blob: e2d28ac96f138cda043b582e75e5792672cf9707 [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.assertTrue;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.util.List;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.concurrent.TimeUnit;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import org.apache.hadoop.hbase.ClusterMetrics;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.HBaseClassTestRule;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.HBaseTestingUtility;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.MiniHBaseCluster;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.ServerName;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.StartMiniClusterOption;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.master.RegionState.State;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.testclassification.FlakeyTests;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.testclassification.LargeTests;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.util.JVMClusterUtil.MasterThread;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.zookeeper.MetaTableLocator;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.junit.ClassRule;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.junit.Rule;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.junit.Test;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.junit.experimental.categories.Category;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.junit.rules.TestName;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.slf4j.Logger;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.slf4j.LoggerFactory;<a name="line.43"></a>
<span class="sourceLineNo">044</span><a name="line.44"></a>
<span class="sourceLineNo">045</span>@Category({FlakeyTests.class, LargeTests.class})<a name="line.45"></a>
<span class="sourceLineNo">046</span>public class TestMasterFailover {<a name="line.46"></a>
<span class="sourceLineNo">047</span><a name="line.47"></a>
<span class="sourceLineNo">048</span> @ClassRule<a name="line.48"></a>
<span class="sourceLineNo">049</span> public static final HBaseClassTestRule CLASS_RULE =<a name="line.49"></a>
<span class="sourceLineNo">050</span> HBaseClassTestRule.forClass(TestMasterFailover.class);<a name="line.50"></a>
<span class="sourceLineNo">051</span><a name="line.51"></a>
<span class="sourceLineNo">052</span> private static final Logger LOG = LoggerFactory.getLogger(TestMasterFailover.class);<a name="line.52"></a>
<span class="sourceLineNo">053</span> @Rule public TestName name = new TestName();<a name="line.53"></a>
<span class="sourceLineNo">054</span><a name="line.54"></a>
<span class="sourceLineNo">055</span> /**<a name="line.55"></a>
<span class="sourceLineNo">056</span> * Simple test of master failover.<a name="line.56"></a>
<span class="sourceLineNo">057</span> * &lt;p&gt;<a name="line.57"></a>
<span class="sourceLineNo">058</span> * Starts with three masters. Kills a backup master. Then kills the active<a name="line.58"></a>
<span class="sourceLineNo">059</span> * master. Ensures the final master becomes active and we can still contact<a name="line.59"></a>
<span class="sourceLineNo">060</span> * the cluster.<a name="line.60"></a>
<span class="sourceLineNo">061</span> */<a name="line.61"></a>
<span class="sourceLineNo">062</span> @Test<a name="line.62"></a>
<span class="sourceLineNo">063</span> public void testSimpleMasterFailover() throws Exception {<a name="line.63"></a>
<span class="sourceLineNo">064</span> final int NUM_MASTERS = 3;<a name="line.64"></a>
<span class="sourceLineNo">065</span> final int NUM_RS = 3;<a name="line.65"></a>
<span class="sourceLineNo">066</span><a name="line.66"></a>
<span class="sourceLineNo">067</span> // Start the cluster<a name="line.67"></a>
<span class="sourceLineNo">068</span> HBaseTestingUtility TEST_UTIL = new HBaseTestingUtility();<a name="line.68"></a>
<span class="sourceLineNo">069</span> try {<a name="line.69"></a>
<span class="sourceLineNo">070</span> StartMiniClusterOption option = StartMiniClusterOption.builder()<a name="line.70"></a>
<span class="sourceLineNo">071</span> .numMasters(NUM_MASTERS).numRegionServers(NUM_RS).numDataNodes(NUM_RS).build();<a name="line.71"></a>
<span class="sourceLineNo">072</span> TEST_UTIL.startMiniCluster(option);<a name="line.72"></a>
<span class="sourceLineNo">073</span> MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();<a name="line.73"></a>
<span class="sourceLineNo">074</span><a name="line.74"></a>
<span class="sourceLineNo">075</span> // get all the master threads<a name="line.75"></a>
<span class="sourceLineNo">076</span> List&lt;MasterThread&gt; masterThreads = cluster.getMasterThreads();<a name="line.76"></a>
<span class="sourceLineNo">077</span><a name="line.77"></a>
<span class="sourceLineNo">078</span> // wait for each to come online<a name="line.78"></a>
<span class="sourceLineNo">079</span> for (MasterThread mt : masterThreads) {<a name="line.79"></a>
<span class="sourceLineNo">080</span> assertTrue(mt.isAlive());<a name="line.80"></a>
<span class="sourceLineNo">081</span> }<a name="line.81"></a>
<span class="sourceLineNo">082</span><a name="line.82"></a>
<span class="sourceLineNo">083</span> // verify only one is the active master and we have right number<a name="line.83"></a>
<span class="sourceLineNo">084</span> int numActive = 0;<a name="line.84"></a>
<span class="sourceLineNo">085</span> int activeIndex = -1;<a name="line.85"></a>
<span class="sourceLineNo">086</span> ServerName activeName = null;<a name="line.86"></a>
<span class="sourceLineNo">087</span> HMaster active = null;<a name="line.87"></a>
<span class="sourceLineNo">088</span> for (int i = 0; i &lt; masterThreads.size(); i++) {<a name="line.88"></a>
<span class="sourceLineNo">089</span> if (masterThreads.get(i).getMaster().isActiveMaster()) {<a name="line.89"></a>
<span class="sourceLineNo">090</span> numActive++;<a name="line.90"></a>
<span class="sourceLineNo">091</span> activeIndex = i;<a name="line.91"></a>
<span class="sourceLineNo">092</span> active = masterThreads.get(activeIndex).getMaster();<a name="line.92"></a>
<span class="sourceLineNo">093</span> activeName = active.getServerName();<a name="line.93"></a>
<span class="sourceLineNo">094</span> }<a name="line.94"></a>
<span class="sourceLineNo">095</span> }<a name="line.95"></a>
<span class="sourceLineNo">096</span> assertEquals(1, numActive);<a name="line.96"></a>
<span class="sourceLineNo">097</span> assertEquals(NUM_MASTERS, masterThreads.size());<a name="line.97"></a>
<span class="sourceLineNo">098</span> LOG.info("Active master " + activeName);<a name="line.98"></a>
<span class="sourceLineNo">099</span><a name="line.99"></a>
<span class="sourceLineNo">100</span> // Check that ClusterStatus reports the correct active and backup masters<a name="line.100"></a>
<span class="sourceLineNo">101</span> assertNotNull(active);<a name="line.101"></a>
<span class="sourceLineNo">102</span> ClusterMetrics status = active.getClusterMetrics();<a name="line.102"></a>
<span class="sourceLineNo">103</span> assertEquals(activeName, status.getMasterName());<a name="line.103"></a>
<span class="sourceLineNo">104</span> assertEquals(2, status.getBackupMasterNames().size());<a name="line.104"></a>
<span class="sourceLineNo">105</span><a name="line.105"></a>
<span class="sourceLineNo">106</span> // attempt to stop one of the inactive masters<a name="line.106"></a>
<span class="sourceLineNo">107</span> int backupIndex = (activeIndex == 0 ? 1 : activeIndex - 1);<a name="line.107"></a>
<span class="sourceLineNo">108</span> HMaster master = cluster.getMaster(backupIndex);<a name="line.108"></a>
<span class="sourceLineNo">109</span> LOG.debug("\n\nStopping a backup master: " + master.getServerName() + "\n");<a name="line.109"></a>
<span class="sourceLineNo">110</span> cluster.stopMaster(backupIndex, false);<a name="line.110"></a>
<span class="sourceLineNo">111</span> cluster.waitOnMaster(backupIndex);<a name="line.111"></a>
<span class="sourceLineNo">112</span><a name="line.112"></a>
<span class="sourceLineNo">113</span> // Verify still one active master and it's the same<a name="line.113"></a>
<span class="sourceLineNo">114</span> for (int i = 0; i &lt; masterThreads.size(); i++) {<a name="line.114"></a>
<span class="sourceLineNo">115</span> if (masterThreads.get(i).getMaster().isActiveMaster()) {<a name="line.115"></a>
<span class="sourceLineNo">116</span> assertEquals(activeName, masterThreads.get(i).getMaster().getServerName());<a name="line.116"></a>
<span class="sourceLineNo">117</span> activeIndex = i;<a name="line.117"></a>
<span class="sourceLineNo">118</span> active = masterThreads.get(activeIndex).getMaster();<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> assertEquals(1, numActive);<a name="line.121"></a>
<span class="sourceLineNo">122</span> assertEquals(2, masterThreads.size());<a name="line.122"></a>
<span class="sourceLineNo">123</span> int rsCount = masterThreads.get(activeIndex).getMaster().getClusterMetrics()<a name="line.123"></a>
<span class="sourceLineNo">124</span> .getLiveServerMetrics().size();<a name="line.124"></a>
<span class="sourceLineNo">125</span> LOG.info("Active master " + active.getServerName() + " managing " + rsCount +<a name="line.125"></a>
<span class="sourceLineNo">126</span> " regions servers");<a name="line.126"></a>
<span class="sourceLineNo">127</span> assertEquals(3, rsCount);<a name="line.127"></a>
<span class="sourceLineNo">128</span><a name="line.128"></a>
<span class="sourceLineNo">129</span> // wait for the active master to acknowledge loss of the backup from ZK<a name="line.129"></a>
<span class="sourceLineNo">130</span> final HMaster activeFinal = active;<a name="line.130"></a>
<span class="sourceLineNo">131</span> TEST_UTIL.waitFor(<a name="line.131"></a>
<span class="sourceLineNo">132</span> TimeUnit.MINUTES.toMillis(5), () -&gt; activeFinal.getBackupMasters().size() == 1);<a name="line.132"></a>
<span class="sourceLineNo">133</span><a name="line.133"></a>
<span class="sourceLineNo">134</span> // Check that ClusterStatus reports the correct active and backup masters<a name="line.134"></a>
<span class="sourceLineNo">135</span> assertNotNull(active);<a name="line.135"></a>
<span class="sourceLineNo">136</span> status = active.getClusterMetrics();<a name="line.136"></a>
<span class="sourceLineNo">137</span> assertEquals(activeName, status.getMasterName());<a name="line.137"></a>
<span class="sourceLineNo">138</span> assertEquals(1, status.getBackupMasterNames().size());<a name="line.138"></a>
<span class="sourceLineNo">139</span><a name="line.139"></a>
<span class="sourceLineNo">140</span> // kill the active master<a name="line.140"></a>
<span class="sourceLineNo">141</span> LOG.debug("\n\nStopping the active master " + active.getServerName() + "\n");<a name="line.141"></a>
<span class="sourceLineNo">142</span> cluster.stopMaster(activeIndex, false);<a name="line.142"></a>
<span class="sourceLineNo">143</span> cluster.waitOnMaster(activeIndex);<a name="line.143"></a>
<span class="sourceLineNo">144</span><a name="line.144"></a>
<span class="sourceLineNo">145</span> // wait for an active master to show up and be ready<a name="line.145"></a>
<span class="sourceLineNo">146</span> assertTrue(cluster.waitForActiveAndReadyMaster());<a name="line.146"></a>
<span class="sourceLineNo">147</span><a name="line.147"></a>
<span class="sourceLineNo">148</span> LOG.debug("\n\nVerifying backup master is now active\n");<a name="line.148"></a>
<span class="sourceLineNo">149</span> // should only have one master now<a name="line.149"></a>
<span class="sourceLineNo">150</span> assertEquals(1, masterThreads.size());<a name="line.150"></a>
<span class="sourceLineNo">151</span><a name="line.151"></a>
<span class="sourceLineNo">152</span> // and he should be active<a name="line.152"></a>
<span class="sourceLineNo">153</span> active = masterThreads.get(0).getMaster();<a name="line.153"></a>
<span class="sourceLineNo">154</span> assertNotNull(active);<a name="line.154"></a>
<span class="sourceLineNo">155</span> status = active.getClusterMetrics();<a name="line.155"></a>
<span class="sourceLineNo">156</span> ServerName masterName = status.getMasterName();<a name="line.156"></a>
<span class="sourceLineNo">157</span> assertNotNull(masterName);<a name="line.157"></a>
<span class="sourceLineNo">158</span> assertEquals(active.getServerName(), masterName);<a name="line.158"></a>
<span class="sourceLineNo">159</span> assertTrue(active.isActiveMaster());<a name="line.159"></a>
<span class="sourceLineNo">160</span> assertEquals(0, status.getBackupMasterNames().size());<a name="line.160"></a>
<span class="sourceLineNo">161</span> int rss = status.getLiveServerMetrics().size();<a name="line.161"></a>
<span class="sourceLineNo">162</span> LOG.info("Active master {} managing {} region servers", masterName.getServerName(), rss);<a name="line.162"></a>
<span class="sourceLineNo">163</span> assertEquals(3, rss);<a name="line.163"></a>
<span class="sourceLineNo">164</span> } finally {<a name="line.164"></a>
<span class="sourceLineNo">165</span> // Stop the cluster<a name="line.165"></a>
<span class="sourceLineNo">166</span> TEST_UTIL.shutdownMiniCluster();<a name="line.166"></a>
<span class="sourceLineNo">167</span> }<a name="line.167"></a>
<span class="sourceLineNo">168</span> }<a name="line.168"></a>
<span class="sourceLineNo">169</span><a name="line.169"></a>
<span class="sourceLineNo">170</span> /**<a name="line.170"></a>
<span class="sourceLineNo">171</span> * Test meta in transition when master failover.<a name="line.171"></a>
<span class="sourceLineNo">172</span> * This test used to manipulate region state up in zk. That is not allowed any more in hbase2<a name="line.172"></a>
<span class="sourceLineNo">173</span> * so I removed that messing. That makes this test anemic.<a name="line.173"></a>
<span class="sourceLineNo">174</span> */<a name="line.174"></a>
<span class="sourceLineNo">175</span> @Test<a name="line.175"></a>
<span class="sourceLineNo">176</span> public void testMetaInTransitionWhenMasterFailover() throws Exception {<a name="line.176"></a>
<span class="sourceLineNo">177</span> // Start the cluster<a name="line.177"></a>
<span class="sourceLineNo">178</span> HBaseTestingUtility TEST_UTIL = new HBaseTestingUtility();<a name="line.178"></a>
<span class="sourceLineNo">179</span> TEST_UTIL.startMiniCluster();<a name="line.179"></a>
<span class="sourceLineNo">180</span> try {<a name="line.180"></a>
<span class="sourceLineNo">181</span> MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();<a name="line.181"></a>
<span class="sourceLineNo">182</span> LOG.info("Cluster started");<a name="line.182"></a>
<span class="sourceLineNo">183</span><a name="line.183"></a>
<span class="sourceLineNo">184</span> HMaster activeMaster = cluster.getMaster();<a name="line.184"></a>
<span class="sourceLineNo">185</span> ServerName metaServerName = cluster.getServerHoldingMeta();<a name="line.185"></a>
<span class="sourceLineNo">186</span> HRegionServer hrs = cluster.getRegionServer(metaServerName);<a name="line.186"></a>
<span class="sourceLineNo">187</span><a name="line.187"></a>
<span class="sourceLineNo">188</span> // Now kill master, meta should remain on rs, where we placed it before.<a name="line.188"></a>
<span class="sourceLineNo">189</span> LOG.info("Aborting master");<a name="line.189"></a>
<span class="sourceLineNo">190</span> activeMaster.abort("test-kill");<a name="line.190"></a>
<span class="sourceLineNo">191</span> cluster.waitForMasterToStop(activeMaster.getServerName(), 30000);<a name="line.191"></a>
<span class="sourceLineNo">192</span> LOG.info("Master has aborted");<a name="line.192"></a>
<span class="sourceLineNo">193</span><a name="line.193"></a>
<span class="sourceLineNo">194</span> // meta should remain where it was<a name="line.194"></a>
<span class="sourceLineNo">195</span> RegionState metaState = MetaTableLocator.getMetaRegionState(hrs.getZooKeeper());<a name="line.195"></a>
<span class="sourceLineNo">196</span> assertEquals("hbase:meta should be online on RS",<a name="line.196"></a>
<span class="sourceLineNo">197</span> metaState.getServerName(), metaServerName);<a name="line.197"></a>
<span class="sourceLineNo">198</span> assertEquals("hbase:meta should be online on RS", State.OPEN, metaState.getState());<a name="line.198"></a>
<span class="sourceLineNo">199</span><a name="line.199"></a>
<span class="sourceLineNo">200</span> // Start up a new master<a name="line.200"></a>
<span class="sourceLineNo">201</span> LOG.info("Starting up a new master");<a name="line.201"></a>
<span class="sourceLineNo">202</span> activeMaster = cluster.startMaster().getMaster();<a name="line.202"></a>
<span class="sourceLineNo">203</span> LOG.info("Waiting for master to be ready");<a name="line.203"></a>
<span class="sourceLineNo">204</span> cluster.waitForActiveAndReadyMaster();<a name="line.204"></a>
<span class="sourceLineNo">205</span> LOG.info("Master is ready");<a name="line.205"></a>
<span class="sourceLineNo">206</span><a name="line.206"></a>
<span class="sourceLineNo">207</span> // ensure meta is still deployed on RS<a name="line.207"></a>
<span class="sourceLineNo">208</span> metaState = MetaTableLocator.getMetaRegionState(activeMaster.getZooKeeper());<a name="line.208"></a>
<span class="sourceLineNo">209</span> assertEquals("hbase:meta should be online on RS",<a name="line.209"></a>
<span class="sourceLineNo">210</span> metaState.getServerName(), metaServerName);<a name="line.210"></a>
<span class="sourceLineNo">211</span> assertEquals("hbase:meta should be online on RS", State.OPEN, metaState.getState());<a name="line.211"></a>
<span class="sourceLineNo">212</span><a name="line.212"></a>
<span class="sourceLineNo">213</span> // Done, shutdown the cluster<a name="line.213"></a>
<span class="sourceLineNo">214</span> } finally {<a name="line.214"></a>
<span class="sourceLineNo">215</span> TEST_UTIL.shutdownMiniCluster();<a name="line.215"></a>
<span class="sourceLineNo">216</span> }<a name="line.216"></a>
<span class="sourceLineNo">217</span> }<a name="line.217"></a>
<span class="sourceLineNo">218</span>}<a name="line.218"></a>
<span class="sourceLineNo">219</span><a name="line.219"></a>
</pre>
</div>
</body>
</html>