blob: cf2a1e5983eeb09f4223e7a5d8987c418f8fd6f2 [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.regionserver;<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.assertTrue;<a name="line.21"></a>
<span class="sourceLineNo">022</span><a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.io.IOException;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.List;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.Map;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import java.util.concurrent.atomic.AtomicReference;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.conf.Configuration;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.HBaseClassTestRule;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.HBaseTestingUtil;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.LocalHBaseCluster;<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.SingleProcessHBaseCluster;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.master.HMaster;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.master.ServerListener;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.master.ServerManager;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.testclassification.MediumTests;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.testclassification.RegionServerTests;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.util.JVMClusterUtil.MasterThread;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.junit.ClassRule;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.junit.Ignore;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.junit.Rule;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.junit.Test;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.junit.experimental.categories.Category;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.junit.rules.TestName;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.slf4j.Logger;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.slf4j.LoggerFactory;<a name="line.51"></a>
<span class="sourceLineNo">052</span><a name="line.52"></a>
<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionServerStartupResponse;<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> * Tests that a regionserver that dies after reporting for duty gets removed from list of online<a name="line.56"></a>
<span class="sourceLineNo">057</span> * regions. See HBASE-9593.<a name="line.57"></a>
<span class="sourceLineNo">058</span> */<a name="line.58"></a>
<span class="sourceLineNo">059</span>@Category({ RegionServerTests.class, MediumTests.class })<a name="line.59"></a>
<span class="sourceLineNo">060</span>@Ignore("See HBASE-19515")<a name="line.60"></a>
<span class="sourceLineNo">061</span>public class TestRSKilledWhenInitializing {<a name="line.61"></a>
<span class="sourceLineNo">062</span><a name="line.62"></a>
<span class="sourceLineNo">063</span> @ClassRule<a name="line.63"></a>
<span class="sourceLineNo">064</span> public static final HBaseClassTestRule CLASS_RULE =<a name="line.64"></a>
<span class="sourceLineNo">065</span> HBaseClassTestRule.forClass(TestRSKilledWhenInitializing.class);<a name="line.65"></a>
<span class="sourceLineNo">066</span><a name="line.66"></a>
<span class="sourceLineNo">067</span> private static final Logger LOG = LoggerFactory.getLogger(TestRSKilledWhenInitializing.class);<a name="line.67"></a>
<span class="sourceLineNo">068</span><a name="line.68"></a>
<span class="sourceLineNo">069</span> @Rule<a name="line.69"></a>
<span class="sourceLineNo">070</span> public TestName testName = new TestName();<a name="line.70"></a>
<span class="sourceLineNo">071</span><a name="line.71"></a>
<span class="sourceLineNo">072</span> // This boolean needs to be globally available. It is used below in our<a name="line.72"></a>
<span class="sourceLineNo">073</span> // mocked up regionserver so it knows when to die.<a name="line.73"></a>
<span class="sourceLineNo">074</span> private static AtomicBoolean masterActive = new AtomicBoolean(false);<a name="line.74"></a>
<span class="sourceLineNo">075</span> // Ditto for this variable. It also is used in the mocked regionserver class.<a name="line.75"></a>
<span class="sourceLineNo">076</span> private static final AtomicReference&lt;ServerName&gt; killedRS = new AtomicReference&lt;ServerName&gt;();<a name="line.76"></a>
<span class="sourceLineNo">077</span><a name="line.77"></a>
<span class="sourceLineNo">078</span> private static final int NUM_MASTERS = 1;<a name="line.78"></a>
<span class="sourceLineNo">079</span> private static final int NUM_RS = 2;<a name="line.79"></a>
<span class="sourceLineNo">080</span><a name="line.80"></a>
<span class="sourceLineNo">081</span> /**<a name="line.81"></a>
<span class="sourceLineNo">082</span> * Test verifies whether a region server is removed from online servers list in master if it went<a name="line.82"></a>
<span class="sourceLineNo">083</span> * down after registering with master. Test will TIMEOUT if an error!!!!<a name="line.83"></a>
<span class="sourceLineNo">084</span> */<a name="line.84"></a>
<span class="sourceLineNo">085</span> @Test<a name="line.85"></a>
<span class="sourceLineNo">086</span> public void testRSTerminationAfterRegisteringToMasterBeforeCreatingEphemeralNode()<a name="line.86"></a>
<span class="sourceLineNo">087</span> throws Exception {<a name="line.87"></a>
<span class="sourceLineNo">088</span> // Create config to use for this cluster<a name="line.88"></a>
<span class="sourceLineNo">089</span> Configuration conf = HBaseConfiguration.create();<a name="line.89"></a>
<span class="sourceLineNo">090</span> conf.setInt(ServerManager.WAIT_ON_REGIONSERVERS_MINTOSTART, 1);<a name="line.90"></a>
<span class="sourceLineNo">091</span> // Start the cluster<a name="line.91"></a>
<span class="sourceLineNo">092</span> final HBaseTestingUtil TEST_UTIL = new HBaseTestingUtil(conf);<a name="line.92"></a>
<span class="sourceLineNo">093</span> TEST_UTIL.startMiniDFSCluster(3);<a name="line.93"></a>
<span class="sourceLineNo">094</span> TEST_UTIL.startMiniZKCluster();<a name="line.94"></a>
<span class="sourceLineNo">095</span> TEST_UTIL.createRootDir();<a name="line.95"></a>
<span class="sourceLineNo">096</span> final LocalHBaseCluster cluster = new LocalHBaseCluster(conf, NUM_MASTERS, NUM_RS,<a name="line.96"></a>
<span class="sourceLineNo">097</span> HMaster.class, RegisterAndDieRegionServer.class);<a name="line.97"></a>
<span class="sourceLineNo">098</span> final MasterThread master = startMaster(cluster.getMasters().get(0));<a name="line.98"></a>
<span class="sourceLineNo">099</span> try {<a name="line.99"></a>
<span class="sourceLineNo">100</span> // Master is up waiting on RegionServers to check in. Now start RegionServers.<a name="line.100"></a>
<span class="sourceLineNo">101</span> for (int i = 0; i &lt; NUM_RS; i++) {<a name="line.101"></a>
<span class="sourceLineNo">102</span> cluster.getRegionServers().get(i).start();<a name="line.102"></a>
<span class="sourceLineNo">103</span> }<a name="line.103"></a>
<span class="sourceLineNo">104</span> // Expected total regionservers depends on whether Master can host regions or not.<a name="line.104"></a>
<span class="sourceLineNo">105</span> int expectedTotalRegionServers = NUM_RS;<a name="line.105"></a>
<span class="sourceLineNo">106</span> List&lt;ServerName&gt; onlineServersList = null;<a name="line.106"></a>
<span class="sourceLineNo">107</span> do {<a name="line.107"></a>
<span class="sourceLineNo">108</span> onlineServersList = master.getMaster().getServerManager().getOnlineServersList();<a name="line.108"></a>
<span class="sourceLineNo">109</span> } while (onlineServersList.size() &lt; expectedTotalRegionServers);<a name="line.109"></a>
<span class="sourceLineNo">110</span> // Wait until killedRS is set. Means RegionServer is starting to go down.<a name="line.110"></a>
<span class="sourceLineNo">111</span> while (killedRS.get() == null) {<a name="line.111"></a>
<span class="sourceLineNo">112</span> Threads.sleep(1);<a name="line.112"></a>
<span class="sourceLineNo">113</span> }<a name="line.113"></a>
<span class="sourceLineNo">114</span> // Wait on the RegionServer to fully die.<a name="line.114"></a>
<span class="sourceLineNo">115</span> while (cluster.getLiveRegionServers().size() &gt;= expectedTotalRegionServers) {<a name="line.115"></a>
<span class="sourceLineNo">116</span> Threads.sleep(1);<a name="line.116"></a>
<span class="sourceLineNo">117</span> }<a name="line.117"></a>
<span class="sourceLineNo">118</span> // Make sure Master is fully up before progressing. Could take a while if regions<a name="line.118"></a>
<span class="sourceLineNo">119</span> // being reassigned.<a name="line.119"></a>
<span class="sourceLineNo">120</span> while (!master.getMaster().isInitialized()) {<a name="line.120"></a>
<span class="sourceLineNo">121</span> Threads.sleep(1);<a name="line.121"></a>
<span class="sourceLineNo">122</span> }<a name="line.122"></a>
<span class="sourceLineNo">123</span><a name="line.123"></a>
<span class="sourceLineNo">124</span> // Now in steady state. How many regions open? Master should have too many regionservers<a name="line.124"></a>
<span class="sourceLineNo">125</span> // showing still. The downed RegionServer should still be showing as registered.<a name="line.125"></a>
<span class="sourceLineNo">126</span> assertTrue(master.getMaster().getServerManager().isServerOnline(killedRS.get()));<a name="line.126"></a>
<span class="sourceLineNo">127</span> // Find non-meta region (namespace?) and assign to the killed server. That'll trigger cleanup.<a name="line.127"></a>
<span class="sourceLineNo">128</span> Map&lt;RegionInfo, ServerName&gt; assignments = null;<a name="line.128"></a>
<span class="sourceLineNo">129</span> do {<a name="line.129"></a>
<span class="sourceLineNo">130</span> assignments =<a name="line.130"></a>
<span class="sourceLineNo">131</span> master.getMaster().getAssignmentManager().getRegionStates().getRegionAssignments();<a name="line.131"></a>
<span class="sourceLineNo">132</span> } while (assignments == null || assignments.size() &lt; 2);<a name="line.132"></a>
<span class="sourceLineNo">133</span> RegionInfo hri = null;<a name="line.133"></a>
<span class="sourceLineNo">134</span> for (Map.Entry&lt;RegionInfo, ServerName&gt; e : assignments.entrySet()) {<a name="line.134"></a>
<span class="sourceLineNo">135</span> if (e.getKey().isMetaRegion()) continue;<a name="line.135"></a>
<span class="sourceLineNo">136</span> hri = e.getKey();<a name="line.136"></a>
<span class="sourceLineNo">137</span> break;<a name="line.137"></a>
<span class="sourceLineNo">138</span> }<a name="line.138"></a>
<span class="sourceLineNo">139</span> // Try moving region to the killed server. It will fail. As by-product, we will<a name="line.139"></a>
<span class="sourceLineNo">140</span> // remove the RS from Master online list because no corresponding znode.<a name="line.140"></a>
<span class="sourceLineNo">141</span> assertEquals(expectedTotalRegionServers,<a name="line.141"></a>
<span class="sourceLineNo">142</span> master.getMaster().getServerManager().getOnlineServersList().size());<a name="line.142"></a>
<span class="sourceLineNo">143</span> LOG.info("Move " + hri.getEncodedName() + " to " + killedRS.get());<a name="line.143"></a>
<span class="sourceLineNo">144</span> master.getMaster().move(hri.getEncodedNameAsBytes(),<a name="line.144"></a>
<span class="sourceLineNo">145</span> Bytes.toBytes(killedRS.get().toString()));<a name="line.145"></a>
<span class="sourceLineNo">146</span><a name="line.146"></a>
<span class="sourceLineNo">147</span> // TODO: This test could do more to verify fix. It could create a table<a name="line.147"></a>
<span class="sourceLineNo">148</span> // and do round-robin assign. It should fail if zombie RS. HBASE-19515.<a name="line.148"></a>
<span class="sourceLineNo">149</span><a name="line.149"></a>
<span class="sourceLineNo">150</span> // Wait until the RS no longer shows as registered in Master.<a name="line.150"></a>
<span class="sourceLineNo">151</span> while (onlineServersList.size() &gt; (NUM_RS + 1)) {<a name="line.151"></a>
<span class="sourceLineNo">152</span> Thread.sleep(100);<a name="line.152"></a>
<span class="sourceLineNo">153</span> onlineServersList = master.getMaster().getServerManager().getOnlineServersList();<a name="line.153"></a>
<span class="sourceLineNo">154</span> }<a name="line.154"></a>
<span class="sourceLineNo">155</span> } finally {<a name="line.155"></a>
<span class="sourceLineNo">156</span> // Shutdown is messy with complaints about fs being closed. Why? TODO.<a name="line.156"></a>
<span class="sourceLineNo">157</span> cluster.shutdown();<a name="line.157"></a>
<span class="sourceLineNo">158</span> cluster.join();<a name="line.158"></a>
<span class="sourceLineNo">159</span> TEST_UTIL.shutdownMiniDFSCluster();<a name="line.159"></a>
<span class="sourceLineNo">160</span> TEST_UTIL.shutdownMiniZKCluster();<a name="line.160"></a>
<span class="sourceLineNo">161</span> TEST_UTIL.cleanupTestDir();<a name="line.161"></a>
<span class="sourceLineNo">162</span> }<a name="line.162"></a>
<span class="sourceLineNo">163</span> }<a name="line.163"></a>
<span class="sourceLineNo">164</span><a name="line.164"></a>
<span class="sourceLineNo">165</span> /**<a name="line.165"></a>
<span class="sourceLineNo">166</span> * Start Master. Get as far as the state where Master is waiting on RegionServers to check in,<a name="line.166"></a>
<span class="sourceLineNo">167</span> * then return.<a name="line.167"></a>
<span class="sourceLineNo">168</span> */<a name="line.168"></a>
<span class="sourceLineNo">169</span> private MasterThread startMaster(MasterThread master) {<a name="line.169"></a>
<span class="sourceLineNo">170</span> master.start();<a name="line.170"></a>
<span class="sourceLineNo">171</span> // It takes a while until ServerManager creation to happen inside Master startup.<a name="line.171"></a>
<span class="sourceLineNo">172</span> while (master.getMaster().getServerManager() == null) {<a name="line.172"></a>
<span class="sourceLineNo">173</span> continue;<a name="line.173"></a>
<span class="sourceLineNo">174</span> }<a name="line.174"></a>
<span class="sourceLineNo">175</span> // Set a listener for the waiting-on-RegionServers state. We want to wait<a name="line.175"></a>
<span class="sourceLineNo">176</span> // until this condition before we leave this method and start regionservers.<a name="line.176"></a>
<span class="sourceLineNo">177</span> final AtomicBoolean waiting = new AtomicBoolean(false);<a name="line.177"></a>
<span class="sourceLineNo">178</span> if (master.getMaster().getServerManager() == null) throw new NullPointerException("SM");<a name="line.178"></a>
<span class="sourceLineNo">179</span> master.getMaster().getServerManager().registerListener(new ServerListener() {<a name="line.179"></a>
<span class="sourceLineNo">180</span> @Override<a name="line.180"></a>
<span class="sourceLineNo">181</span> public void waiting() {<a name="line.181"></a>
<span class="sourceLineNo">182</span> waiting.set(true);<a name="line.182"></a>
<span class="sourceLineNo">183</span> }<a name="line.183"></a>
<span class="sourceLineNo">184</span> });<a name="line.184"></a>
<span class="sourceLineNo">185</span> // Wait until the Master gets to place where it is waiting on RegionServers to check in.<a name="line.185"></a>
<span class="sourceLineNo">186</span> while (!waiting.get()) {<a name="line.186"></a>
<span class="sourceLineNo">187</span> continue;<a name="line.187"></a>
<span class="sourceLineNo">188</span> }<a name="line.188"></a>
<span class="sourceLineNo">189</span> // Set the global master-is-active; gets picked up by regionservers later.<a name="line.189"></a>
<span class="sourceLineNo">190</span> masterActive.set(true);<a name="line.190"></a>
<span class="sourceLineNo">191</span> return master;<a name="line.191"></a>
<span class="sourceLineNo">192</span> }<a name="line.192"></a>
<span class="sourceLineNo">193</span><a name="line.193"></a>
<span class="sourceLineNo">194</span> /**<a name="line.194"></a>
<span class="sourceLineNo">195</span> * A RegionServer that reports for duty and then immediately dies if it is the first to receive<a name="line.195"></a>
<span class="sourceLineNo">196</span> * the response to a reportForDuty. When it dies, it clears its ephemeral znode which the master<a name="line.196"></a>
<span class="sourceLineNo">197</span> * notices and so removes the region from its set of online regionservers.<a name="line.197"></a>
<span class="sourceLineNo">198</span> */<a name="line.198"></a>
<span class="sourceLineNo">199</span> static class RegisterAndDieRegionServer<a name="line.199"></a>
<span class="sourceLineNo">200</span> extends SingleProcessHBaseCluster.MiniHBaseClusterRegionServer {<a name="line.200"></a>
<span class="sourceLineNo">201</span> public RegisterAndDieRegionServer(Configuration conf) throws IOException, InterruptedException {<a name="line.201"></a>
<span class="sourceLineNo">202</span> super(conf);<a name="line.202"></a>
<span class="sourceLineNo">203</span> }<a name="line.203"></a>
<span class="sourceLineNo">204</span><a name="line.204"></a>
<span class="sourceLineNo">205</span> @Override<a name="line.205"></a>
<span class="sourceLineNo">206</span> protected void handleReportForDutyResponse(RegionServerStartupResponse c) throws IOException {<a name="line.206"></a>
<span class="sourceLineNo">207</span> if (killedRS.compareAndSet(null, getServerName())) {<a name="line.207"></a>
<span class="sourceLineNo">208</span> // Make sure Master is up so it will see the removal of the ephemeral znode for this RS.<a name="line.208"></a>
<span class="sourceLineNo">209</span> while (!masterActive.get()) {<a name="line.209"></a>
<span class="sourceLineNo">210</span> Threads.sleep(100);<a name="line.210"></a>
<span class="sourceLineNo">211</span> }<a name="line.211"></a>
<span class="sourceLineNo">212</span> super.kill();<a name="line.212"></a>
<span class="sourceLineNo">213</span> } else {<a name="line.213"></a>
<span class="sourceLineNo">214</span> super.handleReportForDutyResponse(c);<a name="line.214"></a>
<span class="sourceLineNo">215</span> }<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>
</pre>
</div>
</body>
</html>