blob: becc0af4fb03808c77dd72f76faf636b2b0373ce [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.apache.hadoop.hbase.SplitLogCounters.resetCounters;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_heartbeat;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_node_create_queued;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_orphan_task_acquired;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_rescan;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_rescan_deleted;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_resubmit;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_resubmit_dead_server_task;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_resubmit_failed;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_resubmit_force;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_resubmit_threshold_reached;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_resubmit_unassigned;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import static org.apache.hadoop.hbase.SplitLogCounters.tot_mgr_task_deleted;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import static org.junit.Assert.assertEquals;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import static org.junit.Assert.assertFalse;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import static org.junit.Assert.assertTrue;<a name="line.35"></a>
<span class="sourceLineNo">036</span><a name="line.36"></a>
<span class="sourceLineNo">037</span>import java.io.IOException;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import java.util.Map;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import java.util.concurrent.atomic.LongAdder;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.apache.hadoop.conf.Configuration;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.apache.hadoop.fs.FileSystem;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.apache.hadoop.fs.Path;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.CoordinatedStateManager;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.HBaseClassTestRule;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.HBaseTestingUtil;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.HConstants;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.ServerName;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.SplitLogCounters;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.SplitLogTask;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.Waiter;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.coordination.ZKSplitLogManagerCoordination;<a name="line.51"></a>
<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.coordination.ZkCoordinatedStateManager;<a name="line.52"></a>
<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.master.SplitLogManager.Task;<a name="line.53"></a>
<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.master.SplitLogManager.TaskBatch;<a name="line.54"></a>
<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.testclassification.LargeTests;<a name="line.55"></a>
<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.testclassification.MasterTests;<a name="line.56"></a>
<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.57"></a>
<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.zookeeper.TestMasterAddressTracker.NodeCreationListener;<a name="line.58"></a>
<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.zookeeper.ZKSplitLog;<a name="line.59"></a>
<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.60"></a>
<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.61"></a>
<span class="sourceLineNo">062</span>import org.apache.zookeeper.CreateMode;<a name="line.62"></a>
<span class="sourceLineNo">063</span>import org.apache.zookeeper.KeeperException;<a name="line.63"></a>
<span class="sourceLineNo">064</span>import org.apache.zookeeper.ZooDefs.Ids;<a name="line.64"></a>
<span class="sourceLineNo">065</span>import org.junit.After;<a name="line.65"></a>
<span class="sourceLineNo">066</span>import org.junit.Assert;<a name="line.66"></a>
<span class="sourceLineNo">067</span>import org.junit.Before;<a name="line.67"></a>
<span class="sourceLineNo">068</span>import org.junit.ClassRule;<a name="line.68"></a>
<span class="sourceLineNo">069</span>import org.junit.Test;<a name="line.69"></a>
<span class="sourceLineNo">070</span>import org.junit.experimental.categories.Category;<a name="line.70"></a>
<span class="sourceLineNo">071</span>import org.mockito.Mockito;<a name="line.71"></a>
<span class="sourceLineNo">072</span>import org.slf4j.Logger;<a name="line.72"></a>
<span class="sourceLineNo">073</span>import org.slf4j.LoggerFactory;<a name="line.73"></a>
<span class="sourceLineNo">074</span><a name="line.74"></a>
<span class="sourceLineNo">075</span>@Category({ MasterTests.class, LargeTests.class })<a name="line.75"></a>
<span class="sourceLineNo">076</span>public class TestSplitLogManager {<a name="line.76"></a>
<span class="sourceLineNo">077</span><a name="line.77"></a>
<span class="sourceLineNo">078</span> @ClassRule<a name="line.78"></a>
<span class="sourceLineNo">079</span> public static final HBaseClassTestRule CLASS_RULE =<a name="line.79"></a>
<span class="sourceLineNo">080</span> HBaseClassTestRule.forClass(TestSplitLogManager.class);<a name="line.80"></a>
<span class="sourceLineNo">081</span><a name="line.81"></a>
<span class="sourceLineNo">082</span> private static final Logger LOG = LoggerFactory.getLogger(TestSplitLogManager.class);<a name="line.82"></a>
<span class="sourceLineNo">083</span><a name="line.83"></a>
<span class="sourceLineNo">084</span> private final ServerManager sm = Mockito.mock(ServerManager.class);<a name="line.84"></a>
<span class="sourceLineNo">085</span><a name="line.85"></a>
<span class="sourceLineNo">086</span> private ZKWatcher zkw;<a name="line.86"></a>
<span class="sourceLineNo">087</span> private DummyMasterServices master;<a name="line.87"></a>
<span class="sourceLineNo">088</span> private SplitLogManager slm;<a name="line.88"></a>
<span class="sourceLineNo">089</span> private Configuration conf;<a name="line.89"></a>
<span class="sourceLineNo">090</span> private int to;<a name="line.90"></a>
<span class="sourceLineNo">091</span><a name="line.91"></a>
<span class="sourceLineNo">092</span> private static HBaseTestingUtil TEST_UTIL;<a name="line.92"></a>
<span class="sourceLineNo">093</span><a name="line.93"></a>
<span class="sourceLineNo">094</span> class DummyMasterServices extends MockNoopMasterServices {<a name="line.94"></a>
<span class="sourceLineNo">095</span> private ZKWatcher zkw;<a name="line.95"></a>
<span class="sourceLineNo">096</span> private CoordinatedStateManager cm;<a name="line.96"></a>
<span class="sourceLineNo">097</span><a name="line.97"></a>
<span class="sourceLineNo">098</span> public DummyMasterServices(ZKWatcher zkw, Configuration conf) {<a name="line.98"></a>
<span class="sourceLineNo">099</span> super(conf);<a name="line.99"></a>
<span class="sourceLineNo">100</span> this.zkw = zkw;<a name="line.100"></a>
<span class="sourceLineNo">101</span> cm = new ZkCoordinatedStateManager(this);<a name="line.101"></a>
<span class="sourceLineNo">102</span> }<a name="line.102"></a>
<span class="sourceLineNo">103</span><a name="line.103"></a>
<span class="sourceLineNo">104</span> @Override<a name="line.104"></a>
<span class="sourceLineNo">105</span> public ZKWatcher getZooKeeper() {<a name="line.105"></a>
<span class="sourceLineNo">106</span> return zkw;<a name="line.106"></a>
<span class="sourceLineNo">107</span> }<a name="line.107"></a>
<span class="sourceLineNo">108</span><a name="line.108"></a>
<span class="sourceLineNo">109</span> @Override<a name="line.109"></a>
<span class="sourceLineNo">110</span> public CoordinatedStateManager getCoordinatedStateManager() {<a name="line.110"></a>
<span class="sourceLineNo">111</span> return cm;<a name="line.111"></a>
<span class="sourceLineNo">112</span> }<a name="line.112"></a>
<span class="sourceLineNo">113</span><a name="line.113"></a>
<span class="sourceLineNo">114</span> @Override<a name="line.114"></a>
<span class="sourceLineNo">115</span> public ServerManager getServerManager() {<a name="line.115"></a>
<span class="sourceLineNo">116</span> return sm;<a name="line.116"></a>
<span class="sourceLineNo">117</span> }<a name="line.117"></a>
<span class="sourceLineNo">118</span> }<a name="line.118"></a>
<span class="sourceLineNo">119</span><a name="line.119"></a>
<span class="sourceLineNo">120</span> @Before<a name="line.120"></a>
<span class="sourceLineNo">121</span> public void setup() throws Exception {<a name="line.121"></a>
<span class="sourceLineNo">122</span> TEST_UTIL = new HBaseTestingUtil();<a name="line.122"></a>
<span class="sourceLineNo">123</span> TEST_UTIL.startMiniZKCluster();<a name="line.123"></a>
<span class="sourceLineNo">124</span> conf = TEST_UTIL.getConfiguration();<a name="line.124"></a>
<span class="sourceLineNo">125</span> // Use a different ZK wrapper instance for each tests.<a name="line.125"></a>
<span class="sourceLineNo">126</span> zkw =<a name="line.126"></a>
<span class="sourceLineNo">127</span> new ZKWatcher(conf, "split-log-manager-tests" + TEST_UTIL.getRandomUUID().toString(), null);<a name="line.127"></a>
<span class="sourceLineNo">128</span> master = new DummyMasterServices(zkw, conf);<a name="line.128"></a>
<span class="sourceLineNo">129</span><a name="line.129"></a>
<span class="sourceLineNo">130</span> ZKUtil.deleteChildrenRecursively(zkw, zkw.getZNodePaths().baseZNode);<a name="line.130"></a>
<span class="sourceLineNo">131</span> ZKUtil.createAndFailSilent(zkw, zkw.getZNodePaths().baseZNode);<a name="line.131"></a>
<span class="sourceLineNo">132</span> assertTrue(ZKUtil.checkExists(zkw, zkw.getZNodePaths().baseZNode) != -1);<a name="line.132"></a>
<span class="sourceLineNo">133</span> LOG.debug(zkw.getZNodePaths().baseZNode + " created");<a name="line.133"></a>
<span class="sourceLineNo">134</span> ZKUtil.createAndFailSilent(zkw, zkw.getZNodePaths().splitLogZNode);<a name="line.134"></a>
<span class="sourceLineNo">135</span> assertTrue(ZKUtil.checkExists(zkw, zkw.getZNodePaths().splitLogZNode) != -1);<a name="line.135"></a>
<span class="sourceLineNo">136</span> LOG.debug(zkw.getZNodePaths().splitLogZNode + " created");<a name="line.136"></a>
<span class="sourceLineNo">137</span><a name="line.137"></a>
<span class="sourceLineNo">138</span> resetCounters();<a name="line.138"></a>
<span class="sourceLineNo">139</span><a name="line.139"></a>
<span class="sourceLineNo">140</span> // By default, we let the test manage the error as before, so the server<a name="line.140"></a>
<span class="sourceLineNo">141</span> // does not appear as dead from the master point of view, only from the split log pov.<a name="line.141"></a>
<span class="sourceLineNo">142</span> Mockito.when(sm.isServerOnline(Mockito.any())).thenReturn(true);<a name="line.142"></a>
<span class="sourceLineNo">143</span><a name="line.143"></a>
<span class="sourceLineNo">144</span> to = 12000;<a name="line.144"></a>
<span class="sourceLineNo">145</span> conf.setInt(HConstants.HBASE_SPLITLOG_MANAGER_TIMEOUT, to);<a name="line.145"></a>
<span class="sourceLineNo">146</span> conf.setInt("hbase.splitlog.manager.unassigned.timeout", 2 * to);<a name="line.146"></a>
<span class="sourceLineNo">147</span><a name="line.147"></a>
<span class="sourceLineNo">148</span> conf.setInt("hbase.splitlog.manager.timeoutmonitor.period", 100);<a name="line.148"></a>
<span class="sourceLineNo">149</span> to = to + 16 * 100;<a name="line.149"></a>
<span class="sourceLineNo">150</span> }<a name="line.150"></a>
<span class="sourceLineNo">151</span><a name="line.151"></a>
<span class="sourceLineNo">152</span> @After<a name="line.152"></a>
<span class="sourceLineNo">153</span> public void teardown() throws IOException, KeeperException {<a name="line.153"></a>
<span class="sourceLineNo">154</span> master.stop("");<a name="line.154"></a>
<span class="sourceLineNo">155</span> if (slm != null) {<a name="line.155"></a>
<span class="sourceLineNo">156</span> slm.stop();<a name="line.156"></a>
<span class="sourceLineNo">157</span> }<a name="line.157"></a>
<span class="sourceLineNo">158</span> TEST_UTIL.shutdownMiniZKCluster();<a name="line.158"></a>
<span class="sourceLineNo">159</span> }<a name="line.159"></a>
<span class="sourceLineNo">160</span><a name="line.160"></a>
<span class="sourceLineNo">161</span> @Test<a name="line.161"></a>
<span class="sourceLineNo">162</span> public void testBatchWaitMillis() {<a name="line.162"></a>
<span class="sourceLineNo">163</span> assertEquals(100, SplitLogManager.getBatchWaitTimeMillis(0));<a name="line.163"></a>
<span class="sourceLineNo">164</span> assertEquals(100, SplitLogManager.getBatchWaitTimeMillis(1));<a name="line.164"></a>
<span class="sourceLineNo">165</span> assertEquals(1000, SplitLogManager.getBatchWaitTimeMillis(10));<a name="line.165"></a>
<span class="sourceLineNo">166</span> assertEquals(60_000, SplitLogManager.getBatchWaitTimeMillis(101));<a name="line.166"></a>
<span class="sourceLineNo">167</span> assertEquals(60_000, SplitLogManager.getBatchWaitTimeMillis(1011));<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> private interface Expr {<a name="line.170"></a>
<span class="sourceLineNo">171</span> long eval();<a name="line.171"></a>
<span class="sourceLineNo">172</span> }<a name="line.172"></a>
<span class="sourceLineNo">173</span><a name="line.173"></a>
<span class="sourceLineNo">174</span> private void waitForCounter(final LongAdder ctr, long oldval, long newval, long timems)<a name="line.174"></a>
<span class="sourceLineNo">175</span> throws Exception {<a name="line.175"></a>
<span class="sourceLineNo">176</span> Expr e = new Expr() {<a name="line.176"></a>
<span class="sourceLineNo">177</span> @Override<a name="line.177"></a>
<span class="sourceLineNo">178</span> public long eval() {<a name="line.178"></a>
<span class="sourceLineNo">179</span> return ctr.sum();<a name="line.179"></a>
<span class="sourceLineNo">180</span> }<a name="line.180"></a>
<span class="sourceLineNo">181</span> };<a name="line.181"></a>
<span class="sourceLineNo">182</span> waitForCounter(e, oldval, newval, timems);<a name="line.182"></a>
<span class="sourceLineNo">183</span> return;<a name="line.183"></a>
<span class="sourceLineNo">184</span> }<a name="line.184"></a>
<span class="sourceLineNo">185</span><a name="line.185"></a>
<span class="sourceLineNo">186</span> private void waitForCounter(final Expr e, final long oldval, long newval, long timems)<a name="line.186"></a>
<span class="sourceLineNo">187</span> throws Exception {<a name="line.187"></a>
<span class="sourceLineNo">188</span><a name="line.188"></a>
<span class="sourceLineNo">189</span> TEST_UTIL.waitFor(timems, 10, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.189"></a>
<span class="sourceLineNo">190</span> @Override<a name="line.190"></a>
<span class="sourceLineNo">191</span> public boolean evaluate() throws Exception {<a name="line.191"></a>
<span class="sourceLineNo">192</span> return (e.eval() != oldval);<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 name="line.195"></a>
<span class="sourceLineNo">196</span> assertEquals(newval, e.eval());<a name="line.196"></a>
<span class="sourceLineNo">197</span> }<a name="line.197"></a>
<span class="sourceLineNo">198</span><a name="line.198"></a>
<span class="sourceLineNo">199</span> private Task findOrCreateOrphanTask(String path) {<a name="line.199"></a>
<span class="sourceLineNo">200</span> return slm.tasks.computeIfAbsent(path, k -&gt; {<a name="line.200"></a>
<span class="sourceLineNo">201</span> LOG.info("creating orphan task " + k);<a name="line.201"></a>
<span class="sourceLineNo">202</span> SplitLogCounters.tot_mgr_orphan_task_acquired.increment();<a name="line.202"></a>
<span class="sourceLineNo">203</span> return new Task();<a name="line.203"></a>
<span class="sourceLineNo">204</span> });<a name="line.204"></a>
<span class="sourceLineNo">205</span> }<a name="line.205"></a>
<span class="sourceLineNo">206</span><a name="line.206"></a>
<span class="sourceLineNo">207</span> private String submitTaskAndWait(TaskBatch batch, String name)<a name="line.207"></a>
<span class="sourceLineNo">208</span> throws KeeperException, InterruptedException {<a name="line.208"></a>
<span class="sourceLineNo">209</span> String tasknode = ZKSplitLog.getEncodedNodeName(zkw, name);<a name="line.209"></a>
<span class="sourceLineNo">210</span> NodeCreationListener listener = new NodeCreationListener(zkw, tasknode);<a name="line.210"></a>
<span class="sourceLineNo">211</span> zkw.registerListener(listener);<a name="line.211"></a>
<span class="sourceLineNo">212</span> ZKUtil.watchAndCheckExists(zkw, tasknode);<a name="line.212"></a>
<span class="sourceLineNo">213</span><a name="line.213"></a>
<span class="sourceLineNo">214</span> slm.enqueueSplitTask(name, batch);<a name="line.214"></a>
<span class="sourceLineNo">215</span> assertEquals(1, batch.installed);<a name="line.215"></a>
<span class="sourceLineNo">216</span> assertTrue(findOrCreateOrphanTask(tasknode).batch == batch);<a name="line.216"></a>
<span class="sourceLineNo">217</span> assertEquals(1L, tot_mgr_node_create_queued.sum());<a name="line.217"></a>
<span class="sourceLineNo">218</span><a name="line.218"></a>
<span class="sourceLineNo">219</span> LOG.debug("waiting for task node creation");<a name="line.219"></a>
<span class="sourceLineNo">220</span> listener.waitForCreation();<a name="line.220"></a>
<span class="sourceLineNo">221</span> LOG.debug("task created");<a name="line.221"></a>
<span class="sourceLineNo">222</span> return tasknode;<a name="line.222"></a>
<span class="sourceLineNo">223</span> }<a name="line.223"></a>
<span class="sourceLineNo">224</span><a name="line.224"></a>
<span class="sourceLineNo">225</span> /**<a name="line.225"></a>
<span class="sourceLineNo">226</span> * Test whether the splitlog correctly creates a task in zookeeper<a name="line.226"></a>
<span class="sourceLineNo">227</span> */<a name="line.227"></a>
<span class="sourceLineNo">228</span> @Test<a name="line.228"></a>
<span class="sourceLineNo">229</span> public void testTaskCreation() throws Exception {<a name="line.229"></a>
<span class="sourceLineNo">230</span><a name="line.230"></a>
<span class="sourceLineNo">231</span> LOG.info("TestTaskCreation - test the creation of a task in zk");<a name="line.231"></a>
<span class="sourceLineNo">232</span> slm = new SplitLogManager(master, conf);<a name="line.232"></a>
<span class="sourceLineNo">233</span> TaskBatch batch = new TaskBatch();<a name="line.233"></a>
<span class="sourceLineNo">234</span><a name="line.234"></a>
<span class="sourceLineNo">235</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.235"></a>
<span class="sourceLineNo">236</span><a name="line.236"></a>
<span class="sourceLineNo">237</span> byte[] data = ZKUtil.getData(zkw, tasknode);<a name="line.237"></a>
<span class="sourceLineNo">238</span> SplitLogTask slt = SplitLogTask.parseFrom(data);<a name="line.238"></a>
<span class="sourceLineNo">239</span> LOG.info("Task node created " + slt.toString());<a name="line.239"></a>
<span class="sourceLineNo">240</span> assertTrue(slt.isUnassigned(master.getServerName()));<a name="line.240"></a>
<span class="sourceLineNo">241</span> }<a name="line.241"></a>
<span class="sourceLineNo">242</span><a name="line.242"></a>
<span class="sourceLineNo">243</span> @Test<a name="line.243"></a>
<span class="sourceLineNo">244</span> public void testOrphanTaskAcquisition() throws Exception {<a name="line.244"></a>
<span class="sourceLineNo">245</span> LOG.info("TestOrphanTaskAcquisition");<a name="line.245"></a>
<span class="sourceLineNo">246</span><a name="line.246"></a>
<span class="sourceLineNo">247</span> String tasknode = ZKSplitLog.getEncodedNodeName(zkw, "orphan/test/slash");<a name="line.247"></a>
<span class="sourceLineNo">248</span> SplitLogTask slt = new SplitLogTask.Owned(master.getServerName());<a name="line.248"></a>
<span class="sourceLineNo">249</span> zkw.getRecoverableZooKeeper().create(tasknode, slt.toByteArray(), Ids.OPEN_ACL_UNSAFE,<a name="line.249"></a>
<span class="sourceLineNo">250</span> CreateMode.PERSISTENT);<a name="line.250"></a>
<span class="sourceLineNo">251</span><a name="line.251"></a>
<span class="sourceLineNo">252</span> slm = new SplitLogManager(master, conf);<a name="line.252"></a>
<span class="sourceLineNo">253</span> waitForCounter(tot_mgr_orphan_task_acquired, 0, 1, to / 2);<a name="line.253"></a>
<span class="sourceLineNo">254</span> Task task = findOrCreateOrphanTask(tasknode);<a name="line.254"></a>
<span class="sourceLineNo">255</span> assertTrue(task.isOrphan());<a name="line.255"></a>
<span class="sourceLineNo">256</span> waitForCounter(tot_mgr_heartbeat, 0, 1, to / 2);<a name="line.256"></a>
<span class="sourceLineNo">257</span> assertFalse(task.isUnassigned());<a name="line.257"></a>
<span class="sourceLineNo">258</span> long curt = EnvironmentEdgeManager.currentTime();<a name="line.258"></a>
<span class="sourceLineNo">259</span> assertTrue((task.last_update &lt;= curt) &amp;&amp; (task.last_update &gt; (curt - 1000)));<a name="line.259"></a>
<span class="sourceLineNo">260</span> LOG.info("waiting for manager to resubmit the orphan task");<a name="line.260"></a>
<span class="sourceLineNo">261</span> waitForCounter(tot_mgr_resubmit, 0, 1, to + to / 2);<a name="line.261"></a>
<span class="sourceLineNo">262</span> assertTrue(task.isUnassigned());<a name="line.262"></a>
<span class="sourceLineNo">263</span> waitForCounter(tot_mgr_rescan, 0, 1, to + to / 2);<a name="line.263"></a>
<span class="sourceLineNo">264</span> }<a name="line.264"></a>
<span class="sourceLineNo">265</span><a name="line.265"></a>
<span class="sourceLineNo">266</span> @Test<a name="line.266"></a>
<span class="sourceLineNo">267</span> public void testUnassignedOrphan() throws Exception {<a name="line.267"></a>
<span class="sourceLineNo">268</span> LOG.info("TestUnassignedOrphan - an unassigned task is resubmitted at" + " startup");<a name="line.268"></a>
<span class="sourceLineNo">269</span> String tasknode = ZKSplitLog.getEncodedNodeName(zkw, "orphan/test/slash");<a name="line.269"></a>
<span class="sourceLineNo">270</span> // create an unassigned orphan task<a name="line.270"></a>
<span class="sourceLineNo">271</span> SplitLogTask slt = new SplitLogTask.Unassigned(master.getServerName());<a name="line.271"></a>
<span class="sourceLineNo">272</span> zkw.getRecoverableZooKeeper().create(tasknode, slt.toByteArray(), Ids.OPEN_ACL_UNSAFE,<a name="line.272"></a>
<span class="sourceLineNo">273</span> CreateMode.PERSISTENT);<a name="line.273"></a>
<span class="sourceLineNo">274</span> int version = ZKUtil.checkExists(zkw, tasknode);<a name="line.274"></a>
<span class="sourceLineNo">275</span><a name="line.275"></a>
<span class="sourceLineNo">276</span> slm = new SplitLogManager(master, conf);<a name="line.276"></a>
<span class="sourceLineNo">277</span> waitForCounter(tot_mgr_orphan_task_acquired, 0, 1, to / 2);<a name="line.277"></a>
<span class="sourceLineNo">278</span> Task task = findOrCreateOrphanTask(tasknode);<a name="line.278"></a>
<span class="sourceLineNo">279</span> assertTrue(task.isOrphan());<a name="line.279"></a>
<span class="sourceLineNo">280</span> assertTrue(task.isUnassigned());<a name="line.280"></a>
<span class="sourceLineNo">281</span> // wait for RESCAN node to be created<a name="line.281"></a>
<span class="sourceLineNo">282</span> waitForCounter(tot_mgr_rescan, 0, 1, to / 2);<a name="line.282"></a>
<span class="sourceLineNo">283</span> Task task2 = findOrCreateOrphanTask(tasknode);<a name="line.283"></a>
<span class="sourceLineNo">284</span> assertTrue(task == task2);<a name="line.284"></a>
<span class="sourceLineNo">285</span> LOG.debug("task = " + task);<a name="line.285"></a>
<span class="sourceLineNo">286</span> assertEquals(1L, tot_mgr_resubmit.sum());<a name="line.286"></a>
<span class="sourceLineNo">287</span> assertEquals(1, task.incarnation.get());<a name="line.287"></a>
<span class="sourceLineNo">288</span> assertEquals(0, task.unforcedResubmits.get());<a name="line.288"></a>
<span class="sourceLineNo">289</span> assertTrue(task.isOrphan());<a name="line.289"></a>
<span class="sourceLineNo">290</span> assertTrue(task.isUnassigned());<a name="line.290"></a>
<span class="sourceLineNo">291</span> assertTrue(ZKUtil.checkExists(zkw, tasknode) &gt; version);<a name="line.291"></a>
<span class="sourceLineNo">292</span> }<a name="line.292"></a>
<span class="sourceLineNo">293</span><a name="line.293"></a>
<span class="sourceLineNo">294</span> @Test<a name="line.294"></a>
<span class="sourceLineNo">295</span> public void testMultipleResubmits() throws Exception {<a name="line.295"></a>
<span class="sourceLineNo">296</span> LOG.info("TestMultipleResbmits - no indefinite resubmissions");<a name="line.296"></a>
<span class="sourceLineNo">297</span> conf.setInt("hbase.splitlog.max.resubmit", 2);<a name="line.297"></a>
<span class="sourceLineNo">298</span> slm = new SplitLogManager(master, conf);<a name="line.298"></a>
<span class="sourceLineNo">299</span> TaskBatch batch = new TaskBatch();<a name="line.299"></a>
<span class="sourceLineNo">300</span><a name="line.300"></a>
<span class="sourceLineNo">301</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.301"></a>
<span class="sourceLineNo">302</span> int version = ZKUtil.checkExists(zkw, tasknode);<a name="line.302"></a>
<span class="sourceLineNo">303</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.303"></a>
<span class="sourceLineNo">304</span> final ServerName worker2 = ServerName.valueOf("worker2,1,1");<a name="line.304"></a>
<span class="sourceLineNo">305</span> final ServerName worker3 = ServerName.valueOf("worker3,1,1");<a name="line.305"></a>
<span class="sourceLineNo">306</span> SplitLogTask slt = new SplitLogTask.Owned(worker1);<a name="line.306"></a>
<span class="sourceLineNo">307</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.307"></a>
<span class="sourceLineNo">308</span> waitForCounter(tot_mgr_heartbeat, 0, 1, to / 2);<a name="line.308"></a>
<span class="sourceLineNo">309</span> waitForCounter(tot_mgr_resubmit, 0, 1, to + to / 2);<a name="line.309"></a>
<span class="sourceLineNo">310</span> int version1 = ZKUtil.checkExists(zkw, tasknode);<a name="line.310"></a>
<span class="sourceLineNo">311</span> assertTrue(version1 &gt; version);<a name="line.311"></a>
<span class="sourceLineNo">312</span> slt = new SplitLogTask.Owned(worker2);<a name="line.312"></a>
<span class="sourceLineNo">313</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.313"></a>
<span class="sourceLineNo">314</span> waitForCounter(tot_mgr_heartbeat, 1, 2, to / 2);<a name="line.314"></a>
<span class="sourceLineNo">315</span> waitForCounter(tot_mgr_resubmit, 1, 2, to + to / 2);<a name="line.315"></a>
<span class="sourceLineNo">316</span> int version2 = ZKUtil.checkExists(zkw, tasknode);<a name="line.316"></a>
<span class="sourceLineNo">317</span> assertTrue(version2 &gt; version1);<a name="line.317"></a>
<span class="sourceLineNo">318</span> slt = new SplitLogTask.Owned(worker3);<a name="line.318"></a>
<span class="sourceLineNo">319</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.319"></a>
<span class="sourceLineNo">320</span> waitForCounter(tot_mgr_heartbeat, 2, 3, to / 2);<a name="line.320"></a>
<span class="sourceLineNo">321</span> waitForCounter(tot_mgr_resubmit_threshold_reached, 0, 1, to + to / 2);<a name="line.321"></a>
<span class="sourceLineNo">322</span> Thread.sleep(to + to / 2);<a name="line.322"></a>
<span class="sourceLineNo">323</span> assertEquals(2L, tot_mgr_resubmit.sum() - tot_mgr_resubmit_force.sum());<a name="line.323"></a>
<span class="sourceLineNo">324</span> }<a name="line.324"></a>
<span class="sourceLineNo">325</span><a name="line.325"></a>
<span class="sourceLineNo">326</span> @Test<a name="line.326"></a>
<span class="sourceLineNo">327</span> public void testRescanCleanup() throws Exception {<a name="line.327"></a>
<span class="sourceLineNo">328</span> LOG.info("TestRescanCleanup - ensure RESCAN nodes are cleaned up");<a name="line.328"></a>
<span class="sourceLineNo">329</span><a name="line.329"></a>
<span class="sourceLineNo">330</span> slm = new SplitLogManager(master, conf);<a name="line.330"></a>
<span class="sourceLineNo">331</span> TaskBatch batch = new TaskBatch();<a name="line.331"></a>
<span class="sourceLineNo">332</span><a name="line.332"></a>
<span class="sourceLineNo">333</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.333"></a>
<span class="sourceLineNo">334</span> int version = ZKUtil.checkExists(zkw, tasknode);<a name="line.334"></a>
<span class="sourceLineNo">335</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.335"></a>
<span class="sourceLineNo">336</span> SplitLogTask slt = new SplitLogTask.Owned(worker1);<a name="line.336"></a>
<span class="sourceLineNo">337</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.337"></a>
<span class="sourceLineNo">338</span> waitForCounter(tot_mgr_heartbeat, 0, 1, to / 2);<a name="line.338"></a>
<span class="sourceLineNo">339</span> waitForCounter(new Expr() {<a name="line.339"></a>
<span class="sourceLineNo">340</span> @Override<a name="line.340"></a>
<span class="sourceLineNo">341</span> public long eval() {<a name="line.341"></a>
<span class="sourceLineNo">342</span> return (tot_mgr_resubmit.sum() + tot_mgr_resubmit_failed.sum());<a name="line.342"></a>
<span class="sourceLineNo">343</span> }<a name="line.343"></a>
<span class="sourceLineNo">344</span> }, 0, 1, 5 * 60000); // wait long enough<a name="line.344"></a>
<span class="sourceLineNo">345</span> Assert.assertEquals("Could not run test. Lost ZK connection?", 0,<a name="line.345"></a>
<span class="sourceLineNo">346</span> tot_mgr_resubmit_failed.sum());<a name="line.346"></a>
<span class="sourceLineNo">347</span> int version1 = ZKUtil.checkExists(zkw, tasknode);<a name="line.347"></a>
<span class="sourceLineNo">348</span> assertTrue(version1 &gt; version);<a name="line.348"></a>
<span class="sourceLineNo">349</span> byte[] taskstate = ZKUtil.getData(zkw, tasknode);<a name="line.349"></a>
<span class="sourceLineNo">350</span> slt = SplitLogTask.parseFrom(taskstate);<a name="line.350"></a>
<span class="sourceLineNo">351</span> assertTrue(slt.isUnassigned(master.getServerName()));<a name="line.351"></a>
<span class="sourceLineNo">352</span><a name="line.352"></a>
<span class="sourceLineNo">353</span> waitForCounter(tot_mgr_rescan_deleted, 0, 1, to / 2);<a name="line.353"></a>
<span class="sourceLineNo">354</span> }<a name="line.354"></a>
<span class="sourceLineNo">355</span><a name="line.355"></a>
<span class="sourceLineNo">356</span> @Test<a name="line.356"></a>
<span class="sourceLineNo">357</span> public void testTaskDone() throws Exception {<a name="line.357"></a>
<span class="sourceLineNo">358</span> LOG.info("TestTaskDone - cleanup task node once in DONE state");<a name="line.358"></a>
<span class="sourceLineNo">359</span><a name="line.359"></a>
<span class="sourceLineNo">360</span> slm = new SplitLogManager(master, conf);<a name="line.360"></a>
<span class="sourceLineNo">361</span> TaskBatch batch = new TaskBatch();<a name="line.361"></a>
<span class="sourceLineNo">362</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.362"></a>
<span class="sourceLineNo">363</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.363"></a>
<span class="sourceLineNo">364</span> SplitLogTask slt = new SplitLogTask.Done(worker1);<a name="line.364"></a>
<span class="sourceLineNo">365</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.365"></a>
<span class="sourceLineNo">366</span> synchronized (batch) {<a name="line.366"></a>
<span class="sourceLineNo">367</span> while (batch.installed != batch.done) {<a name="line.367"></a>
<span class="sourceLineNo">368</span> batch.wait();<a name="line.368"></a>
<span class="sourceLineNo">369</span> }<a name="line.369"></a>
<span class="sourceLineNo">370</span> }<a name="line.370"></a>
<span class="sourceLineNo">371</span> waitForCounter(tot_mgr_task_deleted, 0, 1, to / 2);<a name="line.371"></a>
<span class="sourceLineNo">372</span> assertTrue(ZKUtil.checkExists(zkw, tasknode) == -1);<a name="line.372"></a>
<span class="sourceLineNo">373</span> }<a name="line.373"></a>
<span class="sourceLineNo">374</span><a name="line.374"></a>
<span class="sourceLineNo">375</span> @Test<a name="line.375"></a>
<span class="sourceLineNo">376</span> public void testTaskErr() throws Exception {<a name="line.376"></a>
<span class="sourceLineNo">377</span> LOG.info("TestTaskErr - cleanup task node once in ERR state");<a name="line.377"></a>
<span class="sourceLineNo">378</span><a name="line.378"></a>
<span class="sourceLineNo">379</span> conf.setInt("hbase.splitlog.max.resubmit", 0);<a name="line.379"></a>
<span class="sourceLineNo">380</span> slm = new SplitLogManager(master, conf);<a name="line.380"></a>
<span class="sourceLineNo">381</span> TaskBatch batch = new TaskBatch();<a name="line.381"></a>
<span class="sourceLineNo">382</span><a name="line.382"></a>
<span class="sourceLineNo">383</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.383"></a>
<span class="sourceLineNo">384</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.384"></a>
<span class="sourceLineNo">385</span> SplitLogTask slt = new SplitLogTask.Err(worker1);<a name="line.385"></a>
<span class="sourceLineNo">386</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.386"></a>
<span class="sourceLineNo">387</span><a name="line.387"></a>
<span class="sourceLineNo">388</span> synchronized (batch) {<a name="line.388"></a>
<span class="sourceLineNo">389</span> while (batch.installed != batch.error) {<a name="line.389"></a>
<span class="sourceLineNo">390</span> batch.wait();<a name="line.390"></a>
<span class="sourceLineNo">391</span> }<a name="line.391"></a>
<span class="sourceLineNo">392</span> }<a name="line.392"></a>
<span class="sourceLineNo">393</span> waitForCounter(tot_mgr_task_deleted, 0, 1, to / 2);<a name="line.393"></a>
<span class="sourceLineNo">394</span> assertTrue(ZKUtil.checkExists(zkw, tasknode) == -1);<a name="line.394"></a>
<span class="sourceLineNo">395</span> conf.setInt("hbase.splitlog.max.resubmit", ZKSplitLogManagerCoordination.DEFAULT_MAX_RESUBMIT);<a name="line.395"></a>
<span class="sourceLineNo">396</span> }<a name="line.396"></a>
<span class="sourceLineNo">397</span><a name="line.397"></a>
<span class="sourceLineNo">398</span> @Test<a name="line.398"></a>
<span class="sourceLineNo">399</span> public void testTaskResigned() throws Exception {<a name="line.399"></a>
<span class="sourceLineNo">400</span> LOG.info("TestTaskResigned - resubmit task node once in RESIGNED state");<a name="line.400"></a>
<span class="sourceLineNo">401</span> assertEquals(0, tot_mgr_resubmit.sum());<a name="line.401"></a>
<span class="sourceLineNo">402</span> slm = new SplitLogManager(master, conf);<a name="line.402"></a>
<span class="sourceLineNo">403</span> assertEquals(0, tot_mgr_resubmit.sum());<a name="line.403"></a>
<span class="sourceLineNo">404</span> TaskBatch batch = new TaskBatch();<a name="line.404"></a>
<span class="sourceLineNo">405</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.405"></a>
<span class="sourceLineNo">406</span> assertEquals(0, tot_mgr_resubmit.sum());<a name="line.406"></a>
<span class="sourceLineNo">407</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.407"></a>
<span class="sourceLineNo">408</span> assertEquals(0, tot_mgr_resubmit.sum());<a name="line.408"></a>
<span class="sourceLineNo">409</span> SplitLogTask slt = new SplitLogTask.Resigned(worker1);<a name="line.409"></a>
<span class="sourceLineNo">410</span> assertEquals(0, tot_mgr_resubmit.sum());<a name="line.410"></a>
<span class="sourceLineNo">411</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.411"></a>
<span class="sourceLineNo">412</span> ZKUtil.checkExists(zkw, tasknode);<a name="line.412"></a>
<span class="sourceLineNo">413</span> // Could be small race here.<a name="line.413"></a>
<span class="sourceLineNo">414</span> if (tot_mgr_resubmit.sum() == 0) {<a name="line.414"></a>
<span class="sourceLineNo">415</span> waitForCounter(tot_mgr_resubmit, 0, 1, to / 2);<a name="line.415"></a>
<span class="sourceLineNo">416</span> }<a name="line.416"></a>
<span class="sourceLineNo">417</span> assertEquals(1, tot_mgr_resubmit.sum());<a name="line.417"></a>
<span class="sourceLineNo">418</span><a name="line.418"></a>
<span class="sourceLineNo">419</span> byte[] taskstate = ZKUtil.getData(zkw, tasknode);<a name="line.419"></a>
<span class="sourceLineNo">420</span> slt = SplitLogTask.parseFrom(taskstate);<a name="line.420"></a>
<span class="sourceLineNo">421</span> assertTrue(slt.isUnassigned(master.getServerName()));<a name="line.421"></a>
<span class="sourceLineNo">422</span> }<a name="line.422"></a>
<span class="sourceLineNo">423</span><a name="line.423"></a>
<span class="sourceLineNo">424</span> @Test<a name="line.424"></a>
<span class="sourceLineNo">425</span> public void testUnassignedTimeout() throws Exception {<a name="line.425"></a>
<span class="sourceLineNo">426</span> LOG.info("TestUnassignedTimeout - iff all tasks are unassigned then" + " resubmit");<a name="line.426"></a>
<span class="sourceLineNo">427</span><a name="line.427"></a>
<span class="sourceLineNo">428</span> // create an orphan task in OWNED state<a name="line.428"></a>
<span class="sourceLineNo">429</span> String tasknode1 = ZKSplitLog.getEncodedNodeName(zkw, "orphan/1");<a name="line.429"></a>
<span class="sourceLineNo">430</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.430"></a>
<span class="sourceLineNo">431</span> SplitLogTask slt = new SplitLogTask.Owned(worker1);<a name="line.431"></a>
<span class="sourceLineNo">432</span> zkw.getRecoverableZooKeeper().create(tasknode1, slt.toByteArray(), Ids.OPEN_ACL_UNSAFE,<a name="line.432"></a>
<span class="sourceLineNo">433</span> CreateMode.PERSISTENT);<a name="line.433"></a>
<span class="sourceLineNo">434</span><a name="line.434"></a>
<span class="sourceLineNo">435</span> slm = new SplitLogManager(master, conf);<a name="line.435"></a>
<span class="sourceLineNo">436</span> waitForCounter(tot_mgr_orphan_task_acquired, 0, 1, to / 2);<a name="line.436"></a>
<span class="sourceLineNo">437</span><a name="line.437"></a>
<span class="sourceLineNo">438</span> // submit another task which will stay in unassigned mode<a name="line.438"></a>
<span class="sourceLineNo">439</span> TaskBatch batch = new TaskBatch();<a name="line.439"></a>
<span class="sourceLineNo">440</span> submitTaskAndWait(batch, "foo/1");<a name="line.440"></a>
<span class="sourceLineNo">441</span><a name="line.441"></a>
<span class="sourceLineNo">442</span> // keep updating the orphan owned node every to/2 seconds<a name="line.442"></a>
<span class="sourceLineNo">443</span> for (int i = 0; i &lt; (3 * to) / 100; i++) {<a name="line.443"></a>
<span class="sourceLineNo">444</span> Thread.sleep(100);<a name="line.444"></a>
<span class="sourceLineNo">445</span> final ServerName worker2 = ServerName.valueOf("worker1,1,1");<a name="line.445"></a>
<span class="sourceLineNo">446</span> slt = new SplitLogTask.Owned(worker2);<a name="line.446"></a>
<span class="sourceLineNo">447</span> ZKUtil.setData(zkw, tasknode1, slt.toByteArray());<a name="line.447"></a>
<span class="sourceLineNo">448</span> }<a name="line.448"></a>
<span class="sourceLineNo">449</span><a name="line.449"></a>
<span class="sourceLineNo">450</span> // since we have stopped heartbeating the owned node therefore it should<a name="line.450"></a>
<span class="sourceLineNo">451</span> // get resubmitted<a name="line.451"></a>
<span class="sourceLineNo">452</span> LOG.info("waiting for manager to resubmit the orphan task");<a name="line.452"></a>
<span class="sourceLineNo">453</span> waitForCounter(tot_mgr_resubmit, 0, 1, to + to / 2);<a name="line.453"></a>
<span class="sourceLineNo">454</span><a name="line.454"></a>
<span class="sourceLineNo">455</span> // now all the nodes are unassigned. manager should post another rescan<a name="line.455"></a>
<span class="sourceLineNo">456</span> waitForCounter(tot_mgr_resubmit_unassigned, 0, 1, 2 * to + to / 2);<a name="line.456"></a>
<span class="sourceLineNo">457</span> }<a name="line.457"></a>
<span class="sourceLineNo">458</span><a name="line.458"></a>
<span class="sourceLineNo">459</span> @Test<a name="line.459"></a>
<span class="sourceLineNo">460</span> public void testDeadWorker() throws Exception {<a name="line.460"></a>
<span class="sourceLineNo">461</span> LOG.info("testDeadWorker");<a name="line.461"></a>
<span class="sourceLineNo">462</span><a name="line.462"></a>
<span class="sourceLineNo">463</span> conf.setLong("hbase.splitlog.max.resubmit", 0);<a name="line.463"></a>
<span class="sourceLineNo">464</span> slm = new SplitLogManager(master, conf);<a name="line.464"></a>
<span class="sourceLineNo">465</span> TaskBatch batch = new TaskBatch();<a name="line.465"></a>
<span class="sourceLineNo">466</span><a name="line.466"></a>
<span class="sourceLineNo">467</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.467"></a>
<span class="sourceLineNo">468</span> int version = ZKUtil.checkExists(zkw, tasknode);<a name="line.468"></a>
<span class="sourceLineNo">469</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.469"></a>
<span class="sourceLineNo">470</span> SplitLogTask slt = new SplitLogTask.Owned(worker1);<a name="line.470"></a>
<span class="sourceLineNo">471</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.471"></a>
<span class="sourceLineNo">472</span> if (tot_mgr_heartbeat.sum() == 0) {<a name="line.472"></a>
<span class="sourceLineNo">473</span> waitForCounter(tot_mgr_heartbeat, 0, 1, to / 2);<a name="line.473"></a>
<span class="sourceLineNo">474</span> }<a name="line.474"></a>
<span class="sourceLineNo">475</span> slm.handleDeadWorker(worker1);<a name="line.475"></a>
<span class="sourceLineNo">476</span> if (tot_mgr_resubmit.sum() == 0) {<a name="line.476"></a>
<span class="sourceLineNo">477</span> waitForCounter(tot_mgr_resubmit, 0, 1, to + to / 2);<a name="line.477"></a>
<span class="sourceLineNo">478</span> }<a name="line.478"></a>
<span class="sourceLineNo">479</span> if (tot_mgr_resubmit_dead_server_task.sum() == 0) {<a name="line.479"></a>
<span class="sourceLineNo">480</span> waitForCounter(tot_mgr_resubmit_dead_server_task, 0, 1, to + to / 2);<a name="line.480"></a>
<span class="sourceLineNo">481</span> }<a name="line.481"></a>
<span class="sourceLineNo">482</span><a name="line.482"></a>
<span class="sourceLineNo">483</span> int version1 = ZKUtil.checkExists(zkw, tasknode);<a name="line.483"></a>
<span class="sourceLineNo">484</span> assertTrue(version1 &gt; version);<a name="line.484"></a>
<span class="sourceLineNo">485</span> byte[] taskstate = ZKUtil.getData(zkw, tasknode);<a name="line.485"></a>
<span class="sourceLineNo">486</span> slt = SplitLogTask.parseFrom(taskstate);<a name="line.486"></a>
<span class="sourceLineNo">487</span> assertTrue(slt.isUnassigned(master.getServerName()));<a name="line.487"></a>
<span class="sourceLineNo">488</span> return;<a name="line.488"></a>
<span class="sourceLineNo">489</span> }<a name="line.489"></a>
<span class="sourceLineNo">490</span><a name="line.490"></a>
<span class="sourceLineNo">491</span> @Test<a name="line.491"></a>
<span class="sourceLineNo">492</span> public void testWorkerCrash() throws Exception {<a name="line.492"></a>
<span class="sourceLineNo">493</span> slm = new SplitLogManager(master, conf);<a name="line.493"></a>
<span class="sourceLineNo">494</span> TaskBatch batch = new TaskBatch();<a name="line.494"></a>
<span class="sourceLineNo">495</span><a name="line.495"></a>
<span class="sourceLineNo">496</span> String tasknode = submitTaskAndWait(batch, "foo/1");<a name="line.496"></a>
<span class="sourceLineNo">497</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.497"></a>
<span class="sourceLineNo">498</span><a name="line.498"></a>
<span class="sourceLineNo">499</span> SplitLogTask slt = new SplitLogTask.Owned(worker1);<a name="line.499"></a>
<span class="sourceLineNo">500</span> ZKUtil.setData(zkw, tasknode, slt.toByteArray());<a name="line.500"></a>
<span class="sourceLineNo">501</span> if (tot_mgr_heartbeat.sum() == 0) {<a name="line.501"></a>
<span class="sourceLineNo">502</span> waitForCounter(tot_mgr_heartbeat, 0, 1, to / 2);<a name="line.502"></a>
<span class="sourceLineNo">503</span> }<a name="line.503"></a>
<span class="sourceLineNo">504</span><a name="line.504"></a>
<span class="sourceLineNo">505</span> // Not yet resubmitted.<a name="line.505"></a>
<span class="sourceLineNo">506</span> Assert.assertEquals(0, tot_mgr_resubmit.sum());<a name="line.506"></a>
<span class="sourceLineNo">507</span><a name="line.507"></a>
<span class="sourceLineNo">508</span> // This server becomes dead<a name="line.508"></a>
<span class="sourceLineNo">509</span> Mockito.when(sm.isServerOnline(worker1)).thenReturn(false);<a name="line.509"></a>
<span class="sourceLineNo">510</span><a name="line.510"></a>
<span class="sourceLineNo">511</span> Thread.sleep(1300); // The timeout checker is done every 1000 ms (hardcoded).<a name="line.511"></a>
<span class="sourceLineNo">512</span><a name="line.512"></a>
<span class="sourceLineNo">513</span> // It has been resubmitted<a name="line.513"></a>
<span class="sourceLineNo">514</span> Assert.assertEquals(1, tot_mgr_resubmit.sum());<a name="line.514"></a>
<span class="sourceLineNo">515</span> }<a name="line.515"></a>
<span class="sourceLineNo">516</span><a name="line.516"></a>
<span class="sourceLineNo">517</span> @Test<a name="line.517"></a>
<span class="sourceLineNo">518</span> public void testEmptyLogDir() throws Exception {<a name="line.518"></a>
<span class="sourceLineNo">519</span> LOG.info("testEmptyLogDir");<a name="line.519"></a>
<span class="sourceLineNo">520</span> slm = new SplitLogManager(master, conf);<a name="line.520"></a>
<span class="sourceLineNo">521</span> FileSystem fs = TEST_UTIL.getTestFileSystem();<a name="line.521"></a>
<span class="sourceLineNo">522</span> Path emptyLogDirPath =<a name="line.522"></a>
<span class="sourceLineNo">523</span> new Path(new Path(fs.getWorkingDirectory(), HConstants.HREGION_LOGDIR_NAME),<a name="line.523"></a>
<span class="sourceLineNo">524</span> ServerName.valueOf("emptyLogDir", 1, 1).toString());<a name="line.524"></a>
<span class="sourceLineNo">525</span> fs.mkdirs(emptyLogDirPath);<a name="line.525"></a>
<span class="sourceLineNo">526</span> slm.splitLogDistributed(emptyLogDirPath);<a name="line.526"></a>
<span class="sourceLineNo">527</span> assertFalse(fs.exists(emptyLogDirPath));<a name="line.527"></a>
<span class="sourceLineNo">528</span> }<a name="line.528"></a>
<span class="sourceLineNo">529</span><a name="line.529"></a>
<span class="sourceLineNo">530</span> @Test<a name="line.530"></a>
<span class="sourceLineNo">531</span> public void testLogFilesAreArchived() throws Exception {<a name="line.531"></a>
<span class="sourceLineNo">532</span> LOG.info("testLogFilesAreArchived");<a name="line.532"></a>
<span class="sourceLineNo">533</span> slm = new SplitLogManager(master, conf);<a name="line.533"></a>
<span class="sourceLineNo">534</span> FileSystem fs = TEST_UTIL.getTestFileSystem();<a name="line.534"></a>
<span class="sourceLineNo">535</span> Path dir = TEST_UTIL.getDataTestDirOnTestFS("testLogFilesAreArchived");<a name="line.535"></a>
<span class="sourceLineNo">536</span> conf.set(HConstants.HBASE_DIR, dir.toString());<a name="line.536"></a>
<span class="sourceLineNo">537</span> String serverName = ServerName.valueOf("foo", 1, 1).toString();<a name="line.537"></a>
<span class="sourceLineNo">538</span> Path logDirPath = new Path(new Path(dir, HConstants.HREGION_LOGDIR_NAME), serverName);<a name="line.538"></a>
<span class="sourceLineNo">539</span> fs.mkdirs(logDirPath);<a name="line.539"></a>
<span class="sourceLineNo">540</span> // create an empty log file<a name="line.540"></a>
<span class="sourceLineNo">541</span> String logFile = new Path(logDirPath, TEST_UTIL.getRandomUUID().toString()).toString();<a name="line.541"></a>
<span class="sourceLineNo">542</span> fs.create(new Path(logDirPath, logFile)).close();<a name="line.542"></a>
<span class="sourceLineNo">543</span><a name="line.543"></a>
<span class="sourceLineNo">544</span> // spin up a thread mocking split done.<a name="line.544"></a>
<span class="sourceLineNo">545</span> new Thread() {<a name="line.545"></a>
<span class="sourceLineNo">546</span> @Override<a name="line.546"></a>
<span class="sourceLineNo">547</span> public void run() {<a name="line.547"></a>
<span class="sourceLineNo">548</span> boolean done = false;<a name="line.548"></a>
<span class="sourceLineNo">549</span> while (!done) {<a name="line.549"></a>
<span class="sourceLineNo">550</span> for (Map.Entry&lt;String, Task&gt; entry : slm.getTasks().entrySet()) {<a name="line.550"></a>
<span class="sourceLineNo">551</span> final ServerName worker1 = ServerName.valueOf("worker1,1,1");<a name="line.551"></a>
<span class="sourceLineNo">552</span> SplitLogTask slt = new SplitLogTask.Done(worker1);<a name="line.552"></a>
<span class="sourceLineNo">553</span> boolean encounteredZKException = false;<a name="line.553"></a>
<span class="sourceLineNo">554</span> try {<a name="line.554"></a>
<span class="sourceLineNo">555</span> ZKUtil.setData(zkw, entry.getKey(), slt.toByteArray());<a name="line.555"></a>
<span class="sourceLineNo">556</span> } catch (KeeperException e) {<a name="line.556"></a>
<span class="sourceLineNo">557</span> LOG.warn(e.toString(), e);<a name="line.557"></a>
<span class="sourceLineNo">558</span> encounteredZKException = true;<a name="line.558"></a>
<span class="sourceLineNo">559</span> }<a name="line.559"></a>
<span class="sourceLineNo">560</span> if (!encounteredZKException) {<a name="line.560"></a>
<span class="sourceLineNo">561</span> done = true;<a name="line.561"></a>
<span class="sourceLineNo">562</span> }<a name="line.562"></a>
<span class="sourceLineNo">563</span> }<a name="line.563"></a>
<span class="sourceLineNo">564</span> }<a name="line.564"></a>
<span class="sourceLineNo">565</span> }<a name="line.565"></a>
<span class="sourceLineNo">566</span> }.start();<a name="line.566"></a>
<span class="sourceLineNo">567</span><a name="line.567"></a>
<span class="sourceLineNo">568</span> slm.splitLogDistributed(logDirPath);<a name="line.568"></a>
<span class="sourceLineNo">569</span><a name="line.569"></a>
<span class="sourceLineNo">570</span> assertFalse(fs.exists(logDirPath));<a name="line.570"></a>
<span class="sourceLineNo">571</span> }<a name="line.571"></a>
<span class="sourceLineNo">572</span>}<a name="line.572"></a>
</pre>
</div>
</body>
</html>