blob: 3a15b1ed8f3e8f8a730ca92cf0d5057e2c3f73da [file] [log] [blame]
<!DOCTYPE HTML>
<html lang="en">
<head>
<!-- Generated by javadoc (17) -->
<title>Source code</title>
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="source: package: org.apache.hadoop.hbase.util, class: HBaseFsckRepair">
<meta name="generator" content="javadoc/SourceToHTMLConverter">
<link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style">
</head>
<body class="source-page">
<main role="main">
<div class="source-container">
<pre><span class="source-line-no">001</span><span id="line-1">/*</span>
<span class="source-line-no">002</span><span id="line-2"> * Licensed to the Apache Software Foundation (ASF) under one</span>
<span class="source-line-no">003</span><span id="line-3"> * or more contributor license agreements. See the NOTICE file</span>
<span class="source-line-no">004</span><span id="line-4"> * distributed with this work for additional information</span>
<span class="source-line-no">005</span><span id="line-5"> * regarding copyright ownership. The ASF licenses this file</span>
<span class="source-line-no">006</span><span id="line-6"> * to you under the Apache License, Version 2.0 (the</span>
<span class="source-line-no">007</span><span id="line-7"> * "License"); you may not use this file except in compliance</span>
<span class="source-line-no">008</span><span id="line-8"> * with the License. You may obtain a copy of the License at</span>
<span class="source-line-no">009</span><span id="line-9"> *</span>
<span class="source-line-no">010</span><span id="line-10"> * http://www.apache.org/licenses/LICENSE-2.0</span>
<span class="source-line-no">011</span><span id="line-11"> *</span>
<span class="source-line-no">012</span><span id="line-12"> * Unless required by applicable law or agreed to in writing, software</span>
<span class="source-line-no">013</span><span id="line-13"> * distributed under the License is distributed on an "AS IS" BASIS,</span>
<span class="source-line-no">014</span><span id="line-14"> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
<span class="source-line-no">015</span><span id="line-15"> * See the License for the specific language governing permissions and</span>
<span class="source-line-no">016</span><span id="line-16"> * limitations under the License.</span>
<span class="source-line-no">017</span><span id="line-17"> */</span>
<span class="source-line-no">018</span><span id="line-18">package org.apache.hadoop.hbase.util;</span>
<span class="source-line-no">019</span><span id="line-19"></span>
<span class="source-line-no">020</span><span id="line-20">import java.io.IOException;</span>
<span class="source-line-no">021</span><span id="line-21">import java.util.Collection;</span>
<span class="source-line-no">022</span><span id="line-22">import java.util.EnumSet;</span>
<span class="source-line-no">023</span><span id="line-23">import java.util.List;</span>
<span class="source-line-no">024</span><span id="line-24">import java.util.Random;</span>
<span class="source-line-no">025</span><span id="line-25">import java.util.concurrent.ThreadLocalRandom;</span>
<span class="source-line-no">026</span><span id="line-26">import org.apache.hadoop.conf.Configuration;</span>
<span class="source-line-no">027</span><span id="line-27">import org.apache.hadoop.fs.Path;</span>
<span class="source-line-no">028</span><span id="line-28">import org.apache.hadoop.hbase.ClusterMetrics.Option;</span>
<span class="source-line-no">029</span><span id="line-29">import org.apache.hadoop.hbase.MetaTableAccessor;</span>
<span class="source-line-no">030</span><span id="line-30">import org.apache.hadoop.hbase.ServerName;</span>
<span class="source-line-no">031</span><span id="line-31">import org.apache.hadoop.hbase.TableName;</span>
<span class="source-line-no">032</span><span id="line-32">import org.apache.hadoop.hbase.ZooKeeperConnectionException;</span>
<span class="source-line-no">033</span><span id="line-33">import org.apache.hadoop.hbase.client.Admin;</span>
<span class="source-line-no">034</span><span id="line-34">import org.apache.hadoop.hbase.client.AsyncClusterConnection;</span>
<span class="source-line-no">035</span><span id="line-35">import org.apache.hadoop.hbase.client.ClusterConnectionFactory;</span>
<span class="source-line-no">036</span><span id="line-36">import org.apache.hadoop.hbase.client.Connection;</span>
<span class="source-line-no">037</span><span id="line-37">import org.apache.hadoop.hbase.client.ConnectionFactory;</span>
<span class="source-line-no">038</span><span id="line-38">import org.apache.hadoop.hbase.client.Put;</span>
<span class="source-line-no">039</span><span id="line-39">import org.apache.hadoop.hbase.client.RegionInfo;</span>
<span class="source-line-no">040</span><span id="line-40">import org.apache.hadoop.hbase.client.Table;</span>
<span class="source-line-no">041</span><span id="line-41">import org.apache.hadoop.hbase.client.TableDescriptor;</span>
<span class="source-line-no">042</span><span id="line-42">import org.apache.hadoop.hbase.master.RegionState;</span>
<span class="source-line-no">043</span><span id="line-43">import org.apache.hadoop.hbase.master.ServerManager;</span>
<span class="source-line-no">044</span><span id="line-44">import org.apache.hadoop.hbase.regionserver.HRegion;</span>
<span class="source-line-no">045</span><span id="line-45">import org.apache.hadoop.hbase.security.User;</span>
<span class="source-line-no">046</span><span id="line-46">import org.apache.yetus.audience.InterfaceAudience;</span>
<span class="source-line-no">047</span><span id="line-47">import org.apache.zookeeper.KeeperException;</span>
<span class="source-line-no">048</span><span id="line-48">import org.slf4j.Logger;</span>
<span class="source-line-no">049</span><span id="line-49">import org.slf4j.LoggerFactory;</span>
<span class="source-line-no">050</span><span id="line-50"></span>
<span class="source-line-no">051</span><span id="line-51">/**</span>
<span class="source-line-no">052</span><span id="line-52"> * This class contains helper methods that repair parts of hbase's filesystem contents.</span>
<span class="source-line-no">053</span><span id="line-53"> */</span>
<span class="source-line-no">054</span><span id="line-54">@InterfaceAudience.Private</span>
<span class="source-line-no">055</span><span id="line-55">public class HBaseFsckRepair {</span>
<span class="source-line-no">056</span><span id="line-56"> private static final Logger LOG = LoggerFactory.getLogger(HBaseFsckRepair.class);</span>
<span class="source-line-no">057</span><span id="line-57"></span>
<span class="source-line-no">058</span><span id="line-58"> /**</span>
<span class="source-line-no">059</span><span id="line-59"> * Fix multiple assignment by doing silent closes on each RS hosting the region and then force ZK</span>
<span class="source-line-no">060</span><span id="line-60"> * unassigned node to OFFLINE to trigger assignment by master.</span>
<span class="source-line-no">061</span><span id="line-61"> * @param connection HBase connection to the cluster</span>
<span class="source-line-no">062</span><span id="line-62"> * @param region Region to undeploy</span>
<span class="source-line-no">063</span><span id="line-63"> * @param servers list of Servers to undeploy from</span>
<span class="source-line-no">064</span><span id="line-64"> */</span>
<span class="source-line-no">065</span><span id="line-65"> public static void fixMultiAssignment(Connection connection, RegionInfo region,</span>
<span class="source-line-no">066</span><span id="line-66"> List&lt;ServerName&gt; servers) throws IOException, KeeperException, InterruptedException {</span>
<span class="source-line-no">067</span><span id="line-67"> // Close region on the servers silently</span>
<span class="source-line-no">068</span><span id="line-68"> for (ServerName server : servers) {</span>
<span class="source-line-no">069</span><span id="line-69"> closeRegionSilentlyAndWait(connection, server, region);</span>
<span class="source-line-no">070</span><span id="line-70"> }</span>
<span class="source-line-no">071</span><span id="line-71"></span>
<span class="source-line-no">072</span><span id="line-72"> // Force ZK node to OFFLINE so master assigns</span>
<span class="source-line-no">073</span><span id="line-73"> forceOfflineInZK(connection.getAdmin(), region);</span>
<span class="source-line-no">074</span><span id="line-74"> }</span>
<span class="source-line-no">075</span><span id="line-75"></span>
<span class="source-line-no">076</span><span id="line-76"> /**</span>
<span class="source-line-no">077</span><span id="line-77"> * Fix unassigned by creating/transition the unassigned ZK node for this region to OFFLINE state</span>
<span class="source-line-no">078</span><span id="line-78"> * with a special flag to tell the master that this is a forced operation by HBCK. This assumes</span>
<span class="source-line-no">079</span><span id="line-79"> * that info is in META.</span>
<span class="source-line-no">080</span><span id="line-80"> */</span>
<span class="source-line-no">081</span><span id="line-81"> public static void fixUnassigned(Admin admin, RegionInfo region)</span>
<span class="source-line-no">082</span><span id="line-82"> throws IOException, KeeperException, InterruptedException {</span>
<span class="source-line-no">083</span><span id="line-83"> // Force ZK node to OFFLINE so master assigns</span>
<span class="source-line-no">084</span><span id="line-84"> forceOfflineInZK(admin, region);</span>
<span class="source-line-no">085</span><span id="line-85"> }</span>
<span class="source-line-no">086</span><span id="line-86"></span>
<span class="source-line-no">087</span><span id="line-87"> /**</span>
<span class="source-line-no">088</span><span id="line-88"> * In 0.90, this forces an HRI offline by setting the RegionTransitionData in ZK to have</span>
<span class="source-line-no">089</span><span id="line-89"> * HBCK_CODE_NAME as the server. This is a special case in the AssignmentManager that attempts an</span>
<span class="source-line-no">090</span><span id="line-90"> * assign call by the master. This doesn't seem to work properly in the updated version of 0.92+'s</span>
<span class="source-line-no">091</span><span id="line-91"> * hbck so we use assign to force the region into transition. This has the side-effect of</span>
<span class="source-line-no">092</span><span id="line-92"> * requiring a RegionInfo that considers regionId (timestamp) in comparators that is addressed by</span>
<span class="source-line-no">093</span><span id="line-93"> * HBASE-5563.</span>
<span class="source-line-no">094</span><span id="line-94"> */</span>
<span class="source-line-no">095</span><span id="line-95"> private static void forceOfflineInZK(Admin admin, final RegionInfo region)</span>
<span class="source-line-no">096</span><span id="line-96"> throws ZooKeeperConnectionException, KeeperException, IOException, InterruptedException {</span>
<span class="source-line-no">097</span><span id="line-97"> admin.assign(region.getRegionName());</span>
<span class="source-line-no">098</span><span id="line-98"> }</span>
<span class="source-line-no">099</span><span id="line-99"></span>
<span class="source-line-no">100</span><span id="line-100"> /*</span>
<span class="source-line-no">101</span><span id="line-101"> * Should we check all assignments or just not in RIT?</span>
<span class="source-line-no">102</span><span id="line-102"> */</span>
<span class="source-line-no">103</span><span id="line-103"> public static void waitUntilAssigned(Admin admin, RegionInfo region)</span>
<span class="source-line-no">104</span><span id="line-104"> throws IOException, InterruptedException {</span>
<span class="source-line-no">105</span><span id="line-105"> long timeout = admin.getConfiguration().getLong("hbase.hbck.assign.timeout", 120000);</span>
<span class="source-line-no">106</span><span id="line-106"> long expiration = timeout + EnvironmentEdgeManager.currentTime();</span>
<span class="source-line-no">107</span><span id="line-107"> while (EnvironmentEdgeManager.currentTime() &lt; expiration) {</span>
<span class="source-line-no">108</span><span id="line-108"> try {</span>
<span class="source-line-no">109</span><span id="line-109"> boolean inTransition = false;</span>
<span class="source-line-no">110</span><span id="line-110"> for (RegionState rs : admin.getClusterMetrics(EnumSet.of(Option.REGIONS_IN_TRANSITION))</span>
<span class="source-line-no">111</span><span id="line-111"> .getRegionStatesInTransition()) {</span>
<span class="source-line-no">112</span><span id="line-112"> if (RegionInfo.COMPARATOR.compare(rs.getRegion(), region) == 0) {</span>
<span class="source-line-no">113</span><span id="line-113"> inTransition = true;</span>
<span class="source-line-no">114</span><span id="line-114"> break;</span>
<span class="source-line-no">115</span><span id="line-115"> }</span>
<span class="source-line-no">116</span><span id="line-116"> }</span>
<span class="source-line-no">117</span><span id="line-117"> if (!inTransition) {</span>
<span class="source-line-no">118</span><span id="line-118"> // yay! no longer RIT</span>
<span class="source-line-no">119</span><span id="line-119"> return;</span>
<span class="source-line-no">120</span><span id="line-120"> }</span>
<span class="source-line-no">121</span><span id="line-121"> // still in rit</span>
<span class="source-line-no">122</span><span id="line-122"> LOG.info("Region still in transition, waiting for " + "it to become assigned: " + region);</span>
<span class="source-line-no">123</span><span id="line-123"> } catch (IOException e) {</span>
<span class="source-line-no">124</span><span id="line-124"> LOG.warn("Exception when waiting for region to become assigned," + " retrying", e);</span>
<span class="source-line-no">125</span><span id="line-125"> }</span>
<span class="source-line-no">126</span><span id="line-126"> Thread.sleep(1000);</span>
<span class="source-line-no">127</span><span id="line-127"> }</span>
<span class="source-line-no">128</span><span id="line-128"> throw new IOException("Region " + region + " failed to move out of "</span>
<span class="source-line-no">129</span><span id="line-129"> + "transition within timeout " + timeout + "ms");</span>
<span class="source-line-no">130</span><span id="line-130"> }</span>
<span class="source-line-no">131</span><span id="line-131"></span>
<span class="source-line-no">132</span><span id="line-132"> /**</span>
<span class="source-line-no">133</span><span id="line-133"> * Contacts a region server and waits up to hbase.hbck.close.timeout ms (default 120s) to close</span>
<span class="source-line-no">134</span><span id="line-134"> * the region. This bypasses the active hmaster.</span>
<span class="source-line-no">135</span><span id="line-135"> */</span>
<span class="source-line-no">136</span><span id="line-136"> public static void closeRegionSilentlyAndWait(Connection connection, ServerName server,</span>
<span class="source-line-no">137</span><span id="line-137"> RegionInfo region) throws IOException, InterruptedException {</span>
<span class="source-line-no">138</span><span id="line-138"> long timeout = connection.getConfiguration().getLong("hbase.hbck.close.timeout", 120000);</span>
<span class="source-line-no">139</span><span id="line-139"> // this is a bit ugly but it is only used in the old hbck and tests, so I think it is fine.</span>
<span class="source-line-no">140</span><span id="line-140"> try (AsyncClusterConnection asyncConn = ClusterConnectionFactory</span>
<span class="source-line-no">141</span><span id="line-141"> .createAsyncClusterConnection(connection.getConfiguration(), null, User.getCurrent())) {</span>
<span class="source-line-no">142</span><span id="line-142"> ServerManager.closeRegionSilentlyAndWait(asyncConn, server, region, timeout);</span>
<span class="source-line-no">143</span><span id="line-143"> }</span>
<span class="source-line-no">144</span><span id="line-144"> }</span>
<span class="source-line-no">145</span><span id="line-145"></span>
<span class="source-line-no">146</span><span id="line-146"> /**</span>
<span class="source-line-no">147</span><span id="line-147"> * Puts the specified RegionInfo into META with replica related columns</span>
<span class="source-line-no">148</span><span id="line-148"> */</span>
<span class="source-line-no">149</span><span id="line-149"> public static void fixMetaHoleOnlineAndAddReplicas(Configuration conf, RegionInfo hri,</span>
<span class="source-line-no">150</span><span id="line-150"> Collection&lt;ServerName&gt; servers, int numReplicas) throws IOException {</span>
<span class="source-line-no">151</span><span id="line-151"> Connection conn = ConnectionFactory.createConnection(conf);</span>
<span class="source-line-no">152</span><span id="line-152"> Table meta = conn.getTable(TableName.META_TABLE_NAME);</span>
<span class="source-line-no">153</span><span id="line-153"> Put put = MetaTableAccessor.makePutFromRegionInfo(hri);</span>
<span class="source-line-no">154</span><span id="line-154"> if (numReplicas &gt; 1) {</span>
<span class="source-line-no">155</span><span id="line-155"> Random rand = ThreadLocalRandom.current();</span>
<span class="source-line-no">156</span><span id="line-156"> ServerName[] serversArr = servers.toArray(new ServerName[servers.size()]);</span>
<span class="source-line-no">157</span><span id="line-157"> for (int i = 1; i &lt; numReplicas; i++) {</span>
<span class="source-line-no">158</span><span id="line-158"> ServerName sn = serversArr[rand.nextInt(serversArr.length)];</span>
<span class="source-line-no">159</span><span id="line-159"> // the column added here is just to make sure the master is able to</span>
<span class="source-line-no">160</span><span id="line-160"> // see the additional replicas when it is asked to assign. The</span>
<span class="source-line-no">161</span><span id="line-161"> // final value of these columns will be different and will be updated</span>
<span class="source-line-no">162</span><span id="line-162"> // by the actual regionservers that start hosting the respective replicas</span>
<span class="source-line-no">163</span><span id="line-163"> MetaTableAccessor.addLocation(put, sn, sn.getStartcode(), i);</span>
<span class="source-line-no">164</span><span id="line-164"> }</span>
<span class="source-line-no">165</span><span id="line-165"> }</span>
<span class="source-line-no">166</span><span id="line-166"> meta.put(put);</span>
<span class="source-line-no">167</span><span id="line-167"> meta.close();</span>
<span class="source-line-no">168</span><span id="line-168"> conn.close();</span>
<span class="source-line-no">169</span><span id="line-169"> }</span>
<span class="source-line-no">170</span><span id="line-170"></span>
<span class="source-line-no">171</span><span id="line-171"> /**</span>
<span class="source-line-no">172</span><span id="line-172"> * Creates, flushes, and closes a new region.</span>
<span class="source-line-no">173</span><span id="line-173"> */</span>
<span class="source-line-no">174</span><span id="line-174"> public static HRegion createHDFSRegionDir(Configuration conf, RegionInfo hri, TableDescriptor htd)</span>
<span class="source-line-no">175</span><span id="line-175"> throws IOException {</span>
<span class="source-line-no">176</span><span id="line-176"> // Create HRegion</span>
<span class="source-line-no">177</span><span id="line-177"> Path root = CommonFSUtils.getRootDir(conf);</span>
<span class="source-line-no">178</span><span id="line-178"> HRegion region = HRegion.createHRegion(hri, root, conf, htd, null);</span>
<span class="source-line-no">179</span><span id="line-179"></span>
<span class="source-line-no">180</span><span id="line-180"> // Close the new region to flush to disk. Close log file too.</span>
<span class="source-line-no">181</span><span id="line-181"> region.close();</span>
<span class="source-line-no">182</span><span id="line-182"> return region;</span>
<span class="source-line-no">183</span><span id="line-183"> }</span>
<span class="source-line-no">184</span><span id="line-184"></span>
<span class="source-line-no">185</span><span id="line-185"> /*</span>
<span class="source-line-no">186</span><span id="line-186"> * Remove parent</span>
<span class="source-line-no">187</span><span id="line-187"> */</span>
<span class="source-line-no">188</span><span id="line-188"> public static void removeParentInMeta(Configuration conf, RegionInfo hri) throws IOException {</span>
<span class="source-line-no">189</span><span id="line-189"> throw new UnsupportedOperationException("HBCK1 is read-only now, use HBCK2 instead");</span>
<span class="source-line-no">190</span><span id="line-190"> }</span>
<span class="source-line-no">191</span><span id="line-191">}</span>
</pre>
</div>
</main>
</body>
</html>