blob: dd29e0a5af4695e26303d1f070ddf158f47adbc9 [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.replication.regionserver;<a name="line.18"></a>
<span class="sourceLineNo">019</span><a name="line.19"></a>
<span class="sourceLineNo">020</span>import java.io.IOException;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import java.util.List;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import java.util.UUID;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.util.concurrent.PriorityBlockingQueue;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import org.apache.hadoop.conf.Configuration;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import org.apache.hadoop.fs.FileStatus;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.hadoop.fs.FileSystem;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.fs.Path;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.Server;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.ServerName;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.util.CommonFSUtils;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.slf4j.Logger;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.slf4j.LoggerFactory;<a name="line.36"></a>
<span class="sourceLineNo">037</span><a name="line.37"></a>
<span class="sourceLineNo">038</span>/**<a name="line.38"></a>
<span class="sourceLineNo">039</span> * Class that handles the recovered source of a replication stream, which is transfered from<a name="line.39"></a>
<span class="sourceLineNo">040</span> * another dead region server. This will be closed when all logs are pushed to peer cluster.<a name="line.40"></a>
<span class="sourceLineNo">041</span> */<a name="line.41"></a>
<span class="sourceLineNo">042</span>@InterfaceAudience.Private<a name="line.42"></a>
<span class="sourceLineNo">043</span>public class RecoveredReplicationSource extends ReplicationSource {<a name="line.43"></a>
<span class="sourceLineNo">044</span><a name="line.44"></a>
<span class="sourceLineNo">045</span> private static final Logger LOG = LoggerFactory.getLogger(RecoveredReplicationSource.class);<a name="line.45"></a>
<span class="sourceLineNo">046</span><a name="line.46"></a>
<span class="sourceLineNo">047</span> private String actualPeerId;<a name="line.47"></a>
<span class="sourceLineNo">048</span><a name="line.48"></a>
<span class="sourceLineNo">049</span> @Override<a name="line.49"></a>
<span class="sourceLineNo">050</span> public void init(Configuration conf, FileSystem fs, ReplicationSourceManager manager,<a name="line.50"></a>
<span class="sourceLineNo">051</span> ReplicationQueueStorage queueStorage, ReplicationPeer replicationPeer, Server server,<a name="line.51"></a>
<span class="sourceLineNo">052</span> String peerClusterZnode, UUID clusterId, WALFileLengthProvider walFileLengthProvider,<a name="line.52"></a>
<span class="sourceLineNo">053</span> MetricsSource metrics) throws IOException {<a name="line.53"></a>
<span class="sourceLineNo">054</span> super.init(conf, fs, manager, queueStorage, replicationPeer, server, peerClusterZnode,<a name="line.54"></a>
<span class="sourceLineNo">055</span> clusterId, walFileLengthProvider, metrics);<a name="line.55"></a>
<span class="sourceLineNo">056</span> this.actualPeerId = this.replicationQueueInfo.getPeerId();<a name="line.56"></a>
<span class="sourceLineNo">057</span> }<a name="line.57"></a>
<span class="sourceLineNo">058</span><a name="line.58"></a>
<span class="sourceLineNo">059</span> @Override<a name="line.59"></a>
<span class="sourceLineNo">060</span> protected RecoveredReplicationSourceShipper createNewShipper(String walGroupId,<a name="line.60"></a>
<span class="sourceLineNo">061</span> PriorityBlockingQueue&lt;Path&gt; queue) {<a name="line.61"></a>
<span class="sourceLineNo">062</span> return new RecoveredReplicationSourceShipper(conf, walGroupId, queue, this, queueStorage);<a name="line.62"></a>
<span class="sourceLineNo">063</span> }<a name="line.63"></a>
<span class="sourceLineNo">064</span><a name="line.64"></a>
<span class="sourceLineNo">065</span> public void locateRecoveredPaths(PriorityBlockingQueue&lt;Path&gt; queue) throws IOException {<a name="line.65"></a>
<span class="sourceLineNo">066</span> boolean hasPathChanged = false;<a name="line.66"></a>
<span class="sourceLineNo">067</span> PriorityBlockingQueue&lt;Path&gt; newPaths =<a name="line.67"></a>
<span class="sourceLineNo">068</span> new PriorityBlockingQueue&lt;Path&gt;(queueSizePerGroup, new LogsComparator());<a name="line.68"></a>
<span class="sourceLineNo">069</span> pathsLoop: for (Path path : queue) {<a name="line.69"></a>
<span class="sourceLineNo">070</span> if (fs.exists(path)) { // still in same location, don't need to do anything<a name="line.70"></a>
<span class="sourceLineNo">071</span> newPaths.add(path);<a name="line.71"></a>
<span class="sourceLineNo">072</span> continue;<a name="line.72"></a>
<span class="sourceLineNo">073</span> }<a name="line.73"></a>
<span class="sourceLineNo">074</span> // Path changed - try to find the right path.<a name="line.74"></a>
<span class="sourceLineNo">075</span> hasPathChanged = true;<a name="line.75"></a>
<span class="sourceLineNo">076</span> if (server instanceof ReplicationSyncUp.DummyServer) {<a name="line.76"></a>
<span class="sourceLineNo">077</span> // In the case of disaster/recovery, HMaster may be shutdown/crashed before flush data<a name="line.77"></a>
<span class="sourceLineNo">078</span> // from .logs to .oldlogs. Loop into .logs folders and check whether a match exists<a name="line.78"></a>
<span class="sourceLineNo">079</span> Path newPath = getReplSyncUpPath(path);<a name="line.79"></a>
<span class="sourceLineNo">080</span> newPaths.add(newPath);<a name="line.80"></a>
<span class="sourceLineNo">081</span> continue;<a name="line.81"></a>
<span class="sourceLineNo">082</span> } else {<a name="line.82"></a>
<span class="sourceLineNo">083</span> // See if Path exists in the dead RS folder (there could be a chain of failures<a name="line.83"></a>
<span class="sourceLineNo">084</span> // to look at)<a name="line.84"></a>
<span class="sourceLineNo">085</span> List&lt;ServerName&gt; deadRegionServers = this.replicationQueueInfo.getDeadRegionServers();<a name="line.85"></a>
<span class="sourceLineNo">086</span> LOG.info("NB dead servers : " + deadRegionServers.size());<a name="line.86"></a>
<span class="sourceLineNo">087</span> final Path walDir = CommonFSUtils.getWALRootDir(conf);<a name="line.87"></a>
<span class="sourceLineNo">088</span> for (ServerName curDeadServerName : deadRegionServers) {<a name="line.88"></a>
<span class="sourceLineNo">089</span> final Path deadRsDirectory =<a name="line.89"></a>
<span class="sourceLineNo">090</span> new Path(walDir, AbstractFSWALProvider.getWALDirectoryName(curDeadServerName<a name="line.90"></a>
<span class="sourceLineNo">091</span> .getServerName()));<a name="line.91"></a>
<span class="sourceLineNo">092</span> Path[] locs = new Path[] { new Path(deadRsDirectory, path.getName()), new Path(<a name="line.92"></a>
<span class="sourceLineNo">093</span> deadRsDirectory.suffix(AbstractFSWALProvider.SPLITTING_EXT), path.getName()) };<a name="line.93"></a>
<span class="sourceLineNo">094</span> for (Path possibleLogLocation : locs) {<a name="line.94"></a>
<span class="sourceLineNo">095</span> LOG.info("Possible location " + possibleLogLocation.toUri().toString());<a name="line.95"></a>
<span class="sourceLineNo">096</span> if (manager.getFs().exists(possibleLogLocation)) {<a name="line.96"></a>
<span class="sourceLineNo">097</span> // We found the right new location<a name="line.97"></a>
<span class="sourceLineNo">098</span> LOG.info("Log " + path + " still exists at " + possibleLogLocation);<a name="line.98"></a>
<span class="sourceLineNo">099</span> newPaths.add(possibleLogLocation);<a name="line.99"></a>
<span class="sourceLineNo">100</span> continue pathsLoop;<a name="line.100"></a>
<span class="sourceLineNo">101</span> }<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> // didn't find a new location<a name="line.104"></a>
<span class="sourceLineNo">105</span> LOG.error(<a name="line.105"></a>
<span class="sourceLineNo">106</span> String.format("WAL Path %s doesn't exist and couldn't find its new location", path));<a name="line.106"></a>
<span class="sourceLineNo">107</span> newPaths.add(path);<a name="line.107"></a>
<span class="sourceLineNo">108</span> }<a name="line.108"></a>
<span class="sourceLineNo">109</span> }<a name="line.109"></a>
<span class="sourceLineNo">110</span><a name="line.110"></a>
<span class="sourceLineNo">111</span> if (hasPathChanged) {<a name="line.111"></a>
<span class="sourceLineNo">112</span> if (newPaths.size() != queue.size()) { // this shouldn't happen<a name="line.112"></a>
<span class="sourceLineNo">113</span> LOG.error("Recovery queue size is incorrect");<a name="line.113"></a>
<span class="sourceLineNo">114</span> throw new IOException("Recovery queue size error");<a name="line.114"></a>
<span class="sourceLineNo">115</span> }<a name="line.115"></a>
<span class="sourceLineNo">116</span> // put the correct locations in the queue<a name="line.116"></a>
<span class="sourceLineNo">117</span> // since this is a recovered queue with no new incoming logs,<a name="line.117"></a>
<span class="sourceLineNo">118</span> // there shouldn't be any concurrency issues<a name="line.118"></a>
<span class="sourceLineNo">119</span> queue.clear();<a name="line.119"></a>
<span class="sourceLineNo">120</span> for (Path path : newPaths) {<a name="line.120"></a>
<span class="sourceLineNo">121</span> queue.add(path);<a name="line.121"></a>
<span class="sourceLineNo">122</span> }<a name="line.122"></a>
<span class="sourceLineNo">123</span> }<a name="line.123"></a>
<span class="sourceLineNo">124</span> }<a name="line.124"></a>
<span class="sourceLineNo">125</span><a name="line.125"></a>
<span class="sourceLineNo">126</span> // N.B. the ReplicationSyncUp tool sets the manager.getWALDir to the root of the wal<a name="line.126"></a>
<span class="sourceLineNo">127</span> // area rather than to the wal area for a particular region server.<a name="line.127"></a>
<span class="sourceLineNo">128</span> private Path getReplSyncUpPath(Path path) throws IOException {<a name="line.128"></a>
<span class="sourceLineNo">129</span> FileStatus[] rss = fs.listStatus(manager.getLogDir());<a name="line.129"></a>
<span class="sourceLineNo">130</span> for (FileStatus rs : rss) {<a name="line.130"></a>
<span class="sourceLineNo">131</span> Path p = rs.getPath();<a name="line.131"></a>
<span class="sourceLineNo">132</span> FileStatus[] logs = fs.listStatus(p);<a name="line.132"></a>
<span class="sourceLineNo">133</span> for (FileStatus log : logs) {<a name="line.133"></a>
<span class="sourceLineNo">134</span> p = new Path(p, log.getPath().getName());<a name="line.134"></a>
<span class="sourceLineNo">135</span> if (p.getName().equals(path.getName())) {<a name="line.135"></a>
<span class="sourceLineNo">136</span> LOG.info("Log " + p.getName() + " found at " + p);<a name="line.136"></a>
<span class="sourceLineNo">137</span> return p;<a name="line.137"></a>
<span class="sourceLineNo">138</span> }<a name="line.138"></a>
<span class="sourceLineNo">139</span> }<a name="line.139"></a>
<span class="sourceLineNo">140</span> }<a name="line.140"></a>
<span class="sourceLineNo">141</span> LOG.error("Didn't find path for: " + path.getName());<a name="line.141"></a>
<span class="sourceLineNo">142</span> return path;<a name="line.142"></a>
<span class="sourceLineNo">143</span> }<a name="line.143"></a>
<span class="sourceLineNo">144</span><a name="line.144"></a>
<span class="sourceLineNo">145</span> void tryFinish() {<a name="line.145"></a>
<span class="sourceLineNo">146</span> if (workerThreads.isEmpty()) {<a name="line.146"></a>
<span class="sourceLineNo">147</span> this.getSourceMetrics().clear();<a name="line.147"></a>
<span class="sourceLineNo">148</span> manager.finishRecoveredSource(this);<a name="line.148"></a>
<span class="sourceLineNo">149</span> }<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> @Override<a name="line.152"></a>
<span class="sourceLineNo">153</span> public String getPeerId() {<a name="line.153"></a>
<span class="sourceLineNo">154</span> return this.actualPeerId;<a name="line.154"></a>
<span class="sourceLineNo">155</span> }<a name="line.155"></a>
<span class="sourceLineNo">156</span><a name="line.156"></a>
<span class="sourceLineNo">157</span> @Override<a name="line.157"></a>
<span class="sourceLineNo">158</span> public ServerName getServerWALsBelongTo() {<a name="line.158"></a>
<span class="sourceLineNo">159</span> return this.replicationQueueInfo.getDeadRegionServers().get(0);<a name="line.159"></a>
<span class="sourceLineNo">160</span> }<a name="line.160"></a>
<span class="sourceLineNo">161</span><a name="line.161"></a>
<span class="sourceLineNo">162</span> @Override<a name="line.162"></a>
<span class="sourceLineNo">163</span> public boolean isRecovered() {<a name="line.163"></a>
<span class="sourceLineNo">164</span> return true;<a name="line.164"></a>
<span class="sourceLineNo">165</span> }<a name="line.165"></a>
<span class="sourceLineNo">166</span>}<a name="line.166"></a>
</pre>
</div>
</body>
</html>