blob: 0c0c87bc15e9be45f0dddaf39bb383323b81e6fe [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.master, class: RegionServerTracker">
<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.master;</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 io.opentelemetry.api.trace.Span;</span>
<span class="source-line-no">021</span><span id="line-21">import io.opentelemetry.api.trace.StatusCode;</span>
<span class="source-line-no">022</span><span id="line-22">import io.opentelemetry.context.Scope;</span>
<span class="source-line-no">023</span><span id="line-23">import java.io.IOException;</span>
<span class="source-line-no">024</span><span id="line-24">import java.io.InterruptedIOException;</span>
<span class="source-line-no">025</span><span id="line-25">import java.util.Collections;</span>
<span class="source-line-no">026</span><span id="line-26">import java.util.List;</span>
<span class="source-line-no">027</span><span id="line-27">import java.util.Set;</span>
<span class="source-line-no">028</span><span id="line-28">import java.util.concurrent.ExecutorService;</span>
<span class="source-line-no">029</span><span id="line-29">import java.util.concurrent.Executors;</span>
<span class="source-line-no">030</span><span id="line-30">import java.util.stream.Collectors;</span>
<span class="source-line-no">031</span><span id="line-31">import org.apache.hadoop.hbase.ServerMetrics;</span>
<span class="source-line-no">032</span><span id="line-32">import org.apache.hadoop.hbase.ServerMetricsBuilder;</span>
<span class="source-line-no">033</span><span id="line-33">import org.apache.hadoop.hbase.ServerName;</span>
<span class="source-line-no">034</span><span id="line-34">import org.apache.hadoop.hbase.client.VersionInfoUtil;</span>
<span class="source-line-no">035</span><span id="line-35">import org.apache.hadoop.hbase.trace.TraceUtil;</span>
<span class="source-line-no">036</span><span id="line-36">import org.apache.hadoop.hbase.zookeeper.ZKListener;</span>
<span class="source-line-no">037</span><span id="line-37">import org.apache.hadoop.hbase.zookeeper.ZKUtil;</span>
<span class="source-line-no">038</span><span id="line-38">import org.apache.hadoop.hbase.zookeeper.ZKWatcher;</span>
<span class="source-line-no">039</span><span id="line-39">import org.apache.yetus.audience.InterfaceAudience;</span>
<span class="source-line-no">040</span><span id="line-40">import org.apache.zookeeper.KeeperException;</span>
<span class="source-line-no">041</span><span id="line-41">import org.slf4j.Logger;</span>
<span class="source-line-no">042</span><span id="line-42">import org.slf4j.LoggerFactory;</span>
<span class="source-line-no">043</span><span id="line-43"></span>
<span class="source-line-no">044</span><span id="line-44">import org.apache.hbase.thirdparty.com.google.common.collect.Sets;</span>
<span class="source-line-no">045</span><span id="line-45">import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;</span>
<span class="source-line-no">046</span><span id="line-46">import org.apache.hbase.thirdparty.org.apache.commons.collections4.CollectionUtils;</span>
<span class="source-line-no">047</span><span id="line-47"></span>
<span class="source-line-no">048</span><span id="line-48">import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;</span>
<span class="source-line-no">049</span><span id="line-49">import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionServerInfo;</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"> * Tracks the online region servers via ZK.</span>
<span class="source-line-no">053</span><span id="line-53"> * &lt;p/&gt;</span>
<span class="source-line-no">054</span><span id="line-54"> * Handling of new RSs checking in is done via RPC. This class is only responsible for watching for</span>
<span class="source-line-no">055</span><span id="line-55"> * expired nodes. It handles listening for changes in the RS node list. The only exception is when</span>
<span class="source-line-no">056</span><span id="line-56"> * master restart, we will use the list fetched from zk to construct the initial set of live region</span>
<span class="source-line-no">057</span><span id="line-57"> * servers.</span>
<span class="source-line-no">058</span><span id="line-58"> * &lt;p/&gt;</span>
<span class="source-line-no">059</span><span id="line-59"> * If an RS node gets deleted, this automatically handles calling of</span>
<span class="source-line-no">060</span><span id="line-60"> * {@link ServerManager#expireServer(ServerName)}</span>
<span class="source-line-no">061</span><span id="line-61"> */</span>
<span class="source-line-no">062</span><span id="line-62">@InterfaceAudience.Private</span>
<span class="source-line-no">063</span><span id="line-63">public class RegionServerTracker extends ZKListener {</span>
<span class="source-line-no">064</span><span id="line-64"> private static final Logger LOG = LoggerFactory.getLogger(RegionServerTracker.class);</span>
<span class="source-line-no">065</span><span id="line-65"> // indicate whether we are active master</span>
<span class="source-line-no">066</span><span id="line-66"> private boolean active;</span>
<span class="source-line-no">067</span><span id="line-67"> private volatile Set&lt;ServerName&gt; regionServers = Collections.emptySet();</span>
<span class="source-line-no">068</span><span id="line-68"> private final MasterServices server;</span>
<span class="source-line-no">069</span><span id="line-69"> // As we need to send request to zk when processing the nodeChildrenChanged event, we'd better</span>
<span class="source-line-no">070</span><span id="line-70"> // move the operation to a single threaded thread pool in order to not block the zk event</span>
<span class="source-line-no">071</span><span id="line-71"> // processing since all the zk listener across HMaster will be called in one thread sequentially.</span>
<span class="source-line-no">072</span><span id="line-72"> private final ExecutorService executor;</span>
<span class="source-line-no">073</span><span id="line-73"></span>
<span class="source-line-no">074</span><span id="line-74"> public RegionServerTracker(ZKWatcher watcher, MasterServices server) {</span>
<span class="source-line-no">075</span><span id="line-75"> super(watcher);</span>
<span class="source-line-no">076</span><span id="line-76"> this.server = server;</span>
<span class="source-line-no">077</span><span id="line-77"> this.executor = Executors.newSingleThreadExecutor(</span>
<span class="source-line-no">078</span><span id="line-78"> new ThreadFactoryBuilder().setDaemon(true).setNameFormat("RegionServerTracker-%d").build());</span>
<span class="source-line-no">079</span><span id="line-79"> watcher.registerListener(this);</span>
<span class="source-line-no">080</span><span id="line-80"> refresh();</span>
<span class="source-line-no">081</span><span id="line-81"> }</span>
<span class="source-line-no">082</span><span id="line-82"></span>
<span class="source-line-no">083</span><span id="line-83"> private RegionServerInfo getServerInfo(ServerName serverName)</span>
<span class="source-line-no">084</span><span id="line-84"> throws KeeperException, IOException {</span>
<span class="source-line-no">085</span><span id="line-85"> String nodePath = watcher.getZNodePaths().getRsPath(serverName);</span>
<span class="source-line-no">086</span><span id="line-86"> byte[] data;</span>
<span class="source-line-no">087</span><span id="line-87"> try {</span>
<span class="source-line-no">088</span><span id="line-88"> data = ZKUtil.getData(watcher, nodePath);</span>
<span class="source-line-no">089</span><span id="line-89"> } catch (InterruptedException e) {</span>
<span class="source-line-no">090</span><span id="line-90"> throw (InterruptedIOException) new InterruptedIOException().initCause(e);</span>
<span class="source-line-no">091</span><span id="line-91"> }</span>
<span class="source-line-no">092</span><span id="line-92"> if (data == null) {</span>
<span class="source-line-no">093</span><span id="line-93"> // we should receive a children changed event later and then we will expire it, so we still</span>
<span class="source-line-no">094</span><span id="line-94"> // need to add it to the region server set.</span>
<span class="source-line-no">095</span><span id="line-95"> LOG.warn("Server node {} does not exist, already dead?", serverName);</span>
<span class="source-line-no">096</span><span id="line-96"> return null;</span>
<span class="source-line-no">097</span><span id="line-97"> }</span>
<span class="source-line-no">098</span><span id="line-98"> if (data.length == 0 || !ProtobufUtil.isPBMagicPrefix(data)) {</span>
<span class="source-line-no">099</span><span id="line-99"> // this should not happen actually, unless we have bugs or someone has messed zk up.</span>
<span class="source-line-no">100</span><span id="line-100"> LOG.warn("Invalid data for region server node {} on zookeeper, data length = {}", serverName,</span>
<span class="source-line-no">101</span><span id="line-101"> data.length);</span>
<span class="source-line-no">102</span><span id="line-102"> return null;</span>
<span class="source-line-no">103</span><span id="line-103"> }</span>
<span class="source-line-no">104</span><span id="line-104"> RegionServerInfo.Builder builder = RegionServerInfo.newBuilder();</span>
<span class="source-line-no">105</span><span id="line-105"> int magicLen = ProtobufUtil.lengthOfPBMagic();</span>
<span class="source-line-no">106</span><span id="line-106"> ProtobufUtil.mergeFrom(builder, data, magicLen, data.length - magicLen);</span>
<span class="source-line-no">107</span><span id="line-107"> return builder.build();</span>
<span class="source-line-no">108</span><span id="line-108"> }</span>
<span class="source-line-no">109</span><span id="line-109"></span>
<span class="source-line-no">110</span><span id="line-110"> /**</span>
<span class="source-line-no">111</span><span id="line-111"> * Upgrade to active master mode, where besides tracking the changes of region server set, we will</span>
<span class="source-line-no">112</span><span id="line-112"> * also started to add new region servers to ServerManager and also schedule SCP if a region</span>
<span class="source-line-no">113</span><span id="line-113"> * server dies. Starts the tracking of online RegionServers. All RSes will be tracked after this</span>
<span class="source-line-no">114</span><span id="line-114"> * method is called.</span>
<span class="source-line-no">115</span><span id="line-115"> * &lt;p/&gt;</span>
<span class="source-line-no">116</span><span id="line-116"> * In this method, we will also construct the region server sets in {@link ServerManager}. If a</span>
<span class="source-line-no">117</span><span id="line-117"> * region server is dead between the crash of the previous master instance and the start of the</span>
<span class="source-line-no">118</span><span id="line-118"> * current master instance, we will schedule a SCP for it. This is done in</span>
<span class="source-line-no">119</span><span id="line-119"> * {@link ServerManager#findDeadServersAndProcess(Set, Set)}, we call it here under the lock</span>
<span class="source-line-no">120</span><span id="line-120"> * protection to prevent concurrency issues with server expiration operation.</span>
<span class="source-line-no">121</span><span id="line-121"> * @param deadServersFromPE the region servers which already have SCP associated.</span>
<span class="source-line-no">122</span><span id="line-122"> * @param liveServersBeforeRestart the live region servers we recorded before master restarts.</span>
<span class="source-line-no">123</span><span id="line-123"> * @param splittingServersFromWALDir Servers whose WALs are being actively 'split'.</span>
<span class="source-line-no">124</span><span id="line-124"> */</span>
<span class="source-line-no">125</span><span id="line-125"> public void upgrade(Set&lt;ServerName&gt; deadServersFromPE, Set&lt;ServerName&gt; liveServersBeforeRestart,</span>
<span class="source-line-no">126</span><span id="line-126"> Set&lt;ServerName&gt; splittingServersFromWALDir) throws KeeperException, IOException {</span>
<span class="source-line-no">127</span><span id="line-127"> LOG.info(</span>
<span class="source-line-no">128</span><span id="line-128"> "Upgrading RegionServerTracker to active master mode; {} have existing"</span>
<span class="source-line-no">129</span><span id="line-129"> + "ServerCrashProcedures, {} possibly 'live' servers, and {} 'splitting'.",</span>
<span class="source-line-no">130</span><span id="line-130"> deadServersFromPE.size(), liveServersBeforeRestart.size(), splittingServersFromWALDir.size());</span>
<span class="source-line-no">131</span><span id="line-131"> // deadServersFromPE is made from a list of outstanding ServerCrashProcedures.</span>
<span class="source-line-no">132</span><span id="line-132"> // splittingServersFromWALDir are being actively split -- the directory in the FS ends in</span>
<span class="source-line-no">133</span><span id="line-133"> // '-SPLITTING'. Each splitting server should have a corresponding SCP. Log if not.</span>
<span class="source-line-no">134</span><span id="line-134"> splittingServersFromWALDir.stream().filter(s -&gt; !deadServersFromPE.contains(s))</span>
<span class="source-line-no">135</span><span id="line-135"> .forEach(s -&gt; LOG.error("{} has no matching ServerCrashProcedure", s));</span>
<span class="source-line-no">136</span><span id="line-136"> // create ServerNode for all possible live servers from wal directory and master local region</span>
<span class="source-line-no">137</span><span id="line-137"> liveServersBeforeRestart</span>
<span class="source-line-no">138</span><span id="line-138"> .forEach(sn -&gt; server.getAssignmentManager().getRegionStates().createServer(sn));</span>
<span class="source-line-no">139</span><span id="line-139"> ServerManager serverManager = server.getServerManager();</span>
<span class="source-line-no">140</span><span id="line-140"> synchronized (this) {</span>
<span class="source-line-no">141</span><span id="line-141"> Set&lt;ServerName&gt; liveServers = regionServers;</span>
<span class="source-line-no">142</span><span id="line-142"> for (ServerName serverName : liveServers) {</span>
<span class="source-line-no">143</span><span id="line-143"> RegionServerInfo info = getServerInfo(serverName);</span>
<span class="source-line-no">144</span><span id="line-144"> ServerMetrics serverMetrics = info != null</span>
<span class="source-line-no">145</span><span id="line-145"> ? ServerMetricsBuilder.of(serverName,</span>
<span class="source-line-no">146</span><span id="line-146"> VersionInfoUtil.getVersionNumber(info.getVersionInfo()),</span>
<span class="source-line-no">147</span><span id="line-147"> info.getVersionInfo().getVersion())</span>
<span class="source-line-no">148</span><span id="line-148"> : ServerMetricsBuilder.of(serverName);</span>
<span class="source-line-no">149</span><span id="line-149"> serverManager.checkAndRecordNewServer(serverName, serverMetrics);</span>
<span class="source-line-no">150</span><span id="line-150"> }</span>
<span class="source-line-no">151</span><span id="line-151"> serverManager.findDeadServersAndProcess(deadServersFromPE, liveServersBeforeRestart);</span>
<span class="source-line-no">152</span><span id="line-152"> active = true;</span>
<span class="source-line-no">153</span><span id="line-153"> }</span>
<span class="source-line-no">154</span><span id="line-154"> }</span>
<span class="source-line-no">155</span><span id="line-155"></span>
<span class="source-line-no">156</span><span id="line-156"> public void stop() {</span>
<span class="source-line-no">157</span><span id="line-157"> executor.shutdownNow();</span>
<span class="source-line-no">158</span><span id="line-158"> }</span>
<span class="source-line-no">159</span><span id="line-159"></span>
<span class="source-line-no">160</span><span id="line-160"> public Set&lt;ServerName&gt; getRegionServers() {</span>
<span class="source-line-no">161</span><span id="line-161"> return regionServers;</span>
<span class="source-line-no">162</span><span id="line-162"> }</span>
<span class="source-line-no">163</span><span id="line-163"></span>
<span class="source-line-no">164</span><span id="line-164"> // execute the operations which are only needed for active masters, such as expire old servers,</span>
<span class="source-line-no">165</span><span id="line-165"> // add new servers, etc.</span>
<span class="source-line-no">166</span><span id="line-166"> private void processAsActiveMaster(Set&lt;ServerName&gt; newServers) {</span>
<span class="source-line-no">167</span><span id="line-167"> Set&lt;ServerName&gt; oldServers = regionServers;</span>
<span class="source-line-no">168</span><span id="line-168"> ServerManager serverManager = server.getServerManager();</span>
<span class="source-line-no">169</span><span id="line-169"> // expire dead servers</span>
<span class="source-line-no">170</span><span id="line-170"> for (ServerName crashedServer : Sets.difference(oldServers, newServers)) {</span>
<span class="source-line-no">171</span><span id="line-171"> LOG.info("RegionServer ephemeral node deleted, processing expiration [{}]", crashedServer);</span>
<span class="source-line-no">172</span><span id="line-172"> serverManager.expireServer(crashedServer);</span>
<span class="source-line-no">173</span><span id="line-173"> }</span>
<span class="source-line-no">174</span><span id="line-174"> // check whether there are new servers, log them</span>
<span class="source-line-no">175</span><span id="line-175"> boolean newServerAdded = false;</span>
<span class="source-line-no">176</span><span id="line-176"> for (ServerName sn : newServers) {</span>
<span class="source-line-no">177</span><span id="line-177"> if (!oldServers.contains(sn)) {</span>
<span class="source-line-no">178</span><span id="line-178"> newServerAdded = true;</span>
<span class="source-line-no">179</span><span id="line-179"> LOG.info("RegionServer ephemeral node created, adding [" + sn + "]");</span>
<span class="source-line-no">180</span><span id="line-180"> }</span>
<span class="source-line-no">181</span><span id="line-181"> }</span>
<span class="source-line-no">182</span><span id="line-182"> if (newServerAdded &amp;&amp; server.isInitialized()) {</span>
<span class="source-line-no">183</span><span id="line-183"> // Only call the check to move servers if a RegionServer was added to the cluster; in this</span>
<span class="source-line-no">184</span><span id="line-184"> // case it could be a server with a new version so it makes sense to run the check.</span>
<span class="source-line-no">185</span><span id="line-185"> server.checkIfShouldMoveSystemRegionAsync();</span>
<span class="source-line-no">186</span><span id="line-186"> }</span>
<span class="source-line-no">187</span><span id="line-187"> }</span>
<span class="source-line-no">188</span><span id="line-188"></span>
<span class="source-line-no">189</span><span id="line-189"> private synchronized void refresh() {</span>
<span class="source-line-no">190</span><span id="line-190"> List&lt;String&gt; names;</span>
<span class="source-line-no">191</span><span id="line-191"> final Span span = TraceUtil.createSpan("RegionServerTracker.refresh");</span>
<span class="source-line-no">192</span><span id="line-192"> try (final Scope ignored = span.makeCurrent()) {</span>
<span class="source-line-no">193</span><span id="line-193"> try {</span>
<span class="source-line-no">194</span><span id="line-194"> names = ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);</span>
<span class="source-line-no">195</span><span id="line-195"> } catch (KeeperException e) {</span>
<span class="source-line-no">196</span><span id="line-196"> // here we need to abort as we failed to set watcher on the rs node which means that we can</span>
<span class="source-line-no">197</span><span id="line-197"> // not track the node deleted event any more.</span>
<span class="source-line-no">198</span><span id="line-198"> server.abort("Unexpected zk exception getting RS nodes", e);</span>
<span class="source-line-no">199</span><span id="line-199"> return;</span>
<span class="source-line-no">200</span><span id="line-200"> }</span>
<span class="source-line-no">201</span><span id="line-201"> Set&lt;ServerName&gt; newServers = CollectionUtils.isEmpty(names)</span>
<span class="source-line-no">202</span><span id="line-202"> ? Collections.emptySet()</span>
<span class="source-line-no">203</span><span id="line-203"> : names.stream().map(ServerName::parseServerName)</span>
<span class="source-line-no">204</span><span id="line-204"> .collect(Collectors.collectingAndThen(Collectors.toSet(), Collections::unmodifiableSet));</span>
<span class="source-line-no">205</span><span id="line-205"> if (active) {</span>
<span class="source-line-no">206</span><span id="line-206"> processAsActiveMaster(newServers);</span>
<span class="source-line-no">207</span><span id="line-207"> }</span>
<span class="source-line-no">208</span><span id="line-208"> this.regionServers = newServers;</span>
<span class="source-line-no">209</span><span id="line-209"> span.setStatus(StatusCode.OK);</span>
<span class="source-line-no">210</span><span id="line-210"> } finally {</span>
<span class="source-line-no">211</span><span id="line-211"> span.end();</span>
<span class="source-line-no">212</span><span id="line-212"> }</span>
<span class="source-line-no">213</span><span id="line-213"> }</span>
<span class="source-line-no">214</span><span id="line-214"></span>
<span class="source-line-no">215</span><span id="line-215"> @Override</span>
<span class="source-line-no">216</span><span id="line-216"> public void nodeChildrenChanged(String path) {</span>
<span class="source-line-no">217</span><span id="line-217"> if (</span>
<span class="source-line-no">218</span><span id="line-218"> path.equals(watcher.getZNodePaths().rsZNode) &amp;&amp; !server.isAborted() &amp;&amp; !server.isStopped()</span>
<span class="source-line-no">219</span><span id="line-219"> ) {</span>
<span class="source-line-no">220</span><span id="line-220"> executor.execute(this::refresh);</span>
<span class="source-line-no">221</span><span id="line-221"> }</span>
<span class="source-line-no">222</span><span id="line-222"> }</span>
<span class="source-line-no">223</span><span id="line-223">}</span>
</pre>
</div>
</main>
</body>
</html>