blob: 3c30b19ace9eaf2c0a1901207ecdf68eaf5b3990 [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.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.util.ArrayList;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.util.Collection;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.Collections;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.HashMap;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import java.util.Iterator;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import java.util.List;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import java.util.Map;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import java.util.NavigableSet;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import java.util.Set;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import java.util.SortedSet;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import java.util.TreeSet;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import java.util.UUID;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import java.util.concurrent.ConcurrentMap;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import java.util.concurrent.Future;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import java.util.concurrent.RejectedExecutionException;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import java.util.concurrent.ThreadLocalRandom;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import java.util.concurrent.TimeUnit;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import java.util.stream.Collectors;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.conf.Configuration;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.fs.FileSystem;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.fs.Path;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.CompatibilitySingletonFactory;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.HConstants;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.Server;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.ServerName;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.TableName;<a name="line.51"></a>
<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.52"></a>
<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.53"></a>
<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.54"></a>
<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.55"></a>
<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.ReplicationPeerConfig;<a name="line.56"></a>
<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.ReplicationPeerImpl;<a name="line.57"></a>
<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.58"></a>
<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.59"></a>
<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.60"></a>
<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.61"></a>
<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.replication.ReplicationUtils;<a name="line.62"></a>
<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.replication.SyncReplicationState;<a name="line.63"></a>
<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.64"></a>
<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.65"></a>
<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.wal.SyncReplicationWALProvider;<a name="line.66"></a>
<span class="sourceLineNo">067</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.67"></a>
<span class="sourceLineNo">068</span>import org.apache.zookeeper.KeeperException;<a name="line.68"></a>
<span class="sourceLineNo">069</span>import org.slf4j.Logger;<a name="line.69"></a>
<span class="sourceLineNo">070</span>import org.slf4j.LoggerFactory;<a name="line.70"></a>
<span class="sourceLineNo">071</span><a name="line.71"></a>
<span class="sourceLineNo">072</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.72"></a>
<span class="sourceLineNo">073</span>import org.apache.hbase.thirdparty.com.google.common.collect.Sets;<a name="line.73"></a>
<span class="sourceLineNo">074</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.74"></a>
<span class="sourceLineNo">075</span><a name="line.75"></a>
<span class="sourceLineNo">076</span>/**<a name="line.76"></a>
<span class="sourceLineNo">077</span> * This class is responsible to manage all the replication sources. There are two classes of<a name="line.77"></a>
<span class="sourceLineNo">078</span> * sources:<a name="line.78"></a>
<span class="sourceLineNo">079</span> * &lt;ul&gt;<a name="line.79"></a>
<span class="sourceLineNo">080</span> * &lt;li&gt;Normal sources are persistent and one per peer cluster&lt;/li&gt;<a name="line.80"></a>
<span class="sourceLineNo">081</span> * &lt;li&gt;Old sources are recovered from a failed region server and our only goal is to finish<a name="line.81"></a>
<span class="sourceLineNo">082</span> * replicating the WAL queue it had&lt;/li&gt;<a name="line.82"></a>
<span class="sourceLineNo">083</span> * &lt;/ul&gt;<a name="line.83"></a>
<span class="sourceLineNo">084</span> * &lt;p&gt;<a name="line.84"></a>
<span class="sourceLineNo">085</span> * When a region server dies, this class uses a watcher to get notified and it tries to grab a lock<a name="line.85"></a>
<span class="sourceLineNo">086</span> * in order to transfer all the queues in a local old source.<a name="line.86"></a>
<span class="sourceLineNo">087</span> * &lt;p&gt;<a name="line.87"></a>
<span class="sourceLineNo">088</span> * Synchronization specification:<a name="line.88"></a>
<span class="sourceLineNo">089</span> * &lt;ul&gt;<a name="line.89"></a>
<span class="sourceLineNo">090</span> * &lt;li&gt;No need synchronized on {@link #sources}. {@link #sources} is a ConcurrentHashMap and there<a name="line.90"></a>
<span class="sourceLineNo">091</span> * is a Lock for peer id in {@link PeerProcedureHandlerImpl}. So there is no race for peer<a name="line.91"></a>
<span class="sourceLineNo">092</span> * operations.&lt;/li&gt;<a name="line.92"></a>
<span class="sourceLineNo">093</span> * &lt;li&gt;Need synchronized on {@link #walsById}. There are four methods which modify it,<a name="line.93"></a>
<span class="sourceLineNo">094</span> * {@link #addPeer(String)}, {@link #removePeer(String)},<a name="line.94"></a>
<span class="sourceLineNo">095</span> * {@link #cleanOldLogs(String, boolean, ReplicationSourceInterface)} and {@link #preLogRoll(Path)}.<a name="line.95"></a>
<span class="sourceLineNo">096</span> * {@link #walsById} is a ConcurrentHashMap and there is a Lock for peer id in<a name="line.96"></a>
<span class="sourceLineNo">097</span> * {@link PeerProcedureHandlerImpl}. So there is no race between {@link #addPeer(String)} and<a name="line.97"></a>
<span class="sourceLineNo">098</span> * {@link #removePeer(String)}. {@link #cleanOldLogs(String, boolean, ReplicationSourceInterface)}<a name="line.98"></a>
<span class="sourceLineNo">099</span> * is called by {@link ReplicationSourceInterface}. So no race with {@link #addPeer(String)}.<a name="line.99"></a>
<span class="sourceLineNo">100</span> * {@link #removePeer(String)} will terminate the {@link ReplicationSourceInterface} firstly, then<a name="line.100"></a>
<span class="sourceLineNo">101</span> * remove the wals from {@link #walsById}. So no race with {@link #removePeer(String)}. The only<a name="line.101"></a>
<span class="sourceLineNo">102</span> * case need synchronized is {@link #cleanOldLogs(String, boolean, ReplicationSourceInterface)} and<a name="line.102"></a>
<span class="sourceLineNo">103</span> * {@link #preLogRoll(Path)}.&lt;/li&gt;<a name="line.103"></a>
<span class="sourceLineNo">104</span> * &lt;li&gt;No need synchronized on {@link #walsByIdRecoveredQueues}. There are three methods which<a name="line.104"></a>
<span class="sourceLineNo">105</span> * modify it, {@link #removePeer(String)} ,<a name="line.105"></a>
<span class="sourceLineNo">106</span> * {@link #cleanOldLogs(String, boolean, ReplicationSourceInterface)} and<a name="line.106"></a>
<span class="sourceLineNo">107</span> * {@link ReplicationSourceManager.NodeFailoverWorker#run()}.<a name="line.107"></a>
<span class="sourceLineNo">108</span> * {@link #cleanOldLogs(String, boolean, ReplicationSourceInterface)} is called by<a name="line.108"></a>
<span class="sourceLineNo">109</span> * {@link ReplicationSourceInterface}. {@link #removePeer(String)} will terminate the<a name="line.109"></a>
<span class="sourceLineNo">110</span> * {@link ReplicationSourceInterface} firstly, then remove the wals from<a name="line.110"></a>
<span class="sourceLineNo">111</span> * {@link #walsByIdRecoveredQueues}. And {@link ReplicationSourceManager.NodeFailoverWorker#run()}<a name="line.111"></a>
<span class="sourceLineNo">112</span> * will add the wals to {@link #walsByIdRecoveredQueues} firstly, then start up a<a name="line.112"></a>
<span class="sourceLineNo">113</span> * {@link ReplicationSourceInterface}. So there is no race here. For<a name="line.113"></a>
<span class="sourceLineNo">114</span> * {@link ReplicationSourceManager.NodeFailoverWorker#run()} and {@link #removePeer(String)}, there<a name="line.114"></a>
<span class="sourceLineNo">115</span> * is already synchronized on {@link #oldsources}. So no need synchronized on<a name="line.115"></a>
<span class="sourceLineNo">116</span> * {@link #walsByIdRecoveredQueues}.&lt;/li&gt;<a name="line.116"></a>
<span class="sourceLineNo">117</span> * &lt;li&gt;Need synchronized on {@link #latestPaths} to avoid the new open source miss new log.&lt;/li&gt;<a name="line.117"></a>
<span class="sourceLineNo">118</span> * &lt;li&gt;Need synchronized on {@link #oldsources} to avoid adding recovered source for the<a name="line.118"></a>
<span class="sourceLineNo">119</span> * to-be-removed peer.&lt;/li&gt;<a name="line.119"></a>
<span class="sourceLineNo">120</span> * &lt;/ul&gt;<a name="line.120"></a>
<span class="sourceLineNo">121</span> */<a name="line.121"></a>
<span class="sourceLineNo">122</span>@InterfaceAudience.Private<a name="line.122"></a>
<span class="sourceLineNo">123</span>public class ReplicationSourceManager implements ReplicationListener {<a name="line.123"></a>
<span class="sourceLineNo">124</span> private static final Logger LOG = LoggerFactory.getLogger(ReplicationSourceManager.class);<a name="line.124"></a>
<span class="sourceLineNo">125</span> // all the sources that read this RS's logs and every peer only has one replication source<a name="line.125"></a>
<span class="sourceLineNo">126</span> private final ConcurrentMap&lt;String, ReplicationSourceInterface&gt; sources;<a name="line.126"></a>
<span class="sourceLineNo">127</span> // List of all the sources we got from died RSs<a name="line.127"></a>
<span class="sourceLineNo">128</span> private final List&lt;ReplicationSourceInterface&gt; oldsources;<a name="line.128"></a>
<span class="sourceLineNo">129</span> private final ReplicationQueueStorage queueStorage;<a name="line.129"></a>
<span class="sourceLineNo">130</span> private final ReplicationTracker replicationTracker;<a name="line.130"></a>
<span class="sourceLineNo">131</span> private final ReplicationPeers replicationPeers;<a name="line.131"></a>
<span class="sourceLineNo">132</span> // UUID for this cluster<a name="line.132"></a>
<span class="sourceLineNo">133</span> private final UUID clusterId;<a name="line.133"></a>
<span class="sourceLineNo">134</span> // All about stopping<a name="line.134"></a>
<span class="sourceLineNo">135</span> private final Server server;<a name="line.135"></a>
<span class="sourceLineNo">136</span><a name="line.136"></a>
<span class="sourceLineNo">137</span> // All logs we are currently tracking<a name="line.137"></a>
<span class="sourceLineNo">138</span> // Index structure of the map is: queue_id-&gt;logPrefix/logGroup-&gt;logs<a name="line.138"></a>
<span class="sourceLineNo">139</span> // For normal replication source, the peer id is same with the queue id<a name="line.139"></a>
<span class="sourceLineNo">140</span> private final ConcurrentMap&lt;String, Map&lt;String, NavigableSet&lt;String&gt;&gt;&gt; walsById;<a name="line.140"></a>
<span class="sourceLineNo">141</span> // Logs for recovered sources we are currently tracking<a name="line.141"></a>
<span class="sourceLineNo">142</span> // the map is: queue_id-&gt;logPrefix/logGroup-&gt;logs<a name="line.142"></a>
<span class="sourceLineNo">143</span> // For recovered source, the queue id's format is peer_id-servername-*<a name="line.143"></a>
<span class="sourceLineNo">144</span> private final ConcurrentMap&lt;String, Map&lt;String, NavigableSet&lt;String&gt;&gt;&gt; walsByIdRecoveredQueues;<a name="line.144"></a>
<span class="sourceLineNo">145</span><a name="line.145"></a>
<span class="sourceLineNo">146</span> private final SyncReplicationPeerMappingManager syncReplicationPeerMappingManager;<a name="line.146"></a>
<span class="sourceLineNo">147</span><a name="line.147"></a>
<span class="sourceLineNo">148</span> private final Configuration conf;<a name="line.148"></a>
<span class="sourceLineNo">149</span> private final FileSystem fs;<a name="line.149"></a>
<span class="sourceLineNo">150</span> // The paths to the latest log of each wal group, for new coming peers<a name="line.150"></a>
<span class="sourceLineNo">151</span> private final Map&lt;String, Path&gt; latestPaths;<a name="line.151"></a>
<span class="sourceLineNo">152</span> // Path to the wals directories<a name="line.152"></a>
<span class="sourceLineNo">153</span> private final Path logDir;<a name="line.153"></a>
<span class="sourceLineNo">154</span> // Path to the wal archive<a name="line.154"></a>
<span class="sourceLineNo">155</span> private final Path oldLogDir;<a name="line.155"></a>
<span class="sourceLineNo">156</span> private final WALFileLengthProvider walFileLengthProvider;<a name="line.156"></a>
<span class="sourceLineNo">157</span> // The number of ms that we wait before moving znodes, HBASE-3596<a name="line.157"></a>
<span class="sourceLineNo">158</span> private final long sleepBeforeFailover;<a name="line.158"></a>
<span class="sourceLineNo">159</span> // Homemade executer service for replication<a name="line.159"></a>
<span class="sourceLineNo">160</span> private final ThreadPoolExecutor executor;<a name="line.160"></a>
<span class="sourceLineNo">161</span><a name="line.161"></a>
<span class="sourceLineNo">162</span> private final boolean replicationForBulkLoadDataEnabled;<a name="line.162"></a>
<span class="sourceLineNo">163</span><a name="line.163"></a>
<span class="sourceLineNo">164</span> private AtomicLong totalBufferUsed = new AtomicLong();<a name="line.164"></a>
<span class="sourceLineNo">165</span><a name="line.165"></a>
<span class="sourceLineNo">166</span> // How long should we sleep for each retry when deleting remote wal files for sync replication<a name="line.166"></a>
<span class="sourceLineNo">167</span> // peer.<a name="line.167"></a>
<span class="sourceLineNo">168</span> private final long sleepForRetries;<a name="line.168"></a>
<span class="sourceLineNo">169</span> // Maximum number of retries before taking bold actions when deleting remote wal files for sync<a name="line.169"></a>
<span class="sourceLineNo">170</span> // replication peer.<a name="line.170"></a>
<span class="sourceLineNo">171</span> private final int maxRetriesMultiplier;<a name="line.171"></a>
<span class="sourceLineNo">172</span> // Total buffer size on this RegionServer for holding batched edits to be shipped.<a name="line.172"></a>
<span class="sourceLineNo">173</span> private final long totalBufferLimit;<a name="line.173"></a>
<span class="sourceLineNo">174</span> private final MetricsReplicationGlobalSourceSource globalMetrics;<a name="line.174"></a>
<span class="sourceLineNo">175</span><a name="line.175"></a>
<span class="sourceLineNo">176</span> /**<a name="line.176"></a>
<span class="sourceLineNo">177</span> * Creates a replication manager and sets the watch on all the other registered region servers<a name="line.177"></a>
<span class="sourceLineNo">178</span> * @param queueStorage the interface for manipulating replication queues<a name="line.178"></a>
<span class="sourceLineNo">179</span> * @param replicationPeers<a name="line.179"></a>
<span class="sourceLineNo">180</span> * @param replicationTracker<a name="line.180"></a>
<span class="sourceLineNo">181</span> * @param conf the configuration to use<a name="line.181"></a>
<span class="sourceLineNo">182</span> * @param server the server for this region server<a name="line.182"></a>
<span class="sourceLineNo">183</span> * @param fs the file system to use<a name="line.183"></a>
<span class="sourceLineNo">184</span> * @param logDir the directory that contains all wal directories of live RSs<a name="line.184"></a>
<span class="sourceLineNo">185</span> * @param oldLogDir the directory where old logs are archived<a name="line.185"></a>
<span class="sourceLineNo">186</span> * @param clusterId<a name="line.186"></a>
<span class="sourceLineNo">187</span> */<a name="line.187"></a>
<span class="sourceLineNo">188</span> public ReplicationSourceManager(ReplicationQueueStorage queueStorage,<a name="line.188"></a>
<span class="sourceLineNo">189</span> ReplicationPeers replicationPeers, ReplicationTracker replicationTracker, Configuration conf,<a name="line.189"></a>
<span class="sourceLineNo">190</span> Server server, FileSystem fs, Path logDir, Path oldLogDir, UUID clusterId,<a name="line.190"></a>
<span class="sourceLineNo">191</span> WALFileLengthProvider walFileLengthProvider,<a name="line.191"></a>
<span class="sourceLineNo">192</span> SyncReplicationPeerMappingManager syncReplicationPeerMappingManager,<a name="line.192"></a>
<span class="sourceLineNo">193</span> MetricsReplicationGlobalSourceSource globalMetrics) throws IOException {<a name="line.193"></a>
<span class="sourceLineNo">194</span> this.sources = new ConcurrentHashMap&lt;&gt;();<a name="line.194"></a>
<span class="sourceLineNo">195</span> this.queueStorage = queueStorage;<a name="line.195"></a>
<span class="sourceLineNo">196</span> this.replicationPeers = replicationPeers;<a name="line.196"></a>
<span class="sourceLineNo">197</span> this.replicationTracker = replicationTracker;<a name="line.197"></a>
<span class="sourceLineNo">198</span> this.server = server;<a name="line.198"></a>
<span class="sourceLineNo">199</span> this.walsById = new ConcurrentHashMap&lt;&gt;();<a name="line.199"></a>
<span class="sourceLineNo">200</span> this.walsByIdRecoveredQueues = new ConcurrentHashMap&lt;&gt;();<a name="line.200"></a>
<span class="sourceLineNo">201</span> this.oldsources = new ArrayList&lt;&gt;();<a name="line.201"></a>
<span class="sourceLineNo">202</span> this.conf = conf;<a name="line.202"></a>
<span class="sourceLineNo">203</span> this.fs = fs;<a name="line.203"></a>
<span class="sourceLineNo">204</span> this.logDir = logDir;<a name="line.204"></a>
<span class="sourceLineNo">205</span> this.oldLogDir = oldLogDir;<a name="line.205"></a>
<span class="sourceLineNo">206</span> // 30 seconds<a name="line.206"></a>
<span class="sourceLineNo">207</span> this.sleepBeforeFailover = conf.getLong("replication.sleep.before.failover", 30000);<a name="line.207"></a>
<span class="sourceLineNo">208</span> this.clusterId = clusterId;<a name="line.208"></a>
<span class="sourceLineNo">209</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.209"></a>
<span class="sourceLineNo">210</span> this.syncReplicationPeerMappingManager = syncReplicationPeerMappingManager;<a name="line.210"></a>
<span class="sourceLineNo">211</span> this.replicationTracker.registerListener(this);<a name="line.211"></a>
<span class="sourceLineNo">212</span> // It's preferable to failover 1 RS at a time, but with good zk servers<a name="line.212"></a>
<span class="sourceLineNo">213</span> // more could be processed at the same time.<a name="line.213"></a>
<span class="sourceLineNo">214</span> int nbWorkers = conf.getInt("replication.executor.workers", 1);<a name="line.214"></a>
<span class="sourceLineNo">215</span> // use a short 100ms sleep since this could be done inline with a RS startup<a name="line.215"></a>
<span class="sourceLineNo">216</span> // even if we fail, other region servers can take care of it<a name="line.216"></a>
<span class="sourceLineNo">217</span> this.executor = new ThreadPoolExecutor(nbWorkers, nbWorkers, 100,<a name="line.217"></a>
<span class="sourceLineNo">218</span> TimeUnit.MILLISECONDS, new LinkedBlockingQueue&lt;&gt;());<a name="line.218"></a>
<span class="sourceLineNo">219</span> ThreadFactoryBuilder tfb = new ThreadFactoryBuilder();<a name="line.219"></a>
<span class="sourceLineNo">220</span> tfb.setNameFormat("ReplicationExecutor-%d");<a name="line.220"></a>
<span class="sourceLineNo">221</span> tfb.setDaemon(true);<a name="line.221"></a>
<span class="sourceLineNo">222</span> this.executor.setThreadFactory(tfb.build());<a name="line.222"></a>
<span class="sourceLineNo">223</span> this.latestPaths = new HashMap&lt;&gt;();<a name="line.223"></a>
<span class="sourceLineNo">224</span> this.replicationForBulkLoadDataEnabled = conf.getBoolean(<a name="line.224"></a>
<span class="sourceLineNo">225</span> HConstants.REPLICATION_BULKLOAD_ENABLE_KEY, HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.225"></a>
<span class="sourceLineNo">226</span> this.sleepForRetries = this.conf.getLong("replication.source.sync.sleepforretries", 1000);<a name="line.226"></a>
<span class="sourceLineNo">227</span> this.maxRetriesMultiplier =<a name="line.227"></a>
<span class="sourceLineNo">228</span> this.conf.getInt("replication.source.sync.maxretriesmultiplier", 60);<a name="line.228"></a>
<span class="sourceLineNo">229</span> this.totalBufferLimit = conf.getLong(HConstants.REPLICATION_SOURCE_TOTAL_BUFFER_KEY,<a name="line.229"></a>
<span class="sourceLineNo">230</span> HConstants.REPLICATION_SOURCE_TOTAL_BUFFER_DFAULT);<a name="line.230"></a>
<span class="sourceLineNo">231</span> this.globalMetrics = globalMetrics;<a name="line.231"></a>
<span class="sourceLineNo">232</span> }<a name="line.232"></a>
<span class="sourceLineNo">233</span><a name="line.233"></a>
<span class="sourceLineNo">234</span> /**<a name="line.234"></a>
<span class="sourceLineNo">235</span> * Adds a normal source per registered peer cluster and tries to process all old region server wal<a name="line.235"></a>
<span class="sourceLineNo">236</span> * queues<a name="line.236"></a>
<span class="sourceLineNo">237</span> * &lt;p&gt;<a name="line.237"></a>
<span class="sourceLineNo">238</span> * The returned future is for adoptAbandonedQueues task.<a name="line.238"></a>
<span class="sourceLineNo">239</span> */<a name="line.239"></a>
<span class="sourceLineNo">240</span> Future&lt;?&gt; init() throws IOException {<a name="line.240"></a>
<span class="sourceLineNo">241</span> for (String id : this.replicationPeers.getAllPeerIds()) {<a name="line.241"></a>
<span class="sourceLineNo">242</span> addSource(id);<a name="line.242"></a>
<span class="sourceLineNo">243</span> if (replicationForBulkLoadDataEnabled) {<a name="line.243"></a>
<span class="sourceLineNo">244</span> // Check if peer exists in hfile-refs queue, if not add it. This can happen in the case<a name="line.244"></a>
<span class="sourceLineNo">245</span> // when a peer was added before replication for bulk loaded data was enabled.<a name="line.245"></a>
<span class="sourceLineNo">246</span> throwIOExceptionWhenFail(() -&gt; this.queueStorage.addPeerToHFileRefs(id));<a name="line.246"></a>
<span class="sourceLineNo">247</span> }<a name="line.247"></a>
<span class="sourceLineNo">248</span> }<a name="line.248"></a>
<span class="sourceLineNo">249</span> return this.executor.submit(this::adoptAbandonedQueues);<a name="line.249"></a>
<span class="sourceLineNo">250</span> }<a name="line.250"></a>
<span class="sourceLineNo">251</span><a name="line.251"></a>
<span class="sourceLineNo">252</span> private void adoptAbandonedQueues() {<a name="line.252"></a>
<span class="sourceLineNo">253</span> List&lt;ServerName&gt; currentReplicators = null;<a name="line.253"></a>
<span class="sourceLineNo">254</span> try {<a name="line.254"></a>
<span class="sourceLineNo">255</span> currentReplicators = queueStorage.getListOfReplicators();<a name="line.255"></a>
<span class="sourceLineNo">256</span> } catch (ReplicationException e) {<a name="line.256"></a>
<span class="sourceLineNo">257</span> server.abort("Failed to get all replicators", e);<a name="line.257"></a>
<span class="sourceLineNo">258</span> return;<a name="line.258"></a>
<span class="sourceLineNo">259</span> }<a name="line.259"></a>
<span class="sourceLineNo">260</span> if (currentReplicators == null || currentReplicators.isEmpty()) {<a name="line.260"></a>
<span class="sourceLineNo">261</span> return;<a name="line.261"></a>
<span class="sourceLineNo">262</span> }<a name="line.262"></a>
<span class="sourceLineNo">263</span> List&lt;ServerName&gt; otherRegionServers = replicationTracker.getListOfRegionServers().stream()<a name="line.263"></a>
<span class="sourceLineNo">264</span> .map(ServerName::valueOf).collect(Collectors.toList());<a name="line.264"></a>
<span class="sourceLineNo">265</span> LOG.info(<a name="line.265"></a>
<span class="sourceLineNo">266</span> "Current list of replicators: " + currentReplicators + " other RSs: " + otherRegionServers);<a name="line.266"></a>
<span class="sourceLineNo">267</span><a name="line.267"></a>
<span class="sourceLineNo">268</span> // Look if there's anything to process after a restart<a name="line.268"></a>
<span class="sourceLineNo">269</span> for (ServerName rs : currentReplicators) {<a name="line.269"></a>
<span class="sourceLineNo">270</span> if (!otherRegionServers.contains(rs)) {<a name="line.270"></a>
<span class="sourceLineNo">271</span> transferQueues(rs);<a name="line.271"></a>
<span class="sourceLineNo">272</span> }<a name="line.272"></a>
<span class="sourceLineNo">273</span> }<a name="line.273"></a>
<span class="sourceLineNo">274</span> }<a name="line.274"></a>
<span class="sourceLineNo">275</span><a name="line.275"></a>
<span class="sourceLineNo">276</span> /**<a name="line.276"></a>
<span class="sourceLineNo">277</span> * &lt;ol&gt;<a name="line.277"></a>
<span class="sourceLineNo">278</span> * &lt;li&gt;Add peer to replicationPeers&lt;/li&gt;<a name="line.278"></a>
<span class="sourceLineNo">279</span> * &lt;li&gt;Add the normal source and related replication queue&lt;/li&gt;<a name="line.279"></a>
<span class="sourceLineNo">280</span> * &lt;li&gt;Add HFile Refs&lt;/li&gt;<a name="line.280"></a>
<span class="sourceLineNo">281</span> * &lt;/ol&gt;<a name="line.281"></a>
<span class="sourceLineNo">282</span> * @param peerId the id of replication peer<a name="line.282"></a>
<span class="sourceLineNo">283</span> */<a name="line.283"></a>
<span class="sourceLineNo">284</span> public void addPeer(String peerId) throws IOException {<a name="line.284"></a>
<span class="sourceLineNo">285</span> boolean added = false;<a name="line.285"></a>
<span class="sourceLineNo">286</span> try {<a name="line.286"></a>
<span class="sourceLineNo">287</span> added = this.replicationPeers.addPeer(peerId);<a name="line.287"></a>
<span class="sourceLineNo">288</span> } catch (ReplicationException e) {<a name="line.288"></a>
<span class="sourceLineNo">289</span> throw new IOException(e);<a name="line.289"></a>
<span class="sourceLineNo">290</span> }<a name="line.290"></a>
<span class="sourceLineNo">291</span> if (added) {<a name="line.291"></a>
<span class="sourceLineNo">292</span> addSource(peerId);<a name="line.292"></a>
<span class="sourceLineNo">293</span> if (replicationForBulkLoadDataEnabled) {<a name="line.293"></a>
<span class="sourceLineNo">294</span> throwIOExceptionWhenFail(() -&gt; this.queueStorage.addPeerToHFileRefs(peerId));<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>
<span class="sourceLineNo">297</span> }<a name="line.297"></a>
<span class="sourceLineNo">298</span><a name="line.298"></a>
<span class="sourceLineNo">299</span> /**<a name="line.299"></a>
<span class="sourceLineNo">300</span> * &lt;ol&gt;<a name="line.300"></a>
<span class="sourceLineNo">301</span> * &lt;li&gt;Remove peer for replicationPeers&lt;/li&gt;<a name="line.301"></a>
<span class="sourceLineNo">302</span> * &lt;li&gt;Remove all the recovered sources for the specified id and related replication queues&lt;/li&gt;<a name="line.302"></a>
<span class="sourceLineNo">303</span> * &lt;li&gt;Remove the normal source and related replication queue&lt;/li&gt;<a name="line.303"></a>
<span class="sourceLineNo">304</span> * &lt;li&gt;Remove HFile Refs&lt;/li&gt;<a name="line.304"></a>
<span class="sourceLineNo">305</span> * &lt;/ol&gt;<a name="line.305"></a>
<span class="sourceLineNo">306</span> * @param peerId the id of the replication peer<a name="line.306"></a>
<span class="sourceLineNo">307</span> */<a name="line.307"></a>
<span class="sourceLineNo">308</span> public void removePeer(String peerId) {<a name="line.308"></a>
<span class="sourceLineNo">309</span> ReplicationPeer peer = replicationPeers.removePeer(peerId);<a name="line.309"></a>
<span class="sourceLineNo">310</span> String terminateMessage = "Replication stream was removed by a user";<a name="line.310"></a>
<span class="sourceLineNo">311</span> List&lt;ReplicationSourceInterface&gt; oldSourcesToDelete = new ArrayList&lt;&gt;();<a name="line.311"></a>
<span class="sourceLineNo">312</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.312"></a>
<span class="sourceLineNo">313</span> // see NodeFailoverWorker.run<a name="line.313"></a>
<span class="sourceLineNo">314</span> synchronized (this.oldsources) {<a name="line.314"></a>
<span class="sourceLineNo">315</span> // First close all the recovered sources for this peer<a name="line.315"></a>
<span class="sourceLineNo">316</span> for (ReplicationSourceInterface src : oldsources) {<a name="line.316"></a>
<span class="sourceLineNo">317</span> if (peerId.equals(src.getPeerId())) {<a name="line.317"></a>
<span class="sourceLineNo">318</span> oldSourcesToDelete.add(src);<a name="line.318"></a>
<span class="sourceLineNo">319</span> }<a name="line.319"></a>
<span class="sourceLineNo">320</span> }<a name="line.320"></a>
<span class="sourceLineNo">321</span> for (ReplicationSourceInterface src : oldSourcesToDelete) {<a name="line.321"></a>
<span class="sourceLineNo">322</span> src.terminate(terminateMessage);<a name="line.322"></a>
<span class="sourceLineNo">323</span> removeRecoveredSource(src);<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> LOG.info(<a name="line.326"></a>
<span class="sourceLineNo">327</span> "Number of deleted recovered sources for " + peerId + ": " + oldSourcesToDelete.size());<a name="line.327"></a>
<span class="sourceLineNo">328</span> // Now close the normal source for this peer<a name="line.328"></a>
<span class="sourceLineNo">329</span> ReplicationSourceInterface srcToRemove = this.sources.get(peerId);<a name="line.329"></a>
<span class="sourceLineNo">330</span> if (srcToRemove != null) {<a name="line.330"></a>
<span class="sourceLineNo">331</span> srcToRemove.terminate(terminateMessage);<a name="line.331"></a>
<span class="sourceLineNo">332</span> removeSource(srcToRemove);<a name="line.332"></a>
<span class="sourceLineNo">333</span> } else {<a name="line.333"></a>
<span class="sourceLineNo">334</span> // This only happened in unit test TestReplicationSourceManager#testPeerRemovalCleanup<a name="line.334"></a>
<span class="sourceLineNo">335</span> // Delete queue from storage and memory and queue id is same with peer id for normal<a name="line.335"></a>
<span class="sourceLineNo">336</span> // source<a name="line.336"></a>
<span class="sourceLineNo">337</span> deleteQueue(peerId);<a name="line.337"></a>
<span class="sourceLineNo">338</span> this.walsById.remove(peerId);<a name="line.338"></a>
<span class="sourceLineNo">339</span> }<a name="line.339"></a>
<span class="sourceLineNo">340</span> ReplicationPeerConfig peerConfig = peer.getPeerConfig();<a name="line.340"></a>
<span class="sourceLineNo">341</span> if (peerConfig.isSyncReplication()) {<a name="line.341"></a>
<span class="sourceLineNo">342</span> syncReplicationPeerMappingManager.remove(peerId, peerConfig);<a name="line.342"></a>
<span class="sourceLineNo">343</span> }<a name="line.343"></a>
<span class="sourceLineNo">344</span> // Remove HFile Refs<a name="line.344"></a>
<span class="sourceLineNo">345</span> abortWhenFail(() -&gt; this.queueStorage.removePeerFromHFileRefs(peerId));<a name="line.345"></a>
<span class="sourceLineNo">346</span> }<a name="line.346"></a>
<span class="sourceLineNo">347</span><a name="line.347"></a>
<span class="sourceLineNo">348</span> /**<a name="line.348"></a>
<span class="sourceLineNo">349</span> * Factory method to create a replication source<a name="line.349"></a>
<span class="sourceLineNo">350</span> * @param queueId the id of the replication queue<a name="line.350"></a>
<span class="sourceLineNo">351</span> * @return the created source<a name="line.351"></a>
<span class="sourceLineNo">352</span> */<a name="line.352"></a>
<span class="sourceLineNo">353</span> private ReplicationSourceInterface createSource(String queueId, ReplicationPeer replicationPeer)<a name="line.353"></a>
<span class="sourceLineNo">354</span> throws IOException {<a name="line.354"></a>
<span class="sourceLineNo">355</span> ReplicationSourceInterface src = ReplicationSourceFactory.create(conf, queueId);<a name="line.355"></a>
<span class="sourceLineNo">356</span><a name="line.356"></a>
<span class="sourceLineNo">357</span> MetricsSource metrics = new MetricsSource(queueId);<a name="line.357"></a>
<span class="sourceLineNo">358</span> // init replication source<a name="line.358"></a>
<span class="sourceLineNo">359</span> src.init(conf, fs, this, queueStorage, replicationPeer, server, queueId, clusterId,<a name="line.359"></a>
<span class="sourceLineNo">360</span> walFileLengthProvider, metrics);<a name="line.360"></a>
<span class="sourceLineNo">361</span> return src;<a name="line.361"></a>
<span class="sourceLineNo">362</span> }<a name="line.362"></a>
<span class="sourceLineNo">363</span><a name="line.363"></a>
<span class="sourceLineNo">364</span> /**<a name="line.364"></a>
<span class="sourceLineNo">365</span> * Add a normal source for the given peer on this region server. Meanwhile, add new replication<a name="line.365"></a>
<span class="sourceLineNo">366</span> * queue to storage. For the newly added peer, we only need to enqueue the latest log of each wal<a name="line.366"></a>
<span class="sourceLineNo">367</span> * group and do replication<a name="line.367"></a>
<span class="sourceLineNo">368</span> * @param peerId the id of the replication peer<a name="line.368"></a>
<span class="sourceLineNo">369</span> * @return the source that was created<a name="line.369"></a>
<span class="sourceLineNo">370</span> */<a name="line.370"></a>
<span class="sourceLineNo">371</span> @VisibleForTesting<a name="line.371"></a>
<span class="sourceLineNo">372</span> ReplicationSourceInterface addSource(String peerId) throws IOException {<a name="line.372"></a>
<span class="sourceLineNo">373</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.373"></a>
<span class="sourceLineNo">374</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.374"></a>
<span class="sourceLineNo">375</span> // synchronized on latestPaths to avoid missing the new log<a name="line.375"></a>
<span class="sourceLineNo">376</span> synchronized (this.latestPaths) {<a name="line.376"></a>
<span class="sourceLineNo">377</span> this.sources.put(peerId, src);<a name="line.377"></a>
<span class="sourceLineNo">378</span> Map&lt;String, NavigableSet&lt;String&gt;&gt; walsByGroup = new HashMap&lt;&gt;();<a name="line.378"></a>
<span class="sourceLineNo">379</span> this.walsById.put(peerId, walsByGroup);<a name="line.379"></a>
<span class="sourceLineNo">380</span> // Add the latest wal to that source's queue<a name="line.380"></a>
<span class="sourceLineNo">381</span> if (!latestPaths.isEmpty()) {<a name="line.381"></a>
<span class="sourceLineNo">382</span> for (Map.Entry&lt;String, Path&gt; walPrefixAndPath : latestPaths.entrySet()) {<a name="line.382"></a>
<span class="sourceLineNo">383</span> Path walPath = walPrefixAndPath.getValue();<a name="line.383"></a>
<span class="sourceLineNo">384</span> NavigableSet&lt;String&gt; wals = new TreeSet&lt;&gt;();<a name="line.384"></a>
<span class="sourceLineNo">385</span> wals.add(walPath.getName());<a name="line.385"></a>
<span class="sourceLineNo">386</span> walsByGroup.put(walPrefixAndPath.getKey(), wals);<a name="line.386"></a>
<span class="sourceLineNo">387</span> // Abort RS and throw exception to make add peer failed<a name="line.387"></a>
<span class="sourceLineNo">388</span> abortAndThrowIOExceptionWhenFail(<a name="line.388"></a>
<span class="sourceLineNo">389</span> () -&gt; this.queueStorage.addWAL(server.getServerName(), peerId, walPath.getName()));<a name="line.389"></a>
<span class="sourceLineNo">390</span> src.enqueueLog(walPath);<a name="line.390"></a>
<span class="sourceLineNo">391</span> LOG.trace("Enqueued {} to source {} during source creation.", walPath, src.getQueueId());<a name="line.391"></a>
<span class="sourceLineNo">392</span> }<a name="line.392"></a>
<span class="sourceLineNo">393</span> }<a name="line.393"></a>
<span class="sourceLineNo">394</span> }<a name="line.394"></a>
<span class="sourceLineNo">395</span> ReplicationPeerConfig peerConfig = peer.getPeerConfig();<a name="line.395"></a>
<span class="sourceLineNo">396</span> if (peerConfig.isSyncReplication()) {<a name="line.396"></a>
<span class="sourceLineNo">397</span> syncReplicationPeerMappingManager.add(peer.getId(), peerConfig);<a name="line.397"></a>
<span class="sourceLineNo">398</span> }<a name="line.398"></a>
<span class="sourceLineNo">399</span> src.startup();<a name="line.399"></a>
<span class="sourceLineNo">400</span> return src;<a name="line.400"></a>
<span class="sourceLineNo">401</span> }<a name="line.401"></a>
<span class="sourceLineNo">402</span><a name="line.402"></a>
<span class="sourceLineNo">403</span> /**<a name="line.403"></a>
<span class="sourceLineNo">404</span> * &lt;p&gt;<a name="line.404"></a>
<span class="sourceLineNo">405</span> * This is used when we transit a sync replication peer to {@link SyncReplicationState#STANDBY}.<a name="line.405"></a>
<span class="sourceLineNo">406</span> * &lt;/p&gt;<a name="line.406"></a>
<span class="sourceLineNo">407</span> * &lt;p&gt;<a name="line.407"></a>
<span class="sourceLineNo">408</span> * When transiting to {@link SyncReplicationState#STANDBY}, we can remove all the pending wal<a name="line.408"></a>
<span class="sourceLineNo">409</span> * files for a replication peer as we do not need to replicate them any more. And this is<a name="line.409"></a>
<span class="sourceLineNo">410</span> * necessary, otherwise when we transit back to {@link SyncReplicationState#DOWNGRADE_ACTIVE}<a name="line.410"></a>
<span class="sourceLineNo">411</span> * later, the stale data will be replicated again and cause inconsistency.<a name="line.411"></a>
<span class="sourceLineNo">412</span> * &lt;/p&gt;<a name="line.412"></a>
<span class="sourceLineNo">413</span> * &lt;p&gt;<a name="line.413"></a>
<span class="sourceLineNo">414</span> * See HBASE-20426 for more details.<a name="line.414"></a>
<span class="sourceLineNo">415</span> * &lt;/p&gt;<a name="line.415"></a>
<span class="sourceLineNo">416</span> * @param peerId the id of the sync replication peer<a name="line.416"></a>
<span class="sourceLineNo">417</span> */<a name="line.417"></a>
<span class="sourceLineNo">418</span> public void drainSources(String peerId) throws IOException, ReplicationException {<a name="line.418"></a>
<span class="sourceLineNo">419</span> String terminateMessage = "Sync replication peer " + peerId +<a name="line.419"></a>
<span class="sourceLineNo">420</span> " is transiting to STANDBY. Will close the previous replication source and open a new one";<a name="line.420"></a>
<span class="sourceLineNo">421</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.421"></a>
<span class="sourceLineNo">422</span> assert peer.getPeerConfig().isSyncReplication();<a name="line.422"></a>
<span class="sourceLineNo">423</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.423"></a>
<span class="sourceLineNo">424</span> // synchronized here to avoid race with preLogRoll where we add new log to source and also<a name="line.424"></a>
<span class="sourceLineNo">425</span> // walsById.<a name="line.425"></a>
<span class="sourceLineNo">426</span> ReplicationSourceInterface toRemove;<a name="line.426"></a>
<span class="sourceLineNo">427</span> Map&lt;String, NavigableSet&lt;String&gt;&gt; wals = new HashMap&lt;&gt;();<a name="line.427"></a>
<span class="sourceLineNo">428</span> synchronized (latestPaths) {<a name="line.428"></a>
<span class="sourceLineNo">429</span> toRemove = sources.put(peerId, src);<a name="line.429"></a>
<span class="sourceLineNo">430</span> if (toRemove != null) {<a name="line.430"></a>
<span class="sourceLineNo">431</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.431"></a>
<span class="sourceLineNo">432</span> toRemove.terminate(terminateMessage);<a name="line.432"></a>
<span class="sourceLineNo">433</span> toRemove.getSourceMetrics().clear();<a name="line.433"></a>
<span class="sourceLineNo">434</span> }<a name="line.434"></a>
<span class="sourceLineNo">435</span> // Here we make a copy of all the remaining wal files and then delete them from the<a name="line.435"></a>
<span class="sourceLineNo">436</span> // replication queue storage after releasing the lock. It is not safe to just remove the old<a name="line.436"></a>
<span class="sourceLineNo">437</span> // map from walsById since later we may fail to delete them from the replication queue<a name="line.437"></a>
<span class="sourceLineNo">438</span> // storage, and when we retry next time, we can not know the wal files that need to be deleted<a name="line.438"></a>
<span class="sourceLineNo">439</span> // from the replication queue storage.<a name="line.439"></a>
<span class="sourceLineNo">440</span> walsById.get(peerId).forEach((k, v) -&gt; wals.put(k, new TreeSet&lt;&gt;(v)));<a name="line.440"></a>
<span class="sourceLineNo">441</span> }<a name="line.441"></a>
<span class="sourceLineNo">442</span> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.442"></a>
<span class="sourceLineNo">443</span> src.startup();<a name="line.443"></a>
<span class="sourceLineNo">444</span> for (NavigableSet&lt;String&gt; walsByGroup : wals.values()) {<a name="line.444"></a>
<span class="sourceLineNo">445</span> for (String wal : walsByGroup) {<a name="line.445"></a>
<span class="sourceLineNo">446</span> queueStorage.removeWAL(server.getServerName(), peerId, wal);<a name="line.446"></a>
<span class="sourceLineNo">447</span> }<a name="line.447"></a>
<span class="sourceLineNo">448</span> }<a name="line.448"></a>
<span class="sourceLineNo">449</span> synchronized (walsById) {<a name="line.449"></a>
<span class="sourceLineNo">450</span> Map&lt;String, NavigableSet&lt;String&gt;&gt; oldWals = walsById.get(peerId);<a name="line.450"></a>
<span class="sourceLineNo">451</span> wals.forEach((k, v) -&gt; {<a name="line.451"></a>
<span class="sourceLineNo">452</span> NavigableSet&lt;String&gt; walsByGroup = oldWals.get(k);<a name="line.452"></a>
<span class="sourceLineNo">453</span> if (walsByGroup != null) {<a name="line.453"></a>
<span class="sourceLineNo">454</span> walsByGroup.removeAll(v);<a name="line.454"></a>
<span class="sourceLineNo">455</span> }<a name="line.455"></a>
<span class="sourceLineNo">456</span> });<a name="line.456"></a>
<span class="sourceLineNo">457</span> }<a name="line.457"></a>
<span class="sourceLineNo">458</span> // synchronized on oldsources to avoid race with NodeFailoverWorker. Since NodeFailoverWorker is<a name="line.458"></a>
<span class="sourceLineNo">459</span> // a background task, we will delete the file from replication queue storage under the lock to<a name="line.459"></a>
<span class="sourceLineNo">460</span> // simplify the logic.<a name="line.460"></a>
<span class="sourceLineNo">461</span> synchronized (this.oldsources) {<a name="line.461"></a>
<span class="sourceLineNo">462</span> for (Iterator&lt;ReplicationSourceInterface&gt; iter = oldsources.iterator(); iter.hasNext();) {<a name="line.462"></a>
<span class="sourceLineNo">463</span> ReplicationSourceInterface oldSource = iter.next();<a name="line.463"></a>
<span class="sourceLineNo">464</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.464"></a>
<span class="sourceLineNo">465</span> String queueId = oldSource.getQueueId();<a name="line.465"></a>
<span class="sourceLineNo">466</span> oldSource.terminate(terminateMessage);<a name="line.466"></a>
<span class="sourceLineNo">467</span> oldSource.getSourceMetrics().clear();<a name="line.467"></a>
<span class="sourceLineNo">468</span> queueStorage.removeQueue(server.getServerName(), queueId);<a name="line.468"></a>
<span class="sourceLineNo">469</span> walsByIdRecoveredQueues.remove(queueId);<a name="line.469"></a>
<span class="sourceLineNo">470</span> iter.remove();<a name="line.470"></a>
<span class="sourceLineNo">471</span> }<a name="line.471"></a>
<span class="sourceLineNo">472</span> }<a name="line.472"></a>
<span class="sourceLineNo">473</span> }<a name="line.473"></a>
<span class="sourceLineNo">474</span> }<a name="line.474"></a>
<span class="sourceLineNo">475</span><a name="line.475"></a>
<span class="sourceLineNo">476</span> /**<a name="line.476"></a>
<span class="sourceLineNo">477</span> * Close the previous replication sources of this peer id and open new sources to trigger the new<a name="line.477"></a>
<span class="sourceLineNo">478</span> * replication state changes or new replication config changes. Here we don't need to change<a name="line.478"></a>
<span class="sourceLineNo">479</span> * replication queue storage and only to enqueue all logs to the new replication source<a name="line.479"></a>
<span class="sourceLineNo">480</span> * @param peerId the id of the replication peer<a name="line.480"></a>
<span class="sourceLineNo">481</span> */<a name="line.481"></a>
<span class="sourceLineNo">482</span> public void refreshSources(String peerId) throws IOException {<a name="line.482"></a>
<span class="sourceLineNo">483</span> String terminateMessage = "Peer " + peerId +<a name="line.483"></a>
<span class="sourceLineNo">484</span> " state or config changed. Will close the previous replication source and open a new one";<a name="line.484"></a>
<span class="sourceLineNo">485</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.485"></a>
<span class="sourceLineNo">486</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.486"></a>
<span class="sourceLineNo">487</span> // synchronized on latestPaths to avoid missing the new log<a name="line.487"></a>
<span class="sourceLineNo">488</span> synchronized (this.latestPaths) {<a name="line.488"></a>
<span class="sourceLineNo">489</span> ReplicationSourceInterface toRemove = this.sources.put(peerId, src);<a name="line.489"></a>
<span class="sourceLineNo">490</span> if (toRemove != null) {<a name="line.490"></a>
<span class="sourceLineNo">491</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.491"></a>
<span class="sourceLineNo">492</span> // Do not clear metrics<a name="line.492"></a>
<span class="sourceLineNo">493</span> toRemove.terminate(terminateMessage, null, false);<a name="line.493"></a>
<span class="sourceLineNo">494</span> }<a name="line.494"></a>
<span class="sourceLineNo">495</span> for (NavigableSet&lt;String&gt; walsByGroup : walsById.get(peerId).values()) {<a name="line.495"></a>
<span class="sourceLineNo">496</span> walsByGroup.forEach(wal -&gt; src.enqueueLog(new Path(this.logDir, wal)));<a name="line.496"></a>
<span class="sourceLineNo">497</span> }<a name="line.497"></a>
<span class="sourceLineNo">498</span> }<a name="line.498"></a>
<span class="sourceLineNo">499</span> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.499"></a>
<span class="sourceLineNo">500</span> src.startup();<a name="line.500"></a>
<span class="sourceLineNo">501</span><a name="line.501"></a>
<span class="sourceLineNo">502</span> List&lt;ReplicationSourceInterface&gt; toStartup = new ArrayList&lt;&gt;();<a name="line.502"></a>
<span class="sourceLineNo">503</span> // synchronized on oldsources to avoid race with NodeFailoverWorker<a name="line.503"></a>
<span class="sourceLineNo">504</span> synchronized (this.oldsources) {<a name="line.504"></a>
<span class="sourceLineNo">505</span> List&lt;String&gt; previousQueueIds = new ArrayList&lt;&gt;();<a name="line.505"></a>
<span class="sourceLineNo">506</span> for (Iterator&lt;ReplicationSourceInterface&gt; iter = this.oldsources.iterator(); iter<a name="line.506"></a>
<span class="sourceLineNo">507</span> .hasNext();) {<a name="line.507"></a>
<span class="sourceLineNo">508</span> ReplicationSourceInterface oldSource = iter.next();<a name="line.508"></a>
<span class="sourceLineNo">509</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.509"></a>
<span class="sourceLineNo">510</span> previousQueueIds.add(oldSource.getQueueId());<a name="line.510"></a>
<span class="sourceLineNo">511</span> oldSource.terminate(terminateMessage);<a name="line.511"></a>
<span class="sourceLineNo">512</span> iter.remove();<a name="line.512"></a>
<span class="sourceLineNo">513</span> }<a name="line.513"></a>
<span class="sourceLineNo">514</span> }<a name="line.514"></a>
<span class="sourceLineNo">515</span> for (String queueId : previousQueueIds) {<a name="line.515"></a>
<span class="sourceLineNo">516</span> ReplicationSourceInterface recoveredReplicationSource = createSource(queueId, peer);<a name="line.516"></a>
<span class="sourceLineNo">517</span> this.oldsources.add(recoveredReplicationSource);<a name="line.517"></a>
<span class="sourceLineNo">518</span> for (SortedSet&lt;String&gt; walsByGroup : walsByIdRecoveredQueues.get(queueId).values()) {<a name="line.518"></a>
<span class="sourceLineNo">519</span> walsByGroup.forEach(wal -&gt; recoveredReplicationSource.enqueueLog(new Path(wal)));<a name="line.519"></a>
<span class="sourceLineNo">520</span> }<a name="line.520"></a>
<span class="sourceLineNo">521</span> toStartup.add(recoveredReplicationSource);<a name="line.521"></a>
<span class="sourceLineNo">522</span> }<a name="line.522"></a>
<span class="sourceLineNo">523</span> }<a name="line.523"></a>
<span class="sourceLineNo">524</span> for (ReplicationSourceInterface replicationSource : toStartup) {<a name="line.524"></a>
<span class="sourceLineNo">525</span> replicationSource.startup();<a name="line.525"></a>
<span class="sourceLineNo">526</span> }<a name="line.526"></a>
<span class="sourceLineNo">527</span> }<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> * Clear the metrics and related replication queue of the specified old source<a name="line.530"></a>
<span class="sourceLineNo">531</span> * @param src source to clear<a name="line.531"></a>
<span class="sourceLineNo">532</span> */<a name="line.532"></a>
<span class="sourceLineNo">533</span> private boolean removeRecoveredSource(ReplicationSourceInterface src) {<a name="line.533"></a>
<span class="sourceLineNo">534</span> if (!this.oldsources.remove(src)) {<a name="line.534"></a>
<span class="sourceLineNo">535</span> return false;<a name="line.535"></a>
<span class="sourceLineNo">536</span> }<a name="line.536"></a>
<span class="sourceLineNo">537</span> LOG.info("Done with the recovered queue {}", src.getQueueId());<a name="line.537"></a>
<span class="sourceLineNo">538</span> // Delete queue from storage and memory<a name="line.538"></a>
<span class="sourceLineNo">539</span> deleteQueue(src.getQueueId());<a name="line.539"></a>
<span class="sourceLineNo">540</span> this.walsByIdRecoveredQueues.remove(src.getQueueId());<a name="line.540"></a>
<span class="sourceLineNo">541</span> return true;<a name="line.541"></a>
<span class="sourceLineNo">542</span> }<a name="line.542"></a>
<span class="sourceLineNo">543</span><a name="line.543"></a>
<span class="sourceLineNo">544</span> void finishRecoveredSource(ReplicationSourceInterface src) {<a name="line.544"></a>
<span class="sourceLineNo">545</span> synchronized (oldsources) {<a name="line.545"></a>
<span class="sourceLineNo">546</span> if (!removeRecoveredSource(src)) {<a name="line.546"></a>
<span class="sourceLineNo">547</span> return;<a name="line.547"></a>
<span class="sourceLineNo">548</span> }<a name="line.548"></a>
<span class="sourceLineNo">549</span> }<a name="line.549"></a>
<span class="sourceLineNo">550</span> LOG.info("Finished recovering queue {} with the following stats: {}", src.getQueueId(),<a name="line.550"></a>
<span class="sourceLineNo">551</span> src.getStats());<a name="line.551"></a>
<span class="sourceLineNo">552</span> }<a name="line.552"></a>
<span class="sourceLineNo">553</span><a name="line.553"></a>
<span class="sourceLineNo">554</span> /**<a name="line.554"></a>
<span class="sourceLineNo">555</span> * Clear the metrics and related replication queue of the specified old source<a name="line.555"></a>
<span class="sourceLineNo">556</span> * @param src source to clear<a name="line.556"></a>
<span class="sourceLineNo">557</span> */<a name="line.557"></a>
<span class="sourceLineNo">558</span> void removeSource(ReplicationSourceInterface src) {<a name="line.558"></a>
<span class="sourceLineNo">559</span> LOG.info("Done with the queue " + src.getQueueId());<a name="line.559"></a>
<span class="sourceLineNo">560</span> this.sources.remove(src.getPeerId());<a name="line.560"></a>
<span class="sourceLineNo">561</span> // Delete queue from storage and memory<a name="line.561"></a>
<span class="sourceLineNo">562</span> deleteQueue(src.getQueueId());<a name="line.562"></a>
<span class="sourceLineNo">563</span> this.walsById.remove(src.getQueueId());<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><a name="line.566"></a>
<span class="sourceLineNo">567</span> /**<a name="line.567"></a>
<span class="sourceLineNo">568</span> * Delete a complete queue of wals associated with a replication source<a name="line.568"></a>
<span class="sourceLineNo">569</span> * @param queueId the id of replication queue to delete<a name="line.569"></a>
<span class="sourceLineNo">570</span> */<a name="line.570"></a>
<span class="sourceLineNo">571</span> private void deleteQueue(String queueId) {<a name="line.571"></a>
<span class="sourceLineNo">572</span> abortWhenFail(() -&gt; this.queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.572"></a>
<span class="sourceLineNo">573</span> }<a name="line.573"></a>
<span class="sourceLineNo">574</span><a name="line.574"></a>
<span class="sourceLineNo">575</span> @FunctionalInterface<a name="line.575"></a>
<span class="sourceLineNo">576</span> private interface ReplicationQueueOperation {<a name="line.576"></a>
<span class="sourceLineNo">577</span> void exec() throws ReplicationException;<a name="line.577"></a>
<span class="sourceLineNo">578</span> }<a name="line.578"></a>
<span class="sourceLineNo">579</span><a name="line.579"></a>
<span class="sourceLineNo">580</span> /**<a name="line.580"></a>
<span class="sourceLineNo">581</span> * Refresh replication source will terminate the old source first, then the source thread will be<a name="line.581"></a>
<span class="sourceLineNo">582</span> * interrupted. Need to handle it instead of abort the region server.<a name="line.582"></a>
<span class="sourceLineNo">583</span> */<a name="line.583"></a>
<span class="sourceLineNo">584</span> private void interruptOrAbortWhenFail(ReplicationQueueOperation op) {<a name="line.584"></a>
<span class="sourceLineNo">585</span> try {<a name="line.585"></a>
<span class="sourceLineNo">586</span> op.exec();<a name="line.586"></a>
<span class="sourceLineNo">587</span> } catch (ReplicationException e) {<a name="line.587"></a>
<span class="sourceLineNo">588</span> if (e.getCause() != null &amp;&amp; e.getCause() instanceof KeeperException.SystemErrorException<a name="line.588"></a>
<span class="sourceLineNo">589</span> &amp;&amp; e.getCause().getCause() != null &amp;&amp; e.getCause()<a name="line.589"></a>
<span class="sourceLineNo">590</span> .getCause() instanceof InterruptedException) {<a name="line.590"></a>
<span class="sourceLineNo">591</span> // ReplicationRuntimeException(a RuntimeException) is thrown out here. The reason is<a name="line.591"></a>
<span class="sourceLineNo">592</span> // that thread is interrupted deep down in the stack, it should pass the following<a name="line.592"></a>
<span class="sourceLineNo">593</span> // processing logic and propagate to the most top layer which can handle this exception<a name="line.593"></a>
<span class="sourceLineNo">594</span> // properly. In this specific case, the top layer is ReplicationSourceShipper#run().<a name="line.594"></a>
<span class="sourceLineNo">595</span> throw new ReplicationRuntimeException(<a name="line.595"></a>
<span class="sourceLineNo">596</span> "Thread is interrupted, the replication source may be terminated",<a name="line.596"></a>
<span class="sourceLineNo">597</span> e.getCause().getCause());<a name="line.597"></a>
<span class="sourceLineNo">598</span> }<a name="line.598"></a>
<span class="sourceLineNo">599</span> server.abort("Failed to operate on replication queue", e);<a name="line.599"></a>
<span class="sourceLineNo">600</span> }<a name="line.600"></a>
<span class="sourceLineNo">601</span> }<a name="line.601"></a>
<span class="sourceLineNo">602</span><a name="line.602"></a>
<span class="sourceLineNo">603</span> private void abortWhenFail(ReplicationQueueOperation op) {<a name="line.603"></a>
<span class="sourceLineNo">604</span> try {<a name="line.604"></a>
<span class="sourceLineNo">605</span> op.exec();<a name="line.605"></a>
<span class="sourceLineNo">606</span> } catch (ReplicationException e) {<a name="line.606"></a>
<span class="sourceLineNo">607</span> server.abort("Failed to operate on replication queue", e);<a name="line.607"></a>
<span class="sourceLineNo">608</span> }<a name="line.608"></a>
<span class="sourceLineNo">609</span> }<a name="line.609"></a>
<span class="sourceLineNo">610</span><a name="line.610"></a>
<span class="sourceLineNo">611</span> private void throwIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.611"></a>
<span class="sourceLineNo">612</span> try {<a name="line.612"></a>
<span class="sourceLineNo">613</span> op.exec();<a name="line.613"></a>
<span class="sourceLineNo">614</span> } catch (ReplicationException e) {<a name="line.614"></a>
<span class="sourceLineNo">615</span> throw new IOException(e);<a name="line.615"></a>
<span class="sourceLineNo">616</span> }<a name="line.616"></a>
<span class="sourceLineNo">617</span> }<a name="line.617"></a>
<span class="sourceLineNo">618</span><a name="line.618"></a>
<span class="sourceLineNo">619</span> private void abortAndThrowIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.619"></a>
<span class="sourceLineNo">620</span> try {<a name="line.620"></a>
<span class="sourceLineNo">621</span> op.exec();<a name="line.621"></a>
<span class="sourceLineNo">622</span> } catch (ReplicationException e) {<a name="line.622"></a>
<span class="sourceLineNo">623</span> server.abort("Failed to operate on replication queue", e);<a name="line.623"></a>
<span class="sourceLineNo">624</span> throw new IOException(e);<a name="line.624"></a>
<span class="sourceLineNo">625</span> }<a name="line.625"></a>
<span class="sourceLineNo">626</span> }<a name="line.626"></a>
<span class="sourceLineNo">627</span><a name="line.627"></a>
<span class="sourceLineNo">628</span> /**<a name="line.628"></a>
<span class="sourceLineNo">629</span> * This method will log the current position to storage. And also clean old logs from the<a name="line.629"></a>
<span class="sourceLineNo">630</span> * replication queue.<a name="line.630"></a>
<span class="sourceLineNo">631</span> * @param source the replication source<a name="line.631"></a>
<span class="sourceLineNo">632</span> * @param entryBatch the wal entry batch we just shipped<a name="line.632"></a>
<span class="sourceLineNo">633</span> */<a name="line.633"></a>
<span class="sourceLineNo">634</span> public void logPositionAndCleanOldLogs(ReplicationSourceInterface source,<a name="line.634"></a>
<span class="sourceLineNo">635</span> WALEntryBatch entryBatch) {<a name="line.635"></a>
<span class="sourceLineNo">636</span> String fileName = entryBatch.getLastWalPath().getName();<a name="line.636"></a>
<span class="sourceLineNo">637</span> interruptOrAbortWhenFail(() -&gt; this.queueStorage.setWALPosition(server.getServerName(),<a name="line.637"></a>
<span class="sourceLineNo">638</span> source.getQueueId(), fileName, entryBatch.getLastWalPosition(), entryBatch.getLastSeqIds()));<a name="line.638"></a>
<span class="sourceLineNo">639</span> cleanOldLogs(fileName, entryBatch.isEndOfFile(), source);<a name="line.639"></a>
<span class="sourceLineNo">640</span> }<a name="line.640"></a>
<span class="sourceLineNo">641</span><a name="line.641"></a>
<span class="sourceLineNo">642</span> /**<a name="line.642"></a>
<span class="sourceLineNo">643</span> * Cleans a log file and all older logs from replication queue. Called when we are sure that a log<a name="line.643"></a>
<span class="sourceLineNo">644</span> * file is closed and has no more entries.<a name="line.644"></a>
<span class="sourceLineNo">645</span> * @param log Path to the log<a name="line.645"></a>
<span class="sourceLineNo">646</span> * @param inclusive whether we should also remove the given log file<a name="line.646"></a>
<span class="sourceLineNo">647</span> * @param source the replication source<a name="line.647"></a>
<span class="sourceLineNo">648</span> */<a name="line.648"></a>
<span class="sourceLineNo">649</span> @VisibleForTesting<a name="line.649"></a>
<span class="sourceLineNo">650</span> void cleanOldLogs(String log, boolean inclusive, ReplicationSourceInterface source) {<a name="line.650"></a>
<span class="sourceLineNo">651</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log);<a name="line.651"></a>
<span class="sourceLineNo">652</span> if (source.isRecovered()) {<a name="line.652"></a>
<span class="sourceLineNo">653</span> NavigableSet&lt;String&gt; wals = walsByIdRecoveredQueues.get(source.getQueueId()).get(logPrefix);<a name="line.653"></a>
<span class="sourceLineNo">654</span> if (wals != null) {<a name="line.654"></a>
<span class="sourceLineNo">655</span> NavigableSet&lt;String&gt; walsToRemove = wals.headSet(log, inclusive);<a name="line.655"></a>
<span class="sourceLineNo">656</span> if (walsToRemove.isEmpty()) {<a name="line.656"></a>
<span class="sourceLineNo">657</span> return;<a name="line.657"></a>
<span class="sourceLineNo">658</span> }<a name="line.658"></a>
<span class="sourceLineNo">659</span> cleanOldLogs(walsToRemove, source);<a name="line.659"></a>
<span class="sourceLineNo">660</span> walsToRemove.clear();<a name="line.660"></a>
<span class="sourceLineNo">661</span> }<a name="line.661"></a>
<span class="sourceLineNo">662</span> } else {<a name="line.662"></a>
<span class="sourceLineNo">663</span> NavigableSet&lt;String&gt; wals;<a name="line.663"></a>
<span class="sourceLineNo">664</span> NavigableSet&lt;String&gt; walsToRemove;<a name="line.664"></a>
<span class="sourceLineNo">665</span> // synchronized on walsById to avoid race with preLogRoll<a name="line.665"></a>
<span class="sourceLineNo">666</span> synchronized (this.walsById) {<a name="line.666"></a>
<span class="sourceLineNo">667</span> wals = walsById.get(source.getQueueId()).get(logPrefix);<a name="line.667"></a>
<span class="sourceLineNo">668</span> if (wals == null) {<a name="line.668"></a>
<span class="sourceLineNo">669</span> return;<a name="line.669"></a>
<span class="sourceLineNo">670</span> }<a name="line.670"></a>
<span class="sourceLineNo">671</span> walsToRemove = wals.headSet(log, inclusive);<a name="line.671"></a>
<span class="sourceLineNo">672</span> if (walsToRemove.isEmpty()) {<a name="line.672"></a>
<span class="sourceLineNo">673</span> return;<a name="line.673"></a>
<span class="sourceLineNo">674</span> }<a name="line.674"></a>
<span class="sourceLineNo">675</span> walsToRemove = new TreeSet&lt;&gt;(walsToRemove);<a name="line.675"></a>
<span class="sourceLineNo">676</span> }<a name="line.676"></a>
<span class="sourceLineNo">677</span> // cleanOldLogs may spend some time, especially for sync replication where we may want to<a name="line.677"></a>
<span class="sourceLineNo">678</span> // remove remote wals as the remote cluster may have already been down, so we do it outside<a name="line.678"></a>
<span class="sourceLineNo">679</span> // the lock to avoid block preLogRoll<a name="line.679"></a>
<span class="sourceLineNo">680</span> cleanOldLogs(walsToRemove, source);<a name="line.680"></a>
<span class="sourceLineNo">681</span> // now let's remove the files in the set<a name="line.681"></a>
<span class="sourceLineNo">682</span> synchronized (this.walsById) {<a name="line.682"></a>
<span class="sourceLineNo">683</span> wals.removeAll(walsToRemove);<a name="line.683"></a>
<span class="sourceLineNo">684</span> }<a name="line.684"></a>
<span class="sourceLineNo">685</span> }<a name="line.685"></a>
<span class="sourceLineNo">686</span> }<a name="line.686"></a>
<span class="sourceLineNo">687</span><a name="line.687"></a>
<span class="sourceLineNo">688</span> private void removeRemoteWALs(String peerId, String remoteWALDir, Collection&lt;String&gt; wals)<a name="line.688"></a>
<span class="sourceLineNo">689</span> throws IOException {<a name="line.689"></a>
<span class="sourceLineNo">690</span> Path remoteWALDirForPeer = ReplicationUtils.getPeerRemoteWALDir(remoteWALDir, peerId);<a name="line.690"></a>
<span class="sourceLineNo">691</span> FileSystem fs = ReplicationUtils.getRemoteWALFileSystem(conf, remoteWALDir);<a name="line.691"></a>
<span class="sourceLineNo">692</span> for (String wal : wals) {<a name="line.692"></a>
<span class="sourceLineNo">693</span> Path walFile = new Path(remoteWALDirForPeer, wal);<a name="line.693"></a>
<span class="sourceLineNo">694</span> try {<a name="line.694"></a>
<span class="sourceLineNo">695</span> if (!fs.delete(walFile, false) &amp;&amp; fs.exists(walFile)) {<a name="line.695"></a>
<span class="sourceLineNo">696</span> throw new IOException("Can not delete " + walFile);<a name="line.696"></a>
<span class="sourceLineNo">697</span> }<a name="line.697"></a>
<span class="sourceLineNo">698</span> } catch (FileNotFoundException e) {<a name="line.698"></a>
<span class="sourceLineNo">699</span> // Just ignore since this means the file has already been deleted.<a name="line.699"></a>
<span class="sourceLineNo">700</span> // The javadoc of the FileSystem.delete methods does not specify the behavior of deleting an<a name="line.700"></a>
<span class="sourceLineNo">701</span> // inexistent file, so here we deal with both, i.e, check the return value of the<a name="line.701"></a>
<span class="sourceLineNo">702</span> // FileSystem.delete, and also catch FNFE.<a name="line.702"></a>
<span class="sourceLineNo">703</span> LOG.debug("The remote wal {} has already been deleted?", walFile, e);<a name="line.703"></a>
<span class="sourceLineNo">704</span> }<a name="line.704"></a>
<span class="sourceLineNo">705</span> }<a name="line.705"></a>
<span class="sourceLineNo">706</span> }<a name="line.706"></a>
<span class="sourceLineNo">707</span><a name="line.707"></a>
<span class="sourceLineNo">708</span> private void cleanOldLogs(NavigableSet&lt;String&gt; wals, ReplicationSourceInterface source) {<a name="line.708"></a>
<span class="sourceLineNo">709</span> LOG.debug("Removing {} logs in the list: {}", wals.size(), wals);<a name="line.709"></a>
<span class="sourceLineNo">710</span> // The intention here is that, we want to delete the remote wal files ASAP as it may effect the<a name="line.710"></a>
<span class="sourceLineNo">711</span> // failover time if you want to transit the remote cluster from S to A. And the infinite retry<a name="line.711"></a>
<span class="sourceLineNo">712</span> // is not a problem, as if we can not contact with the remote HDFS cluster, then usually we can<a name="line.712"></a>
<span class="sourceLineNo">713</span> // not contact with the HBase cluster either, so the replication will be blocked either.<a name="line.713"></a>
<span class="sourceLineNo">714</span> if (source.isSyncReplication()) {<a name="line.714"></a>
<span class="sourceLineNo">715</span> String peerId = source.getPeerId();<a name="line.715"></a>
<span class="sourceLineNo">716</span> String remoteWALDir = source.getPeer().getPeerConfig().getRemoteWALDir();<a name="line.716"></a>
<span class="sourceLineNo">717</span> // Filter out the wals need to be removed from the remote directory. Its name should be the<a name="line.717"></a>
<span class="sourceLineNo">718</span> // special format, and also, the peer id in its name should match the peer id for the<a name="line.718"></a>
<span class="sourceLineNo">719</span> // replication source.<a name="line.719"></a>
<span class="sourceLineNo">720</span> List&lt;String&gt; remoteWals = wals.stream().filter(w -&gt; SyncReplicationWALProvider<a name="line.720"></a>
<span class="sourceLineNo">721</span> .getSyncReplicationPeerIdFromWALName(w).map(peerId::equals).orElse(false))<a name="line.721"></a>
<span class="sourceLineNo">722</span> .collect(Collectors.toList());<a name="line.722"></a>
<span class="sourceLineNo">723</span> LOG.debug("Removing {} logs from remote dir {} in the list: {}", remoteWals.size(),<a name="line.723"></a>
<span class="sourceLineNo">724</span> remoteWALDir, remoteWals);<a name="line.724"></a>
<span class="sourceLineNo">725</span> if (!remoteWals.isEmpty()) {<a name="line.725"></a>
<span class="sourceLineNo">726</span> for (int sleepMultiplier = 0;;) {<a name="line.726"></a>
<span class="sourceLineNo">727</span> try {<a name="line.727"></a>
<span class="sourceLineNo">728</span> removeRemoteWALs(peerId, remoteWALDir, remoteWals);<a name="line.728"></a>
<span class="sourceLineNo">729</span> break;<a name="line.729"></a>
<span class="sourceLineNo">730</span> } catch (IOException e) {<a name="line.730"></a>
<span class="sourceLineNo">731</span> LOG.warn("Failed to delete remote wals from remote dir {} for peer {}", remoteWALDir,<a name="line.731"></a>
<span class="sourceLineNo">732</span> peerId);<a name="line.732"></a>
<span class="sourceLineNo">733</span> }<a name="line.733"></a>
<span class="sourceLineNo">734</span> if (!source.isSourceActive()) {<a name="line.734"></a>
<span class="sourceLineNo">735</span> // skip the following operations<a name="line.735"></a>
<span class="sourceLineNo">736</span> return;<a name="line.736"></a>
<span class="sourceLineNo">737</span> }<a name="line.737"></a>
<span class="sourceLineNo">738</span> if (ReplicationUtils.sleepForRetries("Failed to delete remote wals", sleepForRetries,<a name="line.738"></a>
<span class="sourceLineNo">739</span> sleepMultiplier, maxRetriesMultiplier)) {<a name="line.739"></a>
<span class="sourceLineNo">740</span> sleepMultiplier++;<a name="line.740"></a>
<span class="sourceLineNo">741</span> }<a name="line.741"></a>
<span class="sourceLineNo">742</span> }<a name="line.742"></a>
<span class="sourceLineNo">743</span> }<a name="line.743"></a>
<span class="sourceLineNo">744</span> }<a name="line.744"></a>
<span class="sourceLineNo">745</span> String queueId = source.getQueueId();<a name="line.745"></a>
<span class="sourceLineNo">746</span> for (String wal : wals) {<a name="line.746"></a>
<span class="sourceLineNo">747</span> interruptOrAbortWhenFail(<a name="line.747"></a>
<span class="sourceLineNo">748</span> () -&gt; this.queueStorage.removeWAL(server.getServerName(), queueId, wal));<a name="line.748"></a>
<span class="sourceLineNo">749</span> }<a name="line.749"></a>
<span class="sourceLineNo">750</span> }<a name="line.750"></a>
<span class="sourceLineNo">751</span><a name="line.751"></a>
<span class="sourceLineNo">752</span> // public because of we call it in TestReplicationEmptyWALRecovery<a name="line.752"></a>
<span class="sourceLineNo">753</span> @VisibleForTesting<a name="line.753"></a>
<span class="sourceLineNo">754</span> public void preLogRoll(Path newLog) throws IOException {<a name="line.754"></a>
<span class="sourceLineNo">755</span> String logName = newLog.getName();<a name="line.755"></a>
<span class="sourceLineNo">756</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.756"></a>
<span class="sourceLineNo">757</span> // synchronized on latestPaths to avoid the new open source miss the new log<a name="line.757"></a>
<span class="sourceLineNo">758</span> synchronized (this.latestPaths) {<a name="line.758"></a>
<span class="sourceLineNo">759</span> // Add log to queue storage<a name="line.759"></a>
<span class="sourceLineNo">760</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.760"></a>
<span class="sourceLineNo">761</span> // If record log to queue storage failed, abort RS and throw exception to make log roll<a name="line.761"></a>
<span class="sourceLineNo">762</span> // failed<a name="line.762"></a>
<span class="sourceLineNo">763</span> abortAndThrowIOExceptionWhenFail(<a name="line.763"></a>
<span class="sourceLineNo">764</span> () -&gt; this.queueStorage.addWAL(server.getServerName(), source.getQueueId(), logName));<a name="line.764"></a>
<span class="sourceLineNo">765</span> }<a name="line.765"></a>
<span class="sourceLineNo">766</span><a name="line.766"></a>
<span class="sourceLineNo">767</span> // synchronized on walsById to avoid race with cleanOldLogs<a name="line.767"></a>
<span class="sourceLineNo">768</span> synchronized (this.walsById) {<a name="line.768"></a>
<span class="sourceLineNo">769</span> // Update walsById map<a name="line.769"></a>
<span class="sourceLineNo">770</span> for (Map.Entry&lt;String, Map&lt;String, NavigableSet&lt;String&gt;&gt;&gt; entry : this.walsById<a name="line.770"></a>
<span class="sourceLineNo">771</span> .entrySet()) {<a name="line.771"></a>
<span class="sourceLineNo">772</span> String peerId = entry.getKey();<a name="line.772"></a>
<span class="sourceLineNo">773</span> Map&lt;String, NavigableSet&lt;String&gt;&gt; walsByPrefix = entry.getValue();<a name="line.773"></a>
<span class="sourceLineNo">774</span> boolean existingPrefix = false;<a name="line.774"></a>
<span class="sourceLineNo">775</span> for (Map.Entry&lt;String, NavigableSet&lt;String&gt;&gt; walsEntry : walsByPrefix.entrySet()) {<a name="line.775"></a>
<span class="sourceLineNo">776</span> SortedSet&lt;String&gt; wals = walsEntry.getValue();<a name="line.776"></a>
<span class="sourceLineNo">777</span> if (this.sources.isEmpty()) {<a name="line.777"></a>
<span class="sourceLineNo">778</span> // If there's no slaves, don't need to keep the old wals since<a name="line.778"></a>
<span class="sourceLineNo">779</span> // we only consider the last one when a new slave comes in<a name="line.779"></a>
<span class="sourceLineNo">780</span> wals.clear();<a name="line.780"></a>
<span class="sourceLineNo">781</span> }<a name="line.781"></a>
<span class="sourceLineNo">782</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.782"></a>
<span class="sourceLineNo">783</span> wals.add(logName);<a name="line.783"></a>
<span class="sourceLineNo">784</span> existingPrefix = true;<a name="line.784"></a>
<span class="sourceLineNo">785</span> }<a name="line.785"></a>
<span class="sourceLineNo">786</span> }<a name="line.786"></a>
<span class="sourceLineNo">787</span> if (!existingPrefix) {<a name="line.787"></a>
<span class="sourceLineNo">788</span> // The new log belongs to a new group, add it into this peer<a name="line.788"></a>
<span class="sourceLineNo">789</span> LOG.debug("Start tracking logs for wal group {} for peer {}", logPrefix, peerId);<a name="line.789"></a>
<span class="sourceLineNo">790</span> NavigableSet&lt;String&gt; wals = new TreeSet&lt;&gt;();<a name="line.790"></a>
<span class="sourceLineNo">791</span> wals.add(logName);<a name="line.791"></a>
<span class="sourceLineNo">792</span> walsByPrefix.put(logPrefix, wals);<a name="line.792"></a>
<span class="sourceLineNo">793</span> }<a name="line.793"></a>
<span class="sourceLineNo">794</span> }<a name="line.794"></a>
<span class="sourceLineNo">795</span> }<a name="line.795"></a>
<span class="sourceLineNo">796</span><a name="line.796"></a>
<span class="sourceLineNo">797</span> // Add to latestPaths<a name="line.797"></a>
<span class="sourceLineNo">798</span> latestPaths.put(logPrefix, newLog);<a name="line.798"></a>
<span class="sourceLineNo">799</span> }<a name="line.799"></a>
<span class="sourceLineNo">800</span> }<a name="line.800"></a>
<span class="sourceLineNo">801</span><a name="line.801"></a>
<span class="sourceLineNo">802</span> // public because of we call it in TestReplicationEmptyWALRecovery<a name="line.802"></a>
<span class="sourceLineNo">803</span> @VisibleForTesting<a name="line.803"></a>
<span class="sourceLineNo">804</span> public void postLogRoll(Path newLog) throws IOException {<a name="line.804"></a>
<span class="sourceLineNo">805</span> // This only updates the sources we own, not the recovered ones<a name="line.805"></a>
<span class="sourceLineNo">806</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.806"></a>
<span class="sourceLineNo">807</span> source.enqueueLog(newLog);<a name="line.807"></a>
<span class="sourceLineNo">808</span> LOG.trace("Enqueued {} to source {} while performing postLogRoll operation.",<a name="line.808"></a>
<span class="sourceLineNo">809</span> newLog, source.getQueueId());<a name="line.809"></a>
<span class="sourceLineNo">810</span> }<a name="line.810"></a>
<span class="sourceLineNo">811</span> }<a name="line.811"></a>
<span class="sourceLineNo">812</span><a name="line.812"></a>
<span class="sourceLineNo">813</span> @Override<a name="line.813"></a>
<span class="sourceLineNo">814</span> public void regionServerRemoved(String regionserver) {<a name="line.814"></a>
<span class="sourceLineNo">815</span> transferQueues(ServerName.valueOf(regionserver));<a name="line.815"></a>
<span class="sourceLineNo">816</span> }<a name="line.816"></a>
<span class="sourceLineNo">817</span><a name="line.817"></a>
<span class="sourceLineNo">818</span> /**<a name="line.818"></a>
<span class="sourceLineNo">819</span> * Transfer all the queues of the specified to this region server. First it tries to grab a lock<a name="line.819"></a>
<span class="sourceLineNo">820</span> * and if it works it will move the old queues and finally will delete the old queues.<a name="line.820"></a>
<span class="sourceLineNo">821</span> * &lt;p&gt;<a name="line.821"></a>
<span class="sourceLineNo">822</span> * It creates one old source for any type of source of the old rs.<a name="line.822"></a>
<span class="sourceLineNo">823</span> */<a name="line.823"></a>
<span class="sourceLineNo">824</span> private void transferQueues(ServerName deadRS) {<a name="line.824"></a>
<span class="sourceLineNo">825</span> if (server.getServerName().equals(deadRS)) {<a name="line.825"></a>
<span class="sourceLineNo">826</span> // it's just us, give up<a name="line.826"></a>
<span class="sourceLineNo">827</span> return;<a name="line.827"></a>
<span class="sourceLineNo">828</span> }<a name="line.828"></a>
<span class="sourceLineNo">829</span> NodeFailoverWorker transfer = new NodeFailoverWorker(deadRS);<a name="line.829"></a>
<span class="sourceLineNo">830</span> try {<a name="line.830"></a>
<span class="sourceLineNo">831</span> this.executor.execute(transfer);<a name="line.831"></a>
<span class="sourceLineNo">832</span> } catch (RejectedExecutionException ex) {<a name="line.832"></a>
<span class="sourceLineNo">833</span> CompatibilitySingletonFactory.getInstance(MetricsReplicationSourceFactory.class)<a name="line.833"></a>
<span class="sourceLineNo">834</span> .getGlobalSource().incrFailedRecoveryQueue();<a name="line.834"></a>
<span class="sourceLineNo">835</span> LOG.info("Cancelling the transfer of " + deadRS + " because of " + ex.getMessage());<a name="line.835"></a>
<span class="sourceLineNo">836</span> }<a name="line.836"></a>
<span class="sourceLineNo">837</span> }<a name="line.837"></a>
<span class="sourceLineNo">838</span><a name="line.838"></a>
<span class="sourceLineNo">839</span> /**<a name="line.839"></a>
<span class="sourceLineNo">840</span> * Class responsible to setup new ReplicationSources to take care of the queues from dead region<a name="line.840"></a>
<span class="sourceLineNo">841</span> * servers.<a name="line.841"></a>
<span class="sourceLineNo">842</span> */<a name="line.842"></a>
<span class="sourceLineNo">843</span> class NodeFailoverWorker extends Thread {<a name="line.843"></a>
<span class="sourceLineNo">844</span><a name="line.844"></a>
<span class="sourceLineNo">845</span> private final ServerName deadRS;<a name="line.845"></a>
<span class="sourceLineNo">846</span> // After claim the queues from dead region server, the NodeFailoverWorker will skip to start<a name="line.846"></a>
<span class="sourceLineNo">847</span> // the RecoveredReplicationSource if the peer has been removed. but there's possible that<a name="line.847"></a>
<span class="sourceLineNo">848</span> // remove a peer with peerId = 2 and add a peer with peerId = 2 again during the<a name="line.848"></a>
<span class="sourceLineNo">849</span> // NodeFailoverWorker. So we need a deep copied &lt;peerId, peer&gt; map to decide whether we<a name="line.849"></a>
<span class="sourceLineNo">850</span> // should start the RecoveredReplicationSource. If the latest peer is not the old peer when<a name="line.850"></a>
<span class="sourceLineNo">851</span> // NodeFailoverWorker begin, we should skip to start the RecoveredReplicationSource, Otherwise<a name="line.851"></a>
<span class="sourceLineNo">852</span> // the rs will abort (See HBASE-20475).<a name="line.852"></a>
<span class="sourceLineNo">853</span> private final Map&lt;String, ReplicationPeerImpl&gt; peersSnapshot;<a name="line.853"></a>
<span class="sourceLineNo">854</span><a name="line.854"></a>
<span class="sourceLineNo">855</span> @VisibleForTesting<a name="line.855"></a>
<span class="sourceLineNo">856</span> public NodeFailoverWorker(ServerName deadRS) {<a name="line.856"></a>
<span class="sourceLineNo">857</span> super("Failover-for-" + deadRS);<a name="line.857"></a>
<span class="sourceLineNo">858</span> this.deadRS = deadRS;<a name="line.858"></a>
<span class="sourceLineNo">859</span> peersSnapshot = new HashMap&lt;&gt;(replicationPeers.getPeerCache());<a name="line.859"></a>
<span class="sourceLineNo">860</span> }<a name="line.860"></a>
<span class="sourceLineNo">861</span><a name="line.861"></a>
<span class="sourceLineNo">862</span> private boolean isOldPeer(String peerId, ReplicationPeerImpl newPeerRef) {<a name="line.862"></a>
<span class="sourceLineNo">863</span> ReplicationPeerImpl oldPeerRef = peersSnapshot.get(peerId);<a name="line.863"></a>
<span class="sourceLineNo">864</span> return oldPeerRef != null &amp;&amp; oldPeerRef == newPeerRef;<a name="line.864"></a>
<span class="sourceLineNo">865</span> }<a name="line.865"></a>
<span class="sourceLineNo">866</span><a name="line.866"></a>
<span class="sourceLineNo">867</span> @Override<a name="line.867"></a>
<span class="sourceLineNo">868</span> public void run() {<a name="line.868"></a>
<span class="sourceLineNo">869</span> // Wait a bit before transferring the queues, we may be shutting down.<a name="line.869"></a>
<span class="sourceLineNo">870</span> // This sleep may not be enough in some cases.<a name="line.870"></a>
<span class="sourceLineNo">871</span> try {<a name="line.871"></a>
<span class="sourceLineNo">872</span> Thread.sleep(sleepBeforeFailover +<a name="line.872"></a>
<span class="sourceLineNo">873</span> (long) (ThreadLocalRandom.current().nextFloat() * sleepBeforeFailover));<a name="line.873"></a>
<span class="sourceLineNo">874</span> } catch (InterruptedException e) {<a name="line.874"></a>
<span class="sourceLineNo">875</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.875"></a>
<span class="sourceLineNo">876</span> Thread.currentThread().interrupt();<a name="line.876"></a>
<span class="sourceLineNo">877</span> }<a name="line.877"></a>
<span class="sourceLineNo">878</span> // We try to lock that rs' queue directory<a name="line.878"></a>
<span class="sourceLineNo">879</span> if (server.isStopped()) {<a name="line.879"></a>
<span class="sourceLineNo">880</span> LOG.info("Not transferring queue since we are shutting down");<a name="line.880"></a>
<span class="sourceLineNo">881</span> return;<a name="line.881"></a>
<span class="sourceLineNo">882</span> }<a name="line.882"></a>
<span class="sourceLineNo">883</span> Map&lt;String, Set&lt;String&gt;&gt; newQueues = new HashMap&lt;&gt;();<a name="line.883"></a>
<span class="sourceLineNo">884</span> try {<a name="line.884"></a>
<span class="sourceLineNo">885</span> List&lt;String&gt; queues = queueStorage.getAllQueues(deadRS);<a name="line.885"></a>
<span class="sourceLineNo">886</span> while (!queues.isEmpty()) {<a name="line.886"></a>
<span class="sourceLineNo">887</span> Pair&lt;String, SortedSet&lt;String&gt;&gt; peer = queueStorage.claimQueue(deadRS,<a name="line.887"></a>
<span class="sourceLineNo">888</span> queues.get(ThreadLocalRandom.current().nextInt(queues.size())), server.getServerName());<a name="line.888"></a>
<span class="sourceLineNo">889</span> long sleep = sleepBeforeFailover / 2;<a name="line.889"></a>
<span class="sourceLineNo">890</span> if (!peer.getSecond().isEmpty()) {<a name="line.890"></a>
<span class="sourceLineNo">891</span> newQueues.put(peer.getFirst(), peer.getSecond());<a name="line.891"></a>
<span class="sourceLineNo">892</span> sleep = sleepBeforeFailover;<a name="line.892"></a>
<span class="sourceLineNo">893</span> }<a name="line.893"></a>
<span class="sourceLineNo">894</span> try {<a name="line.894"></a>
<span class="sourceLineNo">895</span> Thread.sleep(sleep);<a name="line.895"></a>
<span class="sourceLineNo">896</span> } catch (InterruptedException e) {<a name="line.896"></a>
<span class="sourceLineNo">897</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.897"></a>
<span class="sourceLineNo">898</span> Thread.currentThread().interrupt();<a name="line.898"></a>
<span class="sourceLineNo">899</span> }<a name="line.899"></a>
<span class="sourceLineNo">900</span> queues = queueStorage.getAllQueues(deadRS);<a name="line.900"></a>
<span class="sourceLineNo">901</span> }<a name="line.901"></a>
<span class="sourceLineNo">902</span> if (queues.isEmpty()) {<a name="line.902"></a>
<span class="sourceLineNo">903</span> queueStorage.removeReplicatorIfQueueIsEmpty(deadRS);<a name="line.903"></a>
<span class="sourceLineNo">904</span> }<a name="line.904"></a>
<span class="sourceLineNo">905</span> } catch (ReplicationException e) {<a name="line.905"></a>
<span class="sourceLineNo">906</span> LOG.error(String.format("ReplicationException: cannot claim dead region (%s)'s " +<a name="line.906"></a>
<span class="sourceLineNo">907</span> "replication queue. Znode : (%s)" +<a name="line.907"></a>
<span class="sourceLineNo">908</span> " Possible solution: check if znode size exceeds jute.maxBuffer value. " +<a name="line.908"></a>
<span class="sourceLineNo">909</span> " If so, increase it for both client and server side." + e), deadRS,<a name="line.909"></a>
<span class="sourceLineNo">910</span> queueStorage.getRsNode(deadRS));<a name="line.910"></a>
<span class="sourceLineNo">911</span> server.abort("Failed to claim queue from dead regionserver.", e);<a name="line.911"></a>
<span class="sourceLineNo">912</span> return;<a name="line.912"></a>
<span class="sourceLineNo">913</span> }<a name="line.913"></a>
<span class="sourceLineNo">914</span> // Copying over the failed queue is completed.<a name="line.914"></a>
<span class="sourceLineNo">915</span> if (newQueues.isEmpty()) {<a name="line.915"></a>
<span class="sourceLineNo">916</span> // We either didn't get the lock or the failed region server didn't have any outstanding<a name="line.916"></a>
<span class="sourceLineNo">917</span> // WALs to replicate, so we are done.<a name="line.917"></a>
<span class="sourceLineNo">918</span> return;<a name="line.918"></a>
<span class="sourceLineNo">919</span> }<a name="line.919"></a>
<span class="sourceLineNo">920</span><a name="line.920"></a>
<span class="sourceLineNo">921</span> for (Map.Entry&lt;String, Set&lt;String&gt;&gt; entry : newQueues.entrySet()) {<a name="line.921"></a>
<span class="sourceLineNo">922</span> String queueId = entry.getKey();<a name="line.922"></a>
<span class="sourceLineNo">923</span> Set&lt;String&gt; walsSet = entry.getValue();<a name="line.923"></a>
<span class="sourceLineNo">924</span> try {<a name="line.924"></a>
<span class="sourceLineNo">925</span> // there is not an actual peer defined corresponding to peerId for the failover.<a name="line.925"></a>
<span class="sourceLineNo">926</span> ReplicationQueueInfo replicationQueueInfo = new ReplicationQueueInfo(queueId);<a name="line.926"></a>
<span class="sourceLineNo">927</span> String actualPeerId = replicationQueueInfo.getPeerId();<a name="line.927"></a>
<span class="sourceLineNo">928</span><a name="line.928"></a>
<span class="sourceLineNo">929</span> ReplicationPeerImpl peer = replicationPeers.getPeer(actualPeerId);<a name="line.929"></a>
<span class="sourceLineNo">930</span> if (peer == null || !isOldPeer(actualPeerId, peer)) {<a name="line.930"></a>
<span class="sourceLineNo">931</span> LOG.warn("Skipping failover for peer {} of node {}, peer is null", actualPeerId,<a name="line.931"></a>
<span class="sourceLineNo">932</span> deadRS);<a name="line.932"></a>
<span class="sourceLineNo">933</span> abortWhenFail(() -&gt; queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.933"></a>
<span class="sourceLineNo">934</span> continue;<a name="line.934"></a>
<span class="sourceLineNo">935</span> }<a name="line.935"></a>
<span class="sourceLineNo">936</span> if (server instanceof ReplicationSyncUp.DummyServer<a name="line.936"></a>
<span class="sourceLineNo">937</span> &amp;&amp; peer.getPeerState().equals(PeerState.DISABLED)) {<a name="line.937"></a>
<span class="sourceLineNo">938</span> LOG.warn("Peer {} is disabled. ReplicationSyncUp tool will skip "<a name="line.938"></a>
<span class="sourceLineNo">939</span> + "replicating data to this peer.",<a name="line.939"></a>
<span class="sourceLineNo">940</span> actualPeerId);<a name="line.940"></a>
<span class="sourceLineNo">941</span> continue;<a name="line.941"></a>
<span class="sourceLineNo">942</span> }<a name="line.942"></a>
<span class="sourceLineNo">943</span><a name="line.943"></a>
<span class="sourceLineNo">944</span> ReplicationSourceInterface src = createSource(queueId, peer);<a name="line.944"></a>
<span class="sourceLineNo">945</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.945"></a>
<span class="sourceLineNo">946</span> synchronized (oldsources) {<a name="line.946"></a>
<span class="sourceLineNo">947</span> peer = replicationPeers.getPeer(src.getPeerId());<a name="line.947"></a>
<span class="sourceLineNo">948</span> if (peer == null || !isOldPeer(src.getPeerId(), peer)) {<a name="line.948"></a>
<span class="sourceLineNo">949</span> src.terminate("Recovered queue doesn't belong to any current peer");<a name="line.949"></a>
<span class="sourceLineNo">950</span> deleteQueue(queueId);<a name="line.950"></a>
<span class="sourceLineNo">951</span> continue;<a name="line.951"></a>
<span class="sourceLineNo">952</span> }<a name="line.952"></a>
<span class="sourceLineNo">953</span> // Do not setup recovered queue if a sync replication peer is in STANDBY state, or is<a name="line.953"></a>
<span class="sourceLineNo">954</span> // transiting to STANDBY state. The only exception is we are in STANDBY state and<a name="line.954"></a>
<span class="sourceLineNo">955</span> // transiting to DA, under this state we will replay the remote WAL and they need to be<a name="line.955"></a>
<span class="sourceLineNo">956</span> // replicated back.<a name="line.956"></a>
<span class="sourceLineNo">957</span> if (peer.getPeerConfig().isSyncReplication()) {<a name="line.957"></a>
<span class="sourceLineNo">958</span> Pair&lt;SyncReplicationState, SyncReplicationState&gt; stateAndNewState =<a name="line.958"></a>
<span class="sourceLineNo">959</span> peer.getSyncReplicationStateAndNewState();<a name="line.959"></a>
<span class="sourceLineNo">960</span> if ((stateAndNewState.getFirst().equals(SyncReplicationState.STANDBY) &amp;&amp;<a name="line.960"></a>
<span class="sourceLineNo">961</span> stateAndNewState.getSecond().equals(SyncReplicationState.NONE)) ||<a name="line.961"></a>
<span class="sourceLineNo">962</span> stateAndNewState.getSecond().equals(SyncReplicationState.STANDBY)) {<a name="line.962"></a>
<span class="sourceLineNo">963</span> src.terminate("Sync replication peer is in STANDBY state");<a name="line.963"></a>
<span class="sourceLineNo">964</span> deleteQueue(queueId);<a name="line.964"></a>
<span class="sourceLineNo">965</span> continue;<a name="line.965"></a>
<span class="sourceLineNo">966</span> }<a name="line.966"></a>
<span class="sourceLineNo">967</span> }<a name="line.967"></a>
<span class="sourceLineNo">968</span> // track sources in walsByIdRecoveredQueues<a name="line.968"></a>
<span class="sourceLineNo">969</span> Map&lt;String, NavigableSet&lt;String&gt;&gt; walsByGroup = new HashMap&lt;&gt;();<a name="line.969"></a>
<span class="sourceLineNo">970</span> walsByIdRecoveredQueues.put(queueId, walsByGroup);<a name="line.970"></a>
<span class="sourceLineNo">971</span> for (String wal : walsSet) {<a name="line.971"></a>
<span class="sourceLineNo">972</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(wal);<a name="line.972"></a>
<span class="sourceLineNo">973</span> NavigableSet&lt;String&gt; wals = walsByGroup.get(walPrefix);<a name="line.973"></a>
<span class="sourceLineNo">974</span> if (wals == null) {<a name="line.974"></a>
<span class="sourceLineNo">975</span> wals = new TreeSet&lt;&gt;();<a name="line.975"></a>
<span class="sourceLineNo">976</span> walsByGroup.put(walPrefix, wals);<a name="line.976"></a>
<span class="sourceLineNo">977</span> }<a name="line.977"></a>
<span class="sourceLineNo">978</span> wals.add(wal);<a name="line.978"></a>
<span class="sourceLineNo">979</span> }<a name="line.979"></a>
<span class="sourceLineNo">980</span> oldsources.add(src);<a name="line.980"></a>
<span class="sourceLineNo">981</span> LOG.trace("Added source for recovered queue: " + src.getQueueId());<a name="line.981"></a>
<span class="sourceLineNo">982</span> for (String wal : walsSet) {<a name="line.982"></a>
<span class="sourceLineNo">983</span> LOG.trace("Enqueueing log from recovered queue for source: " + src.getQueueId());<a name="line.983"></a>
<span class="sourceLineNo">984</span> src.enqueueLog(new Path(oldLogDir, wal));<a name="line.984"></a>
<span class="sourceLineNo">985</span> }<a name="line.985"></a>
<span class="sourceLineNo">986</span> src.startup();<a name="line.986"></a>
<span class="sourceLineNo">987</span> }<a name="line.987"></a>
<span class="sourceLineNo">988</span> } catch (IOException e) {<a name="line.988"></a>
<span class="sourceLineNo">989</span> // TODO manage it<a name="line.989"></a>
<span class="sourceLineNo">990</span> LOG.error("Failed creating a source", e);<a name="line.990"></a>
<span class="sourceLineNo">991</span> }<a name="line.991"></a>
<span class="sourceLineNo">992</span> }<a name="line.992"></a>
<span class="sourceLineNo">993</span> }<a name="line.993"></a>
<span class="sourceLineNo">994</span> }<a name="line.994"></a>
<span class="sourceLineNo">995</span><a name="line.995"></a>
<span class="sourceLineNo">996</span> /**<a name="line.996"></a>
<span class="sourceLineNo">997</span> * Terminate the replication on this region server<a name="line.997"></a>
<span class="sourceLineNo">998</span> */<a name="line.998"></a>
<span class="sourceLineNo">999</span> public void join() {<a name="line.999"></a>
<span class="sourceLineNo">1000</span> this.executor.shutdown();<a name="line.1000"></a>
<span class="sourceLineNo">1001</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.1001"></a>
<span class="sourceLineNo">1002</span> source.terminate("Region server is closing");<a name="line.1002"></a>
<span class="sourceLineNo">1003</span> }<a name="line.1003"></a>
<span class="sourceLineNo">1004</span> }<a name="line.1004"></a>
<span class="sourceLineNo">1005</span><a name="line.1005"></a>
<span class="sourceLineNo">1006</span> /**<a name="line.1006"></a>
<span class="sourceLineNo">1007</span> * Get a copy of the wals of the normal sources on this rs<a name="line.1007"></a>
<span class="sourceLineNo">1008</span> * @return a sorted set of wal names<a name="line.1008"></a>
<span class="sourceLineNo">1009</span> */<a name="line.1009"></a>
<span class="sourceLineNo">1010</span> @VisibleForTesting<a name="line.1010"></a>
<span class="sourceLineNo">1011</span> public Map&lt;String, Map&lt;String, NavigableSet&lt;String&gt;&gt;&gt; getWALs() {<a name="line.1011"></a>
<span class="sourceLineNo">1012</span> return Collections.unmodifiableMap(walsById);<a name="line.1012"></a>
<span class="sourceLineNo">1013</span> }<a name="line.1013"></a>
<span class="sourceLineNo">1014</span><a name="line.1014"></a>
<span class="sourceLineNo">1015</span> /**<a name="line.1015"></a>
<span class="sourceLineNo">1016</span> * Get a copy of the wals of the recovered sources on this rs<a name="line.1016"></a>
<span class="sourceLineNo">1017</span> * @return a sorted set of wal names<a name="line.1017"></a>
<span class="sourceLineNo">1018</span> */<a name="line.1018"></a>
<span class="sourceLineNo">1019</span> @VisibleForTesting<a name="line.1019"></a>
<span class="sourceLineNo">1020</span> Map&lt;String, Map&lt;String, NavigableSet&lt;String&gt;&gt;&gt; getWalsByIdRecoveredQueues() {<a name="line.1020"></a>
<span class="sourceLineNo">1021</span> return Collections.unmodifiableMap(walsByIdRecoveredQueues);<a name="line.1021"></a>
<span class="sourceLineNo">1022</span> }<a name="line.1022"></a>
<span class="sourceLineNo">1023</span><a name="line.1023"></a>
<span class="sourceLineNo">1024</span> /**<a name="line.1024"></a>
<span class="sourceLineNo">1025</span> * Get a list of all the normal sources of this rs<a name="line.1025"></a>
<span class="sourceLineNo">1026</span> * @return list of all normal sources<a name="line.1026"></a>
<span class="sourceLineNo">1027</span> */<a name="line.1027"></a>
<span class="sourceLineNo">1028</span> public List&lt;ReplicationSourceInterface&gt; getSources() {<a name="line.1028"></a>
<span class="sourceLineNo">1029</span> return new ArrayList&lt;&gt;(this.sources.values());<a name="line.1029"></a>
<span class="sourceLineNo">1030</span> }<a name="line.1030"></a>
<span class="sourceLineNo">1031</span><a name="line.1031"></a>
<span class="sourceLineNo">1032</span> /**<a name="line.1032"></a>
<span class="sourceLineNo">1033</span> * Get a list of all the recovered sources of this rs<a name="line.1033"></a>
<span class="sourceLineNo">1034</span> * @return list of all recovered sources<a name="line.1034"></a>
<span class="sourceLineNo">1035</span> */<a name="line.1035"></a>
<span class="sourceLineNo">1036</span> public List&lt;ReplicationSourceInterface&gt; getOldSources() {<a name="line.1036"></a>
<span class="sourceLineNo">1037</span> return this.oldsources;<a name="line.1037"></a>
<span class="sourceLineNo">1038</span> }<a name="line.1038"></a>
<span class="sourceLineNo">1039</span><a name="line.1039"></a>
<span class="sourceLineNo">1040</span> /**<a name="line.1040"></a>
<span class="sourceLineNo">1041</span> * Get the normal source for a given peer<a name="line.1041"></a>
<span class="sourceLineNo">1042</span> * @return the normal source for the give peer if it exists, otherwise null.<a name="line.1042"></a>
<span class="sourceLineNo">1043</span> */<a name="line.1043"></a>
<span class="sourceLineNo">1044</span> @VisibleForTesting<a name="line.1044"></a>
<span class="sourceLineNo">1045</span> public ReplicationSourceInterface getSource(String peerId) {<a name="line.1045"></a>
<span class="sourceLineNo">1046</span> return this.sources.get(peerId);<a name="line.1046"></a>
<span class="sourceLineNo">1047</span> }<a name="line.1047"></a>
<span class="sourceLineNo">1048</span><a name="line.1048"></a>
<span class="sourceLineNo">1049</span> @VisibleForTesting<a name="line.1049"></a>
<span class="sourceLineNo">1050</span> List&lt;String&gt; getAllQueues() throws IOException {<a name="line.1050"></a>
<span class="sourceLineNo">1051</span> List&lt;String&gt; allQueues = Collections.emptyList();<a name="line.1051"></a>
<span class="sourceLineNo">1052</span> try {<a name="line.1052"></a>
<span class="sourceLineNo">1053</span> allQueues = queueStorage.getAllQueues(server.getServerName());<a name="line.1053"></a>
<span class="sourceLineNo">1054</span> } catch (ReplicationException e) {<a name="line.1054"></a>
<span class="sourceLineNo">1055</span> throw new IOException(e);<a name="line.1055"></a>
<span class="sourceLineNo">1056</span> }<a name="line.1056"></a>
<span class="sourceLineNo">1057</span> return allQueues;<a name="line.1057"></a>
<span class="sourceLineNo">1058</span> }<a name="line.1058"></a>
<span class="sourceLineNo">1059</span><a name="line.1059"></a>
<span class="sourceLineNo">1060</span> @VisibleForTesting<a name="line.1060"></a>
<span class="sourceLineNo">1061</span> int getSizeOfLatestPath() {<a name="line.1061"></a>
<span class="sourceLineNo">1062</span> synchronized (latestPaths) {<a name="line.1062"></a>
<span class="sourceLineNo">1063</span> return latestPaths.size();<a name="line.1063"></a>
<span class="sourceLineNo">1064</span> }<a name="line.1064"></a>
<span class="sourceLineNo">1065</span> }<a name="line.1065"></a>
<span class="sourceLineNo">1066</span><a name="line.1066"></a>
<span class="sourceLineNo">1067</span> @VisibleForTesting<a name="line.1067"></a>
<span class="sourceLineNo">1068</span> Set&lt;Path&gt; getLastestPath() {<a name="line.1068"></a>
<span class="sourceLineNo">1069</span> synchronized (latestPaths) {<a name="line.1069"></a>
<span class="sourceLineNo">1070</span> return Sets.newHashSet(latestPaths.values());<a name="line.1070"></a>
<span class="sourceLineNo">1071</span> }<a name="line.1071"></a>
<span class="sourceLineNo">1072</span> }<a name="line.1072"></a>
<span class="sourceLineNo">1073</span><a name="line.1073"></a>
<span class="sourceLineNo">1074</span> @VisibleForTesting<a name="line.1074"></a>
<span class="sourceLineNo">1075</span> public AtomicLong getTotalBufferUsed() {<a name="line.1075"></a>
<span class="sourceLineNo">1076</span> return totalBufferUsed;<a name="line.1076"></a>
<span class="sourceLineNo">1077</span> }<a name="line.1077"></a>
<span class="sourceLineNo">1078</span><a name="line.1078"></a>
<span class="sourceLineNo">1079</span> /**<a name="line.1079"></a>
<span class="sourceLineNo">1080</span> * Returns the maximum size in bytes of edits held in memory which are pending replication<a name="line.1080"></a>
<span class="sourceLineNo">1081</span> * across all sources inside this RegionServer.<a name="line.1081"></a>
<span class="sourceLineNo">1082</span> */<a name="line.1082"></a>
<span class="sourceLineNo">1083</span> public long getTotalBufferLimit() {<a name="line.1083"></a>
<span class="sourceLineNo">1084</span> return totalBufferLimit;<a name="line.1084"></a>
<span class="sourceLineNo">1085</span> }<a name="line.1085"></a>
<span class="sourceLineNo">1086</span><a name="line.1086"></a>
<span class="sourceLineNo">1087</span> /**<a name="line.1087"></a>
<span class="sourceLineNo">1088</span> * Get the directory where wals are archived<a name="line.1088"></a>
<span class="sourceLineNo">1089</span> * @return the directory where wals are archived<a name="line.1089"></a>
<span class="sourceLineNo">1090</span> */<a name="line.1090"></a>
<span class="sourceLineNo">1091</span> public Path getOldLogDir() {<a name="line.1091"></a>
<span class="sourceLineNo">1092</span> return this.oldLogDir;<a name="line.1092"></a>
<span class="sourceLineNo">1093</span> }<a name="line.1093"></a>
<span class="sourceLineNo">1094</span><a name="line.1094"></a>
<span class="sourceLineNo">1095</span> /**<a name="line.1095"></a>
<span class="sourceLineNo">1096</span> * Get the directory where wals are stored by their RSs<a name="line.1096"></a>
<span class="sourceLineNo">1097</span> * @return the directory where wals are stored by their RSs<a name="line.1097"></a>
<span class="sourceLineNo">1098</span> */<a name="line.1098"></a>
<span class="sourceLineNo">1099</span> public Path getLogDir() {<a name="line.1099"></a>
<span class="sourceLineNo">1100</span> return this.logDir;<a name="line.1100"></a>
<span class="sourceLineNo">1101</span> }<a name="line.1101"></a>
<span class="sourceLineNo">1102</span><a name="line.1102"></a>
<span class="sourceLineNo">1103</span> /**<a name="line.1103"></a>
<span class="sourceLineNo">1104</span> * Get the handle on the local file system<a name="line.1104"></a>
<span class="sourceLineNo">1105</span> * @return Handle on the local file system<a name="line.1105"></a>
<span class="sourceLineNo">1106</span> */<a name="line.1106"></a>
<span class="sourceLineNo">1107</span> public FileSystem getFs() {<a name="line.1107"></a>
<span class="sourceLineNo">1108</span> return this.fs;<a name="line.1108"></a>
<span class="sourceLineNo">1109</span> }<a name="line.1109"></a>
<span class="sourceLineNo">1110</span><a name="line.1110"></a>
<span class="sourceLineNo">1111</span> /**<a name="line.1111"></a>
<span class="sourceLineNo">1112</span> * Get the ReplicationPeers used by this ReplicationSourceManager<a name="line.1112"></a>
<span class="sourceLineNo">1113</span> * @return the ReplicationPeers used by this ReplicationSourceManager<a name="line.1113"></a>
<span class="sourceLineNo">1114</span> */<a name="line.1114"></a>
<span class="sourceLineNo">1115</span> public ReplicationPeers getReplicationPeers() {<a name="line.1115"></a>
<span class="sourceLineNo">1116</span> return this.replicationPeers;<a name="line.1116"></a>
<span class="sourceLineNo">1117</span> }<a name="line.1117"></a>
<span class="sourceLineNo">1118</span><a name="line.1118"></a>
<span class="sourceLineNo">1119</span> /**<a name="line.1119"></a>
<span class="sourceLineNo">1120</span> * Get a string representation of all the sources' metrics<a name="line.1120"></a>
<span class="sourceLineNo">1121</span> */<a name="line.1121"></a>
<span class="sourceLineNo">1122</span> public String getStats() {<a name="line.1122"></a>
<span class="sourceLineNo">1123</span> StringBuilder stats = new StringBuilder();<a name="line.1123"></a>
<span class="sourceLineNo">1124</span> // Print stats that apply across all Replication Sources<a name="line.1124"></a>
<span class="sourceLineNo">1125</span> stats.append("Global stats: ");<a name="line.1125"></a>
<span class="sourceLineNo">1126</span> stats.append("WAL Edits Buffer Used=").append(getTotalBufferUsed().get()).append("B, Limit=")<a name="line.1126"></a>
<span class="sourceLineNo">1127</span> .append(getTotalBufferLimit()).append("B\n");<a name="line.1127"></a>
<span class="sourceLineNo">1128</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.1128"></a>
<span class="sourceLineNo">1129</span> stats.append("Normal source for cluster " + source.getPeerId() + ": ");<a name="line.1129"></a>
<span class="sourceLineNo">1130</span> stats.append(source.getStats() + "\n");<a name="line.1130"></a>
<span class="sourceLineNo">1131</span> }<a name="line.1131"></a>
<span class="sourceLineNo">1132</span> for (ReplicationSourceInterface oldSource : oldsources) {<a name="line.1132"></a>
<span class="sourceLineNo">1133</span> stats.append("Recovered source for cluster/machine(s) " + oldSource.getPeerId() + ": ");<a name="line.1133"></a>
<span class="sourceLineNo">1134</span> stats.append(oldSource.getStats() + "\n");<a name="line.1134"></a>
<span class="sourceLineNo">1135</span> }<a name="line.1135"></a>
<span class="sourceLineNo">1136</span> return stats.toString();<a name="line.1136"></a>
<span class="sourceLineNo">1137</span> }<a name="line.1137"></a>
<span class="sourceLineNo">1138</span><a name="line.1138"></a>
<span class="sourceLineNo">1139</span> public void addHFileRefs(TableName tableName, byte[] family, List&lt;Pair&lt;Path, Path&gt;&gt; pairs)<a name="line.1139"></a>
<span class="sourceLineNo">1140</span> throws IOException {<a name="line.1140"></a>
<span class="sourceLineNo">1141</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.1141"></a>
<span class="sourceLineNo">1142</span> throwIOExceptionWhenFail(() -&gt; source.addHFileRefs(tableName, family, pairs));<a name="line.1142"></a>
<span class="sourceLineNo">1143</span> }<a name="line.1143"></a>
<span class="sourceLineNo">1144</span> }<a name="line.1144"></a>
<span class="sourceLineNo">1145</span><a name="line.1145"></a>
<span class="sourceLineNo">1146</span> public void cleanUpHFileRefs(String peerId, List&lt;String&gt; files) {<a name="line.1146"></a>
<span class="sourceLineNo">1147</span> interruptOrAbortWhenFail(() -&gt; this.queueStorage.removeHFileRefs(peerId, files));<a name="line.1147"></a>
<span class="sourceLineNo">1148</span> }<a name="line.1148"></a>
<span class="sourceLineNo">1149</span><a name="line.1149"></a>
<span class="sourceLineNo">1150</span> int activeFailoverTaskCount() {<a name="line.1150"></a>
<span class="sourceLineNo">1151</span> return executor.getActiveCount();<a name="line.1151"></a>
<span class="sourceLineNo">1152</span> }<a name="line.1152"></a>
<span class="sourceLineNo">1153</span><a name="line.1153"></a>
<span class="sourceLineNo">1154</span> MetricsReplicationGlobalSourceSource getGlobalMetrics() {<a name="line.1154"></a>
<span class="sourceLineNo">1155</span> return this.globalMetrics;<a name="line.1155"></a>
<span class="sourceLineNo">1156</span> }<a name="line.1156"></a>
<span class="sourceLineNo">1157</span>}<a name="line.1157"></a>
</pre>
</div>
</body>
</html>