blob: 61dae21bf0dcb57e57d2832a188924bd79aa110c [file] [log] [blame]
<!DOCTYPE HTML>
<html lang="en">
<head>
<!-- Generated by javadoc (17) -->
<title>Source code</title>
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="source: package: org.apache.hadoop.hbase.util, class: RecoverLeaseFSUtils">
<meta name="generator" content="javadoc/SourceToHTMLConverter">
<link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style">
</head>
<body class="source-page">
<main role="main">
<div class="source-container">
<pre><span class="source-line-no">001</span><span id="line-1">/*</span>
<span class="source-line-no">002</span><span id="line-2"> * Licensed to the Apache Software Foundation (ASF) under one</span>
<span class="source-line-no">003</span><span id="line-3"> * or more contributor license agreements. See the NOTICE file</span>
<span class="source-line-no">004</span><span id="line-4"> * distributed with this work for additional information</span>
<span class="source-line-no">005</span><span id="line-5"> * regarding copyright ownership. The ASF licenses this file</span>
<span class="source-line-no">006</span><span id="line-6"> * to you under the Apache License, Version 2.0 (the</span>
<span class="source-line-no">007</span><span id="line-7"> * "License"); you may not use this file except in compliance</span>
<span class="source-line-no">008</span><span id="line-8"> * with the License. You may obtain a copy of the License at</span>
<span class="source-line-no">009</span><span id="line-9"> *</span>
<span class="source-line-no">010</span><span id="line-10"> * http://www.apache.org/licenses/LICENSE-2.0</span>
<span class="source-line-no">011</span><span id="line-11"> *</span>
<span class="source-line-no">012</span><span id="line-12"> * Unless required by applicable law or agreed to in writing, software</span>
<span class="source-line-no">013</span><span id="line-13"> * distributed under the License is distributed on an "AS IS" BASIS,</span>
<span class="source-line-no">014</span><span id="line-14"> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
<span class="source-line-no">015</span><span id="line-15"> * See the License for the specific language governing permissions and</span>
<span class="source-line-no">016</span><span id="line-16"> * limitations under the License.</span>
<span class="source-line-no">017</span><span id="line-17"> */</span>
<span class="source-line-no">018</span><span id="line-18">package org.apache.hadoop.hbase.util;</span>
<span class="source-line-no">019</span><span id="line-19"></span>
<span class="source-line-no">020</span><span id="line-20">import java.io.FileNotFoundException;</span>
<span class="source-line-no">021</span><span id="line-21">import java.io.IOException;</span>
<span class="source-line-no">022</span><span id="line-22">import java.io.InterruptedIOException;</span>
<span class="source-line-no">023</span><span id="line-23">import java.lang.reflect.Method;</span>
<span class="source-line-no">024</span><span id="line-24">import org.apache.hadoop.conf.Configuration;</span>
<span class="source-line-no">025</span><span id="line-25">import org.apache.hadoop.fs.FileSystem;</span>
<span class="source-line-no">026</span><span id="line-26">import org.apache.hadoop.fs.FilterFileSystem;</span>
<span class="source-line-no">027</span><span id="line-27">import org.apache.hadoop.fs.Path;</span>
<span class="source-line-no">028</span><span id="line-28">import org.apache.hadoop.hdfs.DistributedFileSystem;</span>
<span class="source-line-no">029</span><span id="line-29">import org.apache.hadoop.hdfs.server.namenode.LeaseExpiredException;</span>
<span class="source-line-no">030</span><span id="line-30">import org.apache.yetus.audience.InterfaceAudience;</span>
<span class="source-line-no">031</span><span id="line-31">import org.slf4j.Logger;</span>
<span class="source-line-no">032</span><span id="line-32">import org.slf4j.LoggerFactory;</span>
<span class="source-line-no">033</span><span id="line-33"></span>
<span class="source-line-no">034</span><span id="line-34">/**</span>
<span class="source-line-no">035</span><span id="line-35"> * Utility methods for recovering file lease for hdfs.</span>
<span class="source-line-no">036</span><span id="line-36"> */</span>
<span class="source-line-no">037</span><span id="line-37">@InterfaceAudience.Private</span>
<span class="source-line-no">038</span><span id="line-38">public final class RecoverLeaseFSUtils {</span>
<span class="source-line-no">039</span><span id="line-39"></span>
<span class="source-line-no">040</span><span id="line-40"> private static final Logger LOG = LoggerFactory.getLogger(RecoverLeaseFSUtils.class);</span>
<span class="source-line-no">041</span><span id="line-41"></span>
<span class="source-line-no">042</span><span id="line-42"> private RecoverLeaseFSUtils() {</span>
<span class="source-line-no">043</span><span id="line-43"> }</span>
<span class="source-line-no">044</span><span id="line-44"></span>
<span class="source-line-no">045</span><span id="line-45"> public static void recoverFileLease(FileSystem fs, Path p, Configuration conf)</span>
<span class="source-line-no">046</span><span id="line-46"> throws IOException {</span>
<span class="source-line-no">047</span><span id="line-47"> recoverFileLease(fs, p, conf, null);</span>
<span class="source-line-no">048</span><span id="line-48"> }</span>
<span class="source-line-no">049</span><span id="line-49"></span>
<span class="source-line-no">050</span><span id="line-50"> /**</span>
<span class="source-line-no">051</span><span id="line-51"> * Recover the lease from HDFS, retrying multiple times.</span>
<span class="source-line-no">052</span><span id="line-52"> */</span>
<span class="source-line-no">053</span><span id="line-53"> public static void recoverFileLease(FileSystem fs, Path p, Configuration conf,</span>
<span class="source-line-no">054</span><span id="line-54"> CancelableProgressable reporter) throws IOException {</span>
<span class="source-line-no">055</span><span id="line-55"> if (fs instanceof FilterFileSystem) {</span>
<span class="source-line-no">056</span><span id="line-56"> fs = ((FilterFileSystem) fs).getRawFileSystem();</span>
<span class="source-line-no">057</span><span id="line-57"> }</span>
<span class="source-line-no">058</span><span id="line-58"> // lease recovery not needed for local file system case.</span>
<span class="source-line-no">059</span><span id="line-59"> if (!(fs instanceof DistributedFileSystem)) {</span>
<span class="source-line-no">060</span><span id="line-60"> return;</span>
<span class="source-line-no">061</span><span id="line-61"> }</span>
<span class="source-line-no">062</span><span id="line-62"> recoverDFSFileLease((DistributedFileSystem) fs, p, conf, reporter);</span>
<span class="source-line-no">063</span><span id="line-63"> }</span>
<span class="source-line-no">064</span><span id="line-64"></span>
<span class="source-line-no">065</span><span id="line-65"> /*</span>
<span class="source-line-no">066</span><span id="line-66"> * Run the dfs recover lease. recoverLease is asynchronous. It returns: -false when it starts the</span>
<span class="source-line-no">067</span><span id="line-67"> * lease recovery (i.e. lease recovery not *yet* done) - true when the lease recovery has</span>
<span class="source-line-no">068</span><span id="line-68"> * succeeded or the file is closed. But, we have to be careful. Each time we call recoverLease, it</span>
<span class="source-line-no">069</span><span id="line-69"> * starts the recover lease process over from the beginning. We could put ourselves in a situation</span>
<span class="source-line-no">070</span><span id="line-70"> * where we are doing nothing but starting a recovery, interrupting it to start again, and so on.</span>
<span class="source-line-no">071</span><span id="line-71"> * The findings over in HBASE-8354 have it that the namenode will try to recover the lease on the</span>
<span class="source-line-no">072</span><span id="line-72"> * file's primary node. If all is well, it should return near immediately. But, as is common, it</span>
<span class="source-line-no">073</span><span id="line-73"> * is the very primary node that has crashed and so the namenode will be stuck waiting on a socket</span>
<span class="source-line-no">074</span><span id="line-74"> * timeout before it will ask another datanode to start the recovery. It does not help if we call</span>
<span class="source-line-no">075</span><span id="line-75"> * recoverLease in the meantime and in particular, after the socket timeout, a recoverLease</span>
<span class="source-line-no">076</span><span id="line-76"> * invocation will cause us to start over from square one (possibly waiting on socket timeout</span>
<span class="source-line-no">077</span><span id="line-77"> * against primary node). So, in the below, we do the following: 1. Call recoverLease. 2. If it</span>
<span class="source-line-no">078</span><span id="line-78"> * returns true, break. 3. If it returns false, wait a few seconds and then call it again. 4. If</span>
<span class="source-line-no">079</span><span id="line-79"> * it returns true, break. 5. If it returns false, wait for what we think the datanode socket</span>
<span class="source-line-no">080</span><span id="line-80"> * timeout is (configurable) and then try again. 6. If it returns true, break. 7. If it returns</span>
<span class="source-line-no">081</span><span id="line-81"> * false, repeat starting at step 5. above. If HDFS-4525 is available, call it every second, and</span>
<span class="source-line-no">082</span><span id="line-82"> * we might be able to exit early.</span>
<span class="source-line-no">083</span><span id="line-83"> */</span>
<span class="source-line-no">084</span><span id="line-84"> private static boolean recoverDFSFileLease(final DistributedFileSystem dfs, final Path p,</span>
<span class="source-line-no">085</span><span id="line-85"> final Configuration conf, final CancelableProgressable reporter) throws IOException {</span>
<span class="source-line-no">086</span><span id="line-86"> LOG.info("Recover lease on dfs file " + p);</span>
<span class="source-line-no">087</span><span id="line-87"> long startWaiting = EnvironmentEdgeManager.currentTime();</span>
<span class="source-line-no">088</span><span id="line-88"> // Default is 15 minutes. It's huge, but the idea is that if we have a major issue, HDFS</span>
<span class="source-line-no">089</span><span id="line-89"> // usually needs 10 minutes before marking the nodes as dead. So we're putting ourselves</span>
<span class="source-line-no">090</span><span id="line-90"> // beyond that limit 'to be safe'.</span>
<span class="source-line-no">091</span><span id="line-91"> long recoveryTimeout = conf.getInt("hbase.lease.recovery.timeout", 900000) + startWaiting;</span>
<span class="source-line-no">092</span><span id="line-92"> // This setting should be a little above what the cluster dfs heartbeat is set to.</span>
<span class="source-line-no">093</span><span id="line-93"> long firstPause = conf.getInt("hbase.lease.recovery.first.pause", 4000);</span>
<span class="source-line-no">094</span><span id="line-94"> // This should be set to how long it'll take for us to timeout against primary datanode if it</span>
<span class="source-line-no">095</span><span id="line-95"> // is dead. We set it to 64 seconds, 4 seconds than the default READ_TIMEOUT in HDFS, the</span>
<span class="source-line-no">096</span><span id="line-96"> // default value for DFS_CLIENT_SOCKET_TIMEOUT_KEY. If recovery is still failing after this</span>
<span class="source-line-no">097</span><span id="line-97"> // timeout, then further recovery will take liner backoff with this base, to avoid endless</span>
<span class="source-line-no">098</span><span id="line-98"> // preemptions when this value is not properly configured.</span>
<span class="source-line-no">099</span><span id="line-99"> long subsequentPauseBase = conf.getLong("hbase.lease.recovery.dfs.timeout", 64 * 1000);</span>
<span class="source-line-no">100</span><span id="line-100"></span>
<span class="source-line-no">101</span><span id="line-101"> Method isFileClosedMeth = null;</span>
<span class="source-line-no">102</span><span id="line-102"> // whether we need to look for isFileClosed method</span>
<span class="source-line-no">103</span><span id="line-103"> boolean findIsFileClosedMeth = true;</span>
<span class="source-line-no">104</span><span id="line-104"> boolean recovered = false;</span>
<span class="source-line-no">105</span><span id="line-105"> // We break the loop if we succeed the lease recovery, timeout, or we throw an exception.</span>
<span class="source-line-no">106</span><span id="line-106"> for (int nbAttempt = 0; !recovered; nbAttempt++) {</span>
<span class="source-line-no">107</span><span id="line-107"> recovered = recoverLease(dfs, nbAttempt, p, startWaiting);</span>
<span class="source-line-no">108</span><span id="line-108"> if (recovered) {</span>
<span class="source-line-no">109</span><span id="line-109"> break;</span>
<span class="source-line-no">110</span><span id="line-110"> }</span>
<span class="source-line-no">111</span><span id="line-111"> checkIfCancelled(reporter);</span>
<span class="source-line-no">112</span><span id="line-112"> if (checkIfTimedout(conf, recoveryTimeout, nbAttempt, p, startWaiting)) {</span>
<span class="source-line-no">113</span><span id="line-113"> break;</span>
<span class="source-line-no">114</span><span id="line-114"> }</span>
<span class="source-line-no">115</span><span id="line-115"> try {</span>
<span class="source-line-no">116</span><span id="line-116"> // On the first time through wait the short 'firstPause'.</span>
<span class="source-line-no">117</span><span id="line-117"> if (nbAttempt == 0) {</span>
<span class="source-line-no">118</span><span id="line-118"> Thread.sleep(firstPause);</span>
<span class="source-line-no">119</span><span id="line-119"> } else {</span>
<span class="source-line-no">120</span><span id="line-120"> // Cycle here until (subsequentPause * nbAttempt) elapses. While spinning, check</span>
<span class="source-line-no">121</span><span id="line-121"> // isFileClosed if available (should be in hadoop 2.0.5... not in hadoop 1 though).</span>
<span class="source-line-no">122</span><span id="line-122"> long localStartWaiting = EnvironmentEdgeManager.currentTime();</span>
<span class="source-line-no">123</span><span id="line-123"> while (</span>
<span class="source-line-no">124</span><span id="line-124"> (EnvironmentEdgeManager.currentTime() - localStartWaiting)</span>
<span class="source-line-no">125</span><span id="line-125"> &lt; subsequentPauseBase * nbAttempt</span>
<span class="source-line-no">126</span><span id="line-126"> ) {</span>
<span class="source-line-no">127</span><span id="line-127"> Thread.sleep(conf.getInt("hbase.lease.recovery.pause", 1000));</span>
<span class="source-line-no">128</span><span id="line-128"> if (findIsFileClosedMeth) {</span>
<span class="source-line-no">129</span><span id="line-129"> try {</span>
<span class="source-line-no">130</span><span id="line-130"> isFileClosedMeth =</span>
<span class="source-line-no">131</span><span id="line-131"> dfs.getClass().getMethod("isFileClosed", new Class[] { Path.class });</span>
<span class="source-line-no">132</span><span id="line-132"> } catch (NoSuchMethodException nsme) {</span>
<span class="source-line-no">133</span><span id="line-133"> LOG.debug("isFileClosed not available");</span>
<span class="source-line-no">134</span><span id="line-134"> } finally {</span>
<span class="source-line-no">135</span><span id="line-135"> findIsFileClosedMeth = false;</span>
<span class="source-line-no">136</span><span id="line-136"> }</span>
<span class="source-line-no">137</span><span id="line-137"> }</span>
<span class="source-line-no">138</span><span id="line-138"> if (isFileClosedMeth != null &amp;&amp; isFileClosed(dfs, isFileClosedMeth, p)) {</span>
<span class="source-line-no">139</span><span id="line-139"> recovered = true;</span>
<span class="source-line-no">140</span><span id="line-140"> break;</span>
<span class="source-line-no">141</span><span id="line-141"> }</span>
<span class="source-line-no">142</span><span id="line-142"> checkIfCancelled(reporter);</span>
<span class="source-line-no">143</span><span id="line-143"> }</span>
<span class="source-line-no">144</span><span id="line-144"> }</span>
<span class="source-line-no">145</span><span id="line-145"> } catch (InterruptedException ie) {</span>
<span class="source-line-no">146</span><span id="line-146"> InterruptedIOException iioe = new InterruptedIOException();</span>
<span class="source-line-no">147</span><span id="line-147"> iioe.initCause(ie);</span>
<span class="source-line-no">148</span><span id="line-148"> throw iioe;</span>
<span class="source-line-no">149</span><span id="line-149"> }</span>
<span class="source-line-no">150</span><span id="line-150"> }</span>
<span class="source-line-no">151</span><span id="line-151"> return recovered;</span>
<span class="source-line-no">152</span><span id="line-152"> }</span>
<span class="source-line-no">153</span><span id="line-153"></span>
<span class="source-line-no">154</span><span id="line-154"> private static boolean checkIfTimedout(final Configuration conf, final long recoveryTimeout,</span>
<span class="source-line-no">155</span><span id="line-155"> final int nbAttempt, final Path p, final long startWaiting) {</span>
<span class="source-line-no">156</span><span id="line-156"> if (recoveryTimeout &lt; EnvironmentEdgeManager.currentTime()) {</span>
<span class="source-line-no">157</span><span id="line-157"> LOG.warn("Cannot recoverLease after trying for "</span>
<span class="source-line-no">158</span><span id="line-158"> + conf.getInt("hbase.lease.recovery.timeout", 900000)</span>
<span class="source-line-no">159</span><span id="line-159"> + "ms (hbase.lease.recovery.timeout); continuing, but may be DATALOSS!!!; "</span>
<span class="source-line-no">160</span><span id="line-160"> + getLogMessageDetail(nbAttempt, p, startWaiting));</span>
<span class="source-line-no">161</span><span id="line-161"> return true;</span>
<span class="source-line-no">162</span><span id="line-162"> }</span>
<span class="source-line-no">163</span><span id="line-163"> return false;</span>
<span class="source-line-no">164</span><span id="line-164"> }</span>
<span class="source-line-no">165</span><span id="line-165"></span>
<span class="source-line-no">166</span><span id="line-166"> /**</span>
<span class="source-line-no">167</span><span id="line-167"> * Try to recover the lease.</span>
<span class="source-line-no">168</span><span id="line-168"> * @return True if dfs#recoverLease came by true.</span>
<span class="source-line-no">169</span><span id="line-169"> */</span>
<span class="source-line-no">170</span><span id="line-170"> private static boolean recoverLease(final DistributedFileSystem dfs, final int nbAttempt,</span>
<span class="source-line-no">171</span><span id="line-171"> final Path p, final long startWaiting) throws FileNotFoundException {</span>
<span class="source-line-no">172</span><span id="line-172"> boolean recovered = false;</span>
<span class="source-line-no">173</span><span id="line-173"> try {</span>
<span class="source-line-no">174</span><span id="line-174"> recovered = dfs.recoverLease(p);</span>
<span class="source-line-no">175</span><span id="line-175"> LOG.info((recovered ? "Recovered lease, " : "Failed to recover lease, ")</span>
<span class="source-line-no">176</span><span id="line-176"> + getLogMessageDetail(nbAttempt, p, startWaiting));</span>
<span class="source-line-no">177</span><span id="line-177"> } catch (IOException e) {</span>
<span class="source-line-no">178</span><span id="line-178"> if (e instanceof LeaseExpiredException &amp;&amp; e.getMessage().contains("File does not exist")) {</span>
<span class="source-line-no">179</span><span id="line-179"> // This exception comes out instead of FNFE, fix it</span>
<span class="source-line-no">180</span><span id="line-180"> throw new FileNotFoundException("The given WAL wasn't found at " + p);</span>
<span class="source-line-no">181</span><span id="line-181"> } else if (e instanceof FileNotFoundException) {</span>
<span class="source-line-no">182</span><span id="line-182"> throw (FileNotFoundException) e;</span>
<span class="source-line-no">183</span><span id="line-183"> }</span>
<span class="source-line-no">184</span><span id="line-184"> LOG.warn(getLogMessageDetail(nbAttempt, p, startWaiting), e);</span>
<span class="source-line-no">185</span><span id="line-185"> }</span>
<span class="source-line-no">186</span><span id="line-186"> return recovered;</span>
<span class="source-line-no">187</span><span id="line-187"> }</span>
<span class="source-line-no">188</span><span id="line-188"></span>
<span class="source-line-no">189</span><span id="line-189"> /** Returns Detail to append to any log message around lease recovering. */</span>
<span class="source-line-no">190</span><span id="line-190"> private static String getLogMessageDetail(final int nbAttempt, final Path p,</span>
<span class="source-line-no">191</span><span id="line-191"> final long startWaiting) {</span>
<span class="source-line-no">192</span><span id="line-192"> return "attempt=" + nbAttempt + " on file=" + p + " after "</span>
<span class="source-line-no">193</span><span id="line-193"> + (EnvironmentEdgeManager.currentTime() - startWaiting) + "ms";</span>
<span class="source-line-no">194</span><span id="line-194"> }</span>
<span class="source-line-no">195</span><span id="line-195"></span>
<span class="source-line-no">196</span><span id="line-196"> /**</span>
<span class="source-line-no">197</span><span id="line-197"> * Call HDFS-4525 isFileClosed if it is available.</span>
<span class="source-line-no">198</span><span id="line-198"> * @return True if file is closed.</span>
<span class="source-line-no">199</span><span id="line-199"> */</span>
<span class="source-line-no">200</span><span id="line-200"> private static boolean isFileClosed(final DistributedFileSystem dfs, final Method m,</span>
<span class="source-line-no">201</span><span id="line-201"> final Path p) {</span>
<span class="source-line-no">202</span><span id="line-202"> try {</span>
<span class="source-line-no">203</span><span id="line-203"> return (Boolean) m.invoke(dfs, p);</span>
<span class="source-line-no">204</span><span id="line-204"> } catch (SecurityException e) {</span>
<span class="source-line-no">205</span><span id="line-205"> LOG.warn("No access", e);</span>
<span class="source-line-no">206</span><span id="line-206"> } catch (Exception e) {</span>
<span class="source-line-no">207</span><span id="line-207"> LOG.warn("Failed invocation for " + p.toString(), e);</span>
<span class="source-line-no">208</span><span id="line-208"> }</span>
<span class="source-line-no">209</span><span id="line-209"> return false;</span>
<span class="source-line-no">210</span><span id="line-210"> }</span>
<span class="source-line-no">211</span><span id="line-211"></span>
<span class="source-line-no">212</span><span id="line-212"> private static void checkIfCancelled(final CancelableProgressable reporter)</span>
<span class="source-line-no">213</span><span id="line-213"> throws InterruptedIOException {</span>
<span class="source-line-no">214</span><span id="line-214"> if (reporter == null) {</span>
<span class="source-line-no">215</span><span id="line-215"> return;</span>
<span class="source-line-no">216</span><span id="line-216"> }</span>
<span class="source-line-no">217</span><span id="line-217"> if (!reporter.progress()) {</span>
<span class="source-line-no">218</span><span id="line-218"> throw new InterruptedIOException("Operation cancelled");</span>
<span class="source-line-no">219</span><span id="line-219"> }</span>
<span class="source-line-no">220</span><span id="line-220"> }</span>
<span class="source-line-no">221</span><span id="line-221">}</span>
</pre>
</div>
</main>
</body>
</html>