blob: f9261c49fe9b89a42c71d2096590c8a5cc188128 [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: TestClusterRestartFailover, class: HMasterForTest">
<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 static org.junit.Assert.assertEquals;</span>
<span class="source-line-no">021</span><span id="line-21">import static org.junit.Assert.assertNotNull;</span>
<span class="source-line-no">022</span><span id="line-22">import static org.junit.Assert.assertNull;</span>
<span class="source-line-no">023</span><span id="line-23">import static org.junit.Assert.assertTrue;</span>
<span class="source-line-no">024</span><span id="line-24"></span>
<span class="source-line-no">025</span><span id="line-25">import java.io.IOException;</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.Optional;</span>
<span class="source-line-no">028</span><span id="line-28">import java.util.concurrent.CountDownLatch;</span>
<span class="source-line-no">029</span><span id="line-29">import java.util.stream.Collectors;</span>
<span class="source-line-no">030</span><span id="line-30">import org.apache.hadoop.conf.Configuration;</span>
<span class="source-line-no">031</span><span id="line-31">import org.apache.hadoop.hbase.CompatibilityFactory;</span>
<span class="source-line-no">032</span><span id="line-32">import org.apache.hadoop.hbase.HBaseClassTestRule;</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.StartTestingClusterOption;</span>
<span class="source-line-no">035</span><span id="line-35">import org.apache.hadoop.hbase.TableName;</span>
<span class="source-line-no">036</span><span id="line-36">import org.apache.hadoop.hbase.Waiter.ExplainingPredicate;</span>
<span class="source-line-no">037</span><span id="line-37">import org.apache.hadoop.hbase.client.RegionInfo;</span>
<span class="source-line-no">038</span><span id="line-38">import org.apache.hadoop.hbase.client.Table;</span>
<span class="source-line-no">039</span><span id="line-39">import org.apache.hadoop.hbase.master.assignment.AssignmentManager;</span>
<span class="source-line-no">040</span><span id="line-40">import org.apache.hadoop.hbase.master.assignment.ServerState;</span>
<span class="source-line-no">041</span><span id="line-41">import org.apache.hadoop.hbase.master.assignment.ServerStateNode;</span>
<span class="source-line-no">042</span><span id="line-42">import org.apache.hadoop.hbase.master.procedure.ServerCrashProcedure;</span>
<span class="source-line-no">043</span><span id="line-43">import org.apache.hadoop.hbase.master.region.MasterRegion;</span>
<span class="source-line-no">044</span><span id="line-44">import org.apache.hadoop.hbase.procedure2.Procedure;</span>
<span class="source-line-no">045</span><span id="line-45">import org.apache.hadoop.hbase.test.MetricsAssertHelper;</span>
<span class="source-line-no">046</span><span id="line-46">import org.apache.hadoop.hbase.testclassification.LargeTests;</span>
<span class="source-line-no">047</span><span id="line-47">import org.apache.hadoop.hbase.testclassification.MasterTests;</span>
<span class="source-line-no">048</span><span id="line-48">import org.junit.ClassRule;</span>
<span class="source-line-no">049</span><span id="line-49">import org.junit.Test;</span>
<span class="source-line-no">050</span><span id="line-50">import org.junit.experimental.categories.Category;</span>
<span class="source-line-no">051</span><span id="line-51">import org.slf4j.Logger;</span>
<span class="source-line-no">052</span><span id="line-52">import org.slf4j.LoggerFactory;</span>
<span class="source-line-no">053</span><span id="line-53"></span>
<span class="source-line-no">054</span><span id="line-54">@Category({ MasterTests.class, LargeTests.class })</span>
<span class="source-line-no">055</span><span id="line-55">public class TestClusterRestartFailover extends AbstractTestRestartCluster {</span>
<span class="source-line-no">056</span><span id="line-56"></span>
<span class="source-line-no">057</span><span id="line-57"> @ClassRule</span>
<span class="source-line-no">058</span><span id="line-58"> public static final HBaseClassTestRule CLASS_RULE =</span>
<span class="source-line-no">059</span><span id="line-59"> HBaseClassTestRule.forClass(TestClusterRestartFailover.class);</span>
<span class="source-line-no">060</span><span id="line-60"></span>
<span class="source-line-no">061</span><span id="line-61"> private static final Logger LOG = LoggerFactory.getLogger(TestClusterRestartFailover.class);</span>
<span class="source-line-no">062</span><span id="line-62"> private static final MetricsAssertHelper metricsHelper =</span>
<span class="source-line-no">063</span><span id="line-63"> CompatibilityFactory.getInstance(MetricsAssertHelper.class);</span>
<span class="source-line-no">064</span><span id="line-64"></span>
<span class="source-line-no">065</span><span id="line-65"> private volatile static CountDownLatch SCP_LATCH;</span>
<span class="source-line-no">066</span><span id="line-66"> private static ServerName SERVER_FOR_TEST;</span>
<span class="source-line-no">067</span><span id="line-67"></span>
<span class="source-line-no">068</span><span id="line-68"> @Override</span>
<span class="source-line-no">069</span><span id="line-69"> protected boolean splitWALCoordinatedByZk() {</span>
<span class="source-line-no">070</span><span id="line-70"> return true;</span>
<span class="source-line-no">071</span><span id="line-71"> }</span>
<span class="source-line-no">072</span><span id="line-72"></span>
<span class="source-line-no">073</span><span id="line-73"> private ServerStateNode getServerStateNode(ServerName serverName) {</span>
<span class="source-line-no">074</span><span id="line-74"> return UTIL.getHBaseCluster().getMaster().getAssignmentManager().getRegionStates()</span>
<span class="source-line-no">075</span><span id="line-75"> .getServerNode(serverName);</span>
<span class="source-line-no">076</span><span id="line-76"> }</span>
<span class="source-line-no">077</span><span id="line-77"></span>
<span class="source-line-no">078</span><span id="line-78"> /**</span>
<span class="source-line-no">079</span><span id="line-79"> * Test for HBASE-22964</span>
<span class="source-line-no">080</span><span id="line-80"> */</span>
<span class="source-line-no">081</span><span id="line-81"> @Test</span>
<span class="source-line-no">082</span><span id="line-82"> public void test() throws Exception {</span>
<span class="source-line-no">083</span><span id="line-83"> setupCluster();</span>
<span class="source-line-no">084</span><span id="line-84"> setupTable();</span>
<span class="source-line-no">085</span><span id="line-85"></span>
<span class="source-line-no">086</span><span id="line-86"> SERVER_FOR_TEST = UTIL.getHBaseCluster().getRegionServer(0).getServerName();</span>
<span class="source-line-no">087</span><span id="line-87"> UTIL.waitFor(60000, () -&gt; getServerStateNode(SERVER_FOR_TEST) != null);</span>
<span class="source-line-no">088</span><span id="line-88"> ServerStateNode serverNode = getServerStateNode(SERVER_FOR_TEST);</span>
<span class="source-line-no">089</span><span id="line-89"> assertNotNull(serverNode);</span>
<span class="source-line-no">090</span><span id="line-90"> assertTrue("serverNode should be ONLINE when cluster runs normally",</span>
<span class="source-line-no">091</span><span id="line-91"> serverNode.isInState(ServerState.ONLINE));</span>
<span class="source-line-no">092</span><span id="line-92"></span>
<span class="source-line-no">093</span><span id="line-93"> SCP_LATCH = new CountDownLatch(1);</span>
<span class="source-line-no">094</span><span id="line-94"></span>
<span class="source-line-no">095</span><span id="line-95"> // Shutdown cluster and restart</span>
<span class="source-line-no">096</span><span id="line-96"> List&lt;Integer&gt; ports =</span>
<span class="source-line-no">097</span><span id="line-97"> UTIL.getHBaseCluster().getMaster().getServerManager().getOnlineServersList().stream()</span>
<span class="source-line-no">098</span><span id="line-98"> .map(serverName -&gt; serverName.getPort()).collect(Collectors.toList());</span>
<span class="source-line-no">099</span><span id="line-99"> LOG.info("Shutting down cluster");</span>
<span class="source-line-no">100</span><span id="line-100"> UTIL.getHBaseCluster().killAll();</span>
<span class="source-line-no">101</span><span id="line-101"> UTIL.getHBaseCluster().waitUntilShutDown();</span>
<span class="source-line-no">102</span><span id="line-102"> LOG.info("Restarting cluster");</span>
<span class="source-line-no">103</span><span id="line-103"> UTIL.restartHBaseCluster(StartTestingClusterOption.builder().masterClass(HMasterForTest.class)</span>
<span class="source-line-no">104</span><span id="line-104"> .numMasters(1).numRegionServers(3).rsPorts(ports).build());</span>
<span class="source-line-no">105</span><span id="line-105"> LOG.info("Started cluster");</span>
<span class="source-line-no">106</span><span id="line-106"> UTIL.waitFor(60000, () -&gt; UTIL.getHBaseCluster().getMaster().isInitialized());</span>
<span class="source-line-no">107</span><span id="line-107"> LOG.info("Started cluster master, waiting for {}", SERVER_FOR_TEST);</span>
<span class="source-line-no">108</span><span id="line-108"> UTIL.waitFor(60000, () -&gt; getServerStateNode(SERVER_FOR_TEST) != null);</span>
<span class="source-line-no">109</span><span id="line-109"> UTIL.waitFor(30000, new ExplainingPredicate&lt;Exception&gt;() {</span>
<span class="source-line-no">110</span><span id="line-110"></span>
<span class="source-line-no">111</span><span id="line-111"> @Override</span>
<span class="source-line-no">112</span><span id="line-112"> public boolean evaluate() throws Exception {</span>
<span class="source-line-no">113</span><span id="line-113"> return !getServerStateNode(SERVER_FOR_TEST).isInState(ServerState.ONLINE);</span>
<span class="source-line-no">114</span><span id="line-114"> }</span>
<span class="source-line-no">115</span><span id="line-115"></span>
<span class="source-line-no">116</span><span id="line-116"> @Override</span>
<span class="source-line-no">117</span><span id="line-117"> public String explainFailure() throws Exception {</span>
<span class="source-line-no">118</span><span id="line-118"> return "serverNode should not be ONLINE during SCP processing";</span>
<span class="source-line-no">119</span><span id="line-119"> }</span>
<span class="source-line-no">120</span><span id="line-120"> });</span>
<span class="source-line-no">121</span><span id="line-121"> Optional&lt;Procedure&lt;?&gt;&gt; procedure = UTIL.getHBaseCluster().getMaster().getProcedures().stream()</span>
<span class="source-line-no">122</span><span id="line-122"> .filter(p -&gt; (p instanceof ServerCrashProcedure)</span>
<span class="source-line-no">123</span><span id="line-123"> &amp;&amp; ((ServerCrashProcedure) p).getServerName().equals(SERVER_FOR_TEST))</span>
<span class="source-line-no">124</span><span id="line-124"> .findAny();</span>
<span class="source-line-no">125</span><span id="line-125"> assertTrue("Should have one SCP for " + SERVER_FOR_TEST, procedure.isPresent());</span>
<span class="source-line-no">126</span><span id="line-126"> assertEquals("Submit the SCP for the same serverName " + SERVER_FOR_TEST + " which should fail",</span>
<span class="source-line-no">127</span><span id="line-127"> Procedure.NO_PROC_ID,</span>
<span class="source-line-no">128</span><span id="line-128"> UTIL.getHBaseCluster().getMaster().getServerManager().expireServer(SERVER_FOR_TEST));</span>
<span class="source-line-no">129</span><span id="line-129"></span>
<span class="source-line-no">130</span><span id="line-130"> // Wait the SCP to finish</span>
<span class="source-line-no">131</span><span id="line-131"> LOG.info("Waiting on latch");</span>
<span class="source-line-no">132</span><span id="line-132"> SCP_LATCH.countDown();</span>
<span class="source-line-no">133</span><span id="line-133"> UTIL.waitFor(60000, () -&gt; procedure.get().isFinished());</span>
<span class="source-line-no">134</span><span id="line-134"> assertNull("serverNode should be deleted after SCP finished",</span>
<span class="source-line-no">135</span><span id="line-135"> getServerStateNode(SERVER_FOR_TEST));</span>
<span class="source-line-no">136</span><span id="line-136"></span>
<span class="source-line-no">137</span><span id="line-137"> assertEquals(</span>
<span class="source-line-no">138</span><span id="line-138"> "Even when the SCP is finished, the duplicate SCP should not be scheduled for "</span>
<span class="source-line-no">139</span><span id="line-139"> + SERVER_FOR_TEST,</span>
<span class="source-line-no">140</span><span id="line-140"> Procedure.NO_PROC_ID,</span>
<span class="source-line-no">141</span><span id="line-141"> UTIL.getHBaseCluster().getMaster().getServerManager().expireServer(SERVER_FOR_TEST));</span>
<span class="source-line-no">142</span><span id="line-142"></span>
<span class="source-line-no">143</span><span id="line-143"> MetricsMasterSource masterSource =</span>
<span class="source-line-no">144</span><span id="line-144"> UTIL.getHBaseCluster().getMaster().getMasterMetrics().getMetricsSource();</span>
<span class="source-line-no">145</span><span id="line-145"> metricsHelper.assertCounter(MetricsMasterSource.SERVER_CRASH_METRIC_PREFIX + "SubmittedCount",</span>
<span class="source-line-no">146</span><span id="line-146"> 3, masterSource);</span>
<span class="source-line-no">147</span><span id="line-147"> }</span>
<span class="source-line-no">148</span><span id="line-148"></span>
<span class="source-line-no">149</span><span id="line-149"> private void setupCluster() throws Exception {</span>
<span class="source-line-no">150</span><span id="line-150"> LOG.info("Setup cluster");</span>
<span class="source-line-no">151</span><span id="line-151"> UTIL.startMiniCluster(StartTestingClusterOption.builder().masterClass(HMasterForTest.class)</span>
<span class="source-line-no">152</span><span id="line-152"> .numMasters(1).numRegionServers(3).build());</span>
<span class="source-line-no">153</span><span id="line-153"> // this test has been flaky. When it is rerun by surefire, the underlying minicluster isn't</span>
<span class="source-line-no">154</span><span id="line-154"> // completely cleaned. specifically, the metrics system isn't reset. The result is an otherwise</span>
<span class="source-line-no">155</span><span id="line-155"> // successful re-run is failed because there's 8 or 12 SCPcounts instead of the 4 that a</span>
<span class="source-line-no">156</span><span id="line-156"> // single run of the test would otherwise produce. Thus, explicitly reset the metrics source</span>
<span class="source-line-no">157</span><span id="line-157"> // each time we setup the cluster.</span>
<span class="source-line-no">158</span><span id="line-158"> UTIL.getMiniHBaseCluster().getMaster().getMasterMetrics().getMetricsSource().init();</span>
<span class="source-line-no">159</span><span id="line-159"> LOG.info("Cluster is up");</span>
<span class="source-line-no">160</span><span id="line-160"> UTIL.waitFor(60000, () -&gt; UTIL.getMiniHBaseCluster().getMaster().isInitialized());</span>
<span class="source-line-no">161</span><span id="line-161"> LOG.info("Master is up");</span>
<span class="source-line-no">162</span><span id="line-162"> // wait for all SCPs finished</span>
<span class="source-line-no">163</span><span id="line-163"> UTIL.waitFor(60000, () -&gt; UTIL.getHBaseCluster().getMaster().getProcedures().stream()</span>
<span class="source-line-no">164</span><span id="line-164"> .noneMatch(p -&gt; p instanceof ServerCrashProcedure));</span>
<span class="source-line-no">165</span><span id="line-165"> LOG.info("No SCPs");</span>
<span class="source-line-no">166</span><span id="line-166"> }</span>
<span class="source-line-no">167</span><span id="line-167"></span>
<span class="source-line-no">168</span><span id="line-168"> private void setupTable() throws Exception {</span>
<span class="source-line-no">169</span><span id="line-169"> TableName tableName = TABLES[0];</span>
<span class="source-line-no">170</span><span id="line-170"> UTIL.createMultiRegionTable(tableName, FAMILY);</span>
<span class="source-line-no">171</span><span id="line-171"> UTIL.waitTableAvailable(tableName);</span>
<span class="source-line-no">172</span><span id="line-172"> Table table = UTIL.getConnection().getTable(tableName);</span>
<span class="source-line-no">173</span><span id="line-173"> for (int i = 0; i &lt; 100; i++) {</span>
<span class="source-line-no">174</span><span id="line-174"> UTIL.loadTable(table, FAMILY);</span>
<span class="source-line-no">175</span><span id="line-175"> }</span>
<span class="source-line-no">176</span><span id="line-176"> }</span>
<span class="source-line-no">177</span><span id="line-177"></span>
<span class="source-line-no">178</span><span id="line-178"> public static final class HMasterForTest extends HMaster {</span>
<span class="source-line-no">179</span><span id="line-179"></span>
<span class="source-line-no">180</span><span id="line-180"> public HMasterForTest(Configuration conf) throws IOException {</span>
<span class="source-line-no">181</span><span id="line-181"> super(conf);</span>
<span class="source-line-no">182</span><span id="line-182"> }</span>
<span class="source-line-no">183</span><span id="line-183"></span>
<span class="source-line-no">184</span><span id="line-184"> @Override</span>
<span class="source-line-no">185</span><span id="line-185"> protected AssignmentManager createAssignmentManager(MasterServices master,</span>
<span class="source-line-no">186</span><span id="line-186"> MasterRegion masterRegion) {</span>
<span class="source-line-no">187</span><span id="line-187"> return new AssignmentManagerForTest(master, masterRegion);</span>
<span class="source-line-no">188</span><span id="line-188"> }</span>
<span class="source-line-no">189</span><span id="line-189"> }</span>
<span class="source-line-no">190</span><span id="line-190"></span>
<span class="source-line-no">191</span><span id="line-191"> private static final class AssignmentManagerForTest extends AssignmentManager {</span>
<span class="source-line-no">192</span><span id="line-192"></span>
<span class="source-line-no">193</span><span id="line-193"> public AssignmentManagerForTest(MasterServices master, MasterRegion masterRegion) {</span>
<span class="source-line-no">194</span><span id="line-194"> super(master, masterRegion);</span>
<span class="source-line-no">195</span><span id="line-195"> }</span>
<span class="source-line-no">196</span><span id="line-196"></span>
<span class="source-line-no">197</span><span id="line-197"> @Override</span>
<span class="source-line-no">198</span><span id="line-198"> public List&lt;RegionInfo&gt; getRegionsOnServer(ServerName serverName) {</span>
<span class="source-line-no">199</span><span id="line-199"> List&lt;RegionInfo&gt; regions = super.getRegionsOnServer(serverName);</span>
<span class="source-line-no">200</span><span id="line-200"> // ServerCrashProcedure will call this method, so wait the CountDownLatch here</span>
<span class="source-line-no">201</span><span id="line-201"> if (SCP_LATCH != null &amp;&amp; SERVER_FOR_TEST != null &amp;&amp; serverName.equals(SERVER_FOR_TEST)) {</span>
<span class="source-line-no">202</span><span id="line-202"> try {</span>
<span class="source-line-no">203</span><span id="line-203"> LOG.info("ServerCrashProcedure wait the CountDownLatch here");</span>
<span class="source-line-no">204</span><span id="line-204"> SCP_LATCH.await();</span>
<span class="source-line-no">205</span><span id="line-205"> LOG.info("Continue the ServerCrashProcedure");</span>
<span class="source-line-no">206</span><span id="line-206"> SCP_LATCH = null;</span>
<span class="source-line-no">207</span><span id="line-207"> } catch (InterruptedException e) {</span>
<span class="source-line-no">208</span><span id="line-208"> throw new RuntimeException(e);</span>
<span class="source-line-no">209</span><span id="line-209"> }</span>
<span class="source-line-no">210</span><span id="line-210"> }</span>
<span class="source-line-no">211</span><span id="line-211"> return regions;</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>
</pre>
</div>
</main>
</body>
</html>