</span> // check if unsupported option is specified based on server version<a name="line.3798"></a> |
| <span class="sourceLineNo">3799</span> if (!isOptionsSupported(args)) {<a name="line.3799"></a> |
| <span class="sourceLineNo">3800</span> return printUsageAndExit();<a name="line.3800"></a> |
| <span class="sourceLineNo">3801</span> }<a name="line.3801"></a> |
| <span class="sourceLineNo">3802</span><a name="line.3802"></a> |
| <span class="sourceLineNo">3803</span> try {<a name="line.3803"></a> |
| <span class="sourceLineNo">3804</span> // if corrupt file mode is on, first fix them since they may be opened later<a name="line.3804"></a> |
| <span class="sourceLineNo">3805</span> if (checkCorruptHFiles || sidelineCorruptHFiles) {<a name="line.3805"></a> |
| <span class="sourceLineNo">3806</span> LOG.info("Checking all hfiles for corruption");<a name="line.3806"></a> |
| <span class="sourceLineNo">3807</span> HFileCorruptionChecker hfcc = createHFileCorruptionChecker(sidelineCorruptHFiles);<a name="line.3807"></a> |
| <span class="sourceLineNo">3808</span> setHFileCorruptionChecker(hfcc); // so we can get result<a name="line.3808"></a> |
| <span class="sourceLineNo">3809</span> Collection<TableName> tables = getIncludedTables();<a name="line.3809"></a> |
| <span class="sourceLineNo">3810</span> Collection<Path> tableDirs = new ArrayList<>();<a name="line.3810"></a> |
| <span class="sourceLineNo">3811</span> Path rootdir = CommonFSUtils.getRootDir(getConf());<a name="line.3811"></a> |
| <span class="sourceLineNo">3812</span> if (tables.size() > 0) {<a name="line.3812"></a> |
| <span class="sourceLineNo">3813</span> for (TableName t : tables) {<a name="line.3813"></a> |
| <span class="sourceLineNo">3814</span> tableDirs.add(CommonFSUtils.getTableDir(rootdir, t));<a name="line.3814"></a> |
| <span class="sourceLineNo">3815</span> }<a name="line.3815"></a> |
| <span class="sourceLineNo">3816</span> } else {<a name="line.3816"></a> |
| <span class="sourceLineNo">3817</span> tableDirs = FSUtils.getTableDirs(CommonFSUtils.getCurrentFileSystem(getConf()), rootdir);<a name="line.3817"></a> |
| <span class="sourceLineNo">3818</span> }<a name="line.3818"></a> |
| <span class="sourceLineNo">3819</span> hfcc.checkTables(tableDirs);<a name="line.3819"></a> |
| <span class="sourceLineNo">3820</span> hfcc.report(errors);<a name="line.3820"></a> |
| <span class="sourceLineNo">3821</span> }<a name="line.3821"></a> |
| <span class="sourceLineNo">3822</span><a name="line.3822"></a> |
| <span class="sourceLineNo">3823</span> // check and fix table integrity, region consistency.<a name="line.3823"></a> |
| <span class="sourceLineNo">3824</span> int code = onlineHbck();<a name="line.3824"></a> |
| <span class="sourceLineNo">3825</span> setRetCode(code);<a name="line.3825"></a> |
| <span class="sourceLineNo">3826</span> // If we have changed the HBase state it is better to run hbck again<a name="line.3826"></a> |
| <span class="sourceLineNo">3827</span> // to see if we haven't broken something else in the process.<a name="line.3827"></a> |
| <span class="sourceLineNo">3828</span> // We run it only once more because otherwise we can easily fall into<a name="line.3828"></a> |
| <span class="sourceLineNo">3829</span> // an infinite loop.<a name="line.3829"></a> |
| <span class="sourceLineNo">3830</span> if (shouldRerun()) {<a name="line.3830"></a> |
| <span class="sourceLineNo">3831</span> try {<a name="line.3831"></a> |
| <span class="sourceLineNo">3832</span> LOG.info("Sleeping " + sleepBeforeRerun + "ms before re-checking after fix...");<a name="line.3832"></a> |
| <span class="sourceLineNo">3833</span> Thread.sleep(sleepBeforeRerun);<a name="line.3833"></a> |
| <span class="sourceLineNo">3834</span> } catch (InterruptedException ie) {<a name="line.3834"></a> |
| <span class="sourceLineNo">3835</span> LOG.warn("Interrupted while sleeping");<a name="line.3835"></a> |
| <span class="sourceLineNo">3836</span> return this;<a name="line.3836"></a> |
| <span class="sourceLineNo">3837</span> }<a name="line.3837"></a> |
| <span class="sourceLineNo">3838</span> // Just report<a name="line.3838"></a> |
| <span class="sourceLineNo">3839</span> setFixAssignments(false);<a name="line.3839"></a> |
| <span class="sourceLineNo">3840</span> setFixMeta(false);<a name="line.3840"></a> |
| <span class="sourceLineNo">3841</span> setFixHdfsHoles(false);<a name="line.3841"></a> |
| <span class="sourceLineNo">3842</span> setFixHdfsOverlaps(false);<a name="line.3842"></a> |
| <span class="sourceLineNo">3843</span> setFixVersionFile(false);<a name="line.3843"></a> |
| <span class="sourceLineNo">3844</span> setFixTableOrphans(false);<a name="line.3844"></a> |
| <span class="sourceLineNo">3845</span> errors.resetErrors();<a name="line.3845"></a> |
| <span class="sourceLineNo">3846</span> code = onlineHbck();<a name="line.3846"></a> |
| <span class="sourceLineNo">3847</span> setRetCode(code);<a name="line.3847"></a> |
| <span class="sourceLineNo">3848</span> }<a name="line.3848"></a> |
| <span class="sourceLineNo">3849</span> } finally {<a name="line.3849"></a> |
| <span class="sourceLineNo">3850</span> IOUtils.closeQuietly(this);<a name="line.3850"></a> |
| <span class="sourceLineNo">3851</span> }<a name="line.3851"></a> |
| <span class="sourceLineNo">3852</span> return this;<a name="line.3852"></a> |
| <span class="sourceLineNo">3853</span> }<a name="line.3853"></a> |
| <span class="sourceLineNo">3854</span><a name="line.3854"></a> |
| <span class="sourceLineNo">3855</span> private boolean isOptionsSupported(String[] args) {<a name="line.3855"></a> |
| <span class="sourceLineNo">3856</span> boolean result = true;<a name="line.3856"></a> |
| <span class="sourceLineNo">3857</span> String hbaseServerVersion = status.getHBaseVersion();<a name="line.3857"></a> |
| <span class="sourceLineNo">3858</span> if (VersionInfo.compareVersion("2.any.any", hbaseServerVersion) < 0) {<a name="line.3858"></a> |
| <span class="sourceLineNo">3859</span> // Process command-line args.<a name="line.3859"></a> |
| <span class="sourceLineNo">3860</span> for (String arg : args) {<a name="line.3860"></a> |
| <span class="sourceLineNo">3861</span> if (unsupportedOptionsInV2.contains(arg)) {<a name="line.3861"></a> |
| <span class="sourceLineNo">3862</span> errors.reportError(ERROR_CODE.UNSUPPORTED_OPTION,<a name="line.3862"></a> |
| <span class="sourceLineNo">3863</span> "option '" + arg + "' is not " + "supportted!");<a name="line.3863"></a> |
| <span class="sourceLineNo">3864</span> result = false;<a name="line.3864"></a> |
| <span class="sourceLineNo">3865</span> break;<a name="line.3865"></a> |
| <span class="sourceLineNo">3866</span> }<a name="line.3866"></a> |
| <span class="sourceLineNo">3867</span> }<a name="line.3867"></a> |
| <span class="sourceLineNo">3868</span> }<a name="line.3868"></a> |
| <span class="sourceLineNo">3869</span> return result;<a name="line.3869"></a> |
| <span class="sourceLineNo">3870</span> }<a name="line.3870"></a> |
| <span class="sourceLineNo">3871</span><a name="line.3871"></a> |
| <span class="sourceLineNo">3872</span> public void setCleanReplicationBarrierTable(String cleanReplicationBarrierTable) {<a name="line.3872"></a> |
| <span class="sourceLineNo">3873</span> this.cleanReplicationBarrierTable = TableName.valueOf(cleanReplicationBarrierTable);<a name="line.3873"></a> |
| <span class="sourceLineNo">3874</span> }<a name="line.3874"></a> |
| <span class="sourceLineNo">3875</span><a name="line.3875"></a> |
| <span class="sourceLineNo">3876</span> public void cleanReplicationBarrier() throws IOException {<a name="line.3876"></a> |
| <span class="sourceLineNo">3877</span> if (!cleanReplicationBarrier || cleanReplicationBarrierTable == null) {<a name="line.3877"></a> |
| <span class="sourceLineNo">3878</span> return;<a name="line.3878"></a> |
| <span class="sourceLineNo">3879</span> }<a name="line.3879"></a> |
| <span class="sourceLineNo">3880</span> if (cleanReplicationBarrierTable.isSystemTable()) {<a name="line.3880"></a> |
| <span class="sourceLineNo">3881</span> errors.reportError(ERROR_CODE.INVALID_TABLE,<a name="line.3881"></a> |
| <span class="sourceLineNo">3882</span> "invalid table: " + cleanReplicationBarrierTable);<a name="line.3882"></a> |
| <span class="sourceLineNo">3883</span> return;<a name="line.3883"></a> |
| <span class="sourceLineNo">3884</span> }<a name="line.3884"></a> |
| <span class="sourceLineNo">3885</span><a name="line.3885"></a> |
| <span class="sourceLineNo">3886</span> boolean isGlobalScope = false;<a name="line.3886"></a> |
| <span class="sourceLineNo">3887</span> try {<a name="line.3887"></a> |
| <span class="sourceLineNo">3888</span> isGlobalScope = admin.getDescriptor(cleanReplicationBarrierTable).hasGlobalReplicationScope();<a name="line.3888"></a> |
| <span class="sourceLineNo">3889</span> } catch (TableNotFoundException e) {<a name="line.3889"></a> |
| <span class="sourceLineNo">3890</span> LOG.info("we may need to clean some erroneous data due to bugs");<a name="line.3890"></a> |
| <span class="sourceLineNo">3891</span> }<a name="line.3891"></a> |
| <span class="sourceLineNo">3892</span><a name="line.3892"></a> |
| <span class="sourceLineNo">3893</span> if (isGlobalScope) {<a name="line.3893"></a> |
| <span class="sourceLineNo">3894</span> errors.reportError(ERROR_CODE.INVALID_TABLE,<a name="line.3894"></a> |
| <span class="sourceLineNo">3895</span> "table's replication scope is global: " + cleanReplicationBarrierTable);<a name="line.3895"></a> |
| <span class="sourceLineNo">3896</span> return;<a name="line.3896"></a> |
| <span class="sourceLineNo">3897</span> }<a name="line.3897"></a> |
| <span class="sourceLineNo">3898</span> List<byte[]> regionNames = new ArrayList<>();<a name="line.3898"></a> |
| <span class="sourceLineNo">3899</span> Scan barrierScan = new Scan();<a name="line.3899"></a> |
| <span class="sourceLineNo">3900</span> barrierScan.setCaching(100);<a name="line.3900"></a> |
| <span class="sourceLineNo">3901</span> barrierScan.addFamily(HConstants.REPLICATION_BARRIER_FAMILY);<a name="line.3901"></a> |
| <span class="sourceLineNo">3902</span> barrierScan<a name="line.3902"></a> |
| <span class="sourceLineNo">3903</span> .withStartRow(ClientMetaTableAccessor.getTableStartRowForMeta(cleanReplicationBarrierTable,<a name="line.3903"></a> |
| <span class="sourceLineNo">3904</span> ClientMetaTableAccessor.QueryType.REGION))<a name="line.3904"></a> |
| <span class="sourceLineNo">3905</span> .withStopRow(ClientMetaTableAccessor.getTableStopRowForMeta(cleanReplicationBarrierTable,<a name="line.3905"></a> |
| <span class="sourceLineNo">3906</span> ClientMetaTableAccessor.QueryType.REGION));<a name="line.3906"></a> |
| <span class="sourceLineNo">3907</span> Result result;<a name="line.3907"></a> |
| <span class="sourceLineNo">3908</span> try (ResultScanner scanner = meta.getScanner(barrierScan)) {<a name="line.3908"></a> |
| <span class="sourceLineNo">3909</span> while ((result = scanner.next()) != null) {<a name="line.3909"></a> |
| <span class="sourceLineNo">3910</span> regionNames.add(result.getRow());<a name="line.3910"></a> |
| <span class="sourceLineNo">3911</span> }<a name="line.3911"></a> |
| <span class="sourceLineNo">3912</span> }<a name="line.3912"></a> |
| <span class="sourceLineNo">3913</span> if (regionNames.size() <= 0) {<a name="line.3913"></a> |
| <span class="sourceLineNo">3914</span> errors.reportError(ERROR_CODE.INVALID_TABLE,<a name="line.3914"></a> |
| <span class="sourceLineNo">3915</span> "there is no barriers of this table: " + cleanReplicationBarrierTable);<a name="line.3915"></a> |
| <span class="sourceLineNo">3916</span> return;<a name="line.3916"></a> |
| <span class="sourceLineNo">3917</span> }<a name="line.3917"></a> |
| <span class="sourceLineNo">3918</span> ReplicationQueueStorage queueStorage =<a name="line.3918"></a> |
| <span class="sourceLineNo">3919</span> ReplicationStorageFactory.getReplicationQueueStorage(zkw, getConf());<a name="line.3919"></a> |
| <span class="sourceLineNo">3920</span> List<ReplicationPeerDescription> peerDescriptions = admin.listReplicationPeers();<a name="line.3920"></a> |
| <span class="sourceLineNo">3921</span> if (peerDescriptions != null && peerDescriptions.size() > 0) {<a name="line.3921"></a> |
| <span class="sourceLineNo">3922</span> List<String> peers = peerDescriptions.stream()<a name="line.3922"></a> |
| <span class="sourceLineNo">3923</span> .filter(peerConfig -> peerConfig.getPeerConfig()<a name="line.3923"></a> |
| <span class="sourceLineNo">3924</span> .needToReplicate(cleanReplicationBarrierTable))<a name="line.3924"></a> |
| <span class="sourceLineNo">3925</span> .map(peerConfig -> peerConfig.getPeerId()).collect(Collectors.toList());<a name="line.3925"></a> |
| <span class="sourceLineNo">3926</span> try {<a name="line.3926"></a> |
| <span class="sourceLineNo">3927</span> List<String> batch = new ArrayList<>();<a name="line.3927"></a> |
| <span class="sourceLineNo">3928</span> for (String peer : peers) {<a name="line.3928"></a> |
| <span class="sourceLineNo">3929</span> for (byte[] regionName : regionNames) {<a name="line.3929"></a> |
| <span class="sourceLineNo">3930</span> batch.add(RegionInfo.encodeRegionName(regionName));<a name="line.3930"></a> |
| <span class="sourceLineNo">3931</span> if (batch.size() % 100 == 0) {<a name="line.3931"></a> |
| <span class="sourceLineNo">3932</span> queueStorage.removeLastSequenceIds(peer, batch);<a name="line.3932"></a> |
| <span class="sourceLineNo">3933</span> batch.clear();<a name="line.3933"></a> |
| <span class="sourceLineNo">3934</span> }<a name="line.3934"></a> |
| <span class="sourceLineNo">3935</span> }<a name="line.3935"></a> |
| <span class="sourceLineNo">3936</span> if (batch.size() > 0) {<a name="line.3936"></a> |
| <span class="sourceLineNo">3937</span> queueStorage.removeLastSequenceIds(peer, batch);<a name="line.3937"></a> |
| <span class="sourceLineNo">3938</span> batch.clear();<a name="line.3938"></a> |
| <span class="sourceLineNo">3939</span> }<a name="line.3939"></a> |
| <span class="sourceLineNo">3940</span> }<a name="line.3940"></a> |
| <span class="sourceLineNo">3941</span> } catch (ReplicationException re) {<a name="line.3941"></a> |
| <span class="sourceLineNo">3942</span> throw new IOException(re);<a name="line.3942"></a> |
| <span class="sourceLineNo">3943</span> }<a name="line.3943"></a> |
| <span class="sourceLineNo">3944</span> }<a name="line.3944"></a> |
| <span class="sourceLineNo">3945</span> for (byte[] regionName : regionNames) {<a name="line.3945"></a> |
| <span class="sourceLineNo">3946</span> meta.delete(new Delete(regionName).addFamily(HConstants.REPLICATION_BARRIER_FAMILY));<a name="line.3946"></a> |
| <span class="sourceLineNo">3947</span> }<a name="line.3947"></a> |
| <span class="sourceLineNo">3948</span> setShouldRerun();<a name="line.3948"></a> |
| <span class="sourceLineNo">3949</span> }<a name="line.3949"></a> |
| <span class="sourceLineNo">3950</span><a name="line.3950"></a> |
| <span class="sourceLineNo">3951</span> /**<a name="line.3951"></a> |
| <span class="sourceLineNo">3952</span> * ls -r for debugging purposes<a name="line.3952"></a> |
| <span class="sourceLineNo">3953</span> */<a name="line.3953"></a> |
| <span class="sourceLineNo">3954</span> void debugLsr(Path p) throws IOException {<a name="line.3954"></a> |
| <span class="sourceLineNo">3955</span> debugLsr(getConf(), p, errors);<a name="line.3955"></a> |
| <span class="sourceLineNo">3956</span> }<a name="line.3956"></a> |
| <span class="sourceLineNo">3957</span><a name="line.3957"></a> |
| <span class="sourceLineNo">3958</span> /**<a name="line.3958"></a> |
| <span class="sourceLineNo">3959</span> * ls -r for debugging purposes<a name="line.3959"></a> |
| <span class="sourceLineNo">3960</span> */<a name="line.3960"></a> |
| <span class="sourceLineNo">3961</span> public static void debugLsr(Configuration conf,<a name="line.3961"></a> |
| <span class="sourceLineNo">3962</span> Path p) throws IOException {<a name="line.3962"></a> |
| <span class="sourceLineNo">3963</span> debugLsr(conf, p, new PrintingErrorReporter());<a name="line.3963"></a> |
| <span class="sourceLineNo">3964</span> }<a name="line.3964"></a> |
| <span class="sourceLineNo">3965</span><a name="line.3965"></a> |
| <span class="sourceLineNo">3966</span> /**<a name="line.3966"></a> |
| <span class="sourceLineNo">3967</span> * ls -r for debugging purposes<a name="line.3967"></a> |
| <span class="sourceLineNo">3968</span> */<a name="line.3968"></a> |
| <span class="sourceLineNo">3969</span> public static void debugLsr(Configuration conf,<a name="line.3969"></a> |
| <span class="sourceLineNo">3970</span> Path p, HbckErrorReporter errors) throws IOException {<a name="line.3970"></a> |
| <span class="sourceLineNo">3971</span> if (!LOG.isDebugEnabled() || p == null) {<a name="line.3971"></a> |
| <span class="sourceLineNo">3972</span> return;<a name="line.3972"></a> |
| <span class="sourceLineNo">3973</span> }<a name="line.3973"></a> |
| <span class="sourceLineNo">3974</span> FileSystem fs = p.getFileSystem(conf);<a name="line.3974"></a> |
| <span class="sourceLineNo">3975</span><a name="line.3975"></a> |
| <span class="sourceLineNo">3976</span> if (!fs.exists(p)) {<a name="line.3976"></a> |
| <span class="sourceLineNo">3977</span> // nothing<a name="line.3977"></a> |
| <span class="sourceLineNo">3978</span> return;<a name="line.3978"></a> |
| <span class="sourceLineNo">3979</span> }<a name="line.3979"></a> |
| <span class="sourceLineNo">3980</span> errors.print(p.toString());<a name="line.3980"></a> |
| <span class="sourceLineNo">3981</span><a name="line.3981"></a> |
| <span class="sourceLineNo">3982</span> if (fs.isFile(p)) {<a name="line.3982"></a> |
| <span class="sourceLineNo">3983</span> return;<a name="line.3983"></a> |
| <span class="sourceLineNo">3984</span> }<a name="line.3984"></a> |
| <span class="sourceLineNo">3985</span><a name="line.3985"></a> |
| <span class="sourceLineNo">3986</span> if (fs.getFileStatus(p).isDirectory()) {<a name="line.3986"></a> |
| <span class="sourceLineNo">3987</span> FileStatus[] fss= fs.listStatus(p);<a name="line.3987"></a> |
| <span class="sourceLineNo">3988</span> for (FileStatus status : fss) {<a name="line.3988"></a> |
| <span class="sourceLineNo">3989</span> debugLsr(conf, status.getPath(), errors);<a name="line.3989"></a> |
| <span class="sourceLineNo">3990</span> }<a name="line.3990"></a> |
| <span class="sourceLineNo">3991</span> }<a name="line.3991"></a> |
| <span class="sourceLineNo">3992</span> }<a name="line.3992"></a> |
| <span class="sourceLineNo">3993</span>}<a name="line.3993"></a> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |
