blob: 8fdc7a8491ff2c5b27459380703ac151db9674fe [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 java.io.FileNotFoundException;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import java.io.IOException;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import java.io.InterruptedIOException;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.net.ConnectException;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.net.SocketTimeoutException;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.Optional;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.commons.io.IOUtils;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.conf.Configuration;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.fs.FSDataInputStream;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.fs.FileSystem;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.fs.Path;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.NotServingRegionException;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.Server;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.client.RetriesExhaustedException;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.coordination.SplitLogWorkerCoordination;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.regionserver.SplitLogWorker.TaskExecutor.Status;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.replication.ReplicationPeerImpl;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.replication.ReplicationUtils;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.replication.SyncReplicationState;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.util.CancelableProgressable;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.util.CommonFSUtils;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.util.ExceptionUtil;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.wal.SyncReplicationWALProvider;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.wal.WALFactory;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.wal.WALSplitter;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.slf4j.Logger;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.slf4j.LoggerFactory;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.51"></a>
<span class="sourceLineNo">052</span><a name="line.52"></a>
<span class="sourceLineNo">053</span>/**<a name="line.53"></a>
<span class="sourceLineNo">054</span> * This worker is spawned in every regionserver, including master. The Worker waits for log<a name="line.54"></a>
<span class="sourceLineNo">055</span> * splitting tasks to be put up by the {@link org.apache.hadoop.hbase.master.SplitLogManager}<a name="line.55"></a>
<span class="sourceLineNo">056</span> * running in the master and races with other workers in other serves to acquire those tasks.<a name="line.56"></a>
<span class="sourceLineNo">057</span> * The coordination is done via coordination engine.<a name="line.57"></a>
<span class="sourceLineNo">058</span> * &lt;p&gt;<a name="line.58"></a>
<span class="sourceLineNo">059</span> * If a worker has successfully moved the task from state UNASSIGNED to OWNED then it owns the task.<a name="line.59"></a>
<span class="sourceLineNo">060</span> * It keeps heart beating the manager by periodically moving the task from UNASSIGNED to OWNED<a name="line.60"></a>
<span class="sourceLineNo">061</span> * state. On success it moves the task to TASK_DONE. On unrecoverable error it moves task state to<a name="line.61"></a>
<span class="sourceLineNo">062</span> * ERR. If it cannot continue but wants the master to retry the task then it moves the task state to<a name="line.62"></a>
<span class="sourceLineNo">063</span> * RESIGNED.<a name="line.63"></a>
<span class="sourceLineNo">064</span> * &lt;p&gt;<a name="line.64"></a>
<span class="sourceLineNo">065</span> * The manager can take a task away from a worker by moving the task from OWNED to UNASSIGNED. In<a name="line.65"></a>
<span class="sourceLineNo">066</span> * the absence of a global lock there is a unavoidable race here - a worker might have just finished<a name="line.66"></a>
<span class="sourceLineNo">067</span> * its task when it is stripped of its ownership. Here we rely on the idempotency of the log<a name="line.67"></a>
<span class="sourceLineNo">068</span> * splitting task for correctness<a name="line.68"></a>
<span class="sourceLineNo">069</span> * @deprecated since 2.4.0 and in 3.0.0, to be removed in 4.0.0, replaced by procedure-based<a name="line.69"></a>
<span class="sourceLineNo">070</span> * distributed WAL splitter, see SplitWALRemoteProcedure<a name="line.70"></a>
<span class="sourceLineNo">071</span> */<a name="line.71"></a>
<span class="sourceLineNo">072</span>@Deprecated<a name="line.72"></a>
<span class="sourceLineNo">073</span>@InterfaceAudience.Private<a name="line.73"></a>
<span class="sourceLineNo">074</span>public class SplitLogWorker implements Runnable {<a name="line.74"></a>
<span class="sourceLineNo">075</span><a name="line.75"></a>
<span class="sourceLineNo">076</span> private static final Logger LOG = LoggerFactory.getLogger(SplitLogWorker.class);<a name="line.76"></a>
<span class="sourceLineNo">077</span><a name="line.77"></a>
<span class="sourceLineNo">078</span> Thread worker;<a name="line.78"></a>
<span class="sourceLineNo">079</span> // thread pool which executes recovery work<a name="line.79"></a>
<span class="sourceLineNo">080</span> private final SplitLogWorkerCoordination coordination;<a name="line.80"></a>
<span class="sourceLineNo">081</span> private final RegionServerServices server;<a name="line.81"></a>
<span class="sourceLineNo">082</span><a name="line.82"></a>
<span class="sourceLineNo">083</span> public SplitLogWorker(Server hserver, Configuration conf, RegionServerServices server,<a name="line.83"></a>
<span class="sourceLineNo">084</span> TaskExecutor splitTaskExecutor) {<a name="line.84"></a>
<span class="sourceLineNo">085</span> this.server = server;<a name="line.85"></a>
<span class="sourceLineNo">086</span> this.coordination = hserver.getCoordinatedStateManager().getSplitLogWorkerCoordination();<a name="line.86"></a>
<span class="sourceLineNo">087</span> coordination.init(server, conf, splitTaskExecutor, this);<a name="line.87"></a>
<span class="sourceLineNo">088</span> }<a name="line.88"></a>
<span class="sourceLineNo">089</span><a name="line.89"></a>
<span class="sourceLineNo">090</span> public SplitLogWorker(Configuration conf, RegionServerServices server,<a name="line.90"></a>
<span class="sourceLineNo">091</span> LastSequenceId sequenceIdChecker, WALFactory factory) {<a name="line.91"></a>
<span class="sourceLineNo">092</span> this(server, conf, server, (f, p) -&gt; splitLog(f, p, conf, server, sequenceIdChecker, factory));<a name="line.92"></a>
<span class="sourceLineNo">093</span> }<a name="line.93"></a>
<span class="sourceLineNo">094</span><a name="line.94"></a>
<span class="sourceLineNo">095</span> // returns whether we need to continue the split work<a name="line.95"></a>
<span class="sourceLineNo">096</span> private static boolean processSyncReplicationWAL(String name, Configuration conf,<a name="line.96"></a>
<span class="sourceLineNo">097</span> RegionServerServices server, FileSystem fs, Path walDir) throws IOException {<a name="line.97"></a>
<span class="sourceLineNo">098</span> Path walFile = new Path(walDir, name);<a name="line.98"></a>
<span class="sourceLineNo">099</span> String filename = walFile.getName();<a name="line.99"></a>
<span class="sourceLineNo">100</span> Optional&lt;String&gt; optSyncPeerId =<a name="line.100"></a>
<span class="sourceLineNo">101</span> SyncReplicationWALProvider.getSyncReplicationPeerIdFromWALName(filename);<a name="line.101"></a>
<span class="sourceLineNo">102</span> if (!optSyncPeerId.isPresent()) {<a name="line.102"></a>
<span class="sourceLineNo">103</span> return true;<a name="line.103"></a>
<span class="sourceLineNo">104</span> }<a name="line.104"></a>
<span class="sourceLineNo">105</span> String peerId = optSyncPeerId.get();<a name="line.105"></a>
<span class="sourceLineNo">106</span> ReplicationPeerImpl peer =<a name="line.106"></a>
<span class="sourceLineNo">107</span> server.getReplicationSourceService().getReplicationPeers().getPeer(peerId);<a name="line.107"></a>
<span class="sourceLineNo">108</span> if (peer == null || !peer.getPeerConfig().isSyncReplication()) {<a name="line.108"></a>
<span class="sourceLineNo">109</span> return true;<a name="line.109"></a>
<span class="sourceLineNo">110</span> }<a name="line.110"></a>
<span class="sourceLineNo">111</span> Pair&lt;SyncReplicationState, SyncReplicationState&gt; stateAndNewState =<a name="line.111"></a>
<span class="sourceLineNo">112</span> peer.getSyncReplicationStateAndNewState();<a name="line.112"></a>
<span class="sourceLineNo">113</span> if (stateAndNewState.getFirst().equals(SyncReplicationState.ACTIVE) &amp;&amp;<a name="line.113"></a>
<span class="sourceLineNo">114</span> stateAndNewState.getSecond().equals(SyncReplicationState.NONE)) {<a name="line.114"></a>
<span class="sourceLineNo">115</span> // copy the file to remote and overwrite the previous one<a name="line.115"></a>
<span class="sourceLineNo">116</span> String remoteWALDir = peer.getPeerConfig().getRemoteWALDir();<a name="line.116"></a>
<span class="sourceLineNo">117</span> Path remoteWALDirForPeer = ReplicationUtils.getPeerRemoteWALDir(remoteWALDir, peerId);<a name="line.117"></a>
<span class="sourceLineNo">118</span> Path tmpRemoteWAL = new Path(remoteWALDirForPeer, filename + ".tmp");<a name="line.118"></a>
<span class="sourceLineNo">119</span> FileSystem remoteFs = ReplicationUtils.getRemoteWALFileSystem(conf, remoteWALDir);<a name="line.119"></a>
<span class="sourceLineNo">120</span> try (FSDataInputStream in = fs.open(walFile);<a name="line.120"></a>
<span class="sourceLineNo">121</span> FSDataOutputStream out = remoteFs.createNonRecursive(tmpRemoteWAL, true,<a name="line.121"></a>
<span class="sourceLineNo">122</span> CommonFSUtils.getDefaultBufferSize(remoteFs),<a name="line.122"></a>
<span class="sourceLineNo">123</span> remoteFs.getDefaultReplication(tmpRemoteWAL), remoteFs.getDefaultBlockSize(tmpRemoteWAL),<a name="line.123"></a>
<span class="sourceLineNo">124</span> null)) {<a name="line.124"></a>
<span class="sourceLineNo">125</span> IOUtils.copy(in, out);<a name="line.125"></a>
<span class="sourceLineNo">126</span> }<a name="line.126"></a>
<span class="sourceLineNo">127</span> Path toCommitRemoteWAL =<a name="line.127"></a>
<span class="sourceLineNo">128</span> new Path(remoteWALDirForPeer, filename + ReplicationUtils.RENAME_WAL_SUFFIX);<a name="line.128"></a>
<span class="sourceLineNo">129</span> // Some FileSystem implementations may not support atomic rename so we need to do it in two<a name="line.129"></a>
<span class="sourceLineNo">130</span> // phases<a name="line.130"></a>
<span class="sourceLineNo">131</span> FSUtils.renameFile(remoteFs, tmpRemoteWAL, toCommitRemoteWAL);<a name="line.131"></a>
<span class="sourceLineNo">132</span> FSUtils.renameFile(remoteFs, toCommitRemoteWAL, new Path(remoteWALDirForPeer, filename));<a name="line.132"></a>
<span class="sourceLineNo">133</span> } else if ((stateAndNewState.getFirst().equals(SyncReplicationState.ACTIVE) &amp;&amp;<a name="line.133"></a>
<span class="sourceLineNo">134</span> stateAndNewState.getSecond().equals(SyncReplicationState.STANDBY)) ||<a name="line.134"></a>
<span class="sourceLineNo">135</span> stateAndNewState.getFirst().equals(SyncReplicationState.STANDBY)) {<a name="line.135"></a>
<span class="sourceLineNo">136</span> // check whether we still need to process this file<a name="line.136"></a>
<span class="sourceLineNo">137</span> // actually we only write wal file which name is ended with .syncrep in A state, and after<a name="line.137"></a>
<span class="sourceLineNo">138</span> // transiting to a state other than A, we will reopen all the regions so the data in the wal<a name="line.138"></a>
<span class="sourceLineNo">139</span> // will be flushed so the wal file will be archived soon. But it is still possible that there<a name="line.139"></a>
<span class="sourceLineNo">140</span> // is a server crash when we are transiting from A to S, to simplify the logic of the transit<a name="line.140"></a>
<span class="sourceLineNo">141</span> // procedure, here we will also check the remote snapshot directory in state S, so that we do<a name="line.141"></a>
<span class="sourceLineNo">142</span> // not need wait until all the wal files with .syncrep suffix to be archived before finishing<a name="line.142"></a>
<span class="sourceLineNo">143</span> // the procedure.<a name="line.143"></a>
<span class="sourceLineNo">144</span> String remoteWALDir = peer.getPeerConfig().getRemoteWALDir();<a name="line.144"></a>
<span class="sourceLineNo">145</span> Path remoteSnapshotDirForPeer = ReplicationUtils.getPeerSnapshotWALDir(remoteWALDir, peerId);<a name="line.145"></a>
<span class="sourceLineNo">146</span> FileSystem remoteFs = ReplicationUtils.getRemoteWALFileSystem(conf, remoteWALDir);<a name="line.146"></a>
<span class="sourceLineNo">147</span> if (remoteFs.exists(new Path(remoteSnapshotDirForPeer, filename))) {<a name="line.147"></a>
<span class="sourceLineNo">148</span> // the file has been replayed when the remote cluster was transited from S to DA, the<a name="line.148"></a>
<span class="sourceLineNo">149</span> // content will be replicated back to us so give up split it.<a name="line.149"></a>
<span class="sourceLineNo">150</span> LOG.warn("Giveup splitting {} since it has been replayed in the remote cluster and " +<a name="line.150"></a>
<span class="sourceLineNo">151</span> "the content will be replicated back", filename);<a name="line.151"></a>
<span class="sourceLineNo">152</span> return false;<a name="line.152"></a>
<span class="sourceLineNo">153</span> }<a name="line.153"></a>
<span class="sourceLineNo">154</span> }<a name="line.154"></a>
<span class="sourceLineNo">155</span> return true;<a name="line.155"></a>
<span class="sourceLineNo">156</span> }<a name="line.156"></a>
<span class="sourceLineNo">157</span><a name="line.157"></a>
<span class="sourceLineNo">158</span> /**<a name="line.158"></a>
<span class="sourceLineNo">159</span> * @return Result either DONE, RESIGNED, or ERR.<a name="line.159"></a>
<span class="sourceLineNo">160</span> */<a name="line.160"></a>
<span class="sourceLineNo">161</span> static Status splitLog(String filename, CancelableProgressable p, Configuration conf,<a name="line.161"></a>
<span class="sourceLineNo">162</span> RegionServerServices server, LastSequenceId sequenceIdChecker, WALFactory factory) {<a name="line.162"></a>
<span class="sourceLineNo">163</span> Path walDir;<a name="line.163"></a>
<span class="sourceLineNo">164</span> FileSystem fs;<a name="line.164"></a>
<span class="sourceLineNo">165</span> try {<a name="line.165"></a>
<span class="sourceLineNo">166</span> walDir = CommonFSUtils.getWALRootDir(conf);<a name="line.166"></a>
<span class="sourceLineNo">167</span> fs = walDir.getFileSystem(conf);<a name="line.167"></a>
<span class="sourceLineNo">168</span> } catch (IOException e) {<a name="line.168"></a>
<span class="sourceLineNo">169</span> LOG.warn("Resigning, could not find root dir or fs", e);<a name="line.169"></a>
<span class="sourceLineNo">170</span> return Status.RESIGNED;<a name="line.170"></a>
<span class="sourceLineNo">171</span> }<a name="line.171"></a>
<span class="sourceLineNo">172</span> try {<a name="line.172"></a>
<span class="sourceLineNo">173</span> if (!processSyncReplicationWAL(filename, conf, server, fs, walDir)) {<a name="line.173"></a>
<span class="sourceLineNo">174</span> return Status.DONE;<a name="line.174"></a>
<span class="sourceLineNo">175</span> }<a name="line.175"></a>
<span class="sourceLineNo">176</span> } catch (IOException e) {<a name="line.176"></a>
<span class="sourceLineNo">177</span> LOG.warn("failed to process sync replication wal {}", filename, e);<a name="line.177"></a>
<span class="sourceLineNo">178</span> return Status.RESIGNED;<a name="line.178"></a>
<span class="sourceLineNo">179</span> }<a name="line.179"></a>
<span class="sourceLineNo">180</span> // TODO have to correctly figure out when log splitting has been<a name="line.180"></a>
<span class="sourceLineNo">181</span> // interrupted or has encountered a transient error and when it has<a name="line.181"></a>
<span class="sourceLineNo">182</span> // encountered a bad non-retry-able persistent error.<a name="line.182"></a>
<span class="sourceLineNo">183</span> try {<a name="line.183"></a>
<span class="sourceLineNo">184</span> SplitLogWorkerCoordination splitLogWorkerCoordination =<a name="line.184"></a>
<span class="sourceLineNo">185</span> server.getCoordinatedStateManager() == null ? null<a name="line.185"></a>
<span class="sourceLineNo">186</span> : server.getCoordinatedStateManager().getSplitLogWorkerCoordination();<a name="line.186"></a>
<span class="sourceLineNo">187</span> if (!WALSplitter.splitLogFile(walDir, fs.getFileStatus(new Path(walDir, filename)), fs, conf,<a name="line.187"></a>
<span class="sourceLineNo">188</span> p, sequenceIdChecker, splitLogWorkerCoordination, factory, server)) {<a name="line.188"></a>
<span class="sourceLineNo">189</span> return Status.PREEMPTED;<a name="line.189"></a>
<span class="sourceLineNo">190</span> }<a name="line.190"></a>
<span class="sourceLineNo">191</span> } catch (InterruptedIOException iioe) {<a name="line.191"></a>
<span class="sourceLineNo">192</span> LOG.warn("Resigning, interrupted splitting WAL {}", filename, iioe);<a name="line.192"></a>
<span class="sourceLineNo">193</span> return Status.RESIGNED;<a name="line.193"></a>
<span class="sourceLineNo">194</span> } catch (IOException e) {<a name="line.194"></a>
<span class="sourceLineNo">195</span> if (e instanceof FileNotFoundException) {<a name="line.195"></a>
<span class="sourceLineNo">196</span> // A wal file may not exist anymore. Nothing can be recovered so move on<a name="line.196"></a>
<span class="sourceLineNo">197</span> LOG.warn("Done, WAL {} does not exist anymore", filename, e);<a name="line.197"></a>
<span class="sourceLineNo">198</span> return Status.DONE;<a name="line.198"></a>
<span class="sourceLineNo">199</span> }<a name="line.199"></a>
<span class="sourceLineNo">200</span> Throwable cause = e.getCause();<a name="line.200"></a>
<span class="sourceLineNo">201</span> if (e instanceof RetriesExhaustedException &amp;&amp; (cause instanceof NotServingRegionException<a name="line.201"></a>
<span class="sourceLineNo">202</span> || cause instanceof ConnectException || cause instanceof SocketTimeoutException)) {<a name="line.202"></a>
<span class="sourceLineNo">203</span> LOG.warn("Resigning, can't connect to target regionserver splitting WAL {}", filename, e);<a name="line.203"></a>
<span class="sourceLineNo">204</span> return Status.RESIGNED;<a name="line.204"></a>
<span class="sourceLineNo">205</span> } else if (cause instanceof InterruptedException) {<a name="line.205"></a>
<span class="sourceLineNo">206</span> LOG.warn("Resigning, interrupted splitting WAL {}", filename, e);<a name="line.206"></a>
<span class="sourceLineNo">207</span> return Status.RESIGNED;<a name="line.207"></a>
<span class="sourceLineNo">208</span> }<a name="line.208"></a>
<span class="sourceLineNo">209</span> LOG.warn("Error splitting WAL {}", filename, e);<a name="line.209"></a>
<span class="sourceLineNo">210</span> return Status.ERR;<a name="line.210"></a>
<span class="sourceLineNo">211</span> }<a name="line.211"></a>
<span class="sourceLineNo">212</span> LOG.debug("Done splitting WAL {}", filename);<a name="line.212"></a>
<span class="sourceLineNo">213</span> return Status.DONE;<a name="line.213"></a>
<span class="sourceLineNo">214</span> }<a name="line.214"></a>
<span class="sourceLineNo">215</span><a name="line.215"></a>
<span class="sourceLineNo">216</span> @Override<a name="line.216"></a>
<span class="sourceLineNo">217</span> public void run() {<a name="line.217"></a>
<span class="sourceLineNo">218</span> try {<a name="line.218"></a>
<span class="sourceLineNo">219</span> LOG.info("SplitLogWorker " + server.getServerName() + " starting");<a name="line.219"></a>
<span class="sourceLineNo">220</span> coordination.registerListener();<a name="line.220"></a>
<span class="sourceLineNo">221</span> // wait for Coordination Engine is ready<a name="line.221"></a>
<span class="sourceLineNo">222</span> boolean res = false;<a name="line.222"></a>
<span class="sourceLineNo">223</span> while (!res &amp;&amp; !coordination.isStop()) {<a name="line.223"></a>
<span class="sourceLineNo">224</span> res = coordination.isReady();<a name="line.224"></a>
<span class="sourceLineNo">225</span> }<a name="line.225"></a>
<span class="sourceLineNo">226</span> if (!coordination.isStop()) {<a name="line.226"></a>
<span class="sourceLineNo">227</span> coordination.taskLoop();<a name="line.227"></a>
<span class="sourceLineNo">228</span> }<a name="line.228"></a>
<span class="sourceLineNo">229</span> } catch (Throwable t) {<a name="line.229"></a>
<span class="sourceLineNo">230</span> if (ExceptionUtil.isInterrupt(t)) {<a name="line.230"></a>
<span class="sourceLineNo">231</span> LOG.info("SplitLogWorker interrupted. Exiting. " + (coordination.isStop() ? "" :<a name="line.231"></a>
<span class="sourceLineNo">232</span> " (ERROR: exitWorker is not set, exiting anyway)"));<a name="line.232"></a>
<span class="sourceLineNo">233</span> } else {<a name="line.233"></a>
<span class="sourceLineNo">234</span> // only a logical error can cause here. Printing it out<a name="line.234"></a>
<span class="sourceLineNo">235</span> // to make debugging easier<a name="line.235"></a>
<span class="sourceLineNo">236</span> LOG.error("unexpected error ", t);<a name="line.236"></a>
<span class="sourceLineNo">237</span> }<a name="line.237"></a>
<span class="sourceLineNo">238</span> } finally {<a name="line.238"></a>
<span class="sourceLineNo">239</span> coordination.removeListener();<a name="line.239"></a>
<span class="sourceLineNo">240</span> LOG.info("SplitLogWorker " + server.getServerName() + " exiting");<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><a name="line.243"></a>
<span class="sourceLineNo">244</span> /**<a name="line.244"></a>
<span class="sourceLineNo">245</span> * If the worker is doing a task i.e. splitting a log file then stop the task.<a name="line.245"></a>
<span class="sourceLineNo">246</span> * It doesn't exit the worker thread.<a name="line.246"></a>
<span class="sourceLineNo">247</span> */<a name="line.247"></a>
<span class="sourceLineNo">248</span> public void stopTask() {<a name="line.248"></a>
<span class="sourceLineNo">249</span> LOG.info("Sending interrupt to stop the worker thread");<a name="line.249"></a>
<span class="sourceLineNo">250</span> worker.interrupt(); // TODO interrupt often gets swallowed, do what else?<a name="line.250"></a>
<span class="sourceLineNo">251</span> }<a name="line.251"></a>
<span class="sourceLineNo">252</span><a name="line.252"></a>
<span class="sourceLineNo">253</span> /**<a name="line.253"></a>
<span class="sourceLineNo">254</span> * start the SplitLogWorker thread<a name="line.254"></a>
<span class="sourceLineNo">255</span> */<a name="line.255"></a>
<span class="sourceLineNo">256</span> public void start() {<a name="line.256"></a>
<span class="sourceLineNo">257</span> worker = new Thread(null, this, "SplitLogWorker-" + server.getServerName().toShortString());<a name="line.257"></a>
<span class="sourceLineNo">258</span> worker.start();<a name="line.258"></a>
<span class="sourceLineNo">259</span> }<a name="line.259"></a>
<span class="sourceLineNo">260</span><a name="line.260"></a>
<span class="sourceLineNo">261</span> /**<a name="line.261"></a>
<span class="sourceLineNo">262</span> * stop the SplitLogWorker thread<a name="line.262"></a>
<span class="sourceLineNo">263</span> */<a name="line.263"></a>
<span class="sourceLineNo">264</span> public void stop() {<a name="line.264"></a>
<span class="sourceLineNo">265</span> coordination.stopProcessingTasks();<a name="line.265"></a>
<span class="sourceLineNo">266</span> stopTask();<a name="line.266"></a>
<span class="sourceLineNo">267</span> }<a name="line.267"></a>
<span class="sourceLineNo">268</span><a name="line.268"></a>
<span class="sourceLineNo">269</span> /**<a name="line.269"></a>
<span class="sourceLineNo">270</span> * Objects implementing this interface actually do the task that has been<a name="line.270"></a>
<span class="sourceLineNo">271</span> * acquired by a {@link SplitLogWorker}. Since there isn't a water-tight<a name="line.271"></a>
<span class="sourceLineNo">272</span> * guarantee that two workers will not be executing the same task therefore it<a name="line.272"></a>
<span class="sourceLineNo">273</span> * is better to have workers prepare the task and then have the<a name="line.273"></a>
<span class="sourceLineNo">274</span> * {@link org.apache.hadoop.hbase.master.SplitLogManager} commit the work in<a name="line.274"></a>
<span class="sourceLineNo">275</span> * SplitLogManager.TaskFinisher<a name="line.275"></a>
<span class="sourceLineNo">276</span> */<a name="line.276"></a>
<span class="sourceLineNo">277</span> @FunctionalInterface<a name="line.277"></a>
<span class="sourceLineNo">278</span> public interface TaskExecutor {<a name="line.278"></a>
<span class="sourceLineNo">279</span> enum Status {<a name="line.279"></a>
<span class="sourceLineNo">280</span> DONE(),<a name="line.280"></a>
<span class="sourceLineNo">281</span> ERR(),<a name="line.281"></a>
<span class="sourceLineNo">282</span> RESIGNED(),<a name="line.282"></a>
<span class="sourceLineNo">283</span> PREEMPTED()<a name="line.283"></a>
<span class="sourceLineNo">284</span> }<a name="line.284"></a>
<span class="sourceLineNo">285</span> Status exec(String name, CancelableProgressable p);<a name="line.285"></a>
<span class="sourceLineNo">286</span> }<a name="line.286"></a>
<span class="sourceLineNo">287</span><a name="line.287"></a>
<span class="sourceLineNo">288</span> /**<a name="line.288"></a>
<span class="sourceLineNo">289</span> * Returns the number of tasks processed by coordination.<a name="line.289"></a>
<span class="sourceLineNo">290</span> * This method is used by tests only<a name="line.290"></a>
<span class="sourceLineNo">291</span> */<a name="line.291"></a>
<span class="sourceLineNo">292</span> @VisibleForTesting<a name="line.292"></a>
<span class="sourceLineNo">293</span> public int getTaskReadySeq() {<a name="line.293"></a>
<span class="sourceLineNo">294</span> return coordination.getTaskReadySeq();<a name="line.294"></a>
<span class="sourceLineNo">295</span> }<a name="line.295"></a>
<span class="sourceLineNo">296</span>}<a name="line.296"></a>
</pre>
</div>
</body>
</html>