blob: b74d9e98715400bae40bd102f193646db3da70d7 [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: ActiveMasterManager">
<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 java.io.IOException;</span>
<span class="source-line-no">021</span><span id="line-21">import java.io.InterruptedIOException;</span>
<span class="source-line-no">022</span><span id="line-22">import java.util.List;</span>
<span class="source-line-no">023</span><span id="line-23">import java.util.Optional;</span>
<span class="source-line-no">024</span><span id="line-24">import java.util.concurrent.atomic.AtomicBoolean;</span>
<span class="source-line-no">025</span><span id="line-25">import org.apache.hadoop.hbase.Server;</span>
<span class="source-line-no">026</span><span id="line-26">import org.apache.hadoop.hbase.ServerName;</span>
<span class="source-line-no">027</span><span id="line-27">import org.apache.hadoop.hbase.ZNodeClearer;</span>
<span class="source-line-no">028</span><span id="line-28">import org.apache.hadoop.hbase.exceptions.DeserializationException;</span>
<span class="source-line-no">029</span><span id="line-29">import org.apache.hadoop.hbase.monitoring.MonitoredTask;</span>
<span class="source-line-no">030</span><span id="line-30">import org.apache.hadoop.hbase.monitoring.TaskGroup;</span>
<span class="source-line-no">031</span><span id="line-31">import org.apache.hadoop.hbase.zookeeper.MasterAddressTracker;</span>
<span class="source-line-no">032</span><span id="line-32">import org.apache.hadoop.hbase.zookeeper.ZKListener;</span>
<span class="source-line-no">033</span><span id="line-33">import org.apache.hadoop.hbase.zookeeper.ZKUtil;</span>
<span class="source-line-no">034</span><span id="line-34">import org.apache.hadoop.hbase.zookeeper.ZKWatcher;</span>
<span class="source-line-no">035</span><span id="line-35">import org.apache.hadoop.hbase.zookeeper.ZNodePaths;</span>
<span class="source-line-no">036</span><span id="line-36">import org.apache.yetus.audience.InterfaceAudience;</span>
<span class="source-line-no">037</span><span id="line-37">import org.apache.zookeeper.KeeperException;</span>
<span class="source-line-no">038</span><span id="line-38">import org.slf4j.Logger;</span>
<span class="source-line-no">039</span><span id="line-39">import org.slf4j.LoggerFactory;</span>
<span class="source-line-no">040</span><span id="line-40"></span>
<span class="source-line-no">041</span><span id="line-41">import org.apache.hbase.thirdparty.com.google.common.collect.ImmutableList;</span>
<span class="source-line-no">042</span><span id="line-42"></span>
<span class="source-line-no">043</span><span id="line-43">import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;</span>
<span class="source-line-no">044</span><span id="line-44"></span>
<span class="source-line-no">045</span><span id="line-45">/**</span>
<span class="source-line-no">046</span><span id="line-46"> * Handles everything on master-side related to master election. Keeps track of currently active</span>
<span class="source-line-no">047</span><span id="line-47"> * master and registered backup masters.</span>
<span class="source-line-no">048</span><span id="line-48"> * &lt;p&gt;</span>
<span class="source-line-no">049</span><span id="line-49"> * Listens and responds to ZooKeeper notifications on the master znodes, both</span>
<span class="source-line-no">050</span><span id="line-50"> * &lt;code&gt;nodeCreated&lt;/code&gt; and &lt;code&gt;nodeDeleted&lt;/code&gt;.</span>
<span class="source-line-no">051</span><span id="line-51"> * &lt;p&gt;</span>
<span class="source-line-no">052</span><span id="line-52"> * Contains blocking methods which will hold up backup masters, waiting for the active master to</span>
<span class="source-line-no">053</span><span id="line-53"> * fail.</span>
<span class="source-line-no">054</span><span id="line-54"> * &lt;p&gt;</span>
<span class="source-line-no">055</span><span id="line-55"> * This class is instantiated in the HMaster constructor and the method</span>
<span class="source-line-no">056</span><span id="line-56"> * #blockUntilBecomingActiveMaster() is called to wait until becoming the active master of the</span>
<span class="source-line-no">057</span><span id="line-57"> * cluster.</span>
<span class="source-line-no">058</span><span id="line-58"> */</span>
<span class="source-line-no">059</span><span id="line-59">@InterfaceAudience.Private</span>
<span class="source-line-no">060</span><span id="line-60">public class ActiveMasterManager extends ZKListener {</span>
<span class="source-line-no">061</span><span id="line-61"> private static final Logger LOG = LoggerFactory.getLogger(ActiveMasterManager.class);</span>
<span class="source-line-no">062</span><span id="line-62"></span>
<span class="source-line-no">063</span><span id="line-63"> final AtomicBoolean clusterHasActiveMaster = new AtomicBoolean(false);</span>
<span class="source-line-no">064</span><span id="line-64"> final AtomicBoolean clusterShutDown = new AtomicBoolean(false);</span>
<span class="source-line-no">065</span><span id="line-65"></span>
<span class="source-line-no">066</span><span id="line-66"> // This server's information. Package-private for child implementations.</span>
<span class="source-line-no">067</span><span id="line-67"> int infoPort;</span>
<span class="source-line-no">068</span><span id="line-68"> final ServerName sn;</span>
<span class="source-line-no">069</span><span id="line-69"> final Server master;</span>
<span class="source-line-no">070</span><span id="line-70"></span>
<span class="source-line-no">071</span><span id="line-71"> // Active master's server name. Invalidated anytime active master changes (based on ZK</span>
<span class="source-line-no">072</span><span id="line-72"> // notifications) and lazily fetched on-demand.</span>
<span class="source-line-no">073</span><span id="line-73"> // ServerName is immutable, so we don't need heavy synchronization around it.</span>
<span class="source-line-no">074</span><span id="line-74"> volatile ServerName activeMasterServerName;</span>
<span class="source-line-no">075</span><span id="line-75"> // Registered backup masters. List is kept up to date based on ZK change notifications to</span>
<span class="source-line-no">076</span><span id="line-76"> // backup znode.</span>
<span class="source-line-no">077</span><span id="line-77"> private volatile ImmutableList&lt;ServerName&gt; backupMasters;</span>
<span class="source-line-no">078</span><span id="line-78"></span>
<span class="source-line-no">079</span><span id="line-79"> /**</span>
<span class="source-line-no">080</span><span id="line-80"> * @param watcher ZK watcher</span>
<span class="source-line-no">081</span><span id="line-81"> * @param sn ServerName</span>
<span class="source-line-no">082</span><span id="line-82"> * @param master In an instance of a Master.</span>
<span class="source-line-no">083</span><span id="line-83"> */</span>
<span class="source-line-no">084</span><span id="line-84"> ActiveMasterManager(ZKWatcher watcher, ServerName sn, Server master)</span>
<span class="source-line-no">085</span><span id="line-85"> throws InterruptedIOException {</span>
<span class="source-line-no">086</span><span id="line-86"> super(watcher);</span>
<span class="source-line-no">087</span><span id="line-87"> watcher.registerListener(this);</span>
<span class="source-line-no">088</span><span id="line-88"> this.sn = sn;</span>
<span class="source-line-no">089</span><span id="line-89"> this.master = master;</span>
<span class="source-line-no">090</span><span id="line-90"> updateBackupMasters();</span>
<span class="source-line-no">091</span><span id="line-91"> }</span>
<span class="source-line-no">092</span><span id="line-92"></span>
<span class="source-line-no">093</span><span id="line-93"> // will be set after jetty server is started</span>
<span class="source-line-no">094</span><span id="line-94"> public void setInfoPort(int infoPort) {</span>
<span class="source-line-no">095</span><span id="line-95"> this.infoPort = infoPort;</span>
<span class="source-line-no">096</span><span id="line-96"> }</span>
<span class="source-line-no">097</span><span id="line-97"></span>
<span class="source-line-no">098</span><span id="line-98"> @Override</span>
<span class="source-line-no">099</span><span id="line-99"> public void nodeCreated(String path) {</span>
<span class="source-line-no">100</span><span id="line-100"> handle(path);</span>
<span class="source-line-no">101</span><span id="line-101"> }</span>
<span class="source-line-no">102</span><span id="line-102"></span>
<span class="source-line-no">103</span><span id="line-103"> @Override</span>
<span class="source-line-no">104</span><span id="line-104"> public void nodeChildrenChanged(String path) {</span>
<span class="source-line-no">105</span><span id="line-105"> if (path.equals(watcher.getZNodePaths().backupMasterAddressesZNode)) {</span>
<span class="source-line-no">106</span><span id="line-106"> try {</span>
<span class="source-line-no">107</span><span id="line-107"> updateBackupMasters();</span>
<span class="source-line-no">108</span><span id="line-108"> } catch (InterruptedIOException ioe) {</span>
<span class="source-line-no">109</span><span id="line-109"> LOG.error("Error updating backup masters", ioe);</span>
<span class="source-line-no">110</span><span id="line-110"> }</span>
<span class="source-line-no">111</span><span id="line-111"> }</span>
<span class="source-line-no">112</span><span id="line-112"> }</span>
<span class="source-line-no">113</span><span id="line-113"></span>
<span class="source-line-no">114</span><span id="line-114"> @Override</span>
<span class="source-line-no">115</span><span id="line-115"> public void nodeDeleted(String path) {</span>
<span class="source-line-no">116</span><span id="line-116"> // We need to keep track of the cluster's shutdown status while</span>
<span class="source-line-no">117</span><span id="line-117"> // we wait on the current master. We consider that, if the cluster</span>
<span class="source-line-no">118</span><span id="line-118"> // was already in a "shutdown" state when we started, that this master</span>
<span class="source-line-no">119</span><span id="line-119"> // is part of a new cluster that was started shortly after the old cluster</span>
<span class="source-line-no">120</span><span id="line-120"> // shut down, so that state is now irrelevant. This means that the shutdown</span>
<span class="source-line-no">121</span><span id="line-121"> // state must be set while we wait on the active master in order</span>
<span class="source-line-no">122</span><span id="line-122"> // to shutdown this master. See HBASE-8519.</span>
<span class="source-line-no">123</span><span id="line-123"> if (path.equals(watcher.getZNodePaths().clusterStateZNode) &amp;&amp; !master.isStopped()) {</span>
<span class="source-line-no">124</span><span id="line-124"> clusterShutDown.set(true);</span>
<span class="source-line-no">125</span><span id="line-125"> }</span>
<span class="source-line-no">126</span><span id="line-126"> handle(path);</span>
<span class="source-line-no">127</span><span id="line-127"> }</span>
<span class="source-line-no">128</span><span id="line-128"></span>
<span class="source-line-no">129</span><span id="line-129"> void handle(final String path) {</span>
<span class="source-line-no">130</span><span id="line-130"> if (path.equals(watcher.getZNodePaths().masterAddressZNode) &amp;&amp; !master.isStopped()) {</span>
<span class="source-line-no">131</span><span id="line-131"> handleMasterNodeChange();</span>
<span class="source-line-no">132</span><span id="line-132"> }</span>
<span class="source-line-no">133</span><span id="line-133"> }</span>
<span class="source-line-no">134</span><span id="line-134"></span>
<span class="source-line-no">135</span><span id="line-135"> private void updateBackupMasters() throws InterruptedIOException {</span>
<span class="source-line-no">136</span><span id="line-136"> backupMasters =</span>
<span class="source-line-no">137</span><span id="line-137"> ImmutableList.copyOf(MasterAddressTracker.getBackupMastersAndRenewWatch(watcher));</span>
<span class="source-line-no">138</span><span id="line-138"> }</span>
<span class="source-line-no">139</span><span id="line-139"></span>
<span class="source-line-no">140</span><span id="line-140"> /**</span>
<span class="source-line-no">141</span><span id="line-141"> * Fetches the active master's ServerName from zookeeper.</span>
<span class="source-line-no">142</span><span id="line-142"> */</span>
<span class="source-line-no">143</span><span id="line-143"> private void fetchAndSetActiveMasterServerName() {</span>
<span class="source-line-no">144</span><span id="line-144"> LOG.debug("Attempting to fetch active master sn from zk");</span>
<span class="source-line-no">145</span><span id="line-145"> try {</span>
<span class="source-line-no">146</span><span id="line-146"> activeMasterServerName = MasterAddressTracker.getMasterAddress(watcher);</span>
<span class="source-line-no">147</span><span id="line-147"> } catch (IOException | KeeperException e) {</span>
<span class="source-line-no">148</span><span id="line-148"> // Log and ignore for now and re-fetch later if needed.</span>
<span class="source-line-no">149</span><span id="line-149"> LOG.error("Error fetching active master information", e);</span>
<span class="source-line-no">150</span><span id="line-150"> }</span>
<span class="source-line-no">151</span><span id="line-151"> }</span>
<span class="source-line-no">152</span><span id="line-152"></span>
<span class="source-line-no">153</span><span id="line-153"> public Optional&lt;ServerName&gt; getActiveMasterServerName() {</span>
<span class="source-line-no">154</span><span id="line-154"> if (!clusterHasActiveMaster.get()) {</span>
<span class="source-line-no">155</span><span id="line-155"> return Optional.empty();</span>
<span class="source-line-no">156</span><span id="line-156"> }</span>
<span class="source-line-no">157</span><span id="line-157"> if (activeMasterServerName == null) {</span>
<span class="source-line-no">158</span><span id="line-158"> fetchAndSetActiveMasterServerName();</span>
<span class="source-line-no">159</span><span id="line-159"> }</span>
<span class="source-line-no">160</span><span id="line-160"> // It could still be null, but return whatever we have.</span>
<span class="source-line-no">161</span><span id="line-161"> return Optional.ofNullable(activeMasterServerName);</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"> public int getActiveMasterInfoPort() {</span>
<span class="source-line-no">165</span><span id="line-165"> try {</span>
<span class="source-line-no">166</span><span id="line-166"> return MasterAddressTracker.getMasterInfoPort(watcher);</span>
<span class="source-line-no">167</span><span id="line-167"> } catch (Exception e) {</span>
<span class="source-line-no">168</span><span id="line-168"> LOG.warn("Failed to get active master's info port.", e);</span>
<span class="source-line-no">169</span><span id="line-169"> return 0;</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"></span>
<span class="source-line-no">173</span><span id="line-173"> public int getBackupMasterInfoPort(final ServerName sn) {</span>
<span class="source-line-no">174</span><span id="line-174"> try {</span>
<span class="source-line-no">175</span><span id="line-175"> return MasterAddressTracker.getBackupMasterInfoPort(watcher, sn);</span>
<span class="source-line-no">176</span><span id="line-176"> } catch (Exception e) {</span>
<span class="source-line-no">177</span><span id="line-177"> LOG.warn("Failed to get backup master: " + sn + "'s info port.", e);</span>
<span class="source-line-no">178</span><span id="line-178"> return 0;</span>
<span class="source-line-no">179</span><span id="line-179"> }</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"> /**</span>
<span class="source-line-no">183</span><span id="line-183"> * Handle a change in the master node. Doesn't matter whether this was called from a nodeCreated</span>
<span class="source-line-no">184</span><span id="line-184"> * or nodeDeleted event because there are no guarantees that the current state of the master node</span>
<span class="source-line-no">185</span><span id="line-185"> * matches the event at the time of our next ZK request.</span>
<span class="source-line-no">186</span><span id="line-186"> * &lt;p&gt;</span>
<span class="source-line-no">187</span><span id="line-187"> * Uses the watchAndCheckExists method which watches the master address node regardless of whether</span>
<span class="source-line-no">188</span><span id="line-188"> * it exists or not. If it does exist (there is an active master), it returns true. Otherwise it</span>
<span class="source-line-no">189</span><span id="line-189"> * returns false.</span>
<span class="source-line-no">190</span><span id="line-190"> * &lt;p&gt;</span>
<span class="source-line-no">191</span><span id="line-191"> * A watcher is set which guarantees that this method will get called again if there is another</span>
<span class="source-line-no">192</span><span id="line-192"> * change in the master node.</span>
<span class="source-line-no">193</span><span id="line-193"> */</span>
<span class="source-line-no">194</span><span id="line-194"> private void handleMasterNodeChange() {</span>
<span class="source-line-no">195</span><span id="line-195"> // Watch the node and check if it exists.</span>
<span class="source-line-no">196</span><span id="line-196"> try {</span>
<span class="source-line-no">197</span><span id="line-197"> synchronized (clusterHasActiveMaster) {</span>
<span class="source-line-no">198</span><span id="line-198"> if (ZKUtil.watchAndCheckExists(watcher, watcher.getZNodePaths().masterAddressZNode)) {</span>
<span class="source-line-no">199</span><span id="line-199"> // A master node exists, there is an active master</span>
<span class="source-line-no">200</span><span id="line-200"> LOG.trace("A master is now available");</span>
<span class="source-line-no">201</span><span id="line-201"> clusterHasActiveMaster.set(true);</span>
<span class="source-line-no">202</span><span id="line-202"> } else {</span>
<span class="source-line-no">203</span><span id="line-203"> // Node is no longer there, cluster does not have an active master</span>
<span class="source-line-no">204</span><span id="line-204"> LOG.debug("No master available. Notifying waiting threads");</span>
<span class="source-line-no">205</span><span id="line-205"> clusterHasActiveMaster.set(false);</span>
<span class="source-line-no">206</span><span id="line-206"> // Notify any thread waiting to become the active master</span>
<span class="source-line-no">207</span><span id="line-207"> clusterHasActiveMaster.notifyAll();</span>
<span class="source-line-no">208</span><span id="line-208"> }</span>
<span class="source-line-no">209</span><span id="line-209"> // Reset the active master sn. Will be re-fetched later if needed.</span>
<span class="source-line-no">210</span><span id="line-210"> // We don't want to make a synchronous RPC under a monitor.</span>
<span class="source-line-no">211</span><span id="line-211"> activeMasterServerName = null;</span>
<span class="source-line-no">212</span><span id="line-212"> }</span>
<span class="source-line-no">213</span><span id="line-213"> } catch (KeeperException ke) {</span>
<span class="source-line-no">214</span><span id="line-214"> master.abort("Received an unexpected KeeperException, aborting", ke);</span>
<span class="source-line-no">215</span><span id="line-215"> }</span>
<span class="source-line-no">216</span><span id="line-216"> }</span>
<span class="source-line-no">217</span><span id="line-217"></span>
<span class="source-line-no">218</span><span id="line-218"> /**</span>
<span class="source-line-no">219</span><span id="line-219"> * Block until becoming the active master. Method blocks until there is not another active master</span>
<span class="source-line-no">220</span><span id="line-220"> * and our attempt to become the new active master is successful. This also makes sure that we are</span>
<span class="source-line-no">221</span><span id="line-221"> * watching the master znode so will be notified if another master dies.</span>
<span class="source-line-no">222</span><span id="line-222"> * @param checkInterval the interval to check if the master is stopped</span>
<span class="source-line-no">223</span><span id="line-223"> * @param startupTaskGroup the task group for master startup to track the progress</span>
<span class="source-line-no">224</span><span id="line-224"> * @return True if no issue becoming active master else false if another master was running or if</span>
<span class="source-line-no">225</span><span id="line-225"> * some other problem (zookeeper, stop flag has been set on this Master)</span>
<span class="source-line-no">226</span><span id="line-226"> */</span>
<span class="source-line-no">227</span><span id="line-227"> boolean blockUntilBecomingActiveMaster(int checkInterval, TaskGroup startupTaskGroup) {</span>
<span class="source-line-no">228</span><span id="line-228"> MonitoredTask blockUntilActive =</span>
<span class="source-line-no">229</span><span id="line-229"> startupTaskGroup.addTask("Blocking until becoming active master");</span>
<span class="source-line-no">230</span><span id="line-230"> String backupZNode = ZNodePaths</span>
<span class="source-line-no">231</span><span id="line-231"> .joinZNode(this.watcher.getZNodePaths().backupMasterAddressesZNode, this.sn.toString());</span>
<span class="source-line-no">232</span><span id="line-232"> while (!(master.isAborted() || master.isStopped())) {</span>
<span class="source-line-no">233</span><span id="line-233"> blockUntilActive.setStatus("Trying to register in ZK as active master");</span>
<span class="source-line-no">234</span><span id="line-234"> // Try to become the active master, watch if there is another master.</span>
<span class="source-line-no">235</span><span id="line-235"> // Write out our ServerName as versioned bytes.</span>
<span class="source-line-no">236</span><span id="line-236"> try {</span>
<span class="source-line-no">237</span><span id="line-237"> if (</span>
<span class="source-line-no">238</span><span id="line-238"> MasterAddressTracker.setMasterAddress(this.watcher,</span>
<span class="source-line-no">239</span><span id="line-239"> this.watcher.getZNodePaths().masterAddressZNode, this.sn, infoPort)</span>
<span class="source-line-no">240</span><span id="line-240"> ) {</span>
<span class="source-line-no">241</span><span id="line-241"></span>
<span class="source-line-no">242</span><span id="line-242"> // If we were a backup master before, delete our ZNode from the backup</span>
<span class="source-line-no">243</span><span id="line-243"> // master directory since we are the active now)</span>
<span class="source-line-no">244</span><span id="line-244"> if (ZKUtil.checkExists(this.watcher, backupZNode) != -1) {</span>
<span class="source-line-no">245</span><span id="line-245"> LOG.info("Deleting ZNode for " + backupZNode + " from backup master directory");</span>
<span class="source-line-no">246</span><span id="line-246"> ZKUtil.deleteNodeFailSilent(this.watcher, backupZNode);</span>
<span class="source-line-no">247</span><span id="line-247"> }</span>
<span class="source-line-no">248</span><span id="line-248"> // Save the znode in a file, this will allow to check if we crash in the launch scripts</span>
<span class="source-line-no">249</span><span id="line-249"> ZNodeClearer.writeMyEphemeralNodeOnDisk(this.sn.toString());</span>
<span class="source-line-no">250</span><span id="line-250"></span>
<span class="source-line-no">251</span><span id="line-251"> // We are the master, return</span>
<span class="source-line-no">252</span><span id="line-252"> blockUntilActive.setStatus("Successfully registered as active master.");</span>
<span class="source-line-no">253</span><span id="line-253"> this.clusterHasActiveMaster.set(true);</span>
<span class="source-line-no">254</span><span id="line-254"> activeMasterServerName = sn;</span>
<span class="source-line-no">255</span><span id="line-255"> LOG.info("Registered as active master=" + this.sn);</span>
<span class="source-line-no">256</span><span id="line-256"> return true;</span>
<span class="source-line-no">257</span><span id="line-257"> }</span>
<span class="source-line-no">258</span><span id="line-258"></span>
<span class="source-line-no">259</span><span id="line-259"> // Invalidate the active master name so that subsequent requests do not get any stale</span>
<span class="source-line-no">260</span><span id="line-260"> // master information. Will be re-fetched if needed.</span>
<span class="source-line-no">261</span><span id="line-261"> activeMasterServerName = null;</span>
<span class="source-line-no">262</span><span id="line-262"> // There is another active master running elsewhere or this is a restart</span>
<span class="source-line-no">263</span><span id="line-263"> // and the master ephemeral node has not expired yet.</span>
<span class="source-line-no">264</span><span id="line-264"> this.clusterHasActiveMaster.set(true);</span>
<span class="source-line-no">265</span><span id="line-265"></span>
<span class="source-line-no">266</span><span id="line-266"> String msg;</span>
<span class="source-line-no">267</span><span id="line-267"> byte[] bytes =</span>
<span class="source-line-no">268</span><span id="line-268"> ZKUtil.getDataAndWatch(this.watcher, this.watcher.getZNodePaths().masterAddressZNode);</span>
<span class="source-line-no">269</span><span id="line-269"> if (bytes == null) {</span>
<span class="source-line-no">270</span><span id="line-270"> msg = ("A master was detected, but went down before its address "</span>
<span class="source-line-no">271</span><span id="line-271"> + "could be read. Attempting to become the next active master");</span>
<span class="source-line-no">272</span><span id="line-272"> } else {</span>
<span class="source-line-no">273</span><span id="line-273"> ServerName currentMaster;</span>
<span class="source-line-no">274</span><span id="line-274"> try {</span>
<span class="source-line-no">275</span><span id="line-275"> currentMaster = ProtobufUtil.parseServerNameFrom(bytes);</span>
<span class="source-line-no">276</span><span id="line-276"> } catch (DeserializationException e) {</span>
<span class="source-line-no">277</span><span id="line-277"> LOG.warn("Failed parse", e);</span>
<span class="source-line-no">278</span><span id="line-278"> // Hopefully next time around we won't fail the parse. Dangerous.</span>
<span class="source-line-no">279</span><span id="line-279"> continue;</span>
<span class="source-line-no">280</span><span id="line-280"> }</span>
<span class="source-line-no">281</span><span id="line-281"> if (ServerName.isSameAddress(currentMaster, this.sn)) {</span>
<span class="source-line-no">282</span><span id="line-282"> msg = ("Current master has this master's address, " + currentMaster</span>
<span class="source-line-no">283</span><span id="line-283"> + "; master was restarted? Deleting node.");</span>
<span class="source-line-no">284</span><span id="line-284"> // Hurry along the expiration of the znode.</span>
<span class="source-line-no">285</span><span id="line-285"> ZKUtil.deleteNode(this.watcher, this.watcher.getZNodePaths().masterAddressZNode);</span>
<span class="source-line-no">286</span><span id="line-286"></span>
<span class="source-line-no">287</span><span id="line-287"> // We may have failed to delete the znode at the previous step, but</span>
<span class="source-line-no">288</span><span id="line-288"> // we delete the file anyway: a second attempt to delete the znode is likely to fail</span>
<span class="source-line-no">289</span><span id="line-289"> // again.</span>
<span class="source-line-no">290</span><span id="line-290"> ZNodeClearer.deleteMyEphemeralNodeOnDisk();</span>
<span class="source-line-no">291</span><span id="line-291"> } else {</span>
<span class="source-line-no">292</span><span id="line-292"> msg = "Another master is the active master, " + currentMaster</span>
<span class="source-line-no">293</span><span id="line-293"> + "; waiting to become the next active master";</span>
<span class="source-line-no">294</span><span id="line-294"> }</span>
<span class="source-line-no">295</span><span id="line-295"> }</span>
<span class="source-line-no">296</span><span id="line-296"> LOG.info(msg);</span>
<span class="source-line-no">297</span><span id="line-297"> blockUntilActive.setStatus(msg);</span>
<span class="source-line-no">298</span><span id="line-298"> } catch (KeeperException ke) {</span>
<span class="source-line-no">299</span><span id="line-299"> master.abort("Received an unexpected KeeperException, aborting", ke);</span>
<span class="source-line-no">300</span><span id="line-300"> return false;</span>
<span class="source-line-no">301</span><span id="line-301"> }</span>
<span class="source-line-no">302</span><span id="line-302"> synchronized (this.clusterHasActiveMaster) {</span>
<span class="source-line-no">303</span><span id="line-303"> while (clusterHasActiveMaster.get() &amp;&amp; !master.isStopped()) {</span>
<span class="source-line-no">304</span><span id="line-304"> try {</span>
<span class="source-line-no">305</span><span id="line-305"> clusterHasActiveMaster.wait(checkInterval);</span>
<span class="source-line-no">306</span><span id="line-306"> } catch (InterruptedException e) {</span>
<span class="source-line-no">307</span><span id="line-307"> // We expect to be interrupted when a master dies,</span>
<span class="source-line-no">308</span><span id="line-308"> // will fall out if so</span>
<span class="source-line-no">309</span><span id="line-309"> LOG.debug("Interrupted waiting for master to die", e);</span>
<span class="source-line-no">310</span><span id="line-310"> }</span>
<span class="source-line-no">311</span><span id="line-311"> }</span>
<span class="source-line-no">312</span><span id="line-312"> if (clusterShutDown.get()) {</span>
<span class="source-line-no">313</span><span id="line-313"> this.master.stop("Cluster went down before this master became active");</span>
<span class="source-line-no">314</span><span id="line-314"> }</span>
<span class="source-line-no">315</span><span id="line-315"> }</span>
<span class="source-line-no">316</span><span id="line-316"> }</span>
<span class="source-line-no">317</span><span id="line-317"> return false;</span>
<span class="source-line-no">318</span><span id="line-318"> }</span>
<span class="source-line-no">319</span><span id="line-319"></span>
<span class="source-line-no">320</span><span id="line-320"> /** Returns True if cluster has an active master. */</span>
<span class="source-line-no">321</span><span id="line-321"> boolean hasActiveMaster() {</span>
<span class="source-line-no">322</span><span id="line-322"> try {</span>
<span class="source-line-no">323</span><span id="line-323"> if (ZKUtil.checkExists(watcher, watcher.getZNodePaths().masterAddressZNode) &gt;= 0) {</span>
<span class="source-line-no">324</span><span id="line-324"> return true;</span>
<span class="source-line-no">325</span><span id="line-325"> }</span>
<span class="source-line-no">326</span><span id="line-326"> } catch (KeeperException ke) {</span>
<span class="source-line-no">327</span><span id="line-327"> LOG.info("Received an unexpected KeeperException when checking " + "isActiveMaster : " + ke);</span>
<span class="source-line-no">328</span><span id="line-328"> }</span>
<span class="source-line-no">329</span><span id="line-329"> return false;</span>
<span class="source-line-no">330</span><span id="line-330"> }</span>
<span class="source-line-no">331</span><span id="line-331"></span>
<span class="source-line-no">332</span><span id="line-332"> public void stop() {</span>
<span class="source-line-no">333</span><span id="line-333"> try {</span>
<span class="source-line-no">334</span><span id="line-334"> synchronized (clusterHasActiveMaster) {</span>
<span class="source-line-no">335</span><span id="line-335"> // Master is already stopped, wake up the manager</span>
<span class="source-line-no">336</span><span id="line-336"> // thread so that it can shutdown soon.</span>
<span class="source-line-no">337</span><span id="line-337"> clusterHasActiveMaster.notifyAll();</span>
<span class="source-line-no">338</span><span id="line-338"> }</span>
<span class="source-line-no">339</span><span id="line-339"> // If our address is in ZK, delete it on our way out</span>
<span class="source-line-no">340</span><span id="line-340"> ServerName activeMaster = null;</span>
<span class="source-line-no">341</span><span id="line-341"> try {</span>
<span class="source-line-no">342</span><span id="line-342"> activeMaster = MasterAddressTracker.getMasterAddress(this.watcher);</span>
<span class="source-line-no">343</span><span id="line-343"> } catch (IOException e) {</span>
<span class="source-line-no">344</span><span id="line-344"> LOG.warn("Failed get of master address: " + e.toString());</span>
<span class="source-line-no">345</span><span id="line-345"> }</span>
<span class="source-line-no">346</span><span id="line-346"> if (activeMaster != null &amp;&amp; activeMaster.equals(this.sn)) {</span>
<span class="source-line-no">347</span><span id="line-347"> ZKUtil.deleteNode(watcher, watcher.getZNodePaths().masterAddressZNode);</span>
<span class="source-line-no">348</span><span id="line-348"> // We may have failed to delete the znode at the previous step, but</span>
<span class="source-line-no">349</span><span id="line-349"> // we delete the file anyway: a second attempt to delete the znode is likely to fail again.</span>
<span class="source-line-no">350</span><span id="line-350"> ZNodeClearer.deleteMyEphemeralNodeOnDisk();</span>
<span class="source-line-no">351</span><span id="line-351"> }</span>
<span class="source-line-no">352</span><span id="line-352"> } catch (KeeperException e) {</span>
<span class="source-line-no">353</span><span id="line-353"> LOG.debug(this.watcher.prefix("Failed delete of our master address node; " + e.getMessage()));</span>
<span class="source-line-no">354</span><span id="line-354"> }</span>
<span class="source-line-no">355</span><span id="line-355"> }</span>
<span class="source-line-no">356</span><span id="line-356"></span>
<span class="source-line-no">357</span><span id="line-357"> /** Returns list of registered backup masters. */</span>
<span class="source-line-no">358</span><span id="line-358"> public List&lt;ServerName&gt; getBackupMasters() {</span>
<span class="source-line-no">359</span><span id="line-359"> return backupMasters;</span>
<span class="source-line-no">360</span><span id="line-360"> }</span>
<span class="source-line-no">361</span><span id="line-361">}</span>
</pre>
</div>
</main>
</body>
</html>