blob: 38e5546f62eb33b5fa79d2ee0272edd0e037ab15 [file] [log] [blame]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html lang="en">
<head>
<title>Source code</title>
<link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style">
</head>
<body>
<div class="sourceContainer">
<pre><span class="sourceLineNo">001</span>/*<a name="line.1"></a>
<span class="sourceLineNo">002</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a>
<span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a>
<span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a>
<span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a>
<span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a>
<span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a>
<span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a>
<span class="sourceLineNo">009</span> *<a name="line.9"></a>
<span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a>
<span class="sourceLineNo">011</span> *<a name="line.11"></a>
<span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a>
<span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a>
<span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a>
<span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a>
<span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a>
<span class="sourceLineNo">017</span> */<a name="line.17"></a>
<span class="sourceLineNo">018</span>package org.apache.hadoop.hbase.master;<a name="line.18"></a>
<span class="sourceLineNo">019</span><a name="line.19"></a>
<span class="sourceLineNo">020</span>import static org.junit.Assert.assertEquals;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import static org.junit.Assert.assertNotNull;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import static org.junit.Assert.assertNull;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import static org.junit.Assert.assertTrue;<a name="line.23"></a>
<span class="sourceLineNo">024</span><a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.io.IOException;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import java.util.List;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import java.util.Optional;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import java.util.concurrent.CountDownLatch;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import java.util.stream.Collectors;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.conf.Configuration;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.CompatibilityFactory;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.HBaseClassTestRule;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.ServerName;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.StartTestingClusterOption;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.TableName;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.Waiter.ExplainingPredicate;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.client.Table;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.master.assignment.AssignmentManager;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.master.assignment.ServerState;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.master.assignment.ServerStateNode;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.master.procedure.ServerCrashProcedure;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.master.region.MasterRegion;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.procedure2.Procedure;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.test.MetricsAssertHelper;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.testclassification.LargeTests;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.testclassification.MasterTests;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.junit.ClassRule;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.junit.Test;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.junit.experimental.categories.Category;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.slf4j.Logger;<a name="line.51"></a>
<span class="sourceLineNo">052</span>import org.slf4j.LoggerFactory;<a name="line.52"></a>
<span class="sourceLineNo">053</span><a name="line.53"></a>
<span class="sourceLineNo">054</span>@Category({ MasterTests.class, LargeTests.class })<a name="line.54"></a>
<span class="sourceLineNo">055</span>public class TestClusterRestartFailover extends AbstractTestRestartCluster {<a name="line.55"></a>
<span class="sourceLineNo">056</span><a name="line.56"></a>
<span class="sourceLineNo">057</span> @ClassRule<a name="line.57"></a>
<span class="sourceLineNo">058</span> public static final HBaseClassTestRule CLASS_RULE =<a name="line.58"></a>
<span class="sourceLineNo">059</span> HBaseClassTestRule.forClass(TestClusterRestartFailover.class);<a name="line.59"></a>
<span class="sourceLineNo">060</span><a name="line.60"></a>
<span class="sourceLineNo">061</span> private static final Logger LOG = LoggerFactory.getLogger(TestClusterRestartFailover.class);<a name="line.61"></a>
<span class="sourceLineNo">062</span> private static final MetricsAssertHelper metricsHelper =<a name="line.62"></a>
<span class="sourceLineNo">063</span> CompatibilityFactory.getInstance(MetricsAssertHelper.class);<a name="line.63"></a>
<span class="sourceLineNo">064</span><a name="line.64"></a>
<span class="sourceLineNo">065</span> private volatile static CountDownLatch SCP_LATCH;<a name="line.65"></a>
<span class="sourceLineNo">066</span> private static ServerName SERVER_FOR_TEST;<a name="line.66"></a>
<span class="sourceLineNo">067</span><a name="line.67"></a>
<span class="sourceLineNo">068</span> @Override<a name="line.68"></a>
<span class="sourceLineNo">069</span> protected boolean splitWALCoordinatedByZk() {<a name="line.69"></a>
<span class="sourceLineNo">070</span> return true;<a name="line.70"></a>
<span class="sourceLineNo">071</span> }<a name="line.71"></a>
<span class="sourceLineNo">072</span><a name="line.72"></a>
<span class="sourceLineNo">073</span> private ServerStateNode getServerStateNode(ServerName serverName) {<a name="line.73"></a>
<span class="sourceLineNo">074</span> return UTIL.getHBaseCluster().getMaster().getAssignmentManager().getRegionStates()<a name="line.74"></a>
<span class="sourceLineNo">075</span> .getServerNode(serverName);<a name="line.75"></a>
<span class="sourceLineNo">076</span> }<a name="line.76"></a>
<span class="sourceLineNo">077</span><a name="line.77"></a>
<span class="sourceLineNo">078</span> /**<a name="line.78"></a>
<span class="sourceLineNo">079</span> * Test for HBASE-22964<a name="line.79"></a>
<span class="sourceLineNo">080</span> */<a name="line.80"></a>
<span class="sourceLineNo">081</span> @Test<a name="line.81"></a>
<span class="sourceLineNo">082</span> public void test() throws Exception {<a name="line.82"></a>
<span class="sourceLineNo">083</span> setupCluster();<a name="line.83"></a>
<span class="sourceLineNo">084</span> setupTable();<a name="line.84"></a>
<span class="sourceLineNo">085</span><a name="line.85"></a>
<span class="sourceLineNo">086</span> SERVER_FOR_TEST = UTIL.getHBaseCluster().getRegionServer(0).getServerName();<a name="line.86"></a>
<span class="sourceLineNo">087</span> UTIL.waitFor(60000, () -&gt; getServerStateNode(SERVER_FOR_TEST) != null);<a name="line.87"></a>
<span class="sourceLineNo">088</span> ServerStateNode serverNode = getServerStateNode(SERVER_FOR_TEST);<a name="line.88"></a>
<span class="sourceLineNo">089</span> assertNotNull(serverNode);<a name="line.89"></a>
<span class="sourceLineNo">090</span> assertTrue("serverNode should be ONLINE when cluster runs normally",<a name="line.90"></a>
<span class="sourceLineNo">091</span> serverNode.isInState(ServerState.ONLINE));<a name="line.91"></a>
<span class="sourceLineNo">092</span><a name="line.92"></a>
<span class="sourceLineNo">093</span> SCP_LATCH = new CountDownLatch(1);<a name="line.93"></a>
<span class="sourceLineNo">094</span><a name="line.94"></a>
<span class="sourceLineNo">095</span> // Shutdown cluster and restart<a name="line.95"></a>
<span class="sourceLineNo">096</span> List&lt;Integer&gt; ports =<a name="line.96"></a>
<span class="sourceLineNo">097</span> UTIL.getHBaseCluster().getMaster().getServerManager().getOnlineServersList().stream()<a name="line.97"></a>
<span class="sourceLineNo">098</span> .map(serverName -&gt; serverName.getPort()).collect(Collectors.toList());<a name="line.98"></a>
<span class="sourceLineNo">099</span> LOG.info("Shutting down cluster");<a name="line.99"></a>
<span class="sourceLineNo">100</span> UTIL.getHBaseCluster().killAll();<a name="line.100"></a>
<span class="sourceLineNo">101</span> UTIL.getHBaseCluster().waitUntilShutDown();<a name="line.101"></a>
<span class="sourceLineNo">102</span> LOG.info("Restarting cluster");<a name="line.102"></a>
<span class="sourceLineNo">103</span> UTIL.restartHBaseCluster(StartTestingClusterOption.builder().masterClass(HMasterForTest.class)<a name="line.103"></a>
<span class="sourceLineNo">104</span> .numMasters(1).numRegionServers(3).rsPorts(ports).build());<a name="line.104"></a>
<span class="sourceLineNo">105</span> LOG.info("Started cluster");<a name="line.105"></a>
<span class="sourceLineNo">106</span> UTIL.waitFor(60000, () -&gt; UTIL.getHBaseCluster().getMaster().isInitialized());<a name="line.106"></a>
<span class="sourceLineNo">107</span> LOG.info("Started cluster master, waiting for {}", SERVER_FOR_TEST);<a name="line.107"></a>
<span class="sourceLineNo">108</span> UTIL.waitFor(60000, () -&gt; getServerStateNode(SERVER_FOR_TEST) != null);<a name="line.108"></a>
<span class="sourceLineNo">109</span> UTIL.waitFor(30000, new ExplainingPredicate&lt;Exception&gt;() {<a name="line.109"></a>
<span class="sourceLineNo">110</span><a name="line.110"></a>
<span class="sourceLineNo">111</span> @Override<a name="line.111"></a>
<span class="sourceLineNo">112</span> public boolean evaluate() throws Exception {<a name="line.112"></a>
<span class="sourceLineNo">113</span> return !getServerStateNode(SERVER_FOR_TEST).isInState(ServerState.ONLINE);<a name="line.113"></a>
<span class="sourceLineNo">114</span> }<a name="line.114"></a>
<span class="sourceLineNo">115</span><a name="line.115"></a>
<span class="sourceLineNo">116</span> @Override<a name="line.116"></a>
<span class="sourceLineNo">117</span> public String explainFailure() throws Exception {<a name="line.117"></a>
<span class="sourceLineNo">118</span> return "serverNode should not be ONLINE during SCP processing";<a name="line.118"></a>
<span class="sourceLineNo">119</span> }<a name="line.119"></a>
<span class="sourceLineNo">120</span> });<a name="line.120"></a>
<span class="sourceLineNo">121</span> Optional&lt;Procedure&lt;?&gt;&gt; procedure = UTIL.getHBaseCluster().getMaster().getProcedures().stream()<a name="line.121"></a>
<span class="sourceLineNo">122</span> .filter(p -&gt; (p instanceof ServerCrashProcedure)<a name="line.122"></a>
<span class="sourceLineNo">123</span> &amp;&amp; ((ServerCrashProcedure) p).getServerName().equals(SERVER_FOR_TEST))<a name="line.123"></a>
<span class="sourceLineNo">124</span> .findAny();<a name="line.124"></a>
<span class="sourceLineNo">125</span> assertTrue("Should have one SCP for " + SERVER_FOR_TEST, procedure.isPresent());<a name="line.125"></a>
<span class="sourceLineNo">126</span> assertEquals("Submit the SCP for the same serverName " + SERVER_FOR_TEST + " which should fail",<a name="line.126"></a>
<span class="sourceLineNo">127</span> Procedure.NO_PROC_ID,<a name="line.127"></a>
<span class="sourceLineNo">128</span> UTIL.getHBaseCluster().getMaster().getServerManager().expireServer(SERVER_FOR_TEST));<a name="line.128"></a>
<span class="sourceLineNo">129</span><a name="line.129"></a>
<span class="sourceLineNo">130</span> // Wait the SCP to finish<a name="line.130"></a>
<span class="sourceLineNo">131</span> LOG.info("Waiting on latch");<a name="line.131"></a>
<span class="sourceLineNo">132</span> SCP_LATCH.countDown();<a name="line.132"></a>
<span class="sourceLineNo">133</span> UTIL.waitFor(60000, () -&gt; procedure.get().isFinished());<a name="line.133"></a>
<span class="sourceLineNo">134</span> assertNull("serverNode should be deleted after SCP finished",<a name="line.134"></a>
<span class="sourceLineNo">135</span> getServerStateNode(SERVER_FOR_TEST));<a name="line.135"></a>
<span class="sourceLineNo">136</span><a name="line.136"></a>
<span class="sourceLineNo">137</span> assertEquals(<a name="line.137"></a>
<span class="sourceLineNo">138</span> "Even when the SCP is finished, the duplicate SCP should not be scheduled for "<a name="line.138"></a>
<span class="sourceLineNo">139</span> + SERVER_FOR_TEST,<a name="line.139"></a>
<span class="sourceLineNo">140</span> Procedure.NO_PROC_ID,<a name="line.140"></a>
<span class="sourceLineNo">141</span> UTIL.getHBaseCluster().getMaster().getServerManager().expireServer(SERVER_FOR_TEST));<a name="line.141"></a>
<span class="sourceLineNo">142</span><a name="line.142"></a>
<span class="sourceLineNo">143</span> MetricsMasterSource masterSource =<a name="line.143"></a>
<span class="sourceLineNo">144</span> UTIL.getHBaseCluster().getMaster().getMasterMetrics().getMetricsSource();<a name="line.144"></a>
<span class="sourceLineNo">145</span> metricsHelper.assertCounter(MetricsMasterSource.SERVER_CRASH_METRIC_PREFIX + "SubmittedCount",<a name="line.145"></a>
<span class="sourceLineNo">146</span> 3, masterSource);<a name="line.146"></a>
<span class="sourceLineNo">147</span> }<a name="line.147"></a>
<span class="sourceLineNo">148</span><a name="line.148"></a>
<span class="sourceLineNo">149</span> private void setupCluster() throws Exception {<a name="line.149"></a>
<span class="sourceLineNo">150</span> LOG.info("Setup cluster");<a name="line.150"></a>
<span class="sourceLineNo">151</span> UTIL.startMiniCluster(StartTestingClusterOption.builder().masterClass(HMasterForTest.class)<a name="line.151"></a>
<span class="sourceLineNo">152</span> .numMasters(1).numRegionServers(3).build());<a name="line.152"></a>
<span class="sourceLineNo">153</span> // this test has been flaky. When it is rerun by surefire, the underlying minicluster isn't<a name="line.153"></a>
<span class="sourceLineNo">154</span> // completely cleaned. specifically, the metrics system isn't reset. The result is an otherwise<a name="line.154"></a>
<span class="sourceLineNo">155</span> // successful re-run is failed because there's 8 or 12 SCPcounts instead of the 4 that a<a name="line.155"></a>
<span class="sourceLineNo">156</span> // single run of the test would otherwise produce. Thus, explicitly reset the metrics source<a name="line.156"></a>
<span class="sourceLineNo">157</span> // each time we setup the cluster.<a name="line.157"></a>
<span class="sourceLineNo">158</span> UTIL.getMiniHBaseCluster().getMaster().getMasterMetrics().getMetricsSource().init();<a name="line.158"></a>
<span class="sourceLineNo">159</span> LOG.info("Cluster is up");<a name="line.159"></a>
<span class="sourceLineNo">160</span> UTIL.waitFor(60000, () -&gt; UTIL.getMiniHBaseCluster().getMaster().isInitialized());<a name="line.160"></a>
<span class="sourceLineNo">161</span> LOG.info("Master is up");<a name="line.161"></a>
<span class="sourceLineNo">162</span> // wait for all SCPs finished<a name="line.162"></a>
<span class="sourceLineNo">163</span> UTIL.waitFor(60000, () -&gt; UTIL.getHBaseCluster().getMaster().getProcedures().stream()<a name="line.163"></a>
<span class="sourceLineNo">164</span> .noneMatch(p -&gt; p instanceof ServerCrashProcedure));<a name="line.164"></a>
<span class="sourceLineNo">165</span> LOG.info("No SCPs");<a name="line.165"></a>
<span class="sourceLineNo">166</span> }<a name="line.166"></a>
<span class="sourceLineNo">167</span><a name="line.167"></a>
<span class="sourceLineNo">168</span> private void setupTable() throws Exception {<a name="line.168"></a>
<span class="sourceLineNo">169</span> TableName tableName = TABLES[0];<a name="line.169"></a>
<span class="sourceLineNo">170</span> UTIL.createMultiRegionTable(tableName, FAMILY);<a name="line.170"></a>
<span class="sourceLineNo">171</span> UTIL.waitTableAvailable(tableName);<a name="line.171"></a>
<span class="sourceLineNo">172</span> Table table = UTIL.getConnection().getTable(tableName);<a name="line.172"></a>
<span class="sourceLineNo">173</span> for (int i = 0; i &lt; 100; i++) {<a name="line.173"></a>
<span class="sourceLineNo">174</span> UTIL.loadTable(table, FAMILY);<a name="line.174"></a>
<span class="sourceLineNo">175</span> }<a name="line.175"></a>
<span class="sourceLineNo">176</span> }<a name="line.176"></a>
<span class="sourceLineNo">177</span><a name="line.177"></a>
<span class="sourceLineNo">178</span> public static final class HMasterForTest extends HMaster {<a name="line.178"></a>
<span class="sourceLineNo">179</span><a name="line.179"></a>
<span class="sourceLineNo">180</span> public HMasterForTest(Configuration conf) throws IOException {<a name="line.180"></a>
<span class="sourceLineNo">181</span> super(conf);<a name="line.181"></a>
<span class="sourceLineNo">182</span> }<a name="line.182"></a>
<span class="sourceLineNo">183</span><a name="line.183"></a>
<span class="sourceLineNo">184</span> @Override<a name="line.184"></a>
<span class="sourceLineNo">185</span> protected AssignmentManager createAssignmentManager(MasterServices master,<a name="line.185"></a>
<span class="sourceLineNo">186</span> MasterRegion masterRegion) {<a name="line.186"></a>
<span class="sourceLineNo">187</span> return new AssignmentManagerForTest(master, masterRegion);<a name="line.187"></a>
<span class="sourceLineNo">188</span> }<a name="line.188"></a>
<span class="sourceLineNo">189</span> }<a name="line.189"></a>
<span class="sourceLineNo">190</span><a name="line.190"></a>
<span class="sourceLineNo">191</span> private static final class AssignmentManagerForTest extends AssignmentManager {<a name="line.191"></a>
<span class="sourceLineNo">192</span><a name="line.192"></a>
<span class="sourceLineNo">193</span> public AssignmentManagerForTest(MasterServices master, MasterRegion masterRegion) {<a name="line.193"></a>
<span class="sourceLineNo">194</span> super(master, masterRegion);<a name="line.194"></a>
<span class="sourceLineNo">195</span> }<a name="line.195"></a>
<span class="sourceLineNo">196</span><a name="line.196"></a>
<span class="sourceLineNo">197</span> @Override<a name="line.197"></a>
<span class="sourceLineNo">198</span> public List&lt;RegionInfo&gt; getRegionsOnServer(ServerName serverName) {<a name="line.198"></a>
<span class="sourceLineNo">199</span> List&lt;RegionInfo&gt; regions = super.getRegionsOnServer(serverName);<a name="line.199"></a>
<span class="sourceLineNo">200</span> // ServerCrashProcedure will call this method, so wait the CountDownLatch here<a name="line.200"></a>
<span class="sourceLineNo">201</span> if (SCP_LATCH != null &amp;&amp; SERVER_FOR_TEST != null &amp;&amp; serverName.equals(SERVER_FOR_TEST)) {<a name="line.201"></a>
<span class="sourceLineNo">202</span> try {<a name="line.202"></a>
<span class="sourceLineNo">203</span> LOG.info("ServerCrashProcedure wait the CountDownLatch here");<a name="line.203"></a>
<span class="sourceLineNo">204</span> SCP_LATCH.await();<a name="line.204"></a>
<span class="sourceLineNo">205</span> LOG.info("Continue the ServerCrashProcedure");<a name="line.205"></a>
<span class="sourceLineNo">206</span> SCP_LATCH = null;<a name="line.206"></a>
<span class="sourceLineNo">207</span> } catch (InterruptedException e) {<a name="line.207"></a>
<span class="sourceLineNo">208</span> throw new RuntimeException(e);<a name="line.208"></a>
<span class="sourceLineNo">209</span> }<a name="line.209"></a>
<span class="sourceLineNo">210</span> }<a name="line.210"></a>
<span class="sourceLineNo">211</span> return regions;<a name="line.211"></a>
<span class="sourceLineNo">212</span> }<a name="line.212"></a>
<span class="sourceLineNo">213</span> }<a name="line.213"></a>
<span class="sourceLineNo">214</span>}<a name="line.214"></a>
</pre>
</div>
</body>
</html>