Google Git
Sign in
apache / hbase-site / 49c03bf03a4b054096f09d00d923c8010abd2c2a / . / testdevapidocs / src-html / org / apache / hadoop / hbase / master / TestMasterTransitions.html
blob: 31a362f54bc10c6d50f8ffaa8455c73f756365c8 [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 java.io.IOException;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import org.apache.hadoop.hbase.CatalogFamilyFormat;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import org.apache.hadoop.hbase.HBaseClassTestRule;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import org.apache.hadoop.hbase.HBaseTestingUtil;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import org.apache.hadoop.hbase.HConstants;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import org.apache.hadoop.hbase.TableName;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.client.Durability;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.client.Put;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.client.RegionLocator;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.client.Result;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.client.ResultScanner;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.client.Scan;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.client.Table;<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.testclassification.MasterTests;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.junit.AfterClass;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.junit.Assert;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.junit.Before;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.junit.BeforeClass;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.junit.ClassRule;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.junit.Ignore;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.junit.Test;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.junit.experimental.categories.Category;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.slf4j.Logger;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.slf4j.LoggerFactory;<a name="line.46"></a>
<span class="sourceLineNo">047</span><a name="line.47"></a>
<span class="sourceLineNo">048</span>/**<a name="line.48"></a>
<span class="sourceLineNo">049</span> * Test transitions of state across the master. Sets up the cluster once and then runs a couple of<a name="line.49"></a>
<span class="sourceLineNo">050</span> * tests.<a name="line.50"></a>
<span class="sourceLineNo">051</span> */<a name="line.51"></a>
<span class="sourceLineNo">052</span>@Category({ MasterTests.class, LargeTests.class })<a name="line.52"></a>
<span class="sourceLineNo">053</span>public class TestMasterTransitions {<a name="line.53"></a>
<span class="sourceLineNo">054</span><a name="line.54"></a>
<span class="sourceLineNo">055</span> @ClassRule<a name="line.55"></a>
<span class="sourceLineNo">056</span> public static final HBaseClassTestRule CLASS_RULE =<a name="line.56"></a>
<span class="sourceLineNo">057</span> HBaseClassTestRule.forClass(TestMasterTransitions.class);<a name="line.57"></a>
<span class="sourceLineNo">058</span><a name="line.58"></a>
<span class="sourceLineNo">059</span> private static final Logger LOG = LoggerFactory.getLogger(TestMasterTransitions.class);<a name="line.59"></a>
<span class="sourceLineNo">060</span> private static final HBaseTestingUtil TEST_UTIL = new HBaseTestingUtil();<a name="line.60"></a>
<span class="sourceLineNo">061</span> private static final TableName TABLENAME = TableName.valueOf("master_transitions");<a name="line.61"></a>
<span class="sourceLineNo">062</span> private static final byte[][] FAMILIES =<a name="line.62"></a>
<span class="sourceLineNo">063</span> new byte[][] { Bytes.toBytes("a"), Bytes.toBytes("b"), Bytes.toBytes("c") };<a name="line.63"></a>
<span class="sourceLineNo">064</span><a name="line.64"></a>
<span class="sourceLineNo">065</span> /**<a name="line.65"></a>
<span class="sourceLineNo">066</span> * Start up a mini cluster and put a small table of many empty regions into it.<a name="line.66"></a>
<span class="sourceLineNo">067</span> */<a name="line.67"></a>
<span class="sourceLineNo">068</span> @BeforeClass<a name="line.68"></a>
<span class="sourceLineNo">069</span> public static void beforeAllTests() throws Exception {<a name="line.69"></a>
<span class="sourceLineNo">070</span> TEST_UTIL.startMiniCluster(2);<a name="line.70"></a>
<span class="sourceLineNo">071</span> // Create a table of three families. This will assign a region.<a name="line.71"></a>
<span class="sourceLineNo">072</span> TEST_UTIL.createMultiRegionTable(TABLENAME, FAMILIES);<a name="line.72"></a>
<span class="sourceLineNo">073</span> Table t = TEST_UTIL.getConnection().getTable(TABLENAME);<a name="line.73"></a>
<span class="sourceLineNo">074</span> int countOfRegions = -1;<a name="line.74"></a>
<span class="sourceLineNo">075</span> try (RegionLocator r = TEST_UTIL.getConnection().getRegionLocator(TABLENAME)) {<a name="line.75"></a>
<span class="sourceLineNo">076</span> countOfRegions = r.getStartKeys().length;<a name="line.76"></a>
<span class="sourceLineNo">077</span> }<a name="line.77"></a>
<span class="sourceLineNo">078</span> TEST_UTIL.waitUntilAllRegionsAssigned(TABLENAME);<a name="line.78"></a>
<span class="sourceLineNo">079</span> addToEachStartKey(countOfRegions);<a name="line.79"></a>
<span class="sourceLineNo">080</span> t.close();<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> @AfterClass<a name="line.83"></a>
<span class="sourceLineNo">084</span> public static void afterAllTests() throws Exception {<a name="line.84"></a>
<span class="sourceLineNo">085</span> TEST_UTIL.shutdownMiniCluster();<a name="line.85"></a>
<span class="sourceLineNo">086</span> }<a name="line.86"></a>
<span class="sourceLineNo">087</span><a name="line.87"></a>
<span class="sourceLineNo">088</span> @Before<a name="line.88"></a>
<span class="sourceLineNo">089</span> public void setup() throws IOException {<a name="line.89"></a>
<span class="sourceLineNo">090</span> TEST_UTIL.ensureSomeRegionServersAvailable(2);<a name="line.90"></a>
<span class="sourceLineNo">091</span> }<a name="line.91"></a>
<span class="sourceLineNo">092</span><a name="line.92"></a>
<span class="sourceLineNo">093</span> /**<a name="line.93"></a>
<span class="sourceLineNo">094</span> * Listener for regionserver events testing hbase-2428 (Infinite loop of region closes if<a name="line.94"></a>
<span class="sourceLineNo">095</span> * hbase:meta region is offline). In particular, listen for the close of the 'metaServer' and when<a name="line.95"></a>
<span class="sourceLineNo">096</span> * it comes in, requeue it with a delay as though there were an issue processing the shutdown. As<a name="line.96"></a>
<span class="sourceLineNo">097</span> * part of the requeuing, send over a close of a region on 'otherServer' so it comes into a master<a name="line.97"></a>
<span class="sourceLineNo">098</span> * that has its meta region marked as offline.<a name="line.98"></a>
<span class="sourceLineNo">099</span> */<a name="line.99"></a>
<span class="sourceLineNo">100</span> /*<a name="line.100"></a>
<span class="sourceLineNo">101</span> * static class HBase2428Listener implements RegionServerOperationListener { // Map of what we've<a name="line.101"></a>
<span class="sourceLineNo">102</span> * delayed so we don't do do repeated delays. private final Set&lt;RegionServerOperation&gt; postponed =<a name="line.102"></a>
<span class="sourceLineNo">103</span> * new CopyOnWriteArraySet&lt;RegionServerOperation&gt;(); private boolean done = false;; private<a name="line.103"></a>
<span class="sourceLineNo">104</span> * boolean metaShutdownReceived = false; private final HServerAddress metaAddress; private final<a name="line.104"></a>
<span class="sourceLineNo">105</span> * MiniHBaseCluster cluster; private final int otherServerIndex; private final RegionInfo hri;<a name="line.105"></a>
<span class="sourceLineNo">106</span> * private int closeCount = 0; static final int SERVER_DURATION = 3 * 1000; static final int<a name="line.106"></a>
<span class="sourceLineNo">107</span> * CLOSE_DURATION = 1 * 1000; HBase2428Listener(final MiniHBaseCluster c, final HServerAddress<a name="line.107"></a>
<span class="sourceLineNo">108</span> * metaAddress, final RegionInfo closingHRI, final int otherServerIndex) { this.cluster = c;<a name="line.108"></a>
<span class="sourceLineNo">109</span> * this.metaAddress = metaAddress; this.hri = closingHRI; this.otherServerIndex =<a name="line.109"></a>
<span class="sourceLineNo">110</span> * otherServerIndex; }<a name="line.110"></a>
<span class="sourceLineNo">111</span> * @Override public boolean process(final RegionServerOperation op) throws IOException { // If a<a name="line.111"></a>
<span class="sourceLineNo">112</span> * regionserver shutdown and its of the meta server, then we want to // delay the processing of<a name="line.112"></a>
<span class="sourceLineNo">113</span> * the shutdown and send off a close of a region on // the 'otherServer. boolean result = true; if<a name="line.113"></a>
<span class="sourceLineNo">114</span> * (op instanceof ProcessServerShutdown) { ProcessServerShutdown pss = (ProcessServerShutdown)op;<a name="line.114"></a>
<span class="sourceLineNo">115</span> * if (pss.getDeadServerAddress().equals(this.metaAddress)) { // Don't postpone more than once. if<a name="line.115"></a>
<span class="sourceLineNo">116</span> * (!this.postponed.contains(pss)) { // Close some region.<a name="line.116"></a>
<span class="sourceLineNo">117</span> * this.cluster.addMessageToSendRegionServer(this.otherServerIndex, new<a name="line.117"></a>
<span class="sourceLineNo">118</span> * HMsg(HMsg.Type.MSG_REGION_CLOSE, hri, Bytes.toBytes("Forcing close in test")));<a name="line.118"></a>
<span class="sourceLineNo">119</span> * this.postponed.add(pss); // Put off the processing of the regionserver shutdown processing.<a name="line.119"></a>
<span class="sourceLineNo">120</span> * pss.setDelay(SERVER_DURATION); this.metaShutdownReceived = true; // Return false. This will add<a name="line.120"></a>
<span class="sourceLineNo">121</span> * this op to the delayed queue. result = false; } } } else { // Have the close run frequently. if<a name="line.121"></a>
<span class="sourceLineNo">122</span> * (isWantedCloseOperation(op) != null) { op.setDelay(CLOSE_DURATION); // Count how many times it<a name="line.122"></a>
<span class="sourceLineNo">123</span> * comes through here. this.closeCount++; } } return result; } public void processed(final<a name="line.123"></a>
<span class="sourceLineNo">124</span> * RegionServerOperation op) { if (isWantedCloseOperation(op) != null) return; this.done = true; }<a name="line.124"></a>
<span class="sourceLineNo">125</span> */<a name="line.125"></a>
<span class="sourceLineNo">126</span> /*<a name="line.126"></a>
<span class="sourceLineNo">127</span> * @return Null if not the wanted ProcessRegionClose, else &lt;code&gt;op&lt;/code&gt; cast as a<a name="line.127"></a>
<span class="sourceLineNo">128</span> * ProcessRegionClose.<a name="line.128"></a>
<span class="sourceLineNo">129</span> */<a name="line.129"></a>
<span class="sourceLineNo">130</span> /*<a name="line.130"></a>
<span class="sourceLineNo">131</span> * private ProcessRegionClose isWantedCloseOperation(final RegionServerOperation op) { // Count<a name="line.131"></a>
<span class="sourceLineNo">132</span> * every time we get a close operation. if (op instanceof ProcessRegionClose) { ProcessRegionClose<a name="line.132"></a>
<span class="sourceLineNo">133</span> * c = (ProcessRegionClose)op; if (c.regionInfo.equals(hri)) { return c; } } return null; }<a name="line.133"></a>
<span class="sourceLineNo">134</span> * boolean isDone() { return this.done; } boolean isMetaShutdownReceived() { return<a name="line.134"></a>
<span class="sourceLineNo">135</span> * metaShutdownReceived; } int getCloseCount() { return this.closeCount; }<a name="line.135"></a>
<span class="sourceLineNo">136</span> * @Override public boolean process(HServerInfo serverInfo, HMsg incomingMsg) { return true; } }<a name="line.136"></a>
<span class="sourceLineNo">137</span> */<a name="line.137"></a>
<span class="sourceLineNo">138</span> /**<a name="line.138"></a>
<span class="sourceLineNo">139</span> * In 2428, the meta region has just been set offline and then a close comes in.<a name="line.139"></a>
<span class="sourceLineNo">140</span> * @see &lt;a href="https://issues.apache.org/jira/browse/HBASE-2428"&gt;HBASE-2428&lt;/a&gt;<a name="line.140"></a>
<span class="sourceLineNo">141</span> */<a name="line.141"></a>
<span class="sourceLineNo">142</span> @Ignore<a name="line.142"></a>
<span class="sourceLineNo">143</span> @Test<a name="line.143"></a>
<span class="sourceLineNo">144</span> public void testRegionCloseWhenNoMetaHBase2428() throws Exception {<a name="line.144"></a>
<span class="sourceLineNo">145</span> /*<a name="line.145"></a>
<span class="sourceLineNo">146</span> * LOG.info("Running testRegionCloseWhenNoMetaHBase2428"); MiniHBaseCluster cluster =<a name="line.146"></a>
<span class="sourceLineNo">147</span> * TEST_UTIL.getHBaseCluster(); final HMaster master = cluster.getMaster(); int metaIndex =<a name="line.147"></a>
<span class="sourceLineNo">148</span> * cluster.getServerWithMeta(); // Figure the index of the server that is not server the<a name="line.148"></a>
<span class="sourceLineNo">149</span> * hbase:meta int otherServerIndex = -1; for (int i = 0; i &lt;<a name="line.149"></a>
<span class="sourceLineNo">150</span> * cluster.getRegionServerThreads().size(); i++) { if (i == metaIndex) continue;<a name="line.150"></a>
<span class="sourceLineNo">151</span> * otherServerIndex = i; break; } final HRegionServer otherServer =<a name="line.151"></a>
<span class="sourceLineNo">152</span> * cluster.getRegionServer(otherServerIndex); final HRegionServer metaHRS =<a name="line.152"></a>
<span class="sourceLineNo">153</span> * cluster.getRegionServer(metaIndex); // Get a region out on the otherServer. final RegionInfo<a name="line.153"></a>
<span class="sourceLineNo">154</span> * hri = otherServer.getOnlineRegions().iterator().next().getRegionInfo(); // Add our<a name="line.154"></a>
<span class="sourceLineNo">155</span> * RegionServerOperationsListener HBase2428Listener listener = new HBase2428Listener(cluster,<a name="line.155"></a>
<span class="sourceLineNo">156</span> * metaHRS.getHServerInfo().getServerAddress(), hri, otherServerIndex);<a name="line.156"></a>
<span class="sourceLineNo">157</span> * master.getRegionServerOperationQueue(). registerRegionServerOperationListener(listener); try<a name="line.157"></a>
<span class="sourceLineNo">158</span> * { // Now close the server carrying meta. cluster.abortRegionServer(metaIndex); // First wait<a name="line.158"></a>
<span class="sourceLineNo">159</span> * on receipt of meta server shutdown message. while(!listener.metaShutdownReceived)<a name="line.159"></a>
<span class="sourceLineNo">160</span> * Threads.sleep(100); while(!listener.isDone()) Threads.sleep(10); // We should not have<a name="line.160"></a>
<span class="sourceLineNo">161</span> * retried the close more times than it took for the // server shutdown message to exit the<a name="line.161"></a>
<span class="sourceLineNo">162</span> * delay queue and get processed // (Multiple by two to add in some slop in case of GC or<a name="line.162"></a>
<span class="sourceLineNo">163</span> * something). assertTrue(listener.getCloseCount() &gt; 1); assertTrue(listener.getCloseCount() &lt;<a name="line.163"></a>
<span class="sourceLineNo">164</span> * ((HBase2428Listener.SERVER_DURATION/HBase2428Listener.CLOSE_DURATION) * 2)); // Assert the<a name="line.164"></a>
<span class="sourceLineNo">165</span> * closed region came back online assertRegionIsBackOnline(hri); } finally {<a name="line.165"></a>
<span class="sourceLineNo">166</span> * master.getRegionServerOperationQueue(). unregisterRegionServerOperationListener(listener); }<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 adding in a new server before old one on same host+port is dead. Make the test more<a name="line.171"></a>
<span class="sourceLineNo">172</span> * onerous by having the server under test carry the meta. If confusion between old and new,<a name="line.172"></a>
<span class="sourceLineNo">173</span> * purportedly meta never comes back. Test that meta gets redeployed.<a name="line.173"></a>
<span class="sourceLineNo">174</span> */<a name="line.174"></a>
<span class="sourceLineNo">175</span> @Ignore<a name="line.175"></a>
<span class="sourceLineNo">176</span> @Test<a name="line.176"></a>
<span class="sourceLineNo">177</span> public void testAddingServerBeforeOldIsDead2413() throws IOException {<a name="line.177"></a>
<span class="sourceLineNo">178</span> /*<a name="line.178"></a>
<span class="sourceLineNo">179</span> * LOG.info("Running testAddingServerBeforeOldIsDead2413"); MiniHBaseCluster cluster =<a name="line.179"></a>
<span class="sourceLineNo">180</span> * TEST_UTIL.getHBaseCluster(); int count = count(); int metaIndex =<a name="line.180"></a>
<span class="sourceLineNo">181</span> * cluster.getServerWithMeta(); MiniHBaseClusterRegionServer metaHRS =<a name="line.181"></a>
<span class="sourceLineNo">182</span> * (MiniHBaseClusterRegionServer)cluster.getRegionServer(metaIndex); int port =<a name="line.182"></a>
<span class="sourceLineNo">183</span> * metaHRS.getServerInfo().getServerAddress().getPort(); Configuration c =<a name="line.183"></a>
<span class="sourceLineNo">184</span> * TEST_UTIL.getConfiguration(); String oldPort = c.get(HConstants.REGIONSERVER_PORT, "0"); try<a name="line.184"></a>
<span class="sourceLineNo">185</span> * { LOG.info("KILLED=" + metaHRS); metaHRS.kill(); c.set(HConstants.REGIONSERVER_PORT,<a name="line.185"></a>
<span class="sourceLineNo">186</span> * Integer.toString(port)); // Try and start new regionserver. It might clash with the old //<a name="line.186"></a>
<span class="sourceLineNo">187</span> * regionserver port so keep trying to get past the BindException. HRegionServer hrs = null;<a name="line.187"></a>
<span class="sourceLineNo">188</span> * while (true) { try { hrs = cluster.startRegionServer().getRegionServer(); break; } catch<a name="line.188"></a>
<span class="sourceLineNo">189</span> * (IOException e) { if (e.getCause() != null &amp;&amp; e.getCause() instanceof<a name="line.189"></a>
<span class="sourceLineNo">190</span> * InvocationTargetException) { InvocationTargetException ee =<a name="line.190"></a>
<span class="sourceLineNo">191</span> * (InvocationTargetException)e.getCause(); if (ee.getCause() != null &amp;&amp; ee.getCause()<a name="line.191"></a>
<span class="sourceLineNo">192</span> * instanceof BindException) { LOG.info("BindException; retrying: " + e.toString()); } } } }<a name="line.192"></a>
<span class="sourceLineNo">193</span> * LOG.info("STARTED=" + hrs); // Wait until he's been given at least 3 regions before we go on<a name="line.193"></a>
<span class="sourceLineNo">194</span> * to try // and count rows in table. while (hrs.getOnlineRegions().size() &lt; 3)<a name="line.194"></a>
<span class="sourceLineNo">195</span> * Threads.sleep(100); LOG.info(hrs.toString() + " has " + hrs.getOnlineRegions().size() +<a name="line.195"></a>
<span class="sourceLineNo">196</span> * " regions"); assertEquals(count, count()); } finally { c.set(HConstants.REGIONSERVER_PORT,<a name="line.196"></a>
<span class="sourceLineNo">197</span> * oldPort); }<a name="line.197"></a>
<span class="sourceLineNo">198</span> */<a name="line.198"></a>
<span class="sourceLineNo">199</span> }<a name="line.199"></a>
<span class="sourceLineNo">200</span><a name="line.200"></a>
<span class="sourceLineNo">201</span> /**<a name="line.201"></a>
<span class="sourceLineNo">202</span> * HBase2482 is about outstanding region openings. If any are outstanding when a regionserver goes<a name="line.202"></a>
<span class="sourceLineNo">203</span> * down, then they'll never deploy. They'll be stuck in the regions-in-transition list for ever.<a name="line.203"></a>
<span class="sourceLineNo">204</span> * This listener looks for a region opening HMsg and if its from the server passed on<a name="line.204"></a>
<span class="sourceLineNo">205</span> * construction, then we kill it. It also looks out for a close message on the victim server<a name="line.205"></a>
<span class="sourceLineNo">206</span> * because that signifies start of the fireworks.<a name="line.206"></a>
<span class="sourceLineNo">207</span> */<a name="line.207"></a>
<span class="sourceLineNo">208</span> /*<a name="line.208"></a>
<span class="sourceLineNo">209</span> * static class HBase2482Listener implements RegionServerOperationListener { private final<a name="line.209"></a>
<span class="sourceLineNo">210</span> * HRegionServer victim; private boolean abortSent = false; // We closed regions on new server.<a name="line.210"></a>
<span class="sourceLineNo">211</span> * private volatile boolean closed = false; // Copy of regions on new server private final<a name="line.211"></a>
<span class="sourceLineNo">212</span> * Collection&lt;HRegion&gt; copyOfOnlineRegions; // This is the region that was in transition on the<a name="line.212"></a>
<span class="sourceLineNo">213</span> * server we aborted. Test // passes if this region comes back online successfully. private<a name="line.213"></a>
<span class="sourceLineNo">214</span> * RegionInfo regionToFind; HBase2482Listener(final HRegionServer victim) { this.victim = victim;<a name="line.214"></a>
<span class="sourceLineNo">215</span> * // Copy regions currently open on this server so I can notice when // there is a close.<a name="line.215"></a>
<span class="sourceLineNo">216</span> * this.copyOfOnlineRegions = this.victim.getCopyOfOnlineRegionsSortedBySize().values(); }<a name="line.216"></a>
<span class="sourceLineNo">217</span> * @Override public boolean process(HServerInfo serverInfo, HMsg incomingMsg) { if<a name="line.217"></a>
<span class="sourceLineNo">218</span> * (!victim.getServerInfo().equals(serverInfo) || this.abortSent || !this.closed) { return true; }<a name="line.218"></a>
<span class="sourceLineNo">219</span> * if (!incomingMsg.isType(HMsg.Type.MSG_REPORT_PROCESS_OPEN)) return true; // Save the region<a name="line.219"></a>
<span class="sourceLineNo">220</span> * that is in transition so can test later it came back. this.regionToFind =<a name="line.220"></a>
<span class="sourceLineNo">221</span> * incomingMsg.getRegionInfo(); String msg = "ABORTING " + this.victim + " because got a " +<a name="line.221"></a>
<span class="sourceLineNo">222</span> * HMsg.Type.MSG_REPORT_PROCESS_OPEN + " on this server for " +<a name="line.222"></a>
<span class="sourceLineNo">223</span> * incomingMsg.getRegionInfo().getRegionNameAsString(); this.victim.abort(msg); this.abortSent =<a name="line.223"></a>
<span class="sourceLineNo">224</span> * true; return true; }<a name="line.224"></a>
<span class="sourceLineNo">225</span> * @Override public boolean process(RegionServerOperation op) throws IOException { return true; }<a name="line.225"></a>
<span class="sourceLineNo">226</span> * @Override public void processed(RegionServerOperation op) { if (this.closed || !(op instanceof<a name="line.226"></a>
<span class="sourceLineNo">227</span> * ProcessRegionClose)) return; ProcessRegionClose close = (ProcessRegionClose)op; for (HRegion r:<a name="line.227"></a>
<span class="sourceLineNo">228</span> * this.copyOfOnlineRegions) { if (r.getRegionInfo().equals(close.regionInfo)) { // We've closed<a name="line.228"></a>
<span class="sourceLineNo">229</span> * one of the regions that was on the victim server. // Now can start testing for when all regions<a name="line.229"></a>
<span class="sourceLineNo">230</span> * are back online again LOG.info("Found close of " + r.getRegionInfo().getRegionNameAsString() +<a name="line.230"></a>
<span class="sourceLineNo">231</span> * "; setting close happened flag"); this.closed = true; break; } } } }<a name="line.231"></a>
<span class="sourceLineNo">232</span> */<a name="line.232"></a>
<span class="sourceLineNo">233</span> /**<a name="line.233"></a>
<span class="sourceLineNo">234</span> * In 2482, a RS with an opening region on it dies. The said region is then stuck in the master's<a name="line.234"></a>
<span class="sourceLineNo">235</span> * regions-in-transition and never leaves it. This test works by bringing up a new regionserver,<a name="line.235"></a>
<span class="sourceLineNo">236</span> * waiting for the load balancer to give it some regions. Then, we close all on the new server.<a name="line.236"></a>
<span class="sourceLineNo">237</span> * After sending all the close messages, we send the new regionserver the special blocking message<a name="line.237"></a>
<span class="sourceLineNo">238</span> * so it can not process any more messages. Meantime reopening of the just-closed regions is<a name="line.238"></a>
<span class="sourceLineNo">239</span> * backed up on the new server. Soon as master gets an opening region from the new regionserver,<a name="line.239"></a>
<span class="sourceLineNo">240</span> * we kill it. We then wait on all regions to come back on line. If bug is fixed, this should<a name="line.240"></a>
<span class="sourceLineNo">241</span> * happen soon as the processing of the killed server is done.<a name="line.241"></a>
<span class="sourceLineNo">242</span> * @see &lt;a href="https://issues.apache.org/jira/browse/HBASE-2482"&gt;HBASE-2482&lt;/a&gt;<a name="line.242"></a>
<span class="sourceLineNo">243</span> */<a name="line.243"></a>
<span class="sourceLineNo">244</span> @Ignore<a name="line.244"></a>
<span class="sourceLineNo">245</span> @Test<a name="line.245"></a>
<span class="sourceLineNo">246</span> public void testKillRSWithOpeningRegion2482() throws Exception {<a name="line.246"></a>
<span class="sourceLineNo">247</span> /*<a name="line.247"></a>
<span class="sourceLineNo">248</span> * LOG.info("Running testKillRSWithOpeningRegion2482"); MiniHBaseCluster cluster =<a name="line.248"></a>
<span class="sourceLineNo">249</span> * TEST_UTIL.getHBaseCluster(); if (cluster.getLiveRegionServerThreads().size() &lt; 2) { // Need<a name="line.249"></a>
<span class="sourceLineNo">250</span> * at least two servers. cluster.startRegionServer(); } // Count how many regions are online.<a name="line.250"></a>
<span class="sourceLineNo">251</span> * They need to be all back online for // this test to succeed. int countOfMetaRegions =<a name="line.251"></a>
<span class="sourceLineNo">252</span> * countOfMetaRegions(); // Add a listener on the server. HMaster m = cluster.getMaster(); //<a name="line.252"></a>
<span class="sourceLineNo">253</span> * Start new regionserver. MiniHBaseClusterRegionServer hrs =<a name="line.253"></a>
<span class="sourceLineNo">254</span> * (MiniHBaseClusterRegionServer)cluster.startRegionServer().getRegionServer();<a name="line.254"></a>
<span class="sourceLineNo">255</span> * LOG.info("Started new regionserver: " + hrs.toString()); // Wait until has some regions<a name="line.255"></a>
<span class="sourceLineNo">256</span> * before proceeding. Balancer will give it some. int minimumRegions =<a name="line.256"></a>
<span class="sourceLineNo">257</span> * countOfMetaRegions/(cluster.getRegionServerThreads().size() * 2); while<a name="line.257"></a>
<span class="sourceLineNo">258</span> * (hrs.getOnlineRegions().size() &lt; minimumRegions) Threads.sleep(100); // Set the listener only<a name="line.258"></a>
<span class="sourceLineNo">259</span> * after some regions have been opened on new server. HBase2482Listener listener = new<a name="line.259"></a>
<span class="sourceLineNo">260</span> * HBase2482Listener(hrs); m.getRegionServerOperationQueue().<a name="line.260"></a>
<span class="sourceLineNo">261</span> * registerRegionServerOperationListener(listener); try { // Go close all non-catalog regions on<a name="line.261"></a>
<span class="sourceLineNo">262</span> * this new server closeAllNonCatalogRegions(cluster, hrs); // After all closes, add blocking<a name="line.262"></a>
<span class="sourceLineNo">263</span> * message before the region opens start to // come in.<a name="line.263"></a>
<span class="sourceLineNo">264</span> * cluster.addMessageToSendRegionServer(hrs, new HMsg(HMsg.Type.TESTING_BLOCK_REGIONSERVER)); //<a name="line.264"></a>
<span class="sourceLineNo">265</span> * Wait till one of the above close messages has an effect before we start // wait on all<a name="line.265"></a>
<span class="sourceLineNo">266</span> * regions back online. while (!listener.closed) Threads.sleep(100); LOG.info("Past close"); //<a name="line.266"></a>
<span class="sourceLineNo">267</span> * Make sure the abort server message was sent. while(!listener.abortSent) Threads.sleep(100);<a name="line.267"></a>
<span class="sourceLineNo">268</span> * LOG.info("Past abort send; waiting on all regions to redeploy"); // Now wait for regions to<a name="line.268"></a>
<span class="sourceLineNo">269</span> * come back online. assertRegionIsBackOnline(listener.regionToFind); } finally {<a name="line.269"></a>
<span class="sourceLineNo">270</span> * m.getRegionServerOperationQueue(). unregisterRegionServerOperationListener(listener); }<a name="line.270"></a>
<span class="sourceLineNo">271</span> */<a name="line.271"></a>
<span class="sourceLineNo">272</span> }<a name="line.272"></a>
<span class="sourceLineNo">273</span><a name="line.273"></a>
<span class="sourceLineNo">274</span> /*<a name="line.274"></a>
<span class="sourceLineNo">275</span> * @return Count of all non-catalog regions on the designated server<a name="line.275"></a>
<span class="sourceLineNo">276</span> */<a name="line.276"></a>
<span class="sourceLineNo">277</span> /*<a name="line.277"></a>
<span class="sourceLineNo">278</span> * private int closeAllNonCatalogRegions(final MiniHBaseCluster cluster, final<a name="line.278"></a>
<span class="sourceLineNo">279</span> * MiniHBaseCluster.MiniHBaseClusterRegionServer hrs) throws IOException { int countOfRegions = 0;<a name="line.279"></a>
<span class="sourceLineNo">280</span> * for (HRegion r: hrs.getOnlineRegions()) { if (r.getRegionInfo().isMetaRegion()) continue;<a name="line.280"></a>
<span class="sourceLineNo">281</span> * cluster.addMessageToSendRegionServer(hrs, new HMsg(HMsg.Type.MSG_REGION_CLOSE,<a name="line.281"></a>
<span class="sourceLineNo">282</span> * r.getRegionInfo())); LOG.info("Sent close of " + r.getRegionInfo().getRegionNameAsString() +<a name="line.282"></a>
<span class="sourceLineNo">283</span> * " on " + hrs.toString()); countOfRegions++; } return countOfRegions; } private void<a name="line.283"></a>
<span class="sourceLineNo">284</span> * assertRegionIsBackOnline(final RegionInfo hri) throws IOException { // Region should have an<a name="line.284"></a>
<span class="sourceLineNo">285</span> * entry in its startkey because of addRowToEachRegion. byte [] row = getStartKey(hri); HTable t =<a name="line.285"></a>
<span class="sourceLineNo">286</span> * new HTable(TEST_UTIL.getConfiguration(), TABLENAME); Get g = new Get(row);<a name="line.286"></a>
<span class="sourceLineNo">287</span> * assertTrue((t.get(g)).size() &gt; 0); } /*<a name="line.287"></a>
<span class="sourceLineNo">288</span> * @return Count of regions in meta table.<a name="line.288"></a>
<span class="sourceLineNo">289</span> */<a name="line.289"></a>
<span class="sourceLineNo">290</span> /*<a name="line.290"></a>
<span class="sourceLineNo">291</span> * private static int countOfMetaRegions() throws IOException { HTable meta = new<a name="line.291"></a>
<span class="sourceLineNo">292</span> * HTable(TEST_UTIL.getConfiguration(), HConstants.META_TABLE_NAME); int rows = 0; Scan scan = new<a name="line.292"></a>
<span class="sourceLineNo">293</span> * Scan(); scan.addColumn(HConstants.CATALOG_FAMILY, HConstants.SERVER_QUALIFIER); ResultScanner s<a name="line.293"></a>
<span class="sourceLineNo">294</span> * = meta.getScanner(scan); for (Result r = null; (r = s.next()) != null;) { byte [] b =<a name="line.294"></a>
<span class="sourceLineNo">295</span> * r.getValue(HConstants.CATALOG_FAMILY, HConstants.SERVER_QUALIFIER); if (b == null || b.length<a name="line.295"></a>
<span class="sourceLineNo">296</span> * &lt;= 0) break; rows++; } s.close(); return rows; }<a name="line.296"></a>
<span class="sourceLineNo">297</span> */<a name="line.297"></a>
<span class="sourceLineNo">298</span> /*<a name="line.298"></a>
<span class="sourceLineNo">299</span> * Add to each of the regions in hbase:meta a value. Key is the startrow of the region (except its<a name="line.299"></a>
<span class="sourceLineNo">300</span> * 'aaa' for first region). Actual value is the row name.<a name="line.300"></a>
<span class="sourceLineNo">301</span> */<a name="line.301"></a>
<span class="sourceLineNo">302</span> private static int addToEachStartKey(final int expected) throws IOException {<a name="line.302"></a>
<span class="sourceLineNo">303</span> Table t = TEST_UTIL.getConnection().getTable(TABLENAME);<a name="line.303"></a>
<span class="sourceLineNo">304</span> Table meta = TEST_UTIL.getConnection().getTable(TableName.META_TABLE_NAME);<a name="line.304"></a>
<span class="sourceLineNo">305</span> int rows = 0;<a name="line.305"></a>
<span class="sourceLineNo">306</span> Scan scan = new Scan();<a name="line.306"></a>
<span class="sourceLineNo">307</span> scan.addColumn(HConstants.CATALOG_FAMILY, HConstants.REGIONINFO_QUALIFIER);<a name="line.307"></a>
<span class="sourceLineNo">308</span> ResultScanner s = meta.getScanner(scan);<a name="line.308"></a>
<span class="sourceLineNo">309</span> for (Result r = null; (r = s.next()) != null;) {<a name="line.309"></a>
<span class="sourceLineNo">310</span> RegionInfo hri = CatalogFamilyFormat.getRegionInfo(r);<a name="line.310"></a>
<span class="sourceLineNo">311</span> if (hri == null) break;<a name="line.311"></a>
<span class="sourceLineNo">312</span> if (!hri.getTable().equals(TABLENAME)) {<a name="line.312"></a>
<span class="sourceLineNo">313</span> continue;<a name="line.313"></a>
<span class="sourceLineNo">314</span> }<a name="line.314"></a>
<span class="sourceLineNo">315</span><a name="line.315"></a>
<span class="sourceLineNo">316</span> // If start key, add 'aaa'.<a name="line.316"></a>
<span class="sourceLineNo">317</span> if (!hri.getTable().equals(TABLENAME)) {<a name="line.317"></a>
<span class="sourceLineNo">318</span> continue;<a name="line.318"></a>
<span class="sourceLineNo">319</span> }<a name="line.319"></a>
<span class="sourceLineNo">320</span> byte[] row = getStartKey(hri);<a name="line.320"></a>
<span class="sourceLineNo">321</span> Put p = new Put(row);<a name="line.321"></a>
<span class="sourceLineNo">322</span> p.setDurability(Durability.SKIP_WAL);<a name="line.322"></a>
<span class="sourceLineNo">323</span> p.addColumn(getTestFamily(), getTestQualifier(), row);<a name="line.323"></a>
<span class="sourceLineNo">324</span> t.put(p);<a name="line.324"></a>
<span class="sourceLineNo">325</span> rows++;<a name="line.325"></a>
<span class="sourceLineNo">326</span> }<a name="line.326"></a>
<span class="sourceLineNo">327</span> s.close();<a name="line.327"></a>
<span class="sourceLineNo">328</span> Assert.assertEquals(expected, rows);<a name="line.328"></a>
<span class="sourceLineNo">329</span> t.close();<a name="line.329"></a>
<span class="sourceLineNo">330</span> meta.close();<a name="line.330"></a>
<span class="sourceLineNo">331</span> return rows;<a name="line.331"></a>
<span class="sourceLineNo">332</span> }<a name="line.332"></a>
<span class="sourceLineNo">333</span><a name="line.333"></a>
<span class="sourceLineNo">334</span> /*<a name="line.334"></a>
<span class="sourceLineNo">335</span> * @return Start key for hri (If start key is '', then return 'aaa'.<a name="line.335"></a>
<span class="sourceLineNo">336</span> */<a name="line.336"></a>
<span class="sourceLineNo">337</span> private static byte[] getStartKey(final RegionInfo hri) {<a name="line.337"></a>
<span class="sourceLineNo">338</span> return Bytes.equals(HConstants.EMPTY_START_ROW, hri.getStartKey())<a name="line.338"></a>
<span class="sourceLineNo">339</span> ? Bytes.toBytes("aaa")<a name="line.339"></a>
<span class="sourceLineNo">340</span> : hri.getStartKey();<a name="line.340"></a>
<span class="sourceLineNo">341</span> }<a name="line.341"></a>
<span class="sourceLineNo">342</span><a name="line.342"></a>
<span class="sourceLineNo">343</span> private static byte[] getTestFamily() {<a name="line.343"></a>
<span class="sourceLineNo">344</span> return FAMILIES[0];<a name="line.344"></a>
<span class="sourceLineNo">345</span> }<a name="line.345"></a>
<span class="sourceLineNo">346</span><a name="line.346"></a>
<span class="sourceLineNo">347</span> private static byte[] getTestQualifier() {<a name="line.347"></a>
<span class="sourceLineNo">348</span> return getTestFamily();<a name="line.348"></a>
<span class="sourceLineNo">349</span> }<a name="line.349"></a>
<span class="sourceLineNo">350</span>}<a name="line.350"></a>
</pre>
</div>
</body>
</html>