| <!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> *<a name="line.2"></a> |
| <span class="sourceLineNo">003</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.3"></a> |
| <span class="sourceLineNo">004</span> * or more contributor license agreements. See the NOTICE file<a name="line.4"></a> |
| <span class="sourceLineNo">005</span> * distributed with this work for additional information<a name="line.5"></a> |
| <span class="sourceLineNo">006</span> * regarding copyright ownership. The ASF licenses this file<a name="line.6"></a> |
| <span class="sourceLineNo">007</span> * to you under the Apache License, Version 2.0 (the<a name="line.7"></a> |
| <span class="sourceLineNo">008</span> * "License"); you may not use this file except in compliance<a name="line.8"></a> |
| <span class="sourceLineNo">009</span> * with the License. You may obtain a copy of the License at<a name="line.9"></a> |
| <span class="sourceLineNo">010</span> *<a name="line.10"></a> |
| <span class="sourceLineNo">011</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.11"></a> |
| <span class="sourceLineNo">012</span> *<a name="line.12"></a> |
| <span class="sourceLineNo">013</span> * Unless required by applicable law or agreed to in writing, software<a name="line.13"></a> |
| <span class="sourceLineNo">014</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.14"></a> |
| <span class="sourceLineNo">015</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.15"></a> |
| <span class="sourceLineNo">016</span> * See the License for the specific language governing permissions and<a name="line.16"></a> |
| <span class="sourceLineNo">017</span> * limitations under the License.<a name="line.17"></a> |
| <span class="sourceLineNo">018</span> */<a name="line.18"></a> |
| <span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.master;<a name="line.19"></a> |
| <span class="sourceLineNo">020</span><a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import static org.apache.hadoop.hbase.util.ConcurrentMapUtils.computeIfAbsent;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span><a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.io.IOException;<a name="line.23"></a> |
| <span class="sourceLineNo">024</span>import java.net.InetAddress;<a name="line.24"></a> |
| <span class="sourceLineNo">025</span>import java.util.ArrayList;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import java.util.Collections;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import java.util.Iterator;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span>import java.util.List;<a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import java.util.Map;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import java.util.Map.Entry;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import java.util.Set;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import java.util.concurrent.ConcurrentNavigableMap;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import java.util.concurrent.ConcurrentSkipListMap;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import java.util.concurrent.CopyOnWriteArrayList;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import java.util.function.Predicate;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import org.apache.hadoop.conf.Configuration;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import org.apache.hadoop.fs.FSDataInputStream;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import org.apache.hadoop.fs.FileSystem;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span>import org.apache.hadoop.fs.Path;<a name="line.41"></a> |
| <span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.ClockOutOfSyncException;<a name="line.42"></a> |
| <span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.HConstants;<a name="line.43"></a> |
| <span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.NotServingRegionException;<a name="line.44"></a> |
| <span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.RegionMetrics;<a name="line.45"></a> |
| <span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.ScheduledChore;<a name="line.46"></a> |
| <span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.ServerMetrics;<a name="line.47"></a> |
| <span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.ServerMetricsBuilder;<a name="line.48"></a> |
| <span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.ServerName;<a name="line.49"></a> |
| <span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.YouAreDeadException;<a name="line.50"></a> |
| <span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.client.AsyncClusterConnection;<a name="line.51"></a> |
| <span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.client.AsyncRegionServerAdmin;<a name="line.52"></a> |
| <span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.53"></a> |
| <span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.ipc.RemoteWithExtrasException;<a name="line.54"></a> |
| <span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.master.assignment.RegionStates;<a name="line.55"></a> |
| <span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.master.procedure.ServerCrashProcedure;<a name="line.56"></a> |
| <span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.monitoring.MonitoredTask;<a name="line.57"></a> |
| <span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.procedure2.Procedure;<a name="line.58"></a> |
| <span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.59"></a> |
| <span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.util.CommonFSUtils;<a name="line.60"></a> |
| <span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.61"></a> |
| <span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.util.FutureUtils;<a name="line.62"></a> |
| <span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.63"></a> |
| <span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.64"></a> |
| <span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.zookeeper.ZNodePaths;<a name="line.65"></a> |
| <span class="sourceLineNo">066</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.66"></a> |
| <span class="sourceLineNo">067</span>import org.apache.zookeeper.KeeperException;<a name="line.67"></a> |
| <span class="sourceLineNo">068</span>import org.slf4j.Logger;<a name="line.68"></a> |
| <span class="sourceLineNo">069</span>import org.slf4j.LoggerFactory;<a name="line.69"></a> |
| <span class="sourceLineNo">070</span><a name="line.70"></a> |
| <span class="sourceLineNo">071</span>import org.apache.hbase.thirdparty.com.google.protobuf.ByteString;<a name="line.71"></a> |
| <span class="sourceLineNo">072</span>import org.apache.hbase.thirdparty.com.google.protobuf.UnsafeByteOperations;<a name="line.72"></a> |
| <span class="sourceLineNo">073</span><a name="line.73"></a> |
| <span class="sourceLineNo">074</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.74"></a> |
| <span class="sourceLineNo">075</span>import org.apache.hadoop.hbase.shaded.protobuf.RequestConverter;<a name="line.75"></a> |
| <span class="sourceLineNo">076</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.RegionStoreSequenceIds;<a name="line.76"></a> |
| <span class="sourceLineNo">077</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.StoreSequenceId;<a name="line.77"></a> |
| <span class="sourceLineNo">078</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.FlushedRegionSequenceId;<a name="line.78"></a> |
| <span class="sourceLineNo">079</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.FlushedSequenceId;<a name="line.79"></a> |
| <span class="sourceLineNo">080</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.FlushedStoreSequenceId;<a name="line.80"></a> |
| <span class="sourceLineNo">081</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionServerStartupRequest;<a name="line.81"></a> |
| <span class="sourceLineNo">082</span><a name="line.82"></a> |
| <span class="sourceLineNo">083</span>/**<a name="line.83"></a> |
| <span class="sourceLineNo">084</span> * The ServerManager class manages info about region servers.<a name="line.84"></a> |
| <span class="sourceLineNo">085</span> * <p><a name="line.85"></a> |
| <span class="sourceLineNo">086</span> * Maintains lists of online and dead servers. Processes the startups,<a name="line.86"></a> |
| <span class="sourceLineNo">087</span> * shutdowns, and deaths of region servers.<a name="line.87"></a> |
| <span class="sourceLineNo">088</span> * <p><a name="line.88"></a> |
| <span class="sourceLineNo">089</span> * Servers are distinguished in two different ways. A given server has a<a name="line.89"></a> |
| <span class="sourceLineNo">090</span> * location, specified by hostname and port, and of which there can only be one<a name="line.90"></a> |
| <span class="sourceLineNo">091</span> * online at any given time. A server instance is specified by the location<a name="line.91"></a> |
| <span class="sourceLineNo">092</span> * (hostname and port) as well as the startcode (timestamp from when the server<a name="line.92"></a> |
| <span class="sourceLineNo">093</span> * was started). This is used to differentiate a restarted instance of a given<a name="line.93"></a> |
| <span class="sourceLineNo">094</span> * server from the original instance.<a name="line.94"></a> |
| <span class="sourceLineNo">095</span> * <p><a name="line.95"></a> |
| <span class="sourceLineNo">096</span> * If a sever is known not to be running any more, it is called dead. The dead<a name="line.96"></a> |
| <span class="sourceLineNo">097</span> * server needs to be handled by a ServerShutdownHandler. If the handler is not<a name="line.97"></a> |
| <span class="sourceLineNo">098</span> * enabled yet, the server can't be handled right away so it is queued up.<a name="line.98"></a> |
| <span class="sourceLineNo">099</span> * After the handler is enabled, the server will be submitted to a handler to handle.<a name="line.99"></a> |
| <span class="sourceLineNo">100</span> * However, the handler may be just partially enabled. If so,<a name="line.100"></a> |
| <span class="sourceLineNo">101</span> * the server cannot be fully processed, and be queued up for further processing.<a name="line.101"></a> |
| <span class="sourceLineNo">102</span> * A server is fully processed only after the handler is fully enabled<a name="line.102"></a> |
| <span class="sourceLineNo">103</span> * and has completed the handling.<a name="line.103"></a> |
| <span class="sourceLineNo">104</span> */<a name="line.104"></a> |
| <span class="sourceLineNo">105</span>@InterfaceAudience.Private<a name="line.105"></a> |
| <span class="sourceLineNo">106</span>public class ServerManager {<a name="line.106"></a> |
| <span class="sourceLineNo">107</span> public static final String WAIT_ON_REGIONSERVERS_MAXTOSTART =<a name="line.107"></a> |
| <span class="sourceLineNo">108</span> "hbase.master.wait.on.regionservers.maxtostart";<a name="line.108"></a> |
| <span class="sourceLineNo">109</span><a name="line.109"></a> |
| <span class="sourceLineNo">110</span> public static final String WAIT_ON_REGIONSERVERS_MINTOSTART =<a name="line.110"></a> |
| <span class="sourceLineNo">111</span> "hbase.master.wait.on.regionservers.mintostart";<a name="line.111"></a> |
| <span class="sourceLineNo">112</span><a name="line.112"></a> |
| <span class="sourceLineNo">113</span> public static final String WAIT_ON_REGIONSERVERS_TIMEOUT =<a name="line.113"></a> |
| <span class="sourceLineNo">114</span> "hbase.master.wait.on.regionservers.timeout";<a name="line.114"></a> |
| <span class="sourceLineNo">115</span><a name="line.115"></a> |
| <span class="sourceLineNo">116</span> public static final String WAIT_ON_REGIONSERVERS_INTERVAL =<a name="line.116"></a> |
| <span class="sourceLineNo">117</span> "hbase.master.wait.on.regionservers.interval";<a name="line.117"></a> |
| <span class="sourceLineNo">118</span><a name="line.118"></a> |
| <span class="sourceLineNo">119</span> /**<a name="line.119"></a> |
| <span class="sourceLineNo">120</span> * see HBASE-20727<a name="line.120"></a> |
| <span class="sourceLineNo">121</span> * if set to true, flushedSequenceIdByRegion and storeFlushedSequenceIdsByRegion<a name="line.121"></a> |
| <span class="sourceLineNo">122</span> * will be persisted to HDFS and loaded when master restart to speed up log split<a name="line.122"></a> |
| <span class="sourceLineNo">123</span> */<a name="line.123"></a> |
| <span class="sourceLineNo">124</span> public static final String PERSIST_FLUSHEDSEQUENCEID =<a name="line.124"></a> |
| <span class="sourceLineNo">125</span> "hbase.master.persist.flushedsequenceid.enabled";<a name="line.125"></a> |
| <span class="sourceLineNo">126</span><a name="line.126"></a> |
| <span class="sourceLineNo">127</span> public static final boolean PERSIST_FLUSHEDSEQUENCEID_DEFAULT = true;<a name="line.127"></a> |
| <span class="sourceLineNo">128</span><a name="line.128"></a> |
| <span class="sourceLineNo">129</span> public static final String FLUSHEDSEQUENCEID_FLUSHER_INTERVAL =<a name="line.129"></a> |
| <span class="sourceLineNo">130</span> "hbase.master.flushedsequenceid.flusher.interval";<a name="line.130"></a> |
| <span class="sourceLineNo">131</span><a name="line.131"></a> |
| <span class="sourceLineNo">132</span> public static final int FLUSHEDSEQUENCEID_FLUSHER_INTERVAL_DEFAULT =<a name="line.132"></a> |
| <span class="sourceLineNo">133</span> 3 * 60 * 60 * 1000; // 3 hours<a name="line.133"></a> |
| <span class="sourceLineNo">134</span><a name="line.134"></a> |
| <span class="sourceLineNo">135</span> public static final String MAX_CLOCK_SKEW_MS = "hbase.master.maxclockskew";<a name="line.135"></a> |
| <span class="sourceLineNo">136</span><a name="line.136"></a> |
| <span class="sourceLineNo">137</span> private static final Logger LOG = LoggerFactory.getLogger(ServerManager.class);<a name="line.137"></a> |
| <span class="sourceLineNo">138</span><a name="line.138"></a> |
| <span class="sourceLineNo">139</span> // Set if we are to shutdown the cluster.<a name="line.139"></a> |
| <span class="sourceLineNo">140</span> private AtomicBoolean clusterShutdown = new AtomicBoolean(false);<a name="line.140"></a> |
| <span class="sourceLineNo">141</span><a name="line.141"></a> |
| <span class="sourceLineNo">142</span> /**<a name="line.142"></a> |
| <span class="sourceLineNo">143</span> * The last flushed sequence id for a region.<a name="line.143"></a> |
| <span class="sourceLineNo">144</span> */<a name="line.144"></a> |
| <span class="sourceLineNo">145</span> private final ConcurrentNavigableMap<byte[], Long> flushedSequenceIdByRegion =<a name="line.145"></a> |
| <span class="sourceLineNo">146</span> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.146"></a> |
| <span class="sourceLineNo">147</span><a name="line.147"></a> |
| <span class="sourceLineNo">148</span> private boolean persistFlushedSequenceId = true;<a name="line.148"></a> |
| <span class="sourceLineNo">149</span> private volatile boolean isFlushSeqIdPersistInProgress = false;<a name="line.149"></a> |
| <span class="sourceLineNo">150</span> /** File on hdfs to store last flushed sequence id of regions */<a name="line.150"></a> |
| <span class="sourceLineNo">151</span> private static final String LAST_FLUSHED_SEQ_ID_FILE = ".lastflushedseqids";<a name="line.151"></a> |
| <span class="sourceLineNo">152</span> private FlushedSequenceIdFlusher flushedSeqIdFlusher;<a name="line.152"></a> |
| <span class="sourceLineNo">153</span><a name="line.153"></a> |
| <span class="sourceLineNo">154</span><a name="line.154"></a> |
| <span class="sourceLineNo">155</span> /**<a name="line.155"></a> |
| <span class="sourceLineNo">156</span> * The last flushed sequence id for a store in a region.<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> */<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> private final ConcurrentNavigableMap<byte[], ConcurrentNavigableMap<byte[], Long>><a name="line.158"></a> |
| <span class="sourceLineNo">159</span> storeFlushedSequenceIdsByRegion = new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.159"></a> |
| <span class="sourceLineNo">160</span><a name="line.160"></a> |
| <span class="sourceLineNo">161</span> /** Map of registered servers to their current load */<a name="line.161"></a> |
| <span class="sourceLineNo">162</span> private final ConcurrentNavigableMap<ServerName, ServerMetrics> onlineServers =<a name="line.162"></a> |
| <span class="sourceLineNo">163</span> new ConcurrentSkipListMap<>();<a name="line.163"></a> |
| <span class="sourceLineNo">164</span><a name="line.164"></a> |
| <span class="sourceLineNo">165</span> /** List of region servers that should not get any more new regions. */<a name="line.165"></a> |
| <span class="sourceLineNo">166</span> private final ArrayList<ServerName> drainingServers = new ArrayList<>();<a name="line.166"></a> |
| <span class="sourceLineNo">167</span><a name="line.167"></a> |
| <span class="sourceLineNo">168</span> private final MasterServices master;<a name="line.168"></a> |
| <span class="sourceLineNo">169</span><a name="line.169"></a> |
| <span class="sourceLineNo">170</span> private final DeadServer deadservers = new DeadServer();<a name="line.170"></a> |
| <span class="sourceLineNo">171</span><a name="line.171"></a> |
| <span class="sourceLineNo">172</span> private final long maxSkew;<a name="line.172"></a> |
| <span class="sourceLineNo">173</span> private final long warningSkew;<a name="line.173"></a> |
| <span class="sourceLineNo">174</span><a name="line.174"></a> |
| <span class="sourceLineNo">175</span> /** Listeners that are called on server events. */<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> private List<ServerListener> listeners = new CopyOnWriteArrayList<>();<a name="line.176"></a> |
| <span class="sourceLineNo">177</span><a name="line.177"></a> |
| <span class="sourceLineNo">178</span> /**<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> * Constructor.<a name="line.179"></a> |
| <span class="sourceLineNo">180</span> */<a name="line.180"></a> |
| <span class="sourceLineNo">181</span> public ServerManager(final MasterServices master) {<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> this.master = master;<a name="line.182"></a> |
| <span class="sourceLineNo">183</span> Configuration c = master.getConfiguration();<a name="line.183"></a> |
| <span class="sourceLineNo">184</span> maxSkew = c.getLong(MAX_CLOCK_SKEW_MS, 30000);<a name="line.184"></a> |
| <span class="sourceLineNo">185</span> warningSkew = c.getLong("hbase.master.warningclockskew", 10000);<a name="line.185"></a> |
| <span class="sourceLineNo">186</span> persistFlushedSequenceId = c.getBoolean(PERSIST_FLUSHEDSEQUENCEID,<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> PERSIST_FLUSHEDSEQUENCEID_DEFAULT);<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> }<a name="line.188"></a> |
| <span class="sourceLineNo">189</span><a name="line.189"></a> |
| <span class="sourceLineNo">190</span> /**<a name="line.190"></a> |
| <span class="sourceLineNo">191</span> * Add the listener to the notification list.<a name="line.191"></a> |
| <span class="sourceLineNo">192</span> * @param listener The ServerListener to register<a name="line.192"></a> |
| <span class="sourceLineNo">193</span> */<a name="line.193"></a> |
| <span class="sourceLineNo">194</span> public void registerListener(final ServerListener listener) {<a name="line.194"></a> |
| <span class="sourceLineNo">195</span> this.listeners.add(listener);<a name="line.195"></a> |
| <span class="sourceLineNo">196</span> }<a name="line.196"></a> |
| <span class="sourceLineNo">197</span><a name="line.197"></a> |
| <span class="sourceLineNo">198</span> /**<a name="line.198"></a> |
| <span class="sourceLineNo">199</span> * Remove the listener from the notification list.<a name="line.199"></a> |
| <span class="sourceLineNo">200</span> * @param listener The ServerListener to unregister<a name="line.200"></a> |
| <span class="sourceLineNo">201</span> */<a name="line.201"></a> |
| <span class="sourceLineNo">202</span> public boolean unregisterListener(final ServerListener listener) {<a name="line.202"></a> |
| <span class="sourceLineNo">203</span> return this.listeners.remove(listener);<a name="line.203"></a> |
| <span class="sourceLineNo">204</span> }<a name="line.204"></a> |
| <span class="sourceLineNo">205</span><a name="line.205"></a> |
| <span class="sourceLineNo">206</span> /**<a name="line.206"></a> |
| <span class="sourceLineNo">207</span> * Let the server manager know a regionserver is requesting configurations.<a name="line.207"></a> |
| <span class="sourceLineNo">208</span> * Regionserver will not be added here, but in its first report.<a name="line.208"></a> |
| <span class="sourceLineNo">209</span> * @param request the startup request<a name="line.209"></a> |
| <span class="sourceLineNo">210</span> * @param versionNumber the version number of the new regionserver<a name="line.210"></a> |
| <span class="sourceLineNo">211</span> * @param version the version of the new regionserver, could contain strings like "SNAPSHOT"<a name="line.211"></a> |
| <span class="sourceLineNo">212</span> * @param ia the InetAddress from which request is received<a name="line.212"></a> |
| <span class="sourceLineNo">213</span> * @return The ServerName we know this server as.<a name="line.213"></a> |
| <span class="sourceLineNo">214</span> * @throws IOException<a name="line.214"></a> |
| <span class="sourceLineNo">215</span> */<a name="line.215"></a> |
| <span class="sourceLineNo">216</span> ServerName regionServerStartup(RegionServerStartupRequest request, int versionNumber,<a name="line.216"></a> |
| <span class="sourceLineNo">217</span> String version, InetAddress ia) throws IOException {<a name="line.217"></a> |
| <span class="sourceLineNo">218</span> // Test for case where we get a region startup message from a regionserver<a name="line.218"></a> |
| <span class="sourceLineNo">219</span> // that has been quickly restarted but whose znode expiration handler has<a name="line.219"></a> |
| <span class="sourceLineNo">220</span> // not yet run, or from a server whose fail we are currently processing.<a name="line.220"></a> |
| <span class="sourceLineNo">221</span> // Test its host+port combo is present in serverAddressToServerInfo. If it<a name="line.221"></a> |
| <span class="sourceLineNo">222</span> // is, reject the server and trigger its expiration. The next time it comes<a name="line.222"></a> |
| <span class="sourceLineNo">223</span> // in, it should have been removed from serverAddressToServerInfo and queued<a name="line.223"></a> |
| <span class="sourceLineNo">224</span> // for processing by ProcessServerShutdown.<a name="line.224"></a> |
| <span class="sourceLineNo">225</span><a name="line.225"></a> |
| <span class="sourceLineNo">226</span> final String hostname =<a name="line.226"></a> |
| <span class="sourceLineNo">227</span> request.hasUseThisHostnameInstead() ? request.getUseThisHostnameInstead() : ia.getHostName();<a name="line.227"></a> |
| <span class="sourceLineNo">228</span> ServerName sn = ServerName.valueOf(hostname, request.getPort(), request.getServerStartCode());<a name="line.228"></a> |
| <span class="sourceLineNo">229</span> checkClockSkew(sn, request.getServerCurrentTime());<a name="line.229"></a> |
| <span class="sourceLineNo">230</span> checkIsDead(sn, "STARTUP");<a name="line.230"></a> |
| <span class="sourceLineNo">231</span> return sn;<a name="line.231"></a> |
| <span class="sourceLineNo">232</span> }<a name="line.232"></a> |
| <span class="sourceLineNo">233</span><a name="line.233"></a> |
| <span class="sourceLineNo">234</span> /**<a name="line.234"></a> |
| <span class="sourceLineNo">235</span> * Updates last flushed sequence Ids for the regions on server sn<a name="line.235"></a> |
| <span class="sourceLineNo">236</span> * @param sn<a name="line.236"></a> |
| <span class="sourceLineNo">237</span> * @param hsl<a name="line.237"></a> |
| <span class="sourceLineNo">238</span> */<a name="line.238"></a> |
| <span class="sourceLineNo">239</span> private void updateLastFlushedSequenceIds(ServerName sn, ServerMetrics hsl) {<a name="line.239"></a> |
| <span class="sourceLineNo">240</span> for (Entry<byte[], RegionMetrics> entry : hsl.getRegionMetrics().entrySet()) {<a name="line.240"></a> |
| <span class="sourceLineNo">241</span> byte[] encodedRegionName = Bytes.toBytes(RegionInfo.encodeRegionName(entry.getKey()));<a name="line.241"></a> |
| <span class="sourceLineNo">242</span> Long existingValue = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.242"></a> |
| <span class="sourceLineNo">243</span> long l = entry.getValue().getCompletedSequenceId();<a name="line.243"></a> |
| <span class="sourceLineNo">244</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.244"></a> |
| <span class="sourceLineNo">245</span> if (LOG.isTraceEnabled()) {<a name="line.245"></a> |
| <span class="sourceLineNo">246</span> LOG.trace(Bytes.toString(encodedRegionName) + ", existingValue=" + existingValue +<a name="line.246"></a> |
| <span class="sourceLineNo">247</span> ", completeSequenceId=" + l);<a name="line.247"></a> |
| <span class="sourceLineNo">248</span> }<a name="line.248"></a> |
| <span class="sourceLineNo">249</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue)) {<a name="line.249"></a> |
| <span class="sourceLineNo">250</span> flushedSequenceIdByRegion.put(encodedRegionName, l);<a name="line.250"></a> |
| <span class="sourceLineNo">251</span> } else if (l != HConstants.NO_SEQNUM && l < existingValue) {<a name="line.251"></a> |
| <span class="sourceLineNo">252</span> LOG.warn("RegionServer " + sn + " indicates a last flushed sequence id ("<a name="line.252"></a> |
| <span class="sourceLineNo">253</span> + l + ") that is less than the previous last flushed sequence id ("<a name="line.253"></a> |
| <span class="sourceLineNo">254</span> + existingValue + ") for region " + Bytes.toString(entry.getKey()) + " Ignoring.");<a name="line.254"></a> |
| <span class="sourceLineNo">255</span> }<a name="line.255"></a> |
| <span class="sourceLineNo">256</span> ConcurrentNavigableMap<byte[], Long> storeFlushedSequenceId =<a name="line.256"></a> |
| <span class="sourceLineNo">257</span> computeIfAbsent(storeFlushedSequenceIdsByRegion, encodedRegionName,<a name="line.257"></a> |
| <span class="sourceLineNo">258</span> () -> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR));<a name="line.258"></a> |
| <span class="sourceLineNo">259</span> for (Entry<byte[], Long> storeSeqId : entry.getValue().getStoreSequenceId().entrySet()) {<a name="line.259"></a> |
| <span class="sourceLineNo">260</span> byte[] family = storeSeqId.getKey();<a name="line.260"></a> |
| <span class="sourceLineNo">261</span> existingValue = storeFlushedSequenceId.get(family);<a name="line.261"></a> |
| <span class="sourceLineNo">262</span> l = storeSeqId.getValue();<a name="line.262"></a> |
| <span class="sourceLineNo">263</span> if (LOG.isTraceEnabled()) {<a name="line.263"></a> |
| <span class="sourceLineNo">264</span> LOG.trace(Bytes.toString(encodedRegionName) + ", family=" + Bytes.toString(family) +<a name="line.264"></a> |
| <span class="sourceLineNo">265</span> ", existingValue=" + existingValue + ", completeSequenceId=" + l);<a name="line.265"></a> |
| <span class="sourceLineNo">266</span> }<a name="line.266"></a> |
| <span class="sourceLineNo">267</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.267"></a> |
| <span class="sourceLineNo">268</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue.longValue())) {<a name="line.268"></a> |
| <span class="sourceLineNo">269</span> storeFlushedSequenceId.put(family, l);<a name="line.269"></a> |
| <span class="sourceLineNo">270</span> }<a name="line.270"></a> |
| <span class="sourceLineNo">271</span> }<a name="line.271"></a> |
| <span class="sourceLineNo">272</span> }<a name="line.272"></a> |
| <span class="sourceLineNo">273</span> }<a name="line.273"></a> |
| <span class="sourceLineNo">274</span><a name="line.274"></a> |
| <span class="sourceLineNo">275</span> public void regionServerReport(ServerName sn,<a name="line.275"></a> |
| <span class="sourceLineNo">276</span> ServerMetrics sl) throws YouAreDeadException {<a name="line.276"></a> |
| <span class="sourceLineNo">277</span> checkIsDead(sn, "REPORT");<a name="line.277"></a> |
| <span class="sourceLineNo">278</span> if (null == this.onlineServers.replace(sn, sl)) {<a name="line.278"></a> |
| <span class="sourceLineNo">279</span> // Already have this host+port combo and its just different start code?<a name="line.279"></a> |
| <span class="sourceLineNo">280</span> // Just let the server in. Presume master joining a running cluster.<a name="line.280"></a> |
| <span class="sourceLineNo">281</span> // The only thing we are skipping is passing back to the regionserver<a name="line.281"></a> |
| <span class="sourceLineNo">282</span> // the ServerName to use. Here we presume a master has already done<a name="line.282"></a> |
| <span class="sourceLineNo">283</span> // that so we'll press on with whatever it gave us for ServerName.<a name="line.283"></a> |
| <span class="sourceLineNo">284</span> if (!checkAndRecordNewServer(sn, sl)) {<a name="line.284"></a> |
| <span class="sourceLineNo">285</span> LOG.info("RegionServerReport ignored, could not record the server: " + sn);<a name="line.285"></a> |
| <span class="sourceLineNo">286</span> return; // Not recorded, so no need to move on<a name="line.286"></a> |
| <span class="sourceLineNo">287</span> }<a name="line.287"></a> |
| <span class="sourceLineNo">288</span> }<a name="line.288"></a> |
| <span class="sourceLineNo">289</span> updateLastFlushedSequenceIds(sn, sl);<a name="line.289"></a> |
| <span class="sourceLineNo">290</span> }<a name="line.290"></a> |
| <span class="sourceLineNo">291</span><a name="line.291"></a> |
| <span class="sourceLineNo">292</span> /**<a name="line.292"></a> |
| <span class="sourceLineNo">293</span> * Check is a server of same host and port already exists,<a name="line.293"></a> |
| <span class="sourceLineNo">294</span> * if not, or the existed one got a smaller start code, record it.<a name="line.294"></a> |
| <span class="sourceLineNo">295</span> *<a name="line.295"></a> |
| <span class="sourceLineNo">296</span> * @param serverName the server to check and record<a name="line.296"></a> |
| <span class="sourceLineNo">297</span> * @param sl the server load on the server<a name="line.297"></a> |
| <span class="sourceLineNo">298</span> * @return true if the server is recorded, otherwise, false<a name="line.298"></a> |
| <span class="sourceLineNo">299</span> */<a name="line.299"></a> |
| <span class="sourceLineNo">300</span> boolean checkAndRecordNewServer(final ServerName serverName, final ServerMetrics sl) {<a name="line.300"></a> |
| <span class="sourceLineNo">301</span> ServerName existingServer = null;<a name="line.301"></a> |
| <span class="sourceLineNo">302</span> synchronized (this.onlineServers) {<a name="line.302"></a> |
| <span class="sourceLineNo">303</span> existingServer = findServerWithSameHostnamePortWithLock(serverName);<a name="line.303"></a> |
| <span class="sourceLineNo">304</span> if (existingServer != null && (existingServer.getStartcode() > serverName.getStartcode())) {<a name="line.304"></a> |
| <span class="sourceLineNo">305</span> LOG.info("Server serverName=" + serverName + " rejected; we already have "<a name="line.305"></a> |
| <span class="sourceLineNo">306</span> + existingServer.toString() + " registered with same hostname and port");<a name="line.306"></a> |
| <span class="sourceLineNo">307</span> return false;<a name="line.307"></a> |
| <span class="sourceLineNo">308</span> }<a name="line.308"></a> |
| <span class="sourceLineNo">309</span> recordNewServerWithLock(serverName, sl);<a name="line.309"></a> |
| <span class="sourceLineNo">310</span> }<a name="line.310"></a> |
| <span class="sourceLineNo">311</span><a name="line.311"></a> |
| <span class="sourceLineNo">312</span> // Tell our listeners that a server was added<a name="line.312"></a> |
| <span class="sourceLineNo">313</span> if (!this.listeners.isEmpty()) {<a name="line.313"></a> |
| <span class="sourceLineNo">314</span> for (ServerListener listener : this.listeners) {<a name="line.314"></a> |
| <span class="sourceLineNo">315</span> listener.serverAdded(serverName);<a name="line.315"></a> |
| <span class="sourceLineNo">316</span> }<a name="line.316"></a> |
| <span class="sourceLineNo">317</span> }<a name="line.317"></a> |
| <span class="sourceLineNo">318</span><a name="line.318"></a> |
| <span class="sourceLineNo">319</span> // Note that we assume that same ts means same server, and don't expire in that case.<a name="line.319"></a> |
| <span class="sourceLineNo">320</span> // TODO: ts can theoretically collide due to clock shifts, so this is a bit hacky.<a name="line.320"></a> |
| <span class="sourceLineNo">321</span> if (existingServer != null &&<a name="line.321"></a> |
| <span class="sourceLineNo">322</span> (existingServer.getStartcode() < serverName.getStartcode())) {<a name="line.322"></a> |
| <span class="sourceLineNo">323</span> LOG.info("Triggering server recovery; existingServer " +<a name="line.323"></a> |
| <span class="sourceLineNo">324</span> existingServer + " looks stale, new server:" + serverName);<a name="line.324"></a> |
| <span class="sourceLineNo">325</span> expireServer(existingServer);<a name="line.325"></a> |
| <span class="sourceLineNo">326</span> }<a name="line.326"></a> |
| <span class="sourceLineNo">327</span> return true;<a name="line.327"></a> |
| <span class="sourceLineNo">328</span> }<a name="line.328"></a> |
| <span class="sourceLineNo">329</span><a name="line.329"></a> |
| <span class="sourceLineNo">330</span> /**<a name="line.330"></a> |
| <span class="sourceLineNo">331</span> * Find out the region servers crashed between the crash of the previous master instance and the<a name="line.331"></a> |
| <span class="sourceLineNo">332</span> * current master instance and schedule SCP for them.<a name="line.332"></a> |
| <span class="sourceLineNo">333</span> * <p/><a name="line.333"></a> |
| <span class="sourceLineNo">334</span> * Since the {@code RegionServerTracker} has already helped us to construct the online servers set<a name="line.334"></a> |
| <span class="sourceLineNo">335</span> * by scanning zookeeper, now we can compare the online servers with {@code liveServersFromWALDir}<a name="line.335"></a> |
| <span class="sourceLineNo">336</span> * to find out whether there are servers which are already dead.<a name="line.336"></a> |
| <span class="sourceLineNo">337</span> * <p/><a name="line.337"></a> |
| <span class="sourceLineNo">338</span> * Must be called inside the initialization method of {@code RegionServerTracker} to avoid<a name="line.338"></a> |
| <span class="sourceLineNo">339</span> * concurrency issue.<a name="line.339"></a> |
| <span class="sourceLineNo">340</span> * @param deadServersFromPE the region servers which already have a SCP associated.<a name="line.340"></a> |
| <span class="sourceLineNo">341</span> * @param liveServersFromWALDir the live region servers from wal directory.<a name="line.341"></a> |
| <span class="sourceLineNo">342</span> */<a name="line.342"></a> |
| <span class="sourceLineNo">343</span> void findDeadServersAndProcess(Set<ServerName> deadServersFromPE,<a name="line.343"></a> |
| <span class="sourceLineNo">344</span> Set<ServerName> liveServersFromWALDir) {<a name="line.344"></a> |
| <span class="sourceLineNo">345</span> deadServersFromPE.forEach(deadservers::putIfAbsent);<a name="line.345"></a> |
| <span class="sourceLineNo">346</span> liveServersFromWALDir.stream().filter(sn -> !onlineServers.containsKey(sn))<a name="line.346"></a> |
| <span class="sourceLineNo">347</span> .forEach(this::expireServer);<a name="line.347"></a> |
| <span class="sourceLineNo">348</span> }<a name="line.348"></a> |
| <span class="sourceLineNo">349</span><a name="line.349"></a> |
| <span class="sourceLineNo">350</span> /**<a name="line.350"></a> |
| <span class="sourceLineNo">351</span> * Checks if the clock skew between the server and the master. If the clock skew exceeds the<a name="line.351"></a> |
| <span class="sourceLineNo">352</span> * configured max, it will throw an exception; if it exceeds the configured warning threshold,<a name="line.352"></a> |
| <span class="sourceLineNo">353</span> * it will log a warning but start normally.<a name="line.353"></a> |
| <span class="sourceLineNo">354</span> * @param serverName Incoming servers's name<a name="line.354"></a> |
| <span class="sourceLineNo">355</span> * @param serverCurrentTime<a name="line.355"></a> |
| <span class="sourceLineNo">356</span> * @throws ClockOutOfSyncException if the skew exceeds the configured max value<a name="line.356"></a> |
| <span class="sourceLineNo">357</span> */<a name="line.357"></a> |
| <span class="sourceLineNo">358</span> private void checkClockSkew(final ServerName serverName, final long serverCurrentTime)<a name="line.358"></a> |
| <span class="sourceLineNo">359</span> throws ClockOutOfSyncException {<a name="line.359"></a> |
| <span class="sourceLineNo">360</span> long skew = Math.abs(EnvironmentEdgeManager.currentTime() - serverCurrentTime);<a name="line.360"></a> |
| <span class="sourceLineNo">361</span> if (skew > maxSkew) {<a name="line.361"></a> |
| <span class="sourceLineNo">362</span> String message = "Server " + serverName + " has been " +<a name="line.362"></a> |
| <span class="sourceLineNo">363</span> "rejected; Reported time is too far out of sync with master. " +<a name="line.363"></a> |
| <span class="sourceLineNo">364</span> "Time difference of " + skew + "ms > max allowed of " + maxSkew + "ms";<a name="line.364"></a> |
| <span class="sourceLineNo">365</span> LOG.warn(message);<a name="line.365"></a> |
| <span class="sourceLineNo">366</span> throw new ClockOutOfSyncException(message);<a name="line.366"></a> |
| <span class="sourceLineNo">367</span> } else if (skew > warningSkew){<a name="line.367"></a> |
| <span class="sourceLineNo">368</span> String message = "Reported time for server " + serverName + " is out of sync with master " +<a name="line.368"></a> |
| <span class="sourceLineNo">369</span> "by " + skew + "ms. (Warning threshold is " + warningSkew + "ms; " +<a name="line.369"></a> |
| <span class="sourceLineNo">370</span> "error threshold is " + maxSkew + "ms)";<a name="line.370"></a> |
| <span class="sourceLineNo">371</span> LOG.warn(message);<a name="line.371"></a> |
| <span class="sourceLineNo">372</span> }<a name="line.372"></a> |
| <span class="sourceLineNo">373</span> }<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> * Called when RegionServer first reports in for duty and thereafter each<a name="line.376"></a> |
| <span class="sourceLineNo">377</span> * time it heartbeats to make sure it is has not been figured for dead.<a name="line.377"></a> |
| <span class="sourceLineNo">378</span> * If this server is on the dead list, reject it with a YouAreDeadException.<a name="line.378"></a> |
| <span class="sourceLineNo">379</span> * If it was dead but came back with a new start code, remove the old entry<a name="line.379"></a> |
| <span class="sourceLineNo">380</span> * from the dead list.<a name="line.380"></a> |
| <span class="sourceLineNo">381</span> * @param what START or REPORT<a name="line.381"></a> |
| <span class="sourceLineNo">382</span> */<a name="line.382"></a> |
| <span class="sourceLineNo">383</span> private void checkIsDead(final ServerName serverName, final String what)<a name="line.383"></a> |
| <span class="sourceLineNo">384</span> throws YouAreDeadException {<a name="line.384"></a> |
| <span class="sourceLineNo">385</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.385"></a> |
| <span class="sourceLineNo">386</span> // Exact match: host name, port and start code all match with existing one of the<a name="line.386"></a> |
| <span class="sourceLineNo">387</span> // dead servers. So, this server must be dead. Tell it to kill itself.<a name="line.387"></a> |
| <span class="sourceLineNo">388</span> String message = "Server " + what + " rejected; currently processing " +<a name="line.388"></a> |
| <span class="sourceLineNo">389</span> serverName + " as dead server";<a name="line.389"></a> |
| <span class="sourceLineNo">390</span> LOG.debug(message);<a name="line.390"></a> |
| <span class="sourceLineNo">391</span> throw new YouAreDeadException(message);<a name="line.391"></a> |
| <span class="sourceLineNo">392</span> }<a name="line.392"></a> |
| <span class="sourceLineNo">393</span> // Remove dead server with same hostname and port of newly checking in rs after master<a name="line.393"></a> |
| <span class="sourceLineNo">394</span> // initialization. See HBASE-5916 for more information.<a name="line.394"></a> |
| <span class="sourceLineNo">395</span> if ((this.master == null || this.master.isInitialized()) &&<a name="line.395"></a> |
| <span class="sourceLineNo">396</span> this.deadservers.cleanPreviousInstance(serverName)) {<a name="line.396"></a> |
| <span class="sourceLineNo">397</span> // This server has now become alive after we marked it as dead.<a name="line.397"></a> |
| <span class="sourceLineNo">398</span> // We removed it's previous entry from the dead list to reflect it.<a name="line.398"></a> |
| <span class="sourceLineNo">399</span> LOG.debug("{} {} came back up, removed it from the dead servers list", what, serverName);<a name="line.399"></a> |
| <span class="sourceLineNo">400</span> }<a name="line.400"></a> |
| <span class="sourceLineNo">401</span> }<a name="line.401"></a> |
| <span class="sourceLineNo">402</span><a name="line.402"></a> |
| <span class="sourceLineNo">403</span> /**<a name="line.403"></a> |
| <span class="sourceLineNo">404</span> * Assumes onlineServers is locked.<a name="line.404"></a> |
| <span class="sourceLineNo">405</span> * @return ServerName with matching hostname and port.<a name="line.405"></a> |
| <span class="sourceLineNo">406</span> */<a name="line.406"></a> |
| <span class="sourceLineNo">407</span> private ServerName findServerWithSameHostnamePortWithLock(<a name="line.407"></a> |
| <span class="sourceLineNo">408</span> final ServerName serverName) {<a name="line.408"></a> |
| <span class="sourceLineNo">409</span> ServerName end = ServerName.valueOf(serverName.getHostname(), serverName.getPort(),<a name="line.409"></a> |
| <span class="sourceLineNo">410</span> Long.MAX_VALUE);<a name="line.410"></a> |
| <span class="sourceLineNo">411</span><a name="line.411"></a> |
| <span class="sourceLineNo">412</span> ServerName r = onlineServers.lowerKey(end);<a name="line.412"></a> |
| <span class="sourceLineNo">413</span> if (r != null) {<a name="line.413"></a> |
| <span class="sourceLineNo">414</span> if (ServerName.isSameAddress(r, serverName)) {<a name="line.414"></a> |
| <span class="sourceLineNo">415</span> return r;<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> return null;<a name="line.418"></a> |
| <span class="sourceLineNo">419</span> }<a name="line.419"></a> |
| <span class="sourceLineNo">420</span><a name="line.420"></a> |
| <span class="sourceLineNo">421</span> /**<a name="line.421"></a> |
| <span class="sourceLineNo">422</span> * Adds the onlineServers list. onlineServers should be locked.<a name="line.422"></a> |
| <span class="sourceLineNo">423</span> * @param serverName The remote servers name.<a name="line.423"></a> |
| <span class="sourceLineNo">424</span> */<a name="line.424"></a> |
| <span class="sourceLineNo">425</span> void recordNewServerWithLock(final ServerName serverName, final ServerMetrics sl) {<a name="line.425"></a> |
| <span class="sourceLineNo">426</span> LOG.info("Registering regionserver=" + serverName);<a name="line.426"></a> |
| <span class="sourceLineNo">427</span> this.onlineServers.put(serverName, sl);<a name="line.427"></a> |
| <span class="sourceLineNo">428</span> }<a name="line.428"></a> |
| <span class="sourceLineNo">429</span><a name="line.429"></a> |
| <span class="sourceLineNo">430</span> public ConcurrentNavigableMap<byte[], Long> getFlushedSequenceIdByRegion() {<a name="line.430"></a> |
| <span class="sourceLineNo">431</span> return flushedSequenceIdByRegion;<a name="line.431"></a> |
| <span class="sourceLineNo">432</span> }<a name="line.432"></a> |
| <span class="sourceLineNo">433</span><a name="line.433"></a> |
| <span class="sourceLineNo">434</span> public RegionStoreSequenceIds getLastFlushedSequenceId(byte[] encodedRegionName) {<a name="line.434"></a> |
| <span class="sourceLineNo">435</span> RegionStoreSequenceIds.Builder builder = RegionStoreSequenceIds.newBuilder();<a name="line.435"></a> |
| <span class="sourceLineNo">436</span> Long seqId = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.436"></a> |
| <span class="sourceLineNo">437</span> builder.setLastFlushedSequenceId(seqId != null ? seqId.longValue() : HConstants.NO_SEQNUM);<a name="line.437"></a> |
| <span class="sourceLineNo">438</span> Map<byte[], Long> storeFlushedSequenceId =<a name="line.438"></a> |
| <span class="sourceLineNo">439</span> storeFlushedSequenceIdsByRegion.get(encodedRegionName);<a name="line.439"></a> |
| <span class="sourceLineNo">440</span> if (storeFlushedSequenceId != null) {<a name="line.440"></a> |
| <span class="sourceLineNo">441</span> for (Map.Entry<byte[], Long> entry : storeFlushedSequenceId.entrySet()) {<a name="line.441"></a> |
| <span class="sourceLineNo">442</span> builder.addStoreSequenceId(StoreSequenceId.newBuilder()<a name="line.442"></a> |
| <span class="sourceLineNo">443</span> .setFamilyName(UnsafeByteOperations.unsafeWrap(entry.getKey()))<a name="line.443"></a> |
| <span class="sourceLineNo">444</span> .setSequenceId(entry.getValue().longValue()).build());<a name="line.444"></a> |
| <span class="sourceLineNo">445</span> }<a name="line.445"></a> |
| <span class="sourceLineNo">446</span> }<a name="line.446"></a> |
| <span class="sourceLineNo">447</span> return builder.build();<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> * @param serverName<a name="line.451"></a> |
| <span class="sourceLineNo">452</span> * @return ServerMetrics if serverName is known else null<a name="line.452"></a> |
| <span class="sourceLineNo">453</span> */<a name="line.453"></a> |
| <span class="sourceLineNo">454</span> public ServerMetrics getLoad(final ServerName serverName) {<a name="line.454"></a> |
| <span class="sourceLineNo">455</span> return this.onlineServers.get(serverName);<a name="line.455"></a> |
| <span class="sourceLineNo">456</span> }<a name="line.456"></a> |
| <span class="sourceLineNo">457</span><a name="line.457"></a> |
| <span class="sourceLineNo">458</span> /**<a name="line.458"></a> |
| <span class="sourceLineNo">459</span> * Compute the average load across all region servers.<a name="line.459"></a> |
| <span class="sourceLineNo">460</span> * Currently, this uses a very naive computation - just uses the number of<a name="line.460"></a> |
| <span class="sourceLineNo">461</span> * regions being served, ignoring stats about number of requests.<a name="line.461"></a> |
| <span class="sourceLineNo">462</span> * @return the average load<a name="line.462"></a> |
| <span class="sourceLineNo">463</span> */<a name="line.463"></a> |
| <span class="sourceLineNo">464</span> public double getAverageLoad() {<a name="line.464"></a> |
| <span class="sourceLineNo">465</span> int totalLoad = 0;<a name="line.465"></a> |
| <span class="sourceLineNo">466</span> int numServers = 0;<a name="line.466"></a> |
| <span class="sourceLineNo">467</span> for (ServerMetrics sl : this.onlineServers.values()) {<a name="line.467"></a> |
| <span class="sourceLineNo">468</span> numServers++;<a name="line.468"></a> |
| <span class="sourceLineNo">469</span> totalLoad += sl.getRegionMetrics().size();<a name="line.469"></a> |
| <span class="sourceLineNo">470</span> }<a name="line.470"></a> |
| <span class="sourceLineNo">471</span> return numServers == 0 ? 0 :<a name="line.471"></a> |
| <span class="sourceLineNo">472</span> (double)totalLoad / (double)numServers;<a name="line.472"></a> |
| <span class="sourceLineNo">473</span> }<a name="line.473"></a> |
| <span class="sourceLineNo">474</span><a name="line.474"></a> |
| <span class="sourceLineNo">475</span> /** @return the count of active regionservers */<a name="line.475"></a> |
| <span class="sourceLineNo">476</span> public int countOfRegionServers() {<a name="line.476"></a> |
| <span class="sourceLineNo">477</span> // Presumes onlineServers is a concurrent map<a name="line.477"></a> |
| <span class="sourceLineNo">478</span> return this.onlineServers.size();<a name="line.478"></a> |
| <span class="sourceLineNo">479</span> }<a name="line.479"></a> |
| <span class="sourceLineNo">480</span><a name="line.480"></a> |
| <span class="sourceLineNo">481</span> /**<a name="line.481"></a> |
| <span class="sourceLineNo">482</span> * @return Read-only map of servers to serverinfo<a name="line.482"></a> |
| <span class="sourceLineNo">483</span> */<a name="line.483"></a> |
| <span class="sourceLineNo">484</span> public Map<ServerName, ServerMetrics> getOnlineServers() {<a name="line.484"></a> |
| <span class="sourceLineNo">485</span> // Presumption is that iterating the returned Map is OK.<a name="line.485"></a> |
| <span class="sourceLineNo">486</span> synchronized (this.onlineServers) {<a name="line.486"></a> |
| <span class="sourceLineNo">487</span> return Collections.unmodifiableMap(this.onlineServers);<a name="line.487"></a> |
| <span class="sourceLineNo">488</span> }<a name="line.488"></a> |
| <span class="sourceLineNo">489</span> }<a name="line.489"></a> |
| <span class="sourceLineNo">490</span><a name="line.490"></a> |
| <span class="sourceLineNo">491</span> public DeadServer getDeadServers() {<a name="line.491"></a> |
| <span class="sourceLineNo">492</span> return this.deadservers;<a name="line.492"></a> |
| <span class="sourceLineNo">493</span> }<a name="line.493"></a> |
| <span class="sourceLineNo">494</span><a name="line.494"></a> |
| <span class="sourceLineNo">495</span> /**<a name="line.495"></a> |
| <span class="sourceLineNo">496</span> * Checks if any dead servers are currently in progress.<a name="line.496"></a> |
| <span class="sourceLineNo">497</span> * @return true if any RS are being processed as dead, false if not<a name="line.497"></a> |
| <span class="sourceLineNo">498</span> */<a name="line.498"></a> |
| <span class="sourceLineNo">499</span> public boolean areDeadServersInProgress() throws IOException {<a name="line.499"></a> |
| <span class="sourceLineNo">500</span> return master.getProcedures().stream()<a name="line.500"></a> |
| <span class="sourceLineNo">501</span> .anyMatch(p -> !p.isFinished() && p instanceof ServerCrashProcedure);<a name="line.501"></a> |
| <span class="sourceLineNo">502</span> }<a name="line.502"></a> |
| <span class="sourceLineNo">503</span><a name="line.503"></a> |
| <span class="sourceLineNo">504</span> void letRegionServersShutdown() {<a name="line.504"></a> |
| <span class="sourceLineNo">505</span> long previousLogTime = 0;<a name="line.505"></a> |
| <span class="sourceLineNo">506</span> ServerName sn = master.getServerName();<a name="line.506"></a> |
| <span class="sourceLineNo">507</span> ZKWatcher zkw = master.getZooKeeper();<a name="line.507"></a> |
| <span class="sourceLineNo">508</span> int onlineServersCt;<a name="line.508"></a> |
| <span class="sourceLineNo">509</span> while ((onlineServersCt = onlineServers.size()) > 0){<a name="line.509"></a> |
| <span class="sourceLineNo">510</span><a name="line.510"></a> |
| <span class="sourceLineNo">511</span> if (System.currentTimeMillis() > (previousLogTime + 1000)) {<a name="line.511"></a> |
| <span class="sourceLineNo">512</span> Set<ServerName> remainingServers = onlineServers.keySet();<a name="line.512"></a> |
| <span class="sourceLineNo">513</span> synchronized (onlineServers) {<a name="line.513"></a> |
| <span class="sourceLineNo">514</span> if (remainingServers.size() == 1 && remainingServers.contains(sn)) {<a name="line.514"></a> |
| <span class="sourceLineNo">515</span> // Master will delete itself later.<a name="line.515"></a> |
| <span class="sourceLineNo">516</span> return;<a name="line.516"></a> |
| <span class="sourceLineNo">517</span> }<a name="line.517"></a> |
| <span class="sourceLineNo">518</span> }<a name="line.518"></a> |
| <span class="sourceLineNo">519</span> StringBuilder sb = new StringBuilder();<a name="line.519"></a> |
| <span class="sourceLineNo">520</span> // It's ok here to not sync on onlineServers - merely logging<a name="line.520"></a> |
| <span class="sourceLineNo">521</span> for (ServerName key : remainingServers) {<a name="line.521"></a> |
| <span class="sourceLineNo">522</span> if (sb.length() > 0) {<a name="line.522"></a> |
| <span class="sourceLineNo">523</span> sb.append(", ");<a name="line.523"></a> |
| <span class="sourceLineNo">524</span> }<a name="line.524"></a> |
| <span class="sourceLineNo">525</span> sb.append(key);<a name="line.525"></a> |
| <span class="sourceLineNo">526</span> }<a name="line.526"></a> |
| <span class="sourceLineNo">527</span> LOG.info("Waiting on regionserver(s) " + sb.toString());<a name="line.527"></a> |
| <span class="sourceLineNo">528</span> previousLogTime = System.currentTimeMillis();<a name="line.528"></a> |
| <span class="sourceLineNo">529</span> }<a name="line.529"></a> |
| <span class="sourceLineNo">530</span><a name="line.530"></a> |
| <span class="sourceLineNo">531</span> try {<a name="line.531"></a> |
| <span class="sourceLineNo">532</span> List<String> servers = getRegionServersInZK(zkw);<a name="line.532"></a> |
| <span class="sourceLineNo">533</span> if (servers == null || servers.isEmpty() || (servers.size() == 1<a name="line.533"></a> |
| <span class="sourceLineNo">534</span> && servers.contains(sn.toString()))) {<a name="line.534"></a> |
| <span class="sourceLineNo">535</span> LOG.info("ZK shows there is only the master self online, exiting now");<a name="line.535"></a> |
| <span class="sourceLineNo">536</span> // Master could have lost some ZK events, no need to wait more.<a name="line.536"></a> |
| <span class="sourceLineNo">537</span> break;<a name="line.537"></a> |
| <span class="sourceLineNo">538</span> }<a name="line.538"></a> |
| <span class="sourceLineNo">539</span> } catch (KeeperException ke) {<a name="line.539"></a> |
| <span class="sourceLineNo">540</span> LOG.warn("Failed to list regionservers", ke);<a name="line.540"></a> |
| <span class="sourceLineNo">541</span> // ZK is malfunctioning, don't hang here<a name="line.541"></a> |
| <span class="sourceLineNo">542</span> break;<a name="line.542"></a> |
| <span class="sourceLineNo">543</span> }<a name="line.543"></a> |
| <span class="sourceLineNo">544</span> synchronized (onlineServers) {<a name="line.544"></a> |
| <span class="sourceLineNo">545</span> try {<a name="line.545"></a> |
| <span class="sourceLineNo">546</span> if (onlineServersCt == onlineServers.size()) onlineServers.wait(100);<a name="line.546"></a> |
| <span class="sourceLineNo">547</span> } catch (InterruptedException ignored) {<a name="line.547"></a> |
| <span class="sourceLineNo">548</span> // continue<a name="line.548"></a> |
| <span class="sourceLineNo">549</span> }<a name="line.549"></a> |
| <span class="sourceLineNo">550</span> }<a name="line.550"></a> |
| <span class="sourceLineNo">551</span> }<a name="line.551"></a> |
| <span class="sourceLineNo">552</span> }<a name="line.552"></a> |
| <span class="sourceLineNo">553</span><a name="line.553"></a> |
| <span class="sourceLineNo">554</span> private List<String> getRegionServersInZK(final ZKWatcher zkw)<a name="line.554"></a> |
| <span class="sourceLineNo">555</span> throws KeeperException {<a name="line.555"></a> |
| <span class="sourceLineNo">556</span> return ZKUtil.listChildrenNoWatch(zkw, zkw.getZNodePaths().rsZNode);<a name="line.556"></a> |
| <span class="sourceLineNo">557</span> }<a name="line.557"></a> |
| <span class="sourceLineNo">558</span><a name="line.558"></a> |
| <span class="sourceLineNo">559</span> /**<a name="line.559"></a> |
| <span class="sourceLineNo">560</span> * Expire the passed server. Add it to list of dead servers and queue a shutdown processing.<a name="line.560"></a> |
| <span class="sourceLineNo">561</span> * @return pid if we queued a ServerCrashProcedure else {@link Procedure#NO_PROC_ID} if we did<a name="line.561"></a> |
| <span class="sourceLineNo">562</span> * not (could happen for many reasons including the fact that its this server that is<a name="line.562"></a> |
| <span class="sourceLineNo">563</span> * going down or we already have queued an SCP for this server or SCP processing is<a name="line.563"></a> |
| <span class="sourceLineNo">564</span> * currently disabled because we are in startup phase).<a name="line.564"></a> |
| <span class="sourceLineNo">565</span> */<a name="line.565"></a> |
| <span class="sourceLineNo">566</span> // Redo test so we can make this protected.<a name="line.566"></a> |
| <span class="sourceLineNo">567</span> public synchronized long expireServer(final ServerName serverName) {<a name="line.567"></a> |
| <span class="sourceLineNo">568</span> return expireServer(serverName, false);<a name="line.568"></a> |
| <span class="sourceLineNo">569</span><a name="line.569"></a> |
| <span class="sourceLineNo">570</span> }<a name="line.570"></a> |
| <span class="sourceLineNo">571</span><a name="line.571"></a> |
| <span class="sourceLineNo">572</span> synchronized long expireServer(final ServerName serverName, boolean force) {<a name="line.572"></a> |
| <span class="sourceLineNo">573</span> // THIS server is going down... can't handle our own expiration.<a name="line.573"></a> |
| <span class="sourceLineNo">574</span> if (serverName.equals(master.getServerName())) {<a name="line.574"></a> |
| <span class="sourceLineNo">575</span> if (!(master.isAborted() || master.isStopped())) {<a name="line.575"></a> |
| <span class="sourceLineNo">576</span> master.stop("We lost our znode?");<a name="line.576"></a> |
| <span class="sourceLineNo">577</span> }<a name="line.577"></a> |
| <span class="sourceLineNo">578</span> return Procedure.NO_PROC_ID;<a name="line.578"></a> |
| <span class="sourceLineNo">579</span> }<a name="line.579"></a> |
| <span class="sourceLineNo">580</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.580"></a> |
| <span class="sourceLineNo">581</span> LOG.warn("Expiration called on {} but already in DeadServer", serverName);<a name="line.581"></a> |
| <span class="sourceLineNo">582</span> return Procedure.NO_PROC_ID;<a name="line.582"></a> |
| <span class="sourceLineNo">583</span> }<a name="line.583"></a> |
| <span class="sourceLineNo">584</span> moveFromOnlineToDeadServers(serverName);<a name="line.584"></a> |
| <span class="sourceLineNo">585</span><a name="line.585"></a> |
| <span class="sourceLineNo">586</span> // If server is in draining mode, remove corresponding znode<a name="line.586"></a> |
| <span class="sourceLineNo">587</span> // In some tests, the mocked HM may not have ZK Instance, hence null check<a name="line.587"></a> |
| <span class="sourceLineNo">588</span> if (master.getZooKeeper() != null) {<a name="line.588"></a> |
| <span class="sourceLineNo">589</span> String drainingZnode = ZNodePaths<a name="line.589"></a> |
| <span class="sourceLineNo">590</span> .joinZNode(master.getZooKeeper().getZNodePaths().drainingZNode, serverName.getServerName());<a name="line.590"></a> |
| <span class="sourceLineNo">591</span> try {<a name="line.591"></a> |
| <span class="sourceLineNo">592</span> ZKUtil.deleteNodeFailSilent(master.getZooKeeper(), drainingZnode);<a name="line.592"></a> |
| <span class="sourceLineNo">593</span> } catch (KeeperException e) {<a name="line.593"></a> |
| <span class="sourceLineNo">594</span> LOG.warn("Error deleting the draining znode for stopping server " + serverName.getServerName(), e);<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> // If cluster is going down, yes, servers are going to be expiring; don't<a name="line.598"></a> |
| <span class="sourceLineNo">599</span> // process as a dead server<a name="line.599"></a> |
| <span class="sourceLineNo">600</span> if (isClusterShutdown()) {<a name="line.600"></a> |
| <span class="sourceLineNo">601</span> LOG.info("Cluster shutdown set; " + serverName +<a name="line.601"></a> |
| <span class="sourceLineNo">602</span> " expired; onlineServers=" + this.onlineServers.size());<a name="line.602"></a> |
| <span class="sourceLineNo">603</span> if (this.onlineServers.isEmpty()) {<a name="line.603"></a> |
| <span class="sourceLineNo">604</span> master.stop("Cluster shutdown set; onlineServer=0");<a name="line.604"></a> |
| <span class="sourceLineNo">605</span> }<a name="line.605"></a> |
| <span class="sourceLineNo">606</span> return Procedure.NO_PROC_ID;<a name="line.606"></a> |
| <span class="sourceLineNo">607</span> }<a name="line.607"></a> |
| <span class="sourceLineNo">608</span> LOG.info("Processing expiration of " + serverName + " on " + this.master.getServerName());<a name="line.608"></a> |
| <span class="sourceLineNo">609</span> long pid = master.getAssignmentManager().submitServerCrash(serverName, true, force);<a name="line.609"></a> |
| <span class="sourceLineNo">610</span> // Tell our listeners that a server was removed<a name="line.610"></a> |
| <span class="sourceLineNo">611</span> if (!this.listeners.isEmpty()) {<a name="line.611"></a> |
| <span class="sourceLineNo">612</span> this.listeners.stream().forEach(l -> l.serverRemoved(serverName));<a name="line.612"></a> |
| <span class="sourceLineNo">613</span> }<a name="line.613"></a> |
| <span class="sourceLineNo">614</span> // trigger a persist of flushedSeqId<a name="line.614"></a> |
| <span class="sourceLineNo">615</span> if (flushedSeqIdFlusher != null) {<a name="line.615"></a> |
| <span class="sourceLineNo">616</span> flushedSeqIdFlusher.triggerNow();<a name="line.616"></a> |
| <span class="sourceLineNo">617</span> }<a name="line.617"></a> |
| <span class="sourceLineNo">618</span> return pid;<a name="line.618"></a> |
| <span class="sourceLineNo">619</span> }<a name="line.619"></a> |
| <span class="sourceLineNo">620</span><a name="line.620"></a> |
| <span class="sourceLineNo">621</span> /**<a name="line.621"></a> |
| <span class="sourceLineNo">622</span> * Called when server has expired.<a name="line.622"></a> |
| <span class="sourceLineNo">623</span> */<a name="line.623"></a> |
| <span class="sourceLineNo">624</span> // Locking in this class needs cleanup.<a name="line.624"></a> |
| <span class="sourceLineNo">625</span> public synchronized void moveFromOnlineToDeadServers(final ServerName sn) {<a name="line.625"></a> |
| <span class="sourceLineNo">626</span> synchronized (this.onlineServers) {<a name="line.626"></a> |
| <span class="sourceLineNo">627</span> boolean online = this.onlineServers.containsKey(sn);<a name="line.627"></a> |
| <span class="sourceLineNo">628</span> if (online) {<a name="line.628"></a> |
| <span class="sourceLineNo">629</span> // Remove the server from the known servers lists and update load info BUT<a name="line.629"></a> |
| <span class="sourceLineNo">630</span> // add to deadservers first; do this so it'll show in dead servers list if<a name="line.630"></a> |
| <span class="sourceLineNo">631</span> // not in online servers list.<a name="line.631"></a> |
| <span class="sourceLineNo">632</span> this.deadservers.putIfAbsent(sn);<a name="line.632"></a> |
| <span class="sourceLineNo">633</span> this.onlineServers.remove(sn);<a name="line.633"></a> |
| <span class="sourceLineNo">634</span> onlineServers.notifyAll();<a name="line.634"></a> |
| <span class="sourceLineNo">635</span> } else {<a name="line.635"></a> |
| <span class="sourceLineNo">636</span> // If not online, that is odd but may happen if 'Unknown Servers' -- where meta<a name="line.636"></a> |
| <span class="sourceLineNo">637</span> // has references to servers not online nor in dead servers list. If<a name="line.637"></a> |
| <span class="sourceLineNo">638</span> // 'Unknown Server', don't add to DeadServers else will be there for ever.<a name="line.638"></a> |
| <span class="sourceLineNo">639</span> LOG.trace("Expiration of {} but server not online", sn);<a name="line.639"></a> |
| <span class="sourceLineNo">640</span> }<a name="line.640"></a> |
| <span class="sourceLineNo">641</span> }<a name="line.641"></a> |
| <span class="sourceLineNo">642</span> }<a name="line.642"></a> |
| <span class="sourceLineNo">643</span><a name="line.643"></a> |
| <span class="sourceLineNo">644</span> /*<a name="line.644"></a> |
| <span class="sourceLineNo">645</span> * Remove the server from the drain list.<a name="line.645"></a> |
| <span class="sourceLineNo">646</span> */<a name="line.646"></a> |
| <span class="sourceLineNo">647</span> public synchronized boolean removeServerFromDrainList(final ServerName sn) {<a name="line.647"></a> |
| <span class="sourceLineNo">648</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.648"></a> |
| <span class="sourceLineNo">649</span> // <hostname> , <port> , <startcode><a name="line.649"></a> |
| <span class="sourceLineNo">650</span><a name="line.650"></a> |
| <span class="sourceLineNo">651</span> if (!this.isServerOnline(sn)) {<a name="line.651"></a> |
| <span class="sourceLineNo">652</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.652"></a> |
| <span class="sourceLineNo">653</span> "Removing from draining list anyway, as requested.");<a name="line.653"></a> |
| <span class="sourceLineNo">654</span> }<a name="line.654"></a> |
| <span class="sourceLineNo">655</span> // Remove the server from the draining servers lists.<a name="line.655"></a> |
| <span class="sourceLineNo">656</span> return this.drainingServers.remove(sn);<a name="line.656"></a> |
| <span class="sourceLineNo">657</span> }<a name="line.657"></a> |
| <span class="sourceLineNo">658</span><a name="line.658"></a> |
| <span class="sourceLineNo">659</span> /**<a name="line.659"></a> |
| <span class="sourceLineNo">660</span> * Add the server to the drain list.<a name="line.660"></a> |
| <span class="sourceLineNo">661</span> * @param sn<a name="line.661"></a> |
| <span class="sourceLineNo">662</span> * @return True if the server is added or the server is already on the drain list.<a name="line.662"></a> |
| <span class="sourceLineNo">663</span> */<a name="line.663"></a> |
| <span class="sourceLineNo">664</span> public synchronized boolean addServerToDrainList(final ServerName sn) {<a name="line.664"></a> |
| <span class="sourceLineNo">665</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.665"></a> |
| <span class="sourceLineNo">666</span> // <hostname> , <port> , <startcode><a name="line.666"></a> |
| <span class="sourceLineNo">667</span><a name="line.667"></a> |
| <span class="sourceLineNo">668</span> if (!this.isServerOnline(sn)) {<a name="line.668"></a> |
| <span class="sourceLineNo">669</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.669"></a> |
| <span class="sourceLineNo">670</span> "Ignoring request to add it to draining list.");<a name="line.670"></a> |
| <span class="sourceLineNo">671</span> return false;<a name="line.671"></a> |
| <span class="sourceLineNo">672</span> }<a name="line.672"></a> |
| <span class="sourceLineNo">673</span> // Add the server to the draining servers lists, if it's not already in<a name="line.673"></a> |
| <span class="sourceLineNo">674</span> // it.<a name="line.674"></a> |
| <span class="sourceLineNo">675</span> if (this.drainingServers.contains(sn)) {<a name="line.675"></a> |
| <span class="sourceLineNo">676</span> LOG.warn("Server " + sn + " is already in the draining server list." +<a name="line.676"></a> |
| <span class="sourceLineNo">677</span> "Ignoring request to add it again.");<a name="line.677"></a> |
| <span class="sourceLineNo">678</span> return true;<a name="line.678"></a> |
| <span class="sourceLineNo">679</span> }<a name="line.679"></a> |
| <span class="sourceLineNo">680</span> LOG.info("Server " + sn + " added to draining server list.");<a name="line.680"></a> |
| <span class="sourceLineNo">681</span> return this.drainingServers.add(sn);<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> /**<a name="line.684"></a> |
| <span class="sourceLineNo">685</span> * Contacts a region server and waits up to timeout ms<a name="line.685"></a> |
| <span class="sourceLineNo">686</span> * to close the region. This bypasses the active hmaster.<a name="line.686"></a> |
| <span class="sourceLineNo">687</span> * Pass -1 as timeout if you do not want to wait on result.<a name="line.687"></a> |
| <span class="sourceLineNo">688</span> */<a name="line.688"></a> |
| <span class="sourceLineNo">689</span> public static void closeRegionSilentlyAndWait(AsyncClusterConnection connection,<a name="line.689"></a> |
| <span class="sourceLineNo">690</span> ServerName server, RegionInfo region, long timeout) throws IOException, InterruptedException {<a name="line.690"></a> |
| <span class="sourceLineNo">691</span> AsyncRegionServerAdmin admin = connection.getRegionServerAdmin(server);<a name="line.691"></a> |
| <span class="sourceLineNo">692</span> try {<a name="line.692"></a> |
| <span class="sourceLineNo">693</span> FutureUtils.get(<a name="line.693"></a> |
| <span class="sourceLineNo">694</span> admin.closeRegion(ProtobufUtil.buildCloseRegionRequest(server, region.getRegionName())));<a name="line.694"></a> |
| <span class="sourceLineNo">695</span> } catch (IOException e) {<a name="line.695"></a> |
| <span class="sourceLineNo">696</span> LOG.warn("Exception when closing region: " + region.getRegionNameAsString(), e);<a name="line.696"></a> |
| <span class="sourceLineNo">697</span> }<a name="line.697"></a> |
| <span class="sourceLineNo">698</span> if (timeout < 0) {<a name="line.698"></a> |
| <span class="sourceLineNo">699</span> return;<a name="line.699"></a> |
| <span class="sourceLineNo">700</span> }<a name="line.700"></a> |
| <span class="sourceLineNo">701</span> long expiration = timeout + System.currentTimeMillis();<a name="line.701"></a> |
| <span class="sourceLineNo">702</span> while (System.currentTimeMillis() < expiration) {<a name="line.702"></a> |
| <span class="sourceLineNo">703</span> try {<a name="line.703"></a> |
| <span class="sourceLineNo">704</span> RegionInfo rsRegion = ProtobufUtil.toRegionInfo(FutureUtils<a name="line.704"></a> |
| <span class="sourceLineNo">705</span> .get(<a name="line.705"></a> |
| <span class="sourceLineNo">706</span> admin.getRegionInfo(RequestConverter.buildGetRegionInfoRequest(region.getRegionName())))<a name="line.706"></a> |
| <span class="sourceLineNo">707</span> .getRegionInfo());<a name="line.707"></a> |
| <span class="sourceLineNo">708</span> if (rsRegion == null) {<a name="line.708"></a> |
| <span class="sourceLineNo">709</span> return;<a name="line.709"></a> |
| <span class="sourceLineNo">710</span> }<a name="line.710"></a> |
| <span class="sourceLineNo">711</span> } catch (IOException ioe) {<a name="line.711"></a> |
| <span class="sourceLineNo">712</span> if (ioe instanceof NotServingRegionException ||<a name="line.712"></a> |
| <span class="sourceLineNo">713</span> (ioe instanceof RemoteWithExtrasException &&<a name="line.713"></a> |
| <span class="sourceLineNo">714</span> ((RemoteWithExtrasException)ioe).unwrapRemoteException()<a name="line.714"></a> |
| <span class="sourceLineNo">715</span> instanceof NotServingRegionException)) {<a name="line.715"></a> |
| <span class="sourceLineNo">716</span> // no need to retry again<a name="line.716"></a> |
| <span class="sourceLineNo">717</span> return;<a name="line.717"></a> |
| <span class="sourceLineNo">718</span> }<a name="line.718"></a> |
| <span class="sourceLineNo">719</span> LOG.warn("Exception when retrieving regioninfo from: " + region.getRegionNameAsString(),<a name="line.719"></a> |
| <span class="sourceLineNo">720</span> ioe);<a name="line.720"></a> |
| <span class="sourceLineNo">721</span> }<a name="line.721"></a> |
| <span class="sourceLineNo">722</span> Thread.sleep(1000);<a name="line.722"></a> |
| <span class="sourceLineNo">723</span> }<a name="line.723"></a> |
| <span class="sourceLineNo">724</span> throw new IOException("Region " + region + " failed to close within" + " timeout " + timeout);<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> /**<a name="line.727"></a> |
| <span class="sourceLineNo">728</span> * Calculate min necessary to start. This is not an absolute. It is just<a name="line.728"></a> |
| <span class="sourceLineNo">729</span> * a friction that will cause us hang around a bit longer waiting on<a name="line.729"></a> |
| <span class="sourceLineNo">730</span> * RegionServers to check-in.<a name="line.730"></a> |
| <span class="sourceLineNo">731</span> */<a name="line.731"></a> |
| <span class="sourceLineNo">732</span> private int getMinToStart() {<a name="line.732"></a> |
| <span class="sourceLineNo">733</span> if (master.isInMaintenanceMode()) {<a name="line.733"></a> |
| <span class="sourceLineNo">734</span> // If in maintenance mode, then master hosting meta will be the only server available<a name="line.734"></a> |
| <span class="sourceLineNo">735</span> return 1;<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> int minimumRequired = 1;<a name="line.738"></a> |
| <span class="sourceLineNo">739</span> if (LoadBalancer.isTablesOnMaster(master.getConfiguration()) &&<a name="line.739"></a> |
| <span class="sourceLineNo">740</span> LoadBalancer.isSystemTablesOnlyOnMaster(master.getConfiguration())) {<a name="line.740"></a> |
| <span class="sourceLineNo">741</span> // If Master is carrying regions it will show up as a 'server', but is not handling user-<a name="line.741"></a> |
| <span class="sourceLineNo">742</span> // space regions, so we need a second server.<a name="line.742"></a> |
| <span class="sourceLineNo">743</span> minimumRequired = 2;<a name="line.743"></a> |
| <span class="sourceLineNo">744</span> }<a name="line.744"></a> |
| <span class="sourceLineNo">745</span><a name="line.745"></a> |
| <span class="sourceLineNo">746</span> int minToStart = this.master.getConfiguration().getInt(WAIT_ON_REGIONSERVERS_MINTOSTART, -1);<a name="line.746"></a> |
| <span class="sourceLineNo">747</span> // Ensure we are never less than minimumRequired else stuff won't work.<a name="line.747"></a> |
| <span class="sourceLineNo">748</span> return Math.max(minToStart, minimumRequired);<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> /**<a name="line.751"></a> |
| <span class="sourceLineNo">752</span> * Wait for the region servers to report in.<a name="line.752"></a> |
| <span class="sourceLineNo">753</span> * We will wait until one of this condition is met:<a name="line.753"></a> |
| <span class="sourceLineNo">754</span> * - the master is stopped<a name="line.754"></a> |
| <span class="sourceLineNo">755</span> * - the 'hbase.master.wait.on.regionservers.maxtostart' number of<a name="line.755"></a> |
| <span class="sourceLineNo">756</span> * region servers is reached<a name="line.756"></a> |
| <span class="sourceLineNo">757</span> * - the 'hbase.master.wait.on.regionservers.mintostart' is reached AND<a name="line.757"></a> |
| <span class="sourceLineNo">758</span> * there have been no new region server in for<a name="line.758"></a> |
| <span class="sourceLineNo">759</span> * 'hbase.master.wait.on.regionservers.interval' time AND<a name="line.759"></a> |
| <span class="sourceLineNo">760</span> * the 'hbase.master.wait.on.regionservers.timeout' is reached<a name="line.760"></a> |
| <span class="sourceLineNo">761</span> *<a name="line.761"></a> |
| <span class="sourceLineNo">762</span> * @throws InterruptedException<a name="line.762"></a> |
| <span class="sourceLineNo">763</span> */<a name="line.763"></a> |
| <span class="sourceLineNo">764</span> public void waitForRegionServers(MonitoredTask status) throws InterruptedException {<a name="line.764"></a> |
| <span class="sourceLineNo">765</span> final long interval = this.master.getConfiguration().<a name="line.765"></a> |
| <span class="sourceLineNo">766</span> getLong(WAIT_ON_REGIONSERVERS_INTERVAL, 1500);<a name="line.766"></a> |
| <span class="sourceLineNo">767</span> final long timeout = this.master.getConfiguration().<a name="line.767"></a> |
| <span class="sourceLineNo">768</span> getLong(WAIT_ON_REGIONSERVERS_TIMEOUT, 4500);<a name="line.768"></a> |
| <span class="sourceLineNo">769</span> // Min is not an absolute; just a friction making us wait longer on server checkin.<a name="line.769"></a> |
| <span class="sourceLineNo">770</span> int minToStart = getMinToStart();<a name="line.770"></a> |
| <span class="sourceLineNo">771</span> int maxToStart = this.master.getConfiguration().<a name="line.771"></a> |
| <span class="sourceLineNo">772</span> getInt(WAIT_ON_REGIONSERVERS_MAXTOSTART, Integer.MAX_VALUE);<a name="line.772"></a> |
| <span class="sourceLineNo">773</span> if (maxToStart < minToStart) {<a name="line.773"></a> |
| <span class="sourceLineNo">774</span> LOG.warn(String.format("The value of '%s' (%d) is set less than '%s' (%d), ignoring.",<a name="line.774"></a> |
| <span class="sourceLineNo">775</span> WAIT_ON_REGIONSERVERS_MAXTOSTART, maxToStart,<a name="line.775"></a> |
| <span class="sourceLineNo">776</span> WAIT_ON_REGIONSERVERS_MINTOSTART, minToStart));<a name="line.776"></a> |
| <span class="sourceLineNo">777</span> maxToStart = Integer.MAX_VALUE;<a name="line.777"></a> |
| <span class="sourceLineNo">778</span> }<a name="line.778"></a> |
| <span class="sourceLineNo">779</span><a name="line.779"></a> |
| <span class="sourceLineNo">780</span> long now = System.currentTimeMillis();<a name="line.780"></a> |
| <span class="sourceLineNo">781</span> final long startTime = now;<a name="line.781"></a> |
| <span class="sourceLineNo">782</span> long slept = 0;<a name="line.782"></a> |
| <span class="sourceLineNo">783</span> long lastLogTime = 0;<a name="line.783"></a> |
| <span class="sourceLineNo">784</span> long lastCountChange = startTime;<a name="line.784"></a> |
| <span class="sourceLineNo">785</span> int count = countOfRegionServers();<a name="line.785"></a> |
| <span class="sourceLineNo">786</span> int oldCount = 0;<a name="line.786"></a> |
| <span class="sourceLineNo">787</span> // This while test is a little hard to read. We try to comment it in below but in essence:<a name="line.787"></a> |
| <span class="sourceLineNo">788</span> // Wait if Master is not stopped and the number of regionservers that have checked-in is<a name="line.788"></a> |
| <span class="sourceLineNo">789</span> // less than the maxToStart. Both of these conditions will be true near universally.<a name="line.789"></a> |
| <span class="sourceLineNo">790</span> // Next, we will keep cycling if ANY of the following three conditions are true:<a name="line.790"></a> |
| <span class="sourceLineNo">791</span> // 1. The time since a regionserver registered is < interval (means servers are actively checking in).<a name="line.791"></a> |
| <span class="sourceLineNo">792</span> // 2. We are under the total timeout.<a name="line.792"></a> |
| <span class="sourceLineNo">793</span> // 3. The count of servers is < minimum.<a name="line.793"></a> |
| <span class="sourceLineNo">794</span> for (ServerListener listener: this.listeners) {<a name="line.794"></a> |
| <span class="sourceLineNo">795</span> listener.waiting();<a name="line.795"></a> |
| <span class="sourceLineNo">796</span> }<a name="line.796"></a> |
| <span class="sourceLineNo">797</span> while (!this.master.isStopped() && !isClusterShutdown() && count < maxToStart &&<a name="line.797"></a> |
| <span class="sourceLineNo">798</span> ((lastCountChange + interval) > now || timeout > slept || count < minToStart)) {<a name="line.798"></a> |
| <span class="sourceLineNo">799</span> // Log some info at every interval time or if there is a change<a name="line.799"></a> |
| <span class="sourceLineNo">800</span> if (oldCount != count || lastLogTime + interval < now) {<a name="line.800"></a> |
| <span class="sourceLineNo">801</span> lastLogTime = now;<a name="line.801"></a> |
| <span class="sourceLineNo">802</span> String msg =<a name="line.802"></a> |
| <span class="sourceLineNo">803</span> "Waiting on regionserver count=" + count + "; waited="+<a name="line.803"></a> |
| <span class="sourceLineNo">804</span> slept + "ms, expecting min=" + minToStart + " server(s), max="+ getStrForMax(maxToStart) +<a name="line.804"></a> |
| <span class="sourceLineNo">805</span> " server(s), " + "timeout=" + timeout + "ms, lastChange=" + (lastCountChange - now) + "ms";<a name="line.805"></a> |
| <span class="sourceLineNo">806</span> LOG.info(msg);<a name="line.806"></a> |
| <span class="sourceLineNo">807</span> status.setStatus(msg);<a name="line.807"></a> |
| <span class="sourceLineNo">808</span> }<a name="line.808"></a> |
| <span class="sourceLineNo">809</span><a name="line.809"></a> |
| <span class="sourceLineNo">810</span> // We sleep for some time<a name="line.810"></a> |
| <span class="sourceLineNo">811</span> final long sleepTime = 50;<a name="line.811"></a> |
| <span class="sourceLineNo">812</span> Thread.sleep(sleepTime);<a name="line.812"></a> |
| <span class="sourceLineNo">813</span> now = System.currentTimeMillis();<a name="line.813"></a> |
| <span class="sourceLineNo">814</span> slept = now - startTime;<a name="line.814"></a> |
| <span class="sourceLineNo">815</span><a name="line.815"></a> |
| <span class="sourceLineNo">816</span> oldCount = count;<a name="line.816"></a> |
| <span class="sourceLineNo">817</span> count = countOfRegionServers();<a name="line.817"></a> |
| <span class="sourceLineNo">818</span> if (count != oldCount) {<a name="line.818"></a> |
| <span class="sourceLineNo">819</span> lastCountChange = now;<a name="line.819"></a> |
| <span class="sourceLineNo">820</span> }<a name="line.820"></a> |
| <span class="sourceLineNo">821</span> }<a name="line.821"></a> |
| <span class="sourceLineNo">822</span> // Did we exit the loop because cluster is going down?<a name="line.822"></a> |
| <span class="sourceLineNo">823</span> if (isClusterShutdown()) {<a name="line.823"></a> |
| <span class="sourceLineNo">824</span> this.master.stop("Cluster shutdown");<a name="line.824"></a> |
| <span class="sourceLineNo">825</span> }<a name="line.825"></a> |
| <span class="sourceLineNo">826</span> LOG.info("Finished waiting on RegionServer count=" + count + "; waited=" + slept + "ms," +<a name="line.826"></a> |
| <span class="sourceLineNo">827</span> " expected min=" + minToStart + " server(s), max=" + getStrForMax(maxToStart) + " server(s),"+<a name="line.827"></a> |
| <span class="sourceLineNo">828</span> " master is "+ (this.master.isStopped() ? "stopped.": "running"));<a name="line.828"></a> |
| <span class="sourceLineNo">829</span> }<a name="line.829"></a> |
| <span class="sourceLineNo">830</span><a name="line.830"></a> |
| <span class="sourceLineNo">831</span> private String getStrForMax(final int max) {<a name="line.831"></a> |
| <span class="sourceLineNo">832</span> return max == Integer.MAX_VALUE? "NO_LIMIT": Integer.toString(max);<a name="line.832"></a> |
| <span class="sourceLineNo">833</span> }<a name="line.833"></a> |
| <span class="sourceLineNo">834</span><a name="line.834"></a> |
| <span class="sourceLineNo">835</span> /**<a name="line.835"></a> |
| <span class="sourceLineNo">836</span> * @return A copy of the internal list of online servers.<a name="line.836"></a> |
| <span class="sourceLineNo">837</span> */<a name="line.837"></a> |
| <span class="sourceLineNo">838</span> public List<ServerName> getOnlineServersList() {<a name="line.838"></a> |
| <span class="sourceLineNo">839</span> // TODO: optimize the load balancer call so we don't need to make a new list<a name="line.839"></a> |
| <span class="sourceLineNo">840</span> // TODO: FIX. THIS IS POPULAR CALL.<a name="line.840"></a> |
| <span class="sourceLineNo">841</span> return new ArrayList<>(this.onlineServers.keySet());<a name="line.841"></a> |
| <span class="sourceLineNo">842</span> }<a name="line.842"></a> |
| <span class="sourceLineNo">843</span><a name="line.843"></a> |
| <span class="sourceLineNo">844</span> /**<a name="line.844"></a> |
| <span class="sourceLineNo">845</span> * @param keys The target server name<a name="line.845"></a> |
| <span class="sourceLineNo">846</span> * @param idleServerPredicator Evaluates the server on the given load<a name="line.846"></a> |
| <span class="sourceLineNo">847</span> * @return A copy of the internal list of online servers matched by the predicator<a name="line.847"></a> |
| <span class="sourceLineNo">848</span> */<a name="line.848"></a> |
| <span class="sourceLineNo">849</span> public List<ServerName> getOnlineServersListWithPredicator(List<ServerName> keys,<a name="line.849"></a> |
| <span class="sourceLineNo">850</span> Predicate<ServerMetrics> idleServerPredicator) {<a name="line.850"></a> |
| <span class="sourceLineNo">851</span> List<ServerName> names = new ArrayList<>();<a name="line.851"></a> |
| <span class="sourceLineNo">852</span> if (keys != null && idleServerPredicator != null) {<a name="line.852"></a> |
| <span class="sourceLineNo">853</span> keys.forEach(name -> {<a name="line.853"></a> |
| <span class="sourceLineNo">854</span> ServerMetrics load = onlineServers.get(name);<a name="line.854"></a> |
| <span class="sourceLineNo">855</span> if (load != null) {<a name="line.855"></a> |
| <span class="sourceLineNo">856</span> if (idleServerPredicator.test(load)) {<a name="line.856"></a> |
| <span class="sourceLineNo">857</span> names.add(name);<a name="line.857"></a> |
| <span class="sourceLineNo">858</span> }<a name="line.858"></a> |
| <span class="sourceLineNo">859</span> }<a name="line.859"></a> |
| <span class="sourceLineNo">860</span> });<a name="line.860"></a> |
| <span class="sourceLineNo">861</span> }<a name="line.861"></a> |
| <span class="sourceLineNo">862</span> return names;<a name="line.862"></a> |
| <span class="sourceLineNo">863</span> }<a name="line.863"></a> |
| <span class="sourceLineNo">864</span><a name="line.864"></a> |
| <span class="sourceLineNo">865</span> /**<a name="line.865"></a> |
| <span class="sourceLineNo">866</span> * @return A copy of the internal list of draining servers.<a name="line.866"></a> |
| <span class="sourceLineNo">867</span> */<a name="line.867"></a> |
| <span class="sourceLineNo">868</span> public List<ServerName> getDrainingServersList() {<a name="line.868"></a> |
| <span class="sourceLineNo">869</span> return new ArrayList<>(this.drainingServers);<a name="line.869"></a> |
| <span class="sourceLineNo">870</span> }<a name="line.870"></a> |
| <span class="sourceLineNo">871</span><a name="line.871"></a> |
| <span class="sourceLineNo">872</span> public boolean isServerOnline(ServerName serverName) {<a name="line.872"></a> |
| <span class="sourceLineNo">873</span> return serverName != null && onlineServers.containsKey(serverName);<a name="line.873"></a> |
| <span class="sourceLineNo">874</span> }<a name="line.874"></a> |
| <span class="sourceLineNo">875</span><a name="line.875"></a> |
| <span class="sourceLineNo">876</span> public enum ServerLiveState {<a name="line.876"></a> |
| <span class="sourceLineNo">877</span> LIVE,<a name="line.877"></a> |
| <span class="sourceLineNo">878</span> DEAD,<a name="line.878"></a> |
| <span class="sourceLineNo">879</span> UNKNOWN<a name="line.879"></a> |
| <span class="sourceLineNo">880</span> }<a name="line.880"></a> |
| <span class="sourceLineNo">881</span><a name="line.881"></a> |
| <span class="sourceLineNo">882</span> /**<a name="line.882"></a> |
| <span class="sourceLineNo">883</span> * @return whether the server is online, dead, or unknown.<a name="line.883"></a> |
| <span class="sourceLineNo">884</span> */<a name="line.884"></a> |
| <span class="sourceLineNo">885</span> public synchronized ServerLiveState isServerKnownAndOnline(ServerName serverName) {<a name="line.885"></a> |
| <span class="sourceLineNo">886</span> return onlineServers.containsKey(serverName) ? ServerLiveState.LIVE<a name="line.886"></a> |
| <span class="sourceLineNo">887</span> : (deadservers.isDeadServer(serverName) ? ServerLiveState.DEAD : ServerLiveState.UNKNOWN);<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> /**<a name="line.890"></a> |
| <span class="sourceLineNo">891</span> * Check if a server is known to be dead. A server can be online,<a name="line.891"></a> |
| <span class="sourceLineNo">892</span> * or known to be dead, or unknown to this manager (i.e, not online,<a name="line.892"></a> |
| <span class="sourceLineNo">893</span> * not known to be dead either; it is simply not tracked by the<a name="line.893"></a> |
| <span class="sourceLineNo">894</span> * master any more, for example, a very old previous instance).<a name="line.894"></a> |
| <span class="sourceLineNo">895</span> */<a name="line.895"></a> |
| <span class="sourceLineNo">896</span> public synchronized boolean isServerDead(ServerName serverName) {<a name="line.896"></a> |
| <span class="sourceLineNo">897</span> return serverName == null || deadservers.isDeadServer(serverName);<a name="line.897"></a> |
| <span class="sourceLineNo">898</span> }<a name="line.898"></a> |
| <span class="sourceLineNo">899</span><a name="line.899"></a> |
| <span class="sourceLineNo">900</span> /**<a name="line.900"></a> |
| <span class="sourceLineNo">901</span> * Check if a server is unknown. A server can be online,<a name="line.901"></a> |
| <span class="sourceLineNo">902</span> * or known to be dead, or unknown to this manager (i.e, not online,<a name="line.902"></a> |
| <span class="sourceLineNo">903</span> * not known to be dead either; it is simply not tracked by the<a name="line.903"></a> |
| <span class="sourceLineNo">904</span> * master any more, for example, a very old previous instance).<a name="line.904"></a> |
| <span class="sourceLineNo">905</span> */<a name="line.905"></a> |
| <span class="sourceLineNo">906</span> public boolean isServerUnknown(ServerName serverName) {<a name="line.906"></a> |
| <span class="sourceLineNo">907</span> return serverName == null<a name="line.907"></a> |
| <span class="sourceLineNo">908</span> || (!onlineServers.containsKey(serverName) && !deadservers.isDeadServer(serverName));<a name="line.908"></a> |
| <span class="sourceLineNo">909</span> }<a name="line.909"></a> |
| <span class="sourceLineNo">910</span><a name="line.910"></a> |
| <span class="sourceLineNo">911</span> public void shutdownCluster() {<a name="line.911"></a> |
| <span class="sourceLineNo">912</span> String statusStr = "Cluster shutdown requested of master=" + this.master.getServerName();<a name="line.912"></a> |
| <span class="sourceLineNo">913</span> LOG.info(statusStr);<a name="line.913"></a> |
| <span class="sourceLineNo">914</span> this.clusterShutdown.set(true);<a name="line.914"></a> |
| <span class="sourceLineNo">915</span> if (onlineServers.isEmpty()) {<a name="line.915"></a> |
| <span class="sourceLineNo">916</span> // we do not synchronize here so this may cause a double stop, but not a big deal<a name="line.916"></a> |
| <span class="sourceLineNo">917</span> master.stop("OnlineServer=0 right after cluster shutdown set");<a name="line.917"></a> |
| <span class="sourceLineNo">918</span> }<a name="line.918"></a> |
| <span class="sourceLineNo">919</span> }<a name="line.919"></a> |
| <span class="sourceLineNo">920</span><a name="line.920"></a> |
| <span class="sourceLineNo">921</span> public boolean isClusterShutdown() {<a name="line.921"></a> |
| <span class="sourceLineNo">922</span> return this.clusterShutdown.get();<a name="line.922"></a> |
| <span class="sourceLineNo">923</span> }<a name="line.923"></a> |
| <span class="sourceLineNo">924</span><a name="line.924"></a> |
| <span class="sourceLineNo">925</span> /**<a name="line.925"></a> |
| <span class="sourceLineNo">926</span> * start chore in ServerManager<a name="line.926"></a> |
| <span class="sourceLineNo">927</span> */<a name="line.927"></a> |
| <span class="sourceLineNo">928</span> public void startChore() {<a name="line.928"></a> |
| <span class="sourceLineNo">929</span> Configuration c = master.getConfiguration();<a name="line.929"></a> |
| <span class="sourceLineNo">930</span> if (persistFlushedSequenceId) {<a name="line.930"></a> |
| <span class="sourceLineNo">931</span> new Thread(() -> {<a name="line.931"></a> |
| <span class="sourceLineNo">932</span> // after AM#loadMeta, RegionStates should be loaded, and some regions are<a name="line.932"></a> |
| <span class="sourceLineNo">933</span> // deleted by drop/split/merge during removeDeletedRegionFromLoadedFlushedSequenceIds,<a name="line.933"></a> |
| <span class="sourceLineNo">934</span> // but these deleted regions are not added back to RegionStates,<a name="line.934"></a> |
| <span class="sourceLineNo">935</span> // so we can safely remove deleted regions.<a name="line.935"></a> |
| <span class="sourceLineNo">936</span> removeDeletedRegionFromLoadedFlushedSequenceIds();<a name="line.936"></a> |
| <span class="sourceLineNo">937</span> }, "RemoveDeletedRegionSyncThread").start();<a name="line.937"></a> |
| <span class="sourceLineNo">938</span> int flushPeriod = c.getInt(FLUSHEDSEQUENCEID_FLUSHER_INTERVAL,<a name="line.938"></a> |
| <span class="sourceLineNo">939</span> FLUSHEDSEQUENCEID_FLUSHER_INTERVAL_DEFAULT);<a name="line.939"></a> |
| <span class="sourceLineNo">940</span> flushedSeqIdFlusher = new FlushedSequenceIdFlusher(<a name="line.940"></a> |
| <span class="sourceLineNo">941</span> "FlushedSequenceIdFlusher", flushPeriod);<a name="line.941"></a> |
| <span class="sourceLineNo">942</span> master.getChoreService().scheduleChore(flushedSeqIdFlusher);<a name="line.942"></a> |
| <span class="sourceLineNo">943</span> }<a name="line.943"></a> |
| <span class="sourceLineNo">944</span> }<a name="line.944"></a> |
| <span class="sourceLineNo">945</span><a name="line.945"></a> |
| <span class="sourceLineNo">946</span> /**<a name="line.946"></a> |
| <span class="sourceLineNo">947</span> * Stop the ServerManager.<a name="line.947"></a> |
| <span class="sourceLineNo">948</span> */<a name="line.948"></a> |
| <span class="sourceLineNo">949</span> public void stop() {<a name="line.949"></a> |
| <span class="sourceLineNo">950</span> if (flushedSeqIdFlusher != null) {<a name="line.950"></a> |
| <span class="sourceLineNo">951</span> flushedSeqIdFlusher.shutdown();<a name="line.951"></a> |
| <span class="sourceLineNo">952</span> }<a name="line.952"></a> |
| <span class="sourceLineNo">953</span> if (persistFlushedSequenceId) {<a name="line.953"></a> |
| <span class="sourceLineNo">954</span> try {<a name="line.954"></a> |
| <span class="sourceLineNo">955</span> persistRegionLastFlushedSequenceIds();<a name="line.955"></a> |
| <span class="sourceLineNo">956</span> } catch (IOException e) {<a name="line.956"></a> |
| <span class="sourceLineNo">957</span> LOG.warn("Failed to persist last flushed sequence id of regions"<a name="line.957"></a> |
| <span class="sourceLineNo">958</span> + " to file system", e);<a name="line.958"></a> |
| <span class="sourceLineNo">959</span> }<a name="line.959"></a> |
| <span class="sourceLineNo">960</span> }<a name="line.960"></a> |
| <span class="sourceLineNo">961</span> }<a name="line.961"></a> |
| <span class="sourceLineNo">962</span><a name="line.962"></a> |
| <span class="sourceLineNo">963</span> /**<a name="line.963"></a> |
| <span class="sourceLineNo">964</span> * Creates a list of possible destinations for a region. It contains the online servers, but not<a name="line.964"></a> |
| <span class="sourceLineNo">965</span> * the draining or dying servers.<a name="line.965"></a> |
| <span class="sourceLineNo">966</span> * @param serversToExclude can be null if there is no server to exclude<a name="line.966"></a> |
| <span class="sourceLineNo">967</span> */<a name="line.967"></a> |
| <span class="sourceLineNo">968</span> public List<ServerName> createDestinationServersList(final List<ServerName> serversToExclude){<a name="line.968"></a> |
| <span class="sourceLineNo">969</span> final List<ServerName> destServers = getOnlineServersList();<a name="line.969"></a> |
| <span class="sourceLineNo">970</span><a name="line.970"></a> |
| <span class="sourceLineNo">971</span> if (serversToExclude != null) {<a name="line.971"></a> |
| <span class="sourceLineNo">972</span> destServers.removeAll(serversToExclude);<a name="line.972"></a> |
| <span class="sourceLineNo">973</span> }<a name="line.973"></a> |
| <span class="sourceLineNo">974</span><a name="line.974"></a> |
| <span class="sourceLineNo">975</span> // Loop through the draining server list and remove them from the server list<a name="line.975"></a> |
| <span class="sourceLineNo">976</span> final List<ServerName> drainingServersCopy = getDrainingServersList();<a name="line.976"></a> |
| <span class="sourceLineNo">977</span> destServers.removeAll(drainingServersCopy);<a name="line.977"></a> |
| <span class="sourceLineNo">978</span><a name="line.978"></a> |
| <span class="sourceLineNo">979</span> return destServers;<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> * Calls {@link #createDestinationServersList} without server to exclude.<a name="line.983"></a> |
| <span class="sourceLineNo">984</span> */<a name="line.984"></a> |
| <span class="sourceLineNo">985</span> public List<ServerName> createDestinationServersList(){<a name="line.985"></a> |
| <span class="sourceLineNo">986</span> return createDestinationServersList(null);<a name="line.986"></a> |
| <span class="sourceLineNo">987</span> }<a name="line.987"></a> |
| <span class="sourceLineNo">988</span><a name="line.988"></a> |
| <span class="sourceLineNo">989</span> /**<a name="line.989"></a> |
| <span class="sourceLineNo">990</span> * To clear any dead server with same host name and port of any online server<a name="line.990"></a> |
| <span class="sourceLineNo">991</span> */<a name="line.991"></a> |
| <span class="sourceLineNo">992</span> void clearDeadServersWithSameHostNameAndPortOfOnlineServer() {<a name="line.992"></a> |
| <span class="sourceLineNo">993</span> for (ServerName serverName : getOnlineServersList()) {<a name="line.993"></a> |
| <span class="sourceLineNo">994</span> deadservers.cleanAllPreviousInstances(serverName);<a name="line.994"></a> |
| <span class="sourceLineNo">995</span> }<a name="line.995"></a> |
| <span class="sourceLineNo">996</span> }<a name="line.996"></a> |
| <span class="sourceLineNo">997</span><a name="line.997"></a> |
| <span class="sourceLineNo">998</span> /**<a name="line.998"></a> |
| <span class="sourceLineNo">999</span> * Called by delete table and similar to notify the ServerManager that a region was removed.<a name="line.999"></a> |
| <span class="sourceLineNo">1000</span> */<a name="line.1000"></a> |
| <span class="sourceLineNo">1001</span> public void removeRegion(final RegionInfo regionInfo) {<a name="line.1001"></a> |
| <span class="sourceLineNo">1002</span> final byte[] encodedName = regionInfo.getEncodedNameAsBytes();<a name="line.1002"></a> |
| <span class="sourceLineNo">1003</span> storeFlushedSequenceIdsByRegion.remove(encodedName);<a name="line.1003"></a> |
| <span class="sourceLineNo">1004</span> flushedSequenceIdByRegion.remove(encodedName);<a name="line.1004"></a> |
| <span class="sourceLineNo">1005</span> }<a name="line.1005"></a> |
| <span class="sourceLineNo">1006</span><a name="line.1006"></a> |
| <span class="sourceLineNo">1007</span> public boolean isRegionInServerManagerStates(final RegionInfo hri) {<a name="line.1007"></a> |
| <span class="sourceLineNo">1008</span> final byte[] encodedName = hri.getEncodedNameAsBytes();<a name="line.1008"></a> |
| <span class="sourceLineNo">1009</span> return (storeFlushedSequenceIdsByRegion.containsKey(encodedName)<a name="line.1009"></a> |
| <span class="sourceLineNo">1010</span> || flushedSequenceIdByRegion.containsKey(encodedName));<a name="line.1010"></a> |
| <span class="sourceLineNo">1011</span> }<a name="line.1011"></a> |
| <span class="sourceLineNo">1012</span><a name="line.1012"></a> |
| <span class="sourceLineNo">1013</span> /**<a name="line.1013"></a> |
| <span class="sourceLineNo">1014</span> * Called by delete table and similar to notify the ServerManager that a region was removed.<a name="line.1014"></a> |
| <span class="sourceLineNo">1015</span> */<a name="line.1015"></a> |
| <span class="sourceLineNo">1016</span> public void removeRegions(final List<RegionInfo> regions) {<a name="line.1016"></a> |
| <span class="sourceLineNo">1017</span> for (RegionInfo hri: regions) {<a name="line.1017"></a> |
| <span class="sourceLineNo">1018</span> removeRegion(hri);<a name="line.1018"></a> |
| <span class="sourceLineNo">1019</span> }<a name="line.1019"></a> |
| <span class="sourceLineNo">1020</span> }<a name="line.1020"></a> |
| <span class="sourceLineNo">1021</span><a name="line.1021"></a> |
| <span class="sourceLineNo">1022</span> /**<a name="line.1022"></a> |
| <span class="sourceLineNo">1023</span> * May return 0 when server is not online.<a name="line.1023"></a> |
| <span class="sourceLineNo">1024</span> */<a name="line.1024"></a> |
| <span class="sourceLineNo">1025</span> public int getVersionNumber(ServerName serverName) {<a name="line.1025"></a> |
| <span class="sourceLineNo">1026</span> ServerMetrics serverMetrics = onlineServers.get(serverName);<a name="line.1026"></a> |
| <span class="sourceLineNo">1027</span> return serverMetrics != null ? serverMetrics.getVersionNumber() : 0;<a name="line.1027"></a> |
| <span class="sourceLineNo">1028</span> }<a name="line.1028"></a> |
| <span class="sourceLineNo">1029</span><a name="line.1029"></a> |
| <span class="sourceLineNo">1030</span> /**<a name="line.1030"></a> |
| <span class="sourceLineNo">1031</span> * May return "0.0.0" when server is not online<a name="line.1031"></a> |
| <span class="sourceLineNo">1032</span> */<a name="line.1032"></a> |
| <span class="sourceLineNo">1033</span> public String getVersion(ServerName serverName) {<a name="line.1033"></a> |
| <span class="sourceLineNo">1034</span> ServerMetrics serverMetrics = onlineServers.get(serverName);<a name="line.1034"></a> |
| <span class="sourceLineNo">1035</span> return serverMetrics != null ? serverMetrics.getVersion() : "0.0.0";<a name="line.1035"></a> |
| <span class="sourceLineNo">1036</span> }<a name="line.1036"></a> |
| <span class="sourceLineNo">1037</span><a name="line.1037"></a> |
| <span class="sourceLineNo">1038</span> public int getInfoPort(ServerName serverName) {<a name="line.1038"></a> |
| <span class="sourceLineNo">1039</span> ServerMetrics serverMetrics = onlineServers.get(serverName);<a name="line.1039"></a> |
| <span class="sourceLineNo">1040</span> return serverMetrics != null ? serverMetrics.getInfoServerPort() : 0;<a name="line.1040"></a> |
| <span class="sourceLineNo">1041</span> }<a name="line.1041"></a> |
| <span class="sourceLineNo">1042</span><a name="line.1042"></a> |
| <span class="sourceLineNo">1043</span> /**<a name="line.1043"></a> |
| <span class="sourceLineNo">1044</span> * Persist last flushed sequence id of each region to HDFS<a name="line.1044"></a> |
| <span class="sourceLineNo">1045</span> * @throws IOException if persit to HDFS fails<a name="line.1045"></a> |
| <span class="sourceLineNo">1046</span> */<a name="line.1046"></a> |
| <span class="sourceLineNo">1047</span> private void persistRegionLastFlushedSequenceIds() throws IOException {<a name="line.1047"></a> |
| <span class="sourceLineNo">1048</span> if (isFlushSeqIdPersistInProgress) {<a name="line.1048"></a> |
| <span class="sourceLineNo">1049</span> return;<a name="line.1049"></a> |
| <span class="sourceLineNo">1050</span> }<a name="line.1050"></a> |
| <span class="sourceLineNo">1051</span> isFlushSeqIdPersistInProgress = true;<a name="line.1051"></a> |
| <span class="sourceLineNo">1052</span> try {<a name="line.1052"></a> |
| <span class="sourceLineNo">1053</span> Configuration conf = master.getConfiguration();<a name="line.1053"></a> |
| <span class="sourceLineNo">1054</span> Path rootDir = CommonFSUtils.getRootDir(conf);<a name="line.1054"></a> |
| <span class="sourceLineNo">1055</span> Path lastFlushedSeqIdPath = new Path(rootDir, LAST_FLUSHED_SEQ_ID_FILE);<a name="line.1055"></a> |
| <span class="sourceLineNo">1056</span> FileSystem fs = FileSystem.get(conf);<a name="line.1056"></a> |
| <span class="sourceLineNo">1057</span> if (fs.exists(lastFlushedSeqIdPath)) {<a name="line.1057"></a> |
| <span class="sourceLineNo">1058</span> LOG.info("Rewriting .lastflushedseqids file at: "<a name="line.1058"></a> |
| <span class="sourceLineNo">1059</span> + lastFlushedSeqIdPath);<a name="line.1059"></a> |
| <span class="sourceLineNo">1060</span> if (!fs.delete(lastFlushedSeqIdPath, false)) {<a name="line.1060"></a> |
| <span class="sourceLineNo">1061</span> throw new IOException("Unable to remove existing "<a name="line.1061"></a> |
| <span class="sourceLineNo">1062</span> + lastFlushedSeqIdPath);<a name="line.1062"></a> |
| <span class="sourceLineNo">1063</span> }<a name="line.1063"></a> |
| <span class="sourceLineNo">1064</span> } else {<a name="line.1064"></a> |
| <span class="sourceLineNo">1065</span> LOG.info("Writing .lastflushedseqids file at: " + lastFlushedSeqIdPath);<a name="line.1065"></a> |
| <span class="sourceLineNo">1066</span> }<a name="line.1066"></a> |
| <span class="sourceLineNo">1067</span> FSDataOutputStream out = fs.create(lastFlushedSeqIdPath);<a name="line.1067"></a> |
| <span class="sourceLineNo">1068</span> FlushedSequenceId.Builder flushedSequenceIdBuilder =<a name="line.1068"></a> |
| <span class="sourceLineNo">1069</span> FlushedSequenceId.newBuilder();<a name="line.1069"></a> |
| <span class="sourceLineNo">1070</span> try {<a name="line.1070"></a> |
| <span class="sourceLineNo">1071</span> for (Entry<byte[], Long> entry : flushedSequenceIdByRegion.entrySet()) {<a name="line.1071"></a> |
| <span class="sourceLineNo">1072</span> FlushedRegionSequenceId.Builder flushedRegionSequenceIdBuilder =<a name="line.1072"></a> |
| <span class="sourceLineNo">1073</span> FlushedRegionSequenceId.newBuilder();<a name="line.1073"></a> |
| <span class="sourceLineNo">1074</span> flushedRegionSequenceIdBuilder.setRegionEncodedName(<a name="line.1074"></a> |
| <span class="sourceLineNo">1075</span> ByteString.copyFrom(entry.getKey()));<a name="line.1075"></a> |
| <span class="sourceLineNo">1076</span> flushedRegionSequenceIdBuilder.setSeqId(entry.getValue());<a name="line.1076"></a> |
| <span class="sourceLineNo">1077</span> ConcurrentNavigableMap<byte[], Long> storeSeqIds =<a name="line.1077"></a> |
| <span class="sourceLineNo">1078</span> storeFlushedSequenceIdsByRegion.get(entry.getKey());<a name="line.1078"></a> |
| <span class="sourceLineNo">1079</span> if (storeSeqIds != null) {<a name="line.1079"></a> |
| <span class="sourceLineNo">1080</span> for (Entry<byte[], Long> store : storeSeqIds.entrySet()) {<a name="line.1080"></a> |
| <span class="sourceLineNo">1081</span> FlushedStoreSequenceId.Builder flushedStoreSequenceIdBuilder =<a name="line.1081"></a> |
| <span class="sourceLineNo">1082</span> FlushedStoreSequenceId.newBuilder();<a name="line.1082"></a> |
| <span class="sourceLineNo">1083</span> flushedStoreSequenceIdBuilder.setFamily(ByteString.copyFrom(store.getKey()));<a name="line.1083"></a> |
| <span class="sourceLineNo">1084</span> flushedStoreSequenceIdBuilder.setSeqId(store.getValue());<a name="line.1084"></a> |
| <span class="sourceLineNo">1085</span> flushedRegionSequenceIdBuilder.addStores(flushedStoreSequenceIdBuilder);<a name="line.1085"></a> |
| <span class="sourceLineNo">1086</span> }<a name="line.1086"></a> |
| <span class="sourceLineNo">1087</span> }<a name="line.1087"></a> |
| <span class="sourceLineNo">1088</span> flushedSequenceIdBuilder.addRegionSequenceId(flushedRegionSequenceIdBuilder);<a name="line.1088"></a> |
| <span class="sourceLineNo">1089</span> }<a name="line.1089"></a> |
| <span class="sourceLineNo">1090</span> flushedSequenceIdBuilder.build().writeDelimitedTo(out);<a name="line.1090"></a> |
| <span class="sourceLineNo">1091</span> } finally {<a name="line.1091"></a> |
| <span class="sourceLineNo">1092</span> if (out != null) {<a name="line.1092"></a> |
| <span class="sourceLineNo">1093</span> out.close();<a name="line.1093"></a> |
| <span class="sourceLineNo">1094</span> }<a name="line.1094"></a> |
| <span class="sourceLineNo">1095</span> }<a name="line.1095"></a> |
| <span class="sourceLineNo">1096</span> } finally {<a name="line.1096"></a> |
| <span class="sourceLineNo">1097</span> isFlushSeqIdPersistInProgress = false;<a name="line.1097"></a> |
| <span class="sourceLineNo">1098</span> }<a name="line.1098"></a> |
| <span class="sourceLineNo">1099</span> }<a name="line.1099"></a> |
| <span class="sourceLineNo">1100</span><a name="line.1100"></a> |
| <span class="sourceLineNo">1101</span> /**<a name="line.1101"></a> |
| <span class="sourceLineNo">1102</span> * Load last flushed sequence id of each region from HDFS, if persisted<a name="line.1102"></a> |
| <span class="sourceLineNo">1103</span> */<a name="line.1103"></a> |
| <span class="sourceLineNo">1104</span> public void loadLastFlushedSequenceIds() throws IOException {<a name="line.1104"></a> |
| <span class="sourceLineNo">1105</span> if (!persistFlushedSequenceId) {<a name="line.1105"></a> |
| <span class="sourceLineNo">1106</span> return;<a name="line.1106"></a> |
| <span class="sourceLineNo">1107</span> }<a name="line.1107"></a> |
| <span class="sourceLineNo">1108</span> Configuration conf = master.getConfiguration();<a name="line.1108"></a> |
| <span class="sourceLineNo">1109</span> Path rootDir = CommonFSUtils.getRootDir(conf);<a name="line.1109"></a> |
| <span class="sourceLineNo">1110</span> Path lastFlushedSeqIdPath = new Path(rootDir, LAST_FLUSHED_SEQ_ID_FILE);<a name="line.1110"></a> |
| <span class="sourceLineNo">1111</span> FileSystem fs = FileSystem.get(conf);<a name="line.1111"></a> |
| <span class="sourceLineNo">1112</span> if (!fs.exists(lastFlushedSeqIdPath)) {<a name="line.1112"></a> |
| <span class="sourceLineNo">1113</span> LOG.info("No .lastflushedseqids found at" + lastFlushedSeqIdPath<a name="line.1113"></a> |
| <span class="sourceLineNo">1114</span> + " will record last flushed sequence id"<a name="line.1114"></a> |
| <span class="sourceLineNo">1115</span> + " for regions by regionserver report all over again");<a name="line.1115"></a> |
| <span class="sourceLineNo">1116</span> return;<a name="line.1116"></a> |
| <span class="sourceLineNo">1117</span> } else {<a name="line.1117"></a> |
| <span class="sourceLineNo">1118</span> LOG.info("begin to load .lastflushedseqids at " + lastFlushedSeqIdPath);<a name="line.1118"></a> |
| <span class="sourceLineNo">1119</span> }<a name="line.1119"></a> |
| <span class="sourceLineNo">1120</span> FSDataInputStream in = fs.open(lastFlushedSeqIdPath);<a name="line.1120"></a> |
| <span class="sourceLineNo">1121</span> try {<a name="line.1121"></a> |
| <span class="sourceLineNo">1122</span> FlushedSequenceId flushedSequenceId =<a name="line.1122"></a> |
| <span class="sourceLineNo">1123</span> FlushedSequenceId.parseDelimitedFrom(in);<a name="line.1123"></a> |
| <span class="sourceLineNo">1124</span> if (flushedSequenceId == null) {<a name="line.1124"></a> |
| <span class="sourceLineNo">1125</span> LOG.info(".lastflushedseqids found at {} is empty", lastFlushedSeqIdPath);<a name="line.1125"></a> |
| <span class="sourceLineNo">1126</span> return;<a name="line.1126"></a> |
| <span class="sourceLineNo">1127</span> }<a name="line.1127"></a> |
| <span class="sourceLineNo">1128</span> for (FlushedRegionSequenceId flushedRegionSequenceId : flushedSequenceId<a name="line.1128"></a> |
| <span class="sourceLineNo">1129</span> .getRegionSequenceIdList()) {<a name="line.1129"></a> |
| <span class="sourceLineNo">1130</span> byte[] encodedRegionName = flushedRegionSequenceId<a name="line.1130"></a> |
| <span class="sourceLineNo">1131</span> .getRegionEncodedName().toByteArray();<a name="line.1131"></a> |
| <span class="sourceLineNo">1132</span> flushedSequenceIdByRegion<a name="line.1132"></a> |
| <span class="sourceLineNo">1133</span> .putIfAbsent(encodedRegionName, flushedRegionSequenceId.getSeqId());<a name="line.1133"></a> |
| <span class="sourceLineNo">1134</span> if (flushedRegionSequenceId.getStoresList() != null<a name="line.1134"></a> |
| <span class="sourceLineNo">1135</span> && flushedRegionSequenceId.getStoresList().size() != 0) {<a name="line.1135"></a> |
| <span class="sourceLineNo">1136</span> ConcurrentNavigableMap<byte[], Long> storeFlushedSequenceId =<a name="line.1136"></a> |
| <span class="sourceLineNo">1137</span> computeIfAbsent(storeFlushedSequenceIdsByRegion, encodedRegionName,<a name="line.1137"></a> |
| <span class="sourceLineNo">1138</span> () -> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR));<a name="line.1138"></a> |
| <span class="sourceLineNo">1139</span> for (FlushedStoreSequenceId flushedStoreSequenceId : flushedRegionSequenceId<a name="line.1139"></a> |
| <span class="sourceLineNo">1140</span> .getStoresList()) {<a name="line.1140"></a> |
| <span class="sourceLineNo">1141</span> storeFlushedSequenceId<a name="line.1141"></a> |
| <span class="sourceLineNo">1142</span> .put(flushedStoreSequenceId.getFamily().toByteArray(),<a name="line.1142"></a> |
| <span class="sourceLineNo">1143</span> flushedStoreSequenceId.getSeqId());<a name="line.1143"></a> |
| <span class="sourceLineNo">1144</span> }<a name="line.1144"></a> |
| <span class="sourceLineNo">1145</span> }<a name="line.1145"></a> |
| <span class="sourceLineNo">1146</span> }<a name="line.1146"></a> |
| <span class="sourceLineNo">1147</span> } finally {<a name="line.1147"></a> |
| <span class="sourceLineNo">1148</span> in.close();<a name="line.1148"></a> |
| <span class="sourceLineNo">1149</span> }<a name="line.1149"></a> |
| <span class="sourceLineNo">1150</span> }<a name="line.1150"></a> |
| <span class="sourceLineNo">1151</span><a name="line.1151"></a> |
| <span class="sourceLineNo">1152</span> /**<a name="line.1152"></a> |
| <span class="sourceLineNo">1153</span> * Regions may have been removed between latest persist of FlushedSequenceIds<a name="line.1153"></a> |
| <span class="sourceLineNo">1154</span> * and master abort. So after loading FlushedSequenceIds from file, and after<a name="line.1154"></a> |
| <span class="sourceLineNo">1155</span> * meta loaded, we need to remove the deleted region according to RegionStates.<a name="line.1155"></a> |
| <span class="sourceLineNo">1156</span> */<a name="line.1156"></a> |
| <span class="sourceLineNo">1157</span> public void removeDeletedRegionFromLoadedFlushedSequenceIds() {<a name="line.1157"></a> |
| <span class="sourceLineNo">1158</span> RegionStates regionStates = master.getAssignmentManager().getRegionStates();<a name="line.1158"></a> |
| <span class="sourceLineNo">1159</span> Iterator<byte[]> it = flushedSequenceIdByRegion.keySet().iterator();<a name="line.1159"></a> |
| <span class="sourceLineNo">1160</span> while(it.hasNext()) {<a name="line.1160"></a> |
| <span class="sourceLineNo">1161</span> byte[] regionEncodedName = it.next();<a name="line.1161"></a> |
| <span class="sourceLineNo">1162</span> if (regionStates.getRegionState(Bytes.toStringBinary(regionEncodedName)) == null) {<a name="line.1162"></a> |
| <span class="sourceLineNo">1163</span> it.remove();<a name="line.1163"></a> |
| <span class="sourceLineNo">1164</span> storeFlushedSequenceIdsByRegion.remove(regionEncodedName);<a name="line.1164"></a> |
| <span class="sourceLineNo">1165</span> }<a name="line.1165"></a> |
| <span class="sourceLineNo">1166</span> }<a name="line.1166"></a> |
| <span class="sourceLineNo">1167</span> }<a name="line.1167"></a> |
| <span class="sourceLineNo">1168</span><a name="line.1168"></a> |
| <span class="sourceLineNo">1169</span> private class FlushedSequenceIdFlusher extends ScheduledChore {<a name="line.1169"></a> |
| <span class="sourceLineNo">1170</span><a name="line.1170"></a> |
| <span class="sourceLineNo">1171</span> public FlushedSequenceIdFlusher(String name, int p) {<a name="line.1171"></a> |
| <span class="sourceLineNo">1172</span> super(name, master, p, 60 * 1000); //delay one minute before first execute<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> @Override<a name="line.1175"></a> |
| <span class="sourceLineNo">1176</span> protected void chore() {<a name="line.1176"></a> |
| <span class="sourceLineNo">1177</span> try {<a name="line.1177"></a> |
| <span class="sourceLineNo">1178</span> persistRegionLastFlushedSequenceIds();<a name="line.1178"></a> |
| <span class="sourceLineNo">1179</span> } catch (IOException e) {<a name="line.1179"></a> |
| <span class="sourceLineNo">1180</span> LOG.debug("Failed to persist last flushed sequence id of regions"<a name="line.1180"></a> |
| <span class="sourceLineNo">1181</span> + " to file system", e);<a name="line.1181"></a> |
| <span class="sourceLineNo">1182</span> }<a name="line.1182"></a> |
| <span class="sourceLineNo">1183</span> }<a name="line.1183"></a> |
| <span class="sourceLineNo">1184</span> }<a name="line.1184"></a> |
| <span class="sourceLineNo">1185</span>}<a name="line.1185"></a> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |
