blob: 12a15f73a358ddcebf28a0e535b648f391265c0e [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 static org.apache.hadoop.hbase.wal.AbstractFSWALProvider.getArchivedLogPath;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import java.io.FileNotFoundException;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import java.io.IOException;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.lang.reflect.InvocationTargetException;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.ArrayList;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.Collection;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import java.util.Collections;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import java.util.Comparator;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import java.util.HashMap;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import java.util.List;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import java.util.Map;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import java.util.Set;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import java.util.TreeMap;<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.PriorityBlockingQueue;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import java.util.concurrent.TimeUnit;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import java.util.concurrent.TimeoutException;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import java.util.function.Predicate;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.apache.commons.lang3.StringUtils;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.apache.commons.lang3.mutable.MutableBoolean;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.hadoop.conf.Configuration;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.fs.FileSystem;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.fs.Path;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.HConstants;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.Server;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.ServerName;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.TableDescriptors;<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.regionserver.HRegionServer;<a name="line.52"></a>
<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.regionserver.RSRpcServices;<a name="line.53"></a>
<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.regionserver.RegionServerCoprocessorHost;<a name="line.54"></a>
<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.replication.ChainWALEntryFilter;<a name="line.55"></a>
<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.ClusterMarkingEntryFilter;<a name="line.56"></a>
<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.ReplicationEndpoint;<a name="line.57"></a>
<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.58"></a>
<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.59"></a>
<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.60"></a>
<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.61"></a>
<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.replication.SystemTableWALEntryFilter;<a name="line.62"></a>
<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.replication.WALEntryFilter;<a name="line.63"></a>
<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.64"></a>
<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.65"></a>
<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.66"></a>
<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.67"></a>
<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.68"></a>
<span class="sourceLineNo">069</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.69"></a>
<span class="sourceLineNo">070</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.70"></a>
<span class="sourceLineNo">071</span>import org.slf4j.Logger;<a name="line.71"></a>
<span class="sourceLineNo">072</span>import org.slf4j.LoggerFactory;<a name="line.72"></a>
<span class="sourceLineNo">073</span>import org.apache.hbase.thirdparty.com.google.common.collect.Lists;<a name="line.73"></a>
<span class="sourceLineNo">074</span><a name="line.74"></a>
<span class="sourceLineNo">075</span>/**<a name="line.75"></a>
<span class="sourceLineNo">076</span> * Class that handles the source of a replication stream.<a name="line.76"></a>
<span class="sourceLineNo">077</span> * Currently does not handle more than 1 slave cluster.<a name="line.77"></a>
<span class="sourceLineNo">078</span> * For each slave cluster it selects a random number of peers<a name="line.78"></a>
<span class="sourceLineNo">079</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.79"></a>
<span class="sourceLineNo">080</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.80"></a>
<span class="sourceLineNo">081</span> * &lt;p&gt;<a name="line.81"></a>
<span class="sourceLineNo">082</span> * A stream is considered down when we cannot contact a region server on the<a name="line.82"></a>
<span class="sourceLineNo">083</span> * peer cluster for more than 55 seconds by default.<a name="line.83"></a>
<span class="sourceLineNo">084</span> * &lt;/p&gt;<a name="line.84"></a>
<span class="sourceLineNo">085</span> */<a name="line.85"></a>
<span class="sourceLineNo">086</span>@InterfaceAudience.Private<a name="line.86"></a>
<span class="sourceLineNo">087</span>public class ReplicationSource implements ReplicationSourceInterface {<a name="line.87"></a>
<span class="sourceLineNo">088</span><a name="line.88"></a>
<span class="sourceLineNo">089</span> private static final Logger LOG = LoggerFactory.getLogger(ReplicationSource.class);<a name="line.89"></a>
<span class="sourceLineNo">090</span> // Queues of logs to process, entry in format of walGroupId-&gt;queue,<a name="line.90"></a>
<span class="sourceLineNo">091</span> // each presents a queue for one wal group<a name="line.91"></a>
<span class="sourceLineNo">092</span> private Map&lt;String, PriorityBlockingQueue&lt;Path&gt;&gt; queues = new HashMap&lt;&gt;();<a name="line.92"></a>
<span class="sourceLineNo">093</span> // per group queue size, keep no more than this number of logs in each wal group<a name="line.93"></a>
<span class="sourceLineNo">094</span> protected int queueSizePerGroup;<a name="line.94"></a>
<span class="sourceLineNo">095</span> protected ReplicationQueueStorage queueStorage;<a name="line.95"></a>
<span class="sourceLineNo">096</span> protected ReplicationPeer replicationPeer;<a name="line.96"></a>
<span class="sourceLineNo">097</span><a name="line.97"></a>
<span class="sourceLineNo">098</span> protected Configuration conf;<a name="line.98"></a>
<span class="sourceLineNo">099</span> protected ReplicationQueueInfo replicationQueueInfo;<a name="line.99"></a>
<span class="sourceLineNo">100</span><a name="line.100"></a>
<span class="sourceLineNo">101</span> // The manager of all sources to which we ping back our progress<a name="line.101"></a>
<span class="sourceLineNo">102</span> protected ReplicationSourceManager manager;<a name="line.102"></a>
<span class="sourceLineNo">103</span> // Should we stop everything?<a name="line.103"></a>
<span class="sourceLineNo">104</span> protected Server server;<a name="line.104"></a>
<span class="sourceLineNo">105</span> // How long should we sleep for each retry<a name="line.105"></a>
<span class="sourceLineNo">106</span> private long sleepForRetries;<a name="line.106"></a>
<span class="sourceLineNo">107</span> protected FileSystem fs;<a name="line.107"></a>
<span class="sourceLineNo">108</span> // id of this cluster<a name="line.108"></a>
<span class="sourceLineNo">109</span> private UUID clusterId;<a name="line.109"></a>
<span class="sourceLineNo">110</span> // total number of edits we replicated<a name="line.110"></a>
<span class="sourceLineNo">111</span> private AtomicLong totalReplicatedEdits = new AtomicLong(0);<a name="line.111"></a>
<span class="sourceLineNo">112</span> // The znode we currently play with<a name="line.112"></a>
<span class="sourceLineNo">113</span> protected String queueId;<a name="line.113"></a>
<span class="sourceLineNo">114</span> // Maximum number of retries before taking bold actions<a name="line.114"></a>
<span class="sourceLineNo">115</span> private int maxRetriesMultiplier;<a name="line.115"></a>
<span class="sourceLineNo">116</span> // Indicates if this particular source is running<a name="line.116"></a>
<span class="sourceLineNo">117</span> private volatile boolean sourceRunning = false;<a name="line.117"></a>
<span class="sourceLineNo">118</span> // Metrics for this source<a name="line.118"></a>
<span class="sourceLineNo">119</span> private MetricsSource metrics;<a name="line.119"></a>
<span class="sourceLineNo">120</span> // WARN threshold for the number of queued logs, defaults to 2<a name="line.120"></a>
<span class="sourceLineNo">121</span> private int logQueueWarnThreshold;<a name="line.121"></a>
<span class="sourceLineNo">122</span> // ReplicationEndpoint which will handle the actual replication<a name="line.122"></a>
<span class="sourceLineNo">123</span> private volatile ReplicationEndpoint replicationEndpoint;<a name="line.123"></a>
<span class="sourceLineNo">124</span><a name="line.124"></a>
<span class="sourceLineNo">125</span> private boolean abortOnError;<a name="line.125"></a>
<span class="sourceLineNo">126</span> //This is needed for the startup loop to identify when there's already<a name="line.126"></a>
<span class="sourceLineNo">127</span> //an initialization happening (but not finished yet),<a name="line.127"></a>
<span class="sourceLineNo">128</span> //so that it doesn't try submit another initialize thread.<a name="line.128"></a>
<span class="sourceLineNo">129</span> //NOTE: this should only be set to false at the end of initialize method, prior to return.<a name="line.129"></a>
<span class="sourceLineNo">130</span> private AtomicBoolean startupOngoing = new AtomicBoolean(false);<a name="line.130"></a>
<span class="sourceLineNo">131</span><a name="line.131"></a>
<span class="sourceLineNo">132</span><a name="line.132"></a>
<span class="sourceLineNo">133</span> /**<a name="line.133"></a>
<span class="sourceLineNo">134</span> * A filter (or a chain of filters) for WAL entries; filters out edits.<a name="line.134"></a>
<span class="sourceLineNo">135</span> */<a name="line.135"></a>
<span class="sourceLineNo">136</span> protected volatile WALEntryFilter walEntryFilter;<a name="line.136"></a>
<span class="sourceLineNo">137</span><a name="line.137"></a>
<span class="sourceLineNo">138</span> // throttler<a name="line.138"></a>
<span class="sourceLineNo">139</span> private ReplicationThrottler throttler;<a name="line.139"></a>
<span class="sourceLineNo">140</span> private long defaultBandwidth;<a name="line.140"></a>
<span class="sourceLineNo">141</span> private long currentBandwidth;<a name="line.141"></a>
<span class="sourceLineNo">142</span> private WALFileLengthProvider walFileLengthProvider;<a name="line.142"></a>
<span class="sourceLineNo">143</span> @VisibleForTesting<a name="line.143"></a>
<span class="sourceLineNo">144</span> protected final ConcurrentHashMap&lt;String, ReplicationSourceShipper&gt; workerThreads =<a name="line.144"></a>
<span class="sourceLineNo">145</span> new ConcurrentHashMap&lt;&gt;();<a name="line.145"></a>
<span class="sourceLineNo">146</span><a name="line.146"></a>
<span class="sourceLineNo">147</span> private AtomicLong totalBufferUsed;<a name="line.147"></a>
<span class="sourceLineNo">148</span><a name="line.148"></a>
<span class="sourceLineNo">149</span> public static final String WAIT_ON_ENDPOINT_SECONDS =<a name="line.149"></a>
<span class="sourceLineNo">150</span> "hbase.replication.wait.on.endpoint.seconds";<a name="line.150"></a>
<span class="sourceLineNo">151</span> public static final int DEFAULT_WAIT_ON_ENDPOINT_SECONDS = 30;<a name="line.151"></a>
<span class="sourceLineNo">152</span> private int waitOnEndpointSeconds = -1;<a name="line.152"></a>
<span class="sourceLineNo">153</span><a name="line.153"></a>
<span class="sourceLineNo">154</span> private Thread initThread;<a name="line.154"></a>
<span class="sourceLineNo">155</span><a name="line.155"></a>
<span class="sourceLineNo">156</span> /**<a name="line.156"></a>
<span class="sourceLineNo">157</span> * WALs to replicate.<a name="line.157"></a>
<span class="sourceLineNo">158</span> * Predicate that returns 'true' for WALs to replicate and false for WALs to skip.<a name="line.158"></a>
<span class="sourceLineNo">159</span> */<a name="line.159"></a>
<span class="sourceLineNo">160</span> private final Predicate&lt;Path&gt; filterInWALs;<a name="line.160"></a>
<span class="sourceLineNo">161</span><a name="line.161"></a>
<span class="sourceLineNo">162</span> /**<a name="line.162"></a>
<span class="sourceLineNo">163</span> * Base WALEntry filters for this class. Unmodifiable. Set on construction.<a name="line.163"></a>
<span class="sourceLineNo">164</span> * Filters *out* edits we do not want replicated, passed on to replication endpoints.<a name="line.164"></a>
<span class="sourceLineNo">165</span> * This is the basic set. Down in #initializeWALEntryFilter this set is added to the end of<a name="line.165"></a>
<span class="sourceLineNo">166</span> * the WALEntry filter chain. These are put after those that we pick up from the configured<a name="line.166"></a>
<span class="sourceLineNo">167</span> * endpoints and other machinations to create the final {@link #walEntryFilter}.<a name="line.167"></a>
<span class="sourceLineNo">168</span> * @see WALEntryFilter<a name="line.168"></a>
<span class="sourceLineNo">169</span> */<a name="line.169"></a>
<span class="sourceLineNo">170</span> private final List&lt;WALEntryFilter&gt; baseFilterOutWALEntries;<a name="line.170"></a>
<span class="sourceLineNo">171</span><a name="line.171"></a>
<span class="sourceLineNo">172</span> ReplicationSource() {<a name="line.172"></a>
<span class="sourceLineNo">173</span> // Default, filters *in* all WALs but meta WALs &amp; filters *out* all WALEntries of System Tables.<a name="line.173"></a>
<span class="sourceLineNo">174</span> this(p -&gt; !AbstractFSWALProvider.isMetaFile(p),<a name="line.174"></a>
<span class="sourceLineNo">175</span> Lists.newArrayList(new SystemTableWALEntryFilter()));<a name="line.175"></a>
<span class="sourceLineNo">176</span> }<a name="line.176"></a>
<span class="sourceLineNo">177</span><a name="line.177"></a>
<span class="sourceLineNo">178</span> /**<a name="line.178"></a>
<span class="sourceLineNo">179</span> * @param replicateWAL Pass a filter to run against WAL Path; filter *in* WALs to Replicate;<a name="line.179"></a>
<span class="sourceLineNo">180</span> * i.e. return 'true' if you want to replicate the content of the WAL.<a name="line.180"></a>
<span class="sourceLineNo">181</span> * @param baseFilterOutWALEntries Base set of filters you want applied always; filters *out*<a name="line.181"></a>
<span class="sourceLineNo">182</span> * WALEntries so they never make it out of this ReplicationSource.<a name="line.182"></a>
<span class="sourceLineNo">183</span> */<a name="line.183"></a>
<span class="sourceLineNo">184</span> ReplicationSource(Predicate&lt;Path&gt; replicateWAL, List&lt;WALEntryFilter&gt; baseFilterOutWALEntries) {<a name="line.184"></a>
<span class="sourceLineNo">185</span> this.filterInWALs = replicateWAL;<a name="line.185"></a>
<span class="sourceLineNo">186</span> this.baseFilterOutWALEntries = Collections.unmodifiableList(baseFilterOutWALEntries);<a name="line.186"></a>
<span class="sourceLineNo">187</span> }<a name="line.187"></a>
<span class="sourceLineNo">188</span><a name="line.188"></a>
<span class="sourceLineNo">189</span> /**<a name="line.189"></a>
<span class="sourceLineNo">190</span> * Instantiation method used by region servers<a name="line.190"></a>
<span class="sourceLineNo">191</span> * @param conf configuration to use<a name="line.191"></a>
<span class="sourceLineNo">192</span> * @param fs file system to use<a name="line.192"></a>
<span class="sourceLineNo">193</span> * @param manager replication manager to ping to<a name="line.193"></a>
<span class="sourceLineNo">194</span> * @param server the server for this region server<a name="line.194"></a>
<span class="sourceLineNo">195</span> * @param queueId the id of our replication queue<a name="line.195"></a>
<span class="sourceLineNo">196</span> * @param clusterId unique UUID for the cluster<a name="line.196"></a>
<span class="sourceLineNo">197</span> * @param metrics metrics for replication source<a name="line.197"></a>
<span class="sourceLineNo">198</span> */<a name="line.198"></a>
<span class="sourceLineNo">199</span> @Override<a name="line.199"></a>
<span class="sourceLineNo">200</span> public void init(Configuration conf, FileSystem fs, ReplicationSourceManager manager,<a name="line.200"></a>
<span class="sourceLineNo">201</span> ReplicationQueueStorage queueStorage, ReplicationPeer replicationPeer, Server server,<a name="line.201"></a>
<span class="sourceLineNo">202</span> String queueId, UUID clusterId, WALFileLengthProvider walFileLengthProvider,<a name="line.202"></a>
<span class="sourceLineNo">203</span> MetricsSource metrics) throws IOException {<a name="line.203"></a>
<span class="sourceLineNo">204</span> this.server = server;<a name="line.204"></a>
<span class="sourceLineNo">205</span> this.conf = HBaseConfiguration.create(conf);<a name="line.205"></a>
<span class="sourceLineNo">206</span> this.waitOnEndpointSeconds =<a name="line.206"></a>
<span class="sourceLineNo">207</span> this.conf.getInt(WAIT_ON_ENDPOINT_SECONDS, DEFAULT_WAIT_ON_ENDPOINT_SECONDS);<a name="line.207"></a>
<span class="sourceLineNo">208</span> decorateConf();<a name="line.208"></a>
<span class="sourceLineNo">209</span> this.sleepForRetries =<a name="line.209"></a>
<span class="sourceLineNo">210</span> this.conf.getLong("replication.source.sleepforretries", 1000); // 1 second<a name="line.210"></a>
<span class="sourceLineNo">211</span> this.maxRetriesMultiplier =<a name="line.211"></a>
<span class="sourceLineNo">212</span> this.conf.getInt("replication.source.maxretriesmultiplier", 300); // 5 minutes @ 1 sec per<a name="line.212"></a>
<span class="sourceLineNo">213</span> this.queueSizePerGroup = this.conf.getInt("hbase.regionserver.maxlogs", 32);<a name="line.213"></a>
<span class="sourceLineNo">214</span> this.queueStorage = queueStorage;<a name="line.214"></a>
<span class="sourceLineNo">215</span> this.replicationPeer = replicationPeer;<a name="line.215"></a>
<span class="sourceLineNo">216</span> this.manager = manager;<a name="line.216"></a>
<span class="sourceLineNo">217</span> this.fs = fs;<a name="line.217"></a>
<span class="sourceLineNo">218</span> this.metrics = metrics;<a name="line.218"></a>
<span class="sourceLineNo">219</span> this.clusterId = clusterId;<a name="line.219"></a>
<span class="sourceLineNo">220</span><a name="line.220"></a>
<span class="sourceLineNo">221</span> this.queueId = queueId;<a name="line.221"></a>
<span class="sourceLineNo">222</span> this.replicationQueueInfo = new ReplicationQueueInfo(queueId);<a name="line.222"></a>
<span class="sourceLineNo">223</span> this.logQueueWarnThreshold = this.conf.getInt("replication.source.log.queue.warn", 2);<a name="line.223"></a>
<span class="sourceLineNo">224</span><a name="line.224"></a>
<span class="sourceLineNo">225</span> defaultBandwidth = this.conf.getLong("replication.source.per.peer.node.bandwidth", 0);<a name="line.225"></a>
<span class="sourceLineNo">226</span> currentBandwidth = getCurrentBandwidth();<a name="line.226"></a>
<span class="sourceLineNo">227</span> this.throttler = new ReplicationThrottler((double) currentBandwidth / 10.0);<a name="line.227"></a>
<span class="sourceLineNo">228</span> this.totalBufferUsed = manager.getTotalBufferUsed();<a name="line.228"></a>
<span class="sourceLineNo">229</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.229"></a>
<span class="sourceLineNo">230</span><a name="line.230"></a>
<span class="sourceLineNo">231</span> this.abortOnError = this.conf.getBoolean("replication.source.regionserver.abort",<a name="line.231"></a>
<span class="sourceLineNo">232</span> true);<a name="line.232"></a>
<span class="sourceLineNo">233</span><a name="line.233"></a>
<span class="sourceLineNo">234</span> LOG.info("queueId={}, ReplicationSource: {}, currentBandwidth={}", queueId,<a name="line.234"></a>
<span class="sourceLineNo">235</span> replicationPeer.getId(), this.currentBandwidth);<a name="line.235"></a>
<span class="sourceLineNo">236</span> }<a name="line.236"></a>
<span class="sourceLineNo">237</span><a name="line.237"></a>
<span class="sourceLineNo">238</span> private void decorateConf() {<a name="line.238"></a>
<span class="sourceLineNo">239</span> String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.239"></a>
<span class="sourceLineNo">240</span> if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.240"></a>
<span class="sourceLineNo">241</span> this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.241"></a>
<span class="sourceLineNo">242</span> }<a name="line.242"></a>
<span class="sourceLineNo">243</span> }<a name="line.243"></a>
<span class="sourceLineNo">244</span><a name="line.244"></a>
<span class="sourceLineNo">245</span> @Override<a name="line.245"></a>
<span class="sourceLineNo">246</span> public void enqueueLog(Path wal) {<a name="line.246"></a>
<span class="sourceLineNo">247</span> if (!this.filterInWALs.test(wal)) {<a name="line.247"></a>
<span class="sourceLineNo">248</span> LOG.trace("NOT replicating {}", wal);<a name="line.248"></a>
<span class="sourceLineNo">249</span> return;<a name="line.249"></a>
<span class="sourceLineNo">250</span> }<a name="line.250"></a>
<span class="sourceLineNo">251</span> // Use WAL prefix as the WALGroupId for this peer.<a name="line.251"></a>
<span class="sourceLineNo">252</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(wal.getName());<a name="line.252"></a>
<span class="sourceLineNo">253</span> PriorityBlockingQueue&lt;Path&gt; queue = queues.get(walPrefix);<a name="line.253"></a>
<span class="sourceLineNo">254</span> if (queue == null) {<a name="line.254"></a>
<span class="sourceLineNo">255</span> queue = new PriorityBlockingQueue&lt;&gt;(queueSizePerGroup, new LogsComparator());<a name="line.255"></a>
<span class="sourceLineNo">256</span> // make sure that we do not use an empty queue when setting up a ReplicationSource, otherwise<a name="line.256"></a>
<span class="sourceLineNo">257</span> // the shipper may quit immediately<a name="line.257"></a>
<span class="sourceLineNo">258</span> queue.put(wal);<a name="line.258"></a>
<span class="sourceLineNo">259</span> queues.put(walPrefix, queue);<a name="line.259"></a>
<span class="sourceLineNo">260</span> if (this.isSourceActive() &amp;&amp; this.walEntryFilter != null) {<a name="line.260"></a>
<span class="sourceLineNo">261</span> // new wal group observed after source startup, start a new worker thread to track it<a name="line.261"></a>
<span class="sourceLineNo">262</span> // notice: it's possible that wal enqueued when this.running is set but worker thread<a name="line.262"></a>
<span class="sourceLineNo">263</span> // still not launched, so it's necessary to check workerThreads before start the worker<a name="line.263"></a>
<span class="sourceLineNo">264</span> tryStartNewShipper(walPrefix, queue);<a name="line.264"></a>
<span class="sourceLineNo">265</span> }<a name="line.265"></a>
<span class="sourceLineNo">266</span> } else {<a name="line.266"></a>
<span class="sourceLineNo">267</span> queue.put(wal);<a name="line.267"></a>
<span class="sourceLineNo">268</span> }<a name="line.268"></a>
<span class="sourceLineNo">269</span> if (LOG.isTraceEnabled()) {<a name="line.269"></a>
<span class="sourceLineNo">270</span> LOG.trace("{} Added wal {} to queue of source {}.", logPeerId(), walPrefix,<a name="line.270"></a>
<span class="sourceLineNo">271</span> this.replicationQueueInfo.getQueueId());<a name="line.271"></a>
<span class="sourceLineNo">272</span> }<a name="line.272"></a>
<span class="sourceLineNo">273</span> this.metrics.incrSizeOfLogQueue();<a name="line.273"></a>
<span class="sourceLineNo">274</span> // This will wal a warning for each new wal that gets created above the warn threshold<a name="line.274"></a>
<span class="sourceLineNo">275</span> int queueSize = queue.size();<a name="line.275"></a>
<span class="sourceLineNo">276</span> if (queueSize &gt; this.logQueueWarnThreshold) {<a name="line.276"></a>
<span class="sourceLineNo">277</span> LOG.warn("{} WAL group {} queue size: {} exceeds value of " +<a name="line.277"></a>
<span class="sourceLineNo">278</span> "replication.source.log.queue.warn {}", logPeerId(), walPrefix, queueSize,<a name="line.278"></a>
<span class="sourceLineNo">279</span> logQueueWarnThreshold);<a name="line.279"></a>
<span class="sourceLineNo">280</span> }<a name="line.280"></a>
<span class="sourceLineNo">281</span> }<a name="line.281"></a>
<span class="sourceLineNo">282</span><a name="line.282"></a>
<span class="sourceLineNo">283</span> @Override<a name="line.283"></a>
<span class="sourceLineNo">284</span> public void addHFileRefs(TableName tableName, byte[] family, List&lt;Pair&lt;Path, Path&gt;&gt; pairs)<a name="line.284"></a>
<span class="sourceLineNo">285</span> throws ReplicationException {<a name="line.285"></a>
<span class="sourceLineNo">286</span> String peerId = replicationPeer.getId();<a name="line.286"></a>
<span class="sourceLineNo">287</span> Set&lt;String&gt; namespaces = replicationPeer.getNamespaces();<a name="line.287"></a>
<span class="sourceLineNo">288</span> Map&lt;TableName, List&lt;String&gt;&gt; tableCFMap = replicationPeer.getTableCFs();<a name="line.288"></a>
<span class="sourceLineNo">289</span> if (tableCFMap != null) { // All peers with TableCFs<a name="line.289"></a>
<span class="sourceLineNo">290</span> List&lt;String&gt; tableCfs = tableCFMap.get(tableName);<a name="line.290"></a>
<span class="sourceLineNo">291</span> if (tableCFMap.containsKey(tableName)<a name="line.291"></a>
<span class="sourceLineNo">292</span> &amp;&amp; (tableCfs == null || tableCfs.contains(Bytes.toString(family)))) {<a name="line.292"></a>
<span class="sourceLineNo">293</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.293"></a>
<span class="sourceLineNo">294</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.294"></a>
<span class="sourceLineNo">295</span> } else {<a name="line.295"></a>
<span class="sourceLineNo">296</span> LOG.debug("HFiles will not be replicated belonging to the table {} family {} to peer id {}",<a name="line.296"></a>
<span class="sourceLineNo">297</span> tableName, Bytes.toString(family), peerId);<a name="line.297"></a>
<span class="sourceLineNo">298</span> }<a name="line.298"></a>
<span class="sourceLineNo">299</span> } else if (namespaces != null) { // Only for set NAMESPACES peers<a name="line.299"></a>
<span class="sourceLineNo">300</span> if (namespaces.contains(tableName.getNamespaceAsString())) {<a name="line.300"></a>
<span class="sourceLineNo">301</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.301"></a>
<span class="sourceLineNo">302</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.302"></a>
<span class="sourceLineNo">303</span> } else {<a name="line.303"></a>
<span class="sourceLineNo">304</span> LOG.debug("HFiles will not be replicated belonging to the table {} family {} to peer id {}",<a name="line.304"></a>
<span class="sourceLineNo">305</span> tableName, Bytes.toString(family), peerId);<a name="line.305"></a>
<span class="sourceLineNo">306</span> }<a name="line.306"></a>
<span class="sourceLineNo">307</span> } else {<a name="line.307"></a>
<span class="sourceLineNo">308</span> // user has explicitly not defined any table cfs for replication, means replicate all the<a name="line.308"></a>
<span class="sourceLineNo">309</span> // data<a name="line.309"></a>
<span class="sourceLineNo">310</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.310"></a>
<span class="sourceLineNo">311</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.311"></a>
<span class="sourceLineNo">312</span> }<a name="line.312"></a>
<span class="sourceLineNo">313</span> }<a name="line.313"></a>
<span class="sourceLineNo">314</span><a name="line.314"></a>
<span class="sourceLineNo">315</span> private ReplicationEndpoint createReplicationEndpoint()<a name="line.315"></a>
<span class="sourceLineNo">316</span> throws InstantiationException, IllegalAccessException, ClassNotFoundException, IOException {<a name="line.316"></a>
<span class="sourceLineNo">317</span> RegionServerCoprocessorHost rsServerHost = null;<a name="line.317"></a>
<span class="sourceLineNo">318</span> if (server instanceof HRegionServer) {<a name="line.318"></a>
<span class="sourceLineNo">319</span> rsServerHost = ((HRegionServer) server).getRegionServerCoprocessorHost();<a name="line.319"></a>
<span class="sourceLineNo">320</span> }<a name="line.320"></a>
<span class="sourceLineNo">321</span> String replicationEndpointImpl = replicationPeer.getPeerConfig().getReplicationEndpointImpl();<a name="line.321"></a>
<span class="sourceLineNo">322</span><a name="line.322"></a>
<span class="sourceLineNo">323</span> ReplicationEndpoint replicationEndpoint;<a name="line.323"></a>
<span class="sourceLineNo">324</span> if (replicationEndpointImpl == null) {<a name="line.324"></a>
<span class="sourceLineNo">325</span> // Default to HBase inter-cluster replication endpoint; skip reflection<a name="line.325"></a>
<span class="sourceLineNo">326</span> replicationEndpoint = new HBaseInterClusterReplicationEndpoint();<a name="line.326"></a>
<span class="sourceLineNo">327</span> } else {<a name="line.327"></a>
<span class="sourceLineNo">328</span> try {<a name="line.328"></a>
<span class="sourceLineNo">329</span> replicationEndpoint = Class.forName(replicationEndpointImpl)<a name="line.329"></a>
<span class="sourceLineNo">330</span> .asSubclass(ReplicationEndpoint.class)<a name="line.330"></a>
<span class="sourceLineNo">331</span> .getDeclaredConstructor()<a name="line.331"></a>
<span class="sourceLineNo">332</span> .newInstance();<a name="line.332"></a>
<span class="sourceLineNo">333</span> } catch (NoSuchMethodException | InvocationTargetException e) {<a name="line.333"></a>
<span class="sourceLineNo">334</span> throw new IllegalArgumentException(e);<a name="line.334"></a>
<span class="sourceLineNo">335</span> }<a name="line.335"></a>
<span class="sourceLineNo">336</span> }<a name="line.336"></a>
<span class="sourceLineNo">337</span> if (rsServerHost != null) {<a name="line.337"></a>
<span class="sourceLineNo">338</span> ReplicationEndpoint newReplicationEndPoint =<a name="line.338"></a>
<span class="sourceLineNo">339</span> rsServerHost.postCreateReplicationEndPoint(replicationEndpoint);<a name="line.339"></a>
<span class="sourceLineNo">340</span> if (newReplicationEndPoint != null) {<a name="line.340"></a>
<span class="sourceLineNo">341</span> // Override the newly created endpoint from the hook with configured end point<a name="line.341"></a>
<span class="sourceLineNo">342</span> replicationEndpoint = newReplicationEndPoint;<a name="line.342"></a>
<span class="sourceLineNo">343</span> }<a name="line.343"></a>
<span class="sourceLineNo">344</span> }<a name="line.344"></a>
<span class="sourceLineNo">345</span> return replicationEndpoint;<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> private void initAndStartReplicationEndpoint(ReplicationEndpoint replicationEndpoint)<a name="line.348"></a>
<span class="sourceLineNo">349</span> throws IOException, TimeoutException {<a name="line.349"></a>
<span class="sourceLineNo">350</span> TableDescriptors tableDescriptors = null;<a name="line.350"></a>
<span class="sourceLineNo">351</span> if (server instanceof HRegionServer) {<a name="line.351"></a>
<span class="sourceLineNo">352</span> tableDescriptors = ((HRegionServer) server).getTableDescriptors();<a name="line.352"></a>
<span class="sourceLineNo">353</span> }<a name="line.353"></a>
<span class="sourceLineNo">354</span> replicationEndpoint<a name="line.354"></a>
<span class="sourceLineNo">355</span> .init(new ReplicationEndpoint.Context(server, conf, replicationPeer.getConfiguration(), fs,<a name="line.355"></a>
<span class="sourceLineNo">356</span> replicationPeer.getId(), clusterId, replicationPeer, metrics, tableDescriptors, server));<a name="line.356"></a>
<span class="sourceLineNo">357</span> replicationEndpoint.start();<a name="line.357"></a>
<span class="sourceLineNo">358</span> replicationEndpoint.awaitRunning(waitOnEndpointSeconds, TimeUnit.SECONDS);<a name="line.358"></a>
<span class="sourceLineNo">359</span> }<a name="line.359"></a>
<span class="sourceLineNo">360</span><a name="line.360"></a>
<span class="sourceLineNo">361</span> private void initializeWALEntryFilter(UUID peerClusterId) {<a name="line.361"></a>
<span class="sourceLineNo">362</span> // get the WALEntryFilter from ReplicationEndpoint and add it to default filters<a name="line.362"></a>
<span class="sourceLineNo">363</span> List&lt;WALEntryFilter&gt; filters = new ArrayList&lt;&gt;(this.baseFilterOutWALEntries);<a name="line.363"></a>
<span class="sourceLineNo">364</span> WALEntryFilter filterFromEndpoint = this.replicationEndpoint.getWALEntryfilter();<a name="line.364"></a>
<span class="sourceLineNo">365</span> if (filterFromEndpoint != null) {<a name="line.365"></a>
<span class="sourceLineNo">366</span> filters.add(filterFromEndpoint);<a name="line.366"></a>
<span class="sourceLineNo">367</span> }<a name="line.367"></a>
<span class="sourceLineNo">368</span> filters.add(new ClusterMarkingEntryFilter(clusterId, peerClusterId, replicationEndpoint));<a name="line.368"></a>
<span class="sourceLineNo">369</span> this.walEntryFilter = new ChainWALEntryFilter(filters);<a name="line.369"></a>
<span class="sourceLineNo">370</span> }<a name="line.370"></a>
<span class="sourceLineNo">371</span><a name="line.371"></a>
<span class="sourceLineNo">372</span> private void tryStartNewShipper(String walGroupId, PriorityBlockingQueue&lt;Path&gt; queue) {<a name="line.372"></a>
<span class="sourceLineNo">373</span> workerThreads.compute(walGroupId, (key, value) -&gt; {<a name="line.373"></a>
<span class="sourceLineNo">374</span> if (value != null) {<a name="line.374"></a>
<span class="sourceLineNo">375</span> LOG.debug("{} preempted start of worker walGroupId={}", logPeerId(), walGroupId);<a name="line.375"></a>
<span class="sourceLineNo">376</span> return value;<a name="line.376"></a>
<span class="sourceLineNo">377</span> } else {<a name="line.377"></a>
<span class="sourceLineNo">378</span> LOG.debug("{} starting worker for walGroupId={}", logPeerId(), walGroupId);<a name="line.378"></a>
<span class="sourceLineNo">379</span> ReplicationSourceShipper worker = createNewShipper(walGroupId, queue);<a name="line.379"></a>
<span class="sourceLineNo">380</span> ReplicationSourceWALReader walReader =<a name="line.380"></a>
<span class="sourceLineNo">381</span> createNewWALReader(walGroupId, queue, worker.getStartPosition());<a name="line.381"></a>
<span class="sourceLineNo">382</span> Threads.setDaemonThreadRunning(<a name="line.382"></a>
<span class="sourceLineNo">383</span> walReader, Thread.currentThread().getName()<a name="line.383"></a>
<span class="sourceLineNo">384</span> + ".replicationSource.wal-reader." + walGroupId + "," + queueId,<a name="line.384"></a>
<span class="sourceLineNo">385</span> (t,e) -&gt; this.uncaughtException(t, e, this.manager, this.getPeerId()));<a name="line.385"></a>
<span class="sourceLineNo">386</span> worker.setWALReader(walReader);<a name="line.386"></a>
<span class="sourceLineNo">387</span> worker.startup((t,e) -&gt; this.uncaughtException(t, e, this.manager, this.getPeerId()));<a name="line.387"></a>
<span class="sourceLineNo">388</span> return worker;<a name="line.388"></a>
<span class="sourceLineNo">389</span> }<a name="line.389"></a>
<span class="sourceLineNo">390</span> });<a name="line.390"></a>
<span class="sourceLineNo">391</span> }<a name="line.391"></a>
<span class="sourceLineNo">392</span><a name="line.392"></a>
<span class="sourceLineNo">393</span> @Override<a name="line.393"></a>
<span class="sourceLineNo">394</span> public Map&lt;String, ReplicationStatus&gt; getWalGroupStatus() {<a name="line.394"></a>
<span class="sourceLineNo">395</span> Map&lt;String, ReplicationStatus&gt; sourceReplicationStatus = new TreeMap&lt;&gt;();<a name="line.395"></a>
<span class="sourceLineNo">396</span> long ageOfLastShippedOp, replicationDelay, fileSize;<a name="line.396"></a>
<span class="sourceLineNo">397</span> for (Map.Entry&lt;String, ReplicationSourceShipper&gt; walGroupShipper : workerThreads.entrySet()) {<a name="line.397"></a>
<span class="sourceLineNo">398</span> String walGroupId = walGroupShipper.getKey();<a name="line.398"></a>
<span class="sourceLineNo">399</span> ReplicationSourceShipper shipper = walGroupShipper.getValue();<a name="line.399"></a>
<span class="sourceLineNo">400</span> ageOfLastShippedOp = metrics.getAgeOfLastShippedOp(walGroupId);<a name="line.400"></a>
<span class="sourceLineNo">401</span> int queueSize = queues.get(walGroupId).size();<a name="line.401"></a>
<span class="sourceLineNo">402</span> replicationDelay = metrics.getReplicationDelay();<a name="line.402"></a>
<span class="sourceLineNo">403</span> Path currentPath = shipper.getCurrentPath();<a name="line.403"></a>
<span class="sourceLineNo">404</span> fileSize = -1;<a name="line.404"></a>
<span class="sourceLineNo">405</span> if (currentPath != null) {<a name="line.405"></a>
<span class="sourceLineNo">406</span> try {<a name="line.406"></a>
<span class="sourceLineNo">407</span> fileSize = getFileSize(currentPath);<a name="line.407"></a>
<span class="sourceLineNo">408</span> } catch (IOException e) {<a name="line.408"></a>
<span class="sourceLineNo">409</span> LOG.warn("Ignore the exception as the file size of HLog only affects the web ui", e);<a name="line.409"></a>
<span class="sourceLineNo">410</span> }<a name="line.410"></a>
<span class="sourceLineNo">411</span> } else {<a name="line.411"></a>
<span class="sourceLineNo">412</span> currentPath = new Path("NO_LOGS_IN_QUEUE");<a name="line.412"></a>
<span class="sourceLineNo">413</span> LOG.warn("{} No replication ongoing, waiting for new log", logPeerId());<a name="line.413"></a>
<span class="sourceLineNo">414</span> }<a name="line.414"></a>
<span class="sourceLineNo">415</span> ReplicationStatus.ReplicationStatusBuilder statusBuilder = ReplicationStatus.newBuilder();<a name="line.415"></a>
<span class="sourceLineNo">416</span> statusBuilder.withPeerId(this.getPeerId())<a name="line.416"></a>
<span class="sourceLineNo">417</span> .withQueueSize(queueSize)<a name="line.417"></a>
<span class="sourceLineNo">418</span> .withWalGroup(walGroupId)<a name="line.418"></a>
<span class="sourceLineNo">419</span> .withCurrentPath(currentPath)<a name="line.419"></a>
<span class="sourceLineNo">420</span> .withCurrentPosition(shipper.getCurrentPosition())<a name="line.420"></a>
<span class="sourceLineNo">421</span> .withFileSize(fileSize)<a name="line.421"></a>
<span class="sourceLineNo">422</span> .withAgeOfLastShippedOp(ageOfLastShippedOp)<a name="line.422"></a>
<span class="sourceLineNo">423</span> .withReplicationDelay(replicationDelay);<a name="line.423"></a>
<span class="sourceLineNo">424</span> sourceReplicationStatus.put(this.getPeerId() + "=&gt;" + walGroupId, statusBuilder.build());<a name="line.424"></a>
<span class="sourceLineNo">425</span> }<a name="line.425"></a>
<span class="sourceLineNo">426</span> return sourceReplicationStatus;<a name="line.426"></a>
<span class="sourceLineNo">427</span> }<a name="line.427"></a>
<span class="sourceLineNo">428</span><a name="line.428"></a>
<span class="sourceLineNo">429</span> private long getFileSize(Path currentPath) throws IOException {<a name="line.429"></a>
<span class="sourceLineNo">430</span> long fileSize;<a name="line.430"></a>
<span class="sourceLineNo">431</span> try {<a name="line.431"></a>
<span class="sourceLineNo">432</span> fileSize = fs.getContentSummary(currentPath).getLength();<a name="line.432"></a>
<span class="sourceLineNo">433</span> } catch (FileNotFoundException e) {<a name="line.433"></a>
<span class="sourceLineNo">434</span> currentPath = getArchivedLogPath(currentPath, conf);<a name="line.434"></a>
<span class="sourceLineNo">435</span> fileSize = fs.getContentSummary(currentPath).getLength();<a name="line.435"></a>
<span class="sourceLineNo">436</span> }<a name="line.436"></a>
<span class="sourceLineNo">437</span> return fileSize;<a name="line.437"></a>
<span class="sourceLineNo">438</span> }<a name="line.438"></a>
<span class="sourceLineNo">439</span><a name="line.439"></a>
<span class="sourceLineNo">440</span> protected ReplicationSourceShipper createNewShipper(String walGroupId,<a name="line.440"></a>
<span class="sourceLineNo">441</span> PriorityBlockingQueue&lt;Path&gt; queue) {<a name="line.441"></a>
<span class="sourceLineNo">442</span> return new ReplicationSourceShipper(conf, walGroupId, queue, this);<a name="line.442"></a>
<span class="sourceLineNo">443</span> }<a name="line.443"></a>
<span class="sourceLineNo">444</span><a name="line.444"></a>
<span class="sourceLineNo">445</span> private ReplicationSourceWALReader createNewWALReader(String walGroupId,<a name="line.445"></a>
<span class="sourceLineNo">446</span> PriorityBlockingQueue&lt;Path&gt; queue, long startPosition) {<a name="line.446"></a>
<span class="sourceLineNo">447</span> return replicationPeer.getPeerConfig().isSerial()<a name="line.447"></a>
<span class="sourceLineNo">448</span> ? new SerialReplicationSourceWALReader(fs, conf, queue, startPosition, walEntryFilter, this)<a name="line.448"></a>
<span class="sourceLineNo">449</span> : new ReplicationSourceWALReader(fs, conf, queue, startPosition, walEntryFilter, this);<a name="line.449"></a>
<span class="sourceLineNo">450</span> }<a name="line.450"></a>
<span class="sourceLineNo">451</span><a name="line.451"></a>
<span class="sourceLineNo">452</span> /**<a name="line.452"></a>
<span class="sourceLineNo">453</span> * Call after {@link #initializeWALEntryFilter(UUID)} else it will be null.<a name="line.453"></a>
<span class="sourceLineNo">454</span> * @return WAL Entry Filter Chain to use on WAL files filtering *out* WALEntry edits.<a name="line.454"></a>
<span class="sourceLineNo">455</span> */<a name="line.455"></a>
<span class="sourceLineNo">456</span> @VisibleForTesting<a name="line.456"></a>
<span class="sourceLineNo">457</span> WALEntryFilter getWalEntryFilter() {<a name="line.457"></a>
<span class="sourceLineNo">458</span> return walEntryFilter;<a name="line.458"></a>
<span class="sourceLineNo">459</span> }<a name="line.459"></a>
<span class="sourceLineNo">460</span><a name="line.460"></a>
<span class="sourceLineNo">461</span> protected final void uncaughtException(Thread t, Throwable e,<a name="line.461"></a>
<span class="sourceLineNo">462</span> ReplicationSourceManager manager, String peerId) {<a name="line.462"></a>
<span class="sourceLineNo">463</span> RSRpcServices.exitIfOOME(e);<a name="line.463"></a>
<span class="sourceLineNo">464</span> LOG.error("Unexpected exception in {} currentPath={}",<a name="line.464"></a>
<span class="sourceLineNo">465</span> t.getName(), getCurrentPath(), e);<a name="line.465"></a>
<span class="sourceLineNo">466</span> if(abortOnError){<a name="line.466"></a>
<span class="sourceLineNo">467</span> server.abort("Unexpected exception in " + t.getName(), e);<a name="line.467"></a>
<span class="sourceLineNo">468</span> }<a name="line.468"></a>
<span class="sourceLineNo">469</span> if(manager != null){<a name="line.469"></a>
<span class="sourceLineNo">470</span> while (true) {<a name="line.470"></a>
<span class="sourceLineNo">471</span> try {<a name="line.471"></a>
<span class="sourceLineNo">472</span> LOG.info("Refreshing replication sources now due to previous error on thread: {}",<a name="line.472"></a>
<span class="sourceLineNo">473</span> t.getName());<a name="line.473"></a>
<span class="sourceLineNo">474</span> manager.refreshSources(peerId);<a name="line.474"></a>
<span class="sourceLineNo">475</span> break;<a name="line.475"></a>
<span class="sourceLineNo">476</span> } catch (IOException e1) {<a name="line.476"></a>
<span class="sourceLineNo">477</span> LOG.error("Replication sources refresh failed.", e1);<a name="line.477"></a>
<span class="sourceLineNo">478</span> sleepForRetries("Sleeping before try refreshing sources again",<a name="line.478"></a>
<span class="sourceLineNo">479</span> maxRetriesMultiplier);<a name="line.479"></a>
<span class="sourceLineNo">480</span> }<a name="line.480"></a>
<span class="sourceLineNo">481</span> }<a name="line.481"></a>
<span class="sourceLineNo">482</span> }<a name="line.482"></a>
<span class="sourceLineNo">483</span> }<a name="line.483"></a>
<span class="sourceLineNo">484</span><a name="line.484"></a>
<span class="sourceLineNo">485</span> @Override<a name="line.485"></a>
<span class="sourceLineNo">486</span> public ReplicationEndpoint getReplicationEndpoint() {<a name="line.486"></a>
<span class="sourceLineNo">487</span> return this.replicationEndpoint;<a name="line.487"></a>
<span class="sourceLineNo">488</span> }<a name="line.488"></a>
<span class="sourceLineNo">489</span><a name="line.489"></a>
<span class="sourceLineNo">490</span> @Override<a name="line.490"></a>
<span class="sourceLineNo">491</span> public ReplicationSourceManager getSourceManager() {<a name="line.491"></a>
<span class="sourceLineNo">492</span> return this.manager;<a name="line.492"></a>
<span class="sourceLineNo">493</span> }<a name="line.493"></a>
<span class="sourceLineNo">494</span><a name="line.494"></a>
<span class="sourceLineNo">495</span> @Override<a name="line.495"></a>
<span class="sourceLineNo">496</span> public void tryThrottle(int batchSize) throws InterruptedException {<a name="line.496"></a>
<span class="sourceLineNo">497</span> checkBandwidthChangeAndResetThrottler();<a name="line.497"></a>
<span class="sourceLineNo">498</span> if (throttler.isEnabled()) {<a name="line.498"></a>
<span class="sourceLineNo">499</span> long sleepTicks = throttler.getNextSleepInterval(batchSize);<a name="line.499"></a>
<span class="sourceLineNo">500</span> if (sleepTicks &gt; 0) {<a name="line.500"></a>
<span class="sourceLineNo">501</span> if (LOG.isTraceEnabled()) {<a name="line.501"></a>
<span class="sourceLineNo">502</span> LOG.trace("{} To sleep {}ms for throttling control", logPeerId(), sleepTicks);<a name="line.502"></a>
<span class="sourceLineNo">503</span> }<a name="line.503"></a>
<span class="sourceLineNo">504</span> Thread.sleep(sleepTicks);<a name="line.504"></a>
<span class="sourceLineNo">505</span> // reset throttler's cycle start tick when sleep for throttling occurs<a name="line.505"></a>
<span class="sourceLineNo">506</span> throttler.resetStartTick();<a name="line.506"></a>
<span class="sourceLineNo">507</span> }<a name="line.507"></a>
<span class="sourceLineNo">508</span> }<a name="line.508"></a>
<span class="sourceLineNo">509</span> }<a name="line.509"></a>
<span class="sourceLineNo">510</span><a name="line.510"></a>
<span class="sourceLineNo">511</span> private void checkBandwidthChangeAndResetThrottler() {<a name="line.511"></a>
<span class="sourceLineNo">512</span> long peerBandwidth = getCurrentBandwidth();<a name="line.512"></a>
<span class="sourceLineNo">513</span> if (peerBandwidth != currentBandwidth) {<a name="line.513"></a>
<span class="sourceLineNo">514</span> currentBandwidth = peerBandwidth;<a name="line.514"></a>
<span class="sourceLineNo">515</span> throttler.setBandwidth((double) currentBandwidth / 10.0);<a name="line.515"></a>
<span class="sourceLineNo">516</span> LOG.info("ReplicationSource : {} bandwidth throttling changed, currentBandWidth={}",<a name="line.516"></a>
<span class="sourceLineNo">517</span> replicationPeer.getId(), currentBandwidth);<a name="line.517"></a>
<span class="sourceLineNo">518</span> }<a name="line.518"></a>
<span class="sourceLineNo">519</span> }<a name="line.519"></a>
<span class="sourceLineNo">520</span><a name="line.520"></a>
<span class="sourceLineNo">521</span> private long getCurrentBandwidth() {<a name="line.521"></a>
<span class="sourceLineNo">522</span> long peerBandwidth = replicationPeer.getPeerBandwidth();<a name="line.522"></a>
<span class="sourceLineNo">523</span> // user can set peer bandwidth to 0 to use default bandwidth<a name="line.523"></a>
<span class="sourceLineNo">524</span> return peerBandwidth != 0 ? peerBandwidth : defaultBandwidth;<a name="line.524"></a>
<span class="sourceLineNo">525</span> }<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> * Do the sleeping logic<a name="line.528"></a>
<span class="sourceLineNo">529</span> * @param msg Why we sleep<a name="line.529"></a>
<span class="sourceLineNo">530</span> * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.530"></a>
<span class="sourceLineNo">531</span> * @return True if &lt;code&gt;sleepMultiplier&lt;/code&gt; is &amp;lt; &lt;code&gt;maxRetriesMultiplier&lt;/code&gt;<a name="line.531"></a>
<span class="sourceLineNo">532</span> */<a name="line.532"></a>
<span class="sourceLineNo">533</span> protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.533"></a>
<span class="sourceLineNo">534</span> try {<a name="line.534"></a>
<span class="sourceLineNo">535</span> if (LOG.isTraceEnabled()) {<a name="line.535"></a>
<span class="sourceLineNo">536</span> LOG.trace("{} {}, sleeping {} times {}",<a name="line.536"></a>
<span class="sourceLineNo">537</span> logPeerId(), msg, sleepForRetries, sleepMultiplier);<a name="line.537"></a>
<span class="sourceLineNo">538</span> }<a name="line.538"></a>
<span class="sourceLineNo">539</span> Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.539"></a>
<span class="sourceLineNo">540</span> } catch (InterruptedException e) {<a name="line.540"></a>
<span class="sourceLineNo">541</span> if(LOG.isDebugEnabled()) {<a name="line.541"></a>
<span class="sourceLineNo">542</span> LOG.debug("{} Interrupted while sleeping between retries", logPeerId());<a name="line.542"></a>
<span class="sourceLineNo">543</span> }<a name="line.543"></a>
<span class="sourceLineNo">544</span> Thread.currentThread().interrupt();<a name="line.544"></a>
<span class="sourceLineNo">545</span> }<a name="line.545"></a>
<span class="sourceLineNo">546</span> return sleepMultiplier &lt; maxRetriesMultiplier;<a name="line.546"></a>
<span class="sourceLineNo">547</span> }<a name="line.547"></a>
<span class="sourceLineNo">548</span><a name="line.548"></a>
<span class="sourceLineNo">549</span> private void initialize() {<a name="line.549"></a>
<span class="sourceLineNo">550</span> int sleepMultiplier = 1;<a name="line.550"></a>
<span class="sourceLineNo">551</span> while (this.isSourceActive()) {<a name="line.551"></a>
<span class="sourceLineNo">552</span> ReplicationEndpoint replicationEndpoint;<a name="line.552"></a>
<span class="sourceLineNo">553</span> try {<a name="line.553"></a>
<span class="sourceLineNo">554</span> replicationEndpoint = createReplicationEndpoint();<a name="line.554"></a>
<span class="sourceLineNo">555</span> } catch (Exception e) {<a name="line.555"></a>
<span class="sourceLineNo">556</span> LOG.warn("{} error creating ReplicationEndpoint, retry", logPeerId(), e);<a name="line.556"></a>
<span class="sourceLineNo">557</span> if (sleepForRetries("Error creating ReplicationEndpoint", sleepMultiplier)) {<a name="line.557"></a>
<span class="sourceLineNo">558</span> sleepMultiplier++;<a name="line.558"></a>
<span class="sourceLineNo">559</span> }<a name="line.559"></a>
<span class="sourceLineNo">560</span> continue;<a name="line.560"></a>
<span class="sourceLineNo">561</span> }<a name="line.561"></a>
<span class="sourceLineNo">562</span><a name="line.562"></a>
<span class="sourceLineNo">563</span> try {<a name="line.563"></a>
<span class="sourceLineNo">564</span> initAndStartReplicationEndpoint(replicationEndpoint);<a name="line.564"></a>
<span class="sourceLineNo">565</span> this.replicationEndpoint = replicationEndpoint;<a name="line.565"></a>
<span class="sourceLineNo">566</span> break;<a name="line.566"></a>
<span class="sourceLineNo">567</span> } catch (Exception e) {<a name="line.567"></a>
<span class="sourceLineNo">568</span> LOG.warn("{} Error starting ReplicationEndpoint, retry", logPeerId(), e);<a name="line.568"></a>
<span class="sourceLineNo">569</span> replicationEndpoint.stop();<a name="line.569"></a>
<span class="sourceLineNo">570</span> if (sleepForRetries("Error starting ReplicationEndpoint", sleepMultiplier)) {<a name="line.570"></a>
<span class="sourceLineNo">571</span> sleepMultiplier++;<a name="line.571"></a>
<span class="sourceLineNo">572</span> } else {<a name="line.572"></a>
<span class="sourceLineNo">573</span> this.startupOngoing.set(false);<a name="line.573"></a>
<span class="sourceLineNo">574</span> throw new RuntimeException("Exhausted retries to start replication endpoint.");<a name="line.574"></a>
<span class="sourceLineNo">575</span> }<a name="line.575"></a>
<span class="sourceLineNo">576</span> }<a name="line.576"></a>
<span class="sourceLineNo">577</span> }<a name="line.577"></a>
<span class="sourceLineNo">578</span><a name="line.578"></a>
<span class="sourceLineNo">579</span> if (!this.isSourceActive()) {<a name="line.579"></a>
<span class="sourceLineNo">580</span> this.startupOngoing.set(false);<a name="line.580"></a>
<span class="sourceLineNo">581</span> throw new IllegalStateException("Source should be active.");<a name="line.581"></a>
<span class="sourceLineNo">582</span> }<a name="line.582"></a>
<span class="sourceLineNo">583</span><a name="line.583"></a>
<span class="sourceLineNo">584</span> sleepMultiplier = 1;<a name="line.584"></a>
<span class="sourceLineNo">585</span> UUID peerClusterId;<a name="line.585"></a>
<span class="sourceLineNo">586</span> // delay this until we are in an asynchronous thread<a name="line.586"></a>
<span class="sourceLineNo">587</span> for (;;) {<a name="line.587"></a>
<span class="sourceLineNo">588</span> peerClusterId = replicationEndpoint.getPeerUUID();<a name="line.588"></a>
<span class="sourceLineNo">589</span> if (this.isSourceActive() &amp;&amp; peerClusterId == null) {<a name="line.589"></a>
<span class="sourceLineNo">590</span> if(LOG.isDebugEnabled()) {<a name="line.590"></a>
<span class="sourceLineNo">591</span> LOG.debug("{} Could not connect to Peer ZK. Sleeping for {} millis", logPeerId(),<a name="line.591"></a>
<span class="sourceLineNo">592</span> (this.sleepForRetries * sleepMultiplier));<a name="line.592"></a>
<span class="sourceLineNo">593</span> }<a name="line.593"></a>
<span class="sourceLineNo">594</span> if (sleepForRetries("Cannot contact the peer's zk ensemble", sleepMultiplier)) {<a name="line.594"></a>
<span class="sourceLineNo">595</span> sleepMultiplier++;<a name="line.595"></a>
<span class="sourceLineNo">596</span> }<a name="line.596"></a>
<span class="sourceLineNo">597</span> } else {<a name="line.597"></a>
<span class="sourceLineNo">598</span> break;<a name="line.598"></a>
<span class="sourceLineNo">599</span> }<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> if(!this.isSourceActive()) {<a name="line.602"></a>
<span class="sourceLineNo">603</span> this.startupOngoing.set(false);<a name="line.603"></a>
<span class="sourceLineNo">604</span> throw new IllegalStateException("Source should be active.");<a name="line.604"></a>
<span class="sourceLineNo">605</span> }<a name="line.605"></a>
<span class="sourceLineNo">606</span> LOG.info("{} queueId={} is replicating from cluster={} to cluster={}",<a name="line.606"></a>
<span class="sourceLineNo">607</span> logPeerId(), this.replicationQueueInfo.getQueueId(), clusterId, peerClusterId);<a name="line.607"></a>
<span class="sourceLineNo">608</span><a name="line.608"></a>
<span class="sourceLineNo">609</span> initializeWALEntryFilter(peerClusterId);<a name="line.609"></a>
<span class="sourceLineNo">610</span> // start workers<a name="line.610"></a>
<span class="sourceLineNo">611</span> for (Map.Entry&lt;String, PriorityBlockingQueue&lt;Path&gt;&gt; entry : queues.entrySet()) {<a name="line.611"></a>
<span class="sourceLineNo">612</span> String walGroupId = entry.getKey();<a name="line.612"></a>
<span class="sourceLineNo">613</span> PriorityBlockingQueue&lt;Path&gt; queue = entry.getValue();<a name="line.613"></a>
<span class="sourceLineNo">614</span> tryStartNewShipper(walGroupId, queue);<a name="line.614"></a>
<span class="sourceLineNo">615</span> }<a name="line.615"></a>
<span class="sourceLineNo">616</span> this.startupOngoing.set(false);<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> @Override<a name="line.619"></a>
<span class="sourceLineNo">620</span> public void startup() {<a name="line.620"></a>
<span class="sourceLineNo">621</span> if (this.sourceRunning) {<a name="line.621"></a>
<span class="sourceLineNo">622</span> return;<a name="line.622"></a>
<span class="sourceLineNo">623</span> }<a name="line.623"></a>
<span class="sourceLineNo">624</span> this.sourceRunning = true;<a name="line.624"></a>
<span class="sourceLineNo">625</span> //Flag that signalizes uncaught error happening while starting up the source<a name="line.625"></a>
<span class="sourceLineNo">626</span> // and a retry should be attempted<a name="line.626"></a>
<span class="sourceLineNo">627</span> MutableBoolean retryStartup = new MutableBoolean(true);<a name="line.627"></a>
<span class="sourceLineNo">628</span> do {<a name="line.628"></a>
<span class="sourceLineNo">629</span> if(retryStartup.booleanValue()) {<a name="line.629"></a>
<span class="sourceLineNo">630</span> retryStartup.setValue(false);<a name="line.630"></a>
<span class="sourceLineNo">631</span> startupOngoing.set(true);<a name="line.631"></a>
<span class="sourceLineNo">632</span> // mark we are running now<a name="line.632"></a>
<span class="sourceLineNo">633</span> initThread = new Thread(this::initialize);<a name="line.633"></a>
<span class="sourceLineNo">634</span> Threads.setDaemonThreadRunning(initThread,<a name="line.634"></a>
<span class="sourceLineNo">635</span> Thread.currentThread().getName() + ".replicationSource," + this.queueId,<a name="line.635"></a>
<span class="sourceLineNo">636</span> (t,e) -&gt; {<a name="line.636"></a>
<span class="sourceLineNo">637</span> sourceRunning = false;<a name="line.637"></a>
<span class="sourceLineNo">638</span> uncaughtException(t, e, null, null);<a name="line.638"></a>
<span class="sourceLineNo">639</span> retryStartup.setValue(!this.abortOnError);<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> } while (this.startupOngoing.get() &amp;&amp; !this.abortOnError);<a name="line.642"></a>
<span class="sourceLineNo">643</span> }<a name="line.643"></a>
<span class="sourceLineNo">644</span><a name="line.644"></a>
<span class="sourceLineNo">645</span> @Override<a name="line.645"></a>
<span class="sourceLineNo">646</span> public void terminate(String reason) {<a name="line.646"></a>
<span class="sourceLineNo">647</span> terminate(reason, null);<a name="line.647"></a>
<span class="sourceLineNo">648</span> }<a name="line.648"></a>
<span class="sourceLineNo">649</span><a name="line.649"></a>
<span class="sourceLineNo">650</span> @Override<a name="line.650"></a>
<span class="sourceLineNo">651</span> public void terminate(String reason, Exception cause) {<a name="line.651"></a>
<span class="sourceLineNo">652</span> terminate(reason, cause, true);<a name="line.652"></a>
<span class="sourceLineNo">653</span> }<a name="line.653"></a>
<span class="sourceLineNo">654</span><a name="line.654"></a>
<span class="sourceLineNo">655</span> @Override<a name="line.655"></a>
<span class="sourceLineNo">656</span> public void terminate(String reason, Exception cause, boolean clearMetrics) {<a name="line.656"></a>
<span class="sourceLineNo">657</span> terminate(reason, cause, clearMetrics, true);<a name="line.657"></a>
<span class="sourceLineNo">658</span> }<a name="line.658"></a>
<span class="sourceLineNo">659</span><a name="line.659"></a>
<span class="sourceLineNo">660</span> public void terminate(String reason, Exception cause, boolean clearMetrics,<a name="line.660"></a>
<span class="sourceLineNo">661</span> boolean join) {<a name="line.661"></a>
<span class="sourceLineNo">662</span> if (cause == null) {<a name="line.662"></a>
<span class="sourceLineNo">663</span> LOG.info("{} Closing source {} because: {}", logPeerId(), this.queueId, reason);<a name="line.663"></a>
<span class="sourceLineNo">664</span> } else {<a name="line.664"></a>
<span class="sourceLineNo">665</span> LOG.error("{} Closing source {} because an error occurred: {}",<a name="line.665"></a>
<span class="sourceLineNo">666</span> logPeerId(), this.queueId, reason, cause);<a name="line.666"></a>
<span class="sourceLineNo">667</span> }<a name="line.667"></a>
<span class="sourceLineNo">668</span> this.sourceRunning = false;<a name="line.668"></a>
<span class="sourceLineNo">669</span> if (initThread != null &amp;&amp; Thread.currentThread() != initThread) {<a name="line.669"></a>
<span class="sourceLineNo">670</span> // This usually won't happen but anyway, let's wait until the initialization thread exits.<a name="line.670"></a>
<span class="sourceLineNo">671</span> // And notice that we may call terminate directly from the initThread so here we need to<a name="line.671"></a>
<span class="sourceLineNo">672</span> // avoid join on ourselves.<a name="line.672"></a>
<span class="sourceLineNo">673</span> initThread.interrupt();<a name="line.673"></a>
<span class="sourceLineNo">674</span> Threads.shutdown(initThread, this.sleepForRetries);<a name="line.674"></a>
<span class="sourceLineNo">675</span> }<a name="line.675"></a>
<span class="sourceLineNo">676</span> Collection&lt;ReplicationSourceShipper&gt; workers = workerThreads.values();<a name="line.676"></a>
<span class="sourceLineNo">677</span><a name="line.677"></a>
<span class="sourceLineNo">678</span> for (ReplicationSourceShipper worker : workers) {<a name="line.678"></a>
<span class="sourceLineNo">679</span> worker.stopWorker();<a name="line.679"></a>
<span class="sourceLineNo">680</span> if (worker.entryReader != null) {<a name="line.680"></a>
<span class="sourceLineNo">681</span> worker.entryReader.setReaderRunning(false);<a name="line.681"></a>
<span class="sourceLineNo">682</span> }<a name="line.682"></a>
<span class="sourceLineNo">683</span> if (worker.isAlive() || worker.entryReader.isAlive()) {<a name="line.683"></a>
<span class="sourceLineNo">684</span> try {<a name="line.684"></a>
<span class="sourceLineNo">685</span> // Wait worker to stop<a name="line.685"></a>
<span class="sourceLineNo">686</span> Thread.sleep(this.sleepForRetries);<a name="line.686"></a>
<span class="sourceLineNo">687</span> } catch (InterruptedException e) {<a name="line.687"></a>
<span class="sourceLineNo">688</span> LOG.info("{} Interrupted while waiting {} to stop", logPeerId(), worker.getName());<a name="line.688"></a>
<span class="sourceLineNo">689</span> Thread.currentThread().interrupt();<a name="line.689"></a>
<span class="sourceLineNo">690</span> }<a name="line.690"></a>
<span class="sourceLineNo">691</span> // If worker still is alive after waiting, interrupt it<a name="line.691"></a>
<span class="sourceLineNo">692</span> if (worker.isAlive()) {<a name="line.692"></a>
<span class="sourceLineNo">693</span> worker.interrupt();<a name="line.693"></a>
<span class="sourceLineNo">694</span> }<a name="line.694"></a>
<span class="sourceLineNo">695</span> // If entry reader is alive after waiting, interrupt it<a name="line.695"></a>
<span class="sourceLineNo">696</span> if (worker.entryReader.isAlive()) {<a name="line.696"></a>
<span class="sourceLineNo">697</span> worker.entryReader.interrupt();<a name="line.697"></a>
<span class="sourceLineNo">698</span> }<a name="line.698"></a>
<span class="sourceLineNo">699</span> }<a name="line.699"></a>
<span class="sourceLineNo">700</span> //If worker is already stopped but there was still entries batched,<a name="line.700"></a>
<span class="sourceLineNo">701</span> //we need to clear buffer used for non processed entries<a name="line.701"></a>
<span class="sourceLineNo">702</span> worker.clearWALEntryBatch();<a name="line.702"></a>
<span class="sourceLineNo">703</span> }<a name="line.703"></a>
<span class="sourceLineNo">704</span><a name="line.704"></a>
<span class="sourceLineNo">705</span> if (this.replicationEndpoint != null) {<a name="line.705"></a>
<span class="sourceLineNo">706</span> this.replicationEndpoint.stop();<a name="line.706"></a>
<span class="sourceLineNo">707</span> }<a name="line.707"></a>
<span class="sourceLineNo">708</span> if (join) {<a name="line.708"></a>
<span class="sourceLineNo">709</span> for (ReplicationSourceShipper worker : workers) {<a name="line.709"></a>
<span class="sourceLineNo">710</span> Threads.shutdown(worker, this.sleepForRetries);<a name="line.710"></a>
<span class="sourceLineNo">711</span> LOG.info("{} ReplicationSourceWorker {} terminated", logPeerId(), worker.getName());<a name="line.711"></a>
<span class="sourceLineNo">712</span> }<a name="line.712"></a>
<span class="sourceLineNo">713</span> if (this.replicationEndpoint != null) {<a name="line.713"></a>
<span class="sourceLineNo">714</span> try {<a name="line.714"></a>
<span class="sourceLineNo">715</span> this.replicationEndpoint.awaitTerminated(sleepForRetries * maxRetriesMultiplier,<a name="line.715"></a>
<span class="sourceLineNo">716</span> TimeUnit.MILLISECONDS);<a name="line.716"></a>
<span class="sourceLineNo">717</span> } catch (TimeoutException te) {<a name="line.717"></a>
<span class="sourceLineNo">718</span> LOG.warn("{} Got exception while waiting for endpoint to shutdown "<a name="line.718"></a>
<span class="sourceLineNo">719</span> + "for replication source : {}", logPeerId(), this.queueId, te);<a name="line.719"></a>
<span class="sourceLineNo">720</span> }<a name="line.720"></a>
<span class="sourceLineNo">721</span> }<a name="line.721"></a>
<span class="sourceLineNo">722</span> }<a name="line.722"></a>
<span class="sourceLineNo">723</span> if (clearMetrics) {<a name="line.723"></a>
<span class="sourceLineNo">724</span> // Can be null in test context.<a name="line.724"></a>
<span class="sourceLineNo">725</span> if (this.metrics != null) {<a name="line.725"></a>
<span class="sourceLineNo">726</span> this.metrics.clear();<a name="line.726"></a>
<span class="sourceLineNo">727</span> }<a name="line.727"></a>
<span class="sourceLineNo">728</span> }<a name="line.728"></a>
<span class="sourceLineNo">729</span> }<a name="line.729"></a>
<span class="sourceLineNo">730</span><a name="line.730"></a>
<span class="sourceLineNo">731</span> @Override<a name="line.731"></a>
<span class="sourceLineNo">732</span> public String getQueueId() {<a name="line.732"></a>
<span class="sourceLineNo">733</span> return this.queueId;<a name="line.733"></a>
<span class="sourceLineNo">734</span> }<a name="line.734"></a>
<span class="sourceLineNo">735</span><a name="line.735"></a>
<span class="sourceLineNo">736</span> @Override<a name="line.736"></a>
<span class="sourceLineNo">737</span> public Path getCurrentPath() {<a name="line.737"></a>
<span class="sourceLineNo">738</span> // only for testing<a name="line.738"></a>
<span class="sourceLineNo">739</span> for (ReplicationSourceShipper worker : workerThreads.values()) {<a name="line.739"></a>
<span class="sourceLineNo">740</span> if (worker.getCurrentPath() != null) {<a name="line.740"></a>
<span class="sourceLineNo">741</span> return worker.getCurrentPath();<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> return null;<a name="line.744"></a>
<span class="sourceLineNo">745</span> }<a name="line.745"></a>
<span class="sourceLineNo">746</span><a name="line.746"></a>
<span class="sourceLineNo">747</span> @Override<a name="line.747"></a>
<span class="sourceLineNo">748</span> public boolean isSourceActive() {<a name="line.748"></a>
<span class="sourceLineNo">749</span> return !this.server.isStopped() &amp;&amp; this.sourceRunning;<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> /**<a name="line.752"></a>
<span class="sourceLineNo">753</span> * Comparator used to compare logs together based on their start time<a name="line.753"></a>
<span class="sourceLineNo">754</span> */<a name="line.754"></a>
<span class="sourceLineNo">755</span> public static class LogsComparator implements Comparator&lt;Path&gt; {<a name="line.755"></a>
<span class="sourceLineNo">756</span><a name="line.756"></a>
<span class="sourceLineNo">757</span> @Override<a name="line.757"></a>
<span class="sourceLineNo">758</span> public int compare(Path o1, Path o2) {<a name="line.758"></a>
<span class="sourceLineNo">759</span> return Long.compare(getTS(o1), getTS(o2));<a name="line.759"></a>
<span class="sourceLineNo">760</span> }<a name="line.760"></a>
<span class="sourceLineNo">761</span><a name="line.761"></a>
<span class="sourceLineNo">762</span> /**<a name="line.762"></a>
<span class="sourceLineNo">763</span> * &lt;p&gt;<a name="line.763"></a>
<span class="sourceLineNo">764</span> * Split a path to get the start time<a name="line.764"></a>
<span class="sourceLineNo">765</span> * &lt;/p&gt;<a name="line.765"></a>
<span class="sourceLineNo">766</span> * &lt;p&gt;<a name="line.766"></a>
<span class="sourceLineNo">767</span> * For example: 10.20.20.171%3A60020.1277499063250<a name="line.767"></a>
<span class="sourceLineNo">768</span> * &lt;/p&gt;<a name="line.768"></a>
<span class="sourceLineNo">769</span> * @param p path to split<a name="line.769"></a>
<span class="sourceLineNo">770</span> * @return start time<a name="line.770"></a>
<span class="sourceLineNo">771</span> */<a name="line.771"></a>
<span class="sourceLineNo">772</span> private static long getTS(Path p) {<a name="line.772"></a>
<span class="sourceLineNo">773</span> return AbstractFSWALProvider.getWALStartTimeFromWALName(p.getName());<a name="line.773"></a>
<span class="sourceLineNo">774</span> }<a name="line.774"></a>
<span class="sourceLineNo">775</span> }<a name="line.775"></a>
<span class="sourceLineNo">776</span><a name="line.776"></a>
<span class="sourceLineNo">777</span> public ReplicationQueueInfo getReplicationQueueInfo() {<a name="line.777"></a>
<span class="sourceLineNo">778</span> return replicationQueueInfo;<a name="line.778"></a>
<span class="sourceLineNo">779</span> }<a name="line.779"></a>
<span class="sourceLineNo">780</span><a name="line.780"></a>
<span class="sourceLineNo">781</span> public boolean isWorkerRunning(){<a name="line.781"></a>
<span class="sourceLineNo">782</span> for(ReplicationSourceShipper worker : this.workerThreads.values()){<a name="line.782"></a>
<span class="sourceLineNo">783</span> if(worker.isActive()){<a name="line.783"></a>
<span class="sourceLineNo">784</span> return worker.isActive();<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> return false;<a name="line.787"></a>
<span class="sourceLineNo">788</span> }<a name="line.788"></a>
<span class="sourceLineNo">789</span><a name="line.789"></a>
<span class="sourceLineNo">790</span> @Override<a name="line.790"></a>
<span class="sourceLineNo">791</span> public String getStats() {<a name="line.791"></a>
<span class="sourceLineNo">792</span> StringBuilder sb = new StringBuilder();<a name="line.792"></a>
<span class="sourceLineNo">793</span> sb.append("Total replicated edits: ").append(totalReplicatedEdits)<a name="line.793"></a>
<span class="sourceLineNo">794</span> .append(", current progress: \n");<a name="line.794"></a>
<span class="sourceLineNo">795</span> for (Map.Entry&lt;String, ReplicationSourceShipper&gt; entry : workerThreads.entrySet()) {<a name="line.795"></a>
<span class="sourceLineNo">796</span> String walGroupId = entry.getKey();<a name="line.796"></a>
<span class="sourceLineNo">797</span> ReplicationSourceShipper worker = entry.getValue();<a name="line.797"></a>
<span class="sourceLineNo">798</span> long position = worker.getCurrentPosition();<a name="line.798"></a>
<span class="sourceLineNo">799</span> Path currentPath = worker.getCurrentPath();<a name="line.799"></a>
<span class="sourceLineNo">800</span> sb.append("walGroup [").append(walGroupId).append("]: ");<a name="line.800"></a>
<span class="sourceLineNo">801</span> if (currentPath != null) {<a name="line.801"></a>
<span class="sourceLineNo">802</span> sb.append("currently replicating from: ").append(currentPath).append(" at position: ")<a name="line.802"></a>
<span class="sourceLineNo">803</span> .append(position).append("\n");<a name="line.803"></a>
<span class="sourceLineNo">804</span> } else {<a name="line.804"></a>
<span class="sourceLineNo">805</span> sb.append("no replication ongoing, waiting for new log");<a name="line.805"></a>
<span class="sourceLineNo">806</span> }<a name="line.806"></a>
<span class="sourceLineNo">807</span> }<a name="line.807"></a>
<span class="sourceLineNo">808</span> return sb.toString();<a name="line.808"></a>
<span class="sourceLineNo">809</span> }<a name="line.809"></a>
<span class="sourceLineNo">810</span><a name="line.810"></a>
<span class="sourceLineNo">811</span> @Override<a name="line.811"></a>
<span class="sourceLineNo">812</span> public MetricsSource getSourceMetrics() {<a name="line.812"></a>
<span class="sourceLineNo">813</span> return this.metrics;<a name="line.813"></a>
<span class="sourceLineNo">814</span> }<a name="line.814"></a>
<span class="sourceLineNo">815</span><a name="line.815"></a>
<span class="sourceLineNo">816</span> @Override<a name="line.816"></a>
<span class="sourceLineNo">817</span> //offsets totalBufferUsed by deducting shipped batchSize.<a name="line.817"></a>
<span class="sourceLineNo">818</span> public void postShipEdits(List&lt;Entry&gt; entries, int batchSize) {<a name="line.818"></a>
<span class="sourceLineNo">819</span> if (throttler.isEnabled()) {<a name="line.819"></a>
<span class="sourceLineNo">820</span> throttler.addPushSize(batchSize);<a name="line.820"></a>
<span class="sourceLineNo">821</span> }<a name="line.821"></a>
<span class="sourceLineNo">822</span> totalReplicatedEdits.addAndGet(entries.size());<a name="line.822"></a>
<span class="sourceLineNo">823</span> long newBufferUsed = totalBufferUsed.addAndGet(-batchSize);<a name="line.823"></a>
<span class="sourceLineNo">824</span> // Record the new buffer usage<a name="line.824"></a>
<span class="sourceLineNo">825</span> this.manager.getGlobalMetrics().setWALReaderEditsBufferBytes(newBufferUsed);<a name="line.825"></a>
<span class="sourceLineNo">826</span> }<a name="line.826"></a>
<span class="sourceLineNo">827</span><a name="line.827"></a>
<span class="sourceLineNo">828</span> @Override<a name="line.828"></a>
<span class="sourceLineNo">829</span> public WALFileLengthProvider getWALFileLengthProvider() {<a name="line.829"></a>
<span class="sourceLineNo">830</span> return walFileLengthProvider;<a name="line.830"></a>
<span class="sourceLineNo">831</span> }<a name="line.831"></a>
<span class="sourceLineNo">832</span><a name="line.832"></a>
<span class="sourceLineNo">833</span> @Override<a name="line.833"></a>
<span class="sourceLineNo">834</span> public ServerName getServerWALsBelongTo() {<a name="line.834"></a>
<span class="sourceLineNo">835</span> return server.getServerName();<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> @Override<a name="line.838"></a>
<span class="sourceLineNo">839</span> public ReplicationPeer getPeer() {<a name="line.839"></a>
<span class="sourceLineNo">840</span> return replicationPeer;<a name="line.840"></a>
<span class="sourceLineNo">841</span> }<a name="line.841"></a>
<span class="sourceLineNo">842</span><a name="line.842"></a>
<span class="sourceLineNo">843</span> Server getServer() {<a name="line.843"></a>
<span class="sourceLineNo">844</span> return server;<a name="line.844"></a>
<span class="sourceLineNo">845</span> }<a name="line.845"></a>
<span class="sourceLineNo">846</span><a name="line.846"></a>
<span class="sourceLineNo">847</span> ReplicationQueueStorage getQueueStorage() {<a name="line.847"></a>
<span class="sourceLineNo">848</span> return queueStorage;<a name="line.848"></a>
<span class="sourceLineNo">849</span> }<a name="line.849"></a>
<span class="sourceLineNo">850</span><a name="line.850"></a>
<span class="sourceLineNo">851</span> void removeWorker(ReplicationSourceShipper worker) {<a name="line.851"></a>
<span class="sourceLineNo">852</span> workerThreads.remove(worker.walGroupId, worker);<a name="line.852"></a>
<span class="sourceLineNo">853</span> }<a name="line.853"></a>
<span class="sourceLineNo">854</span><a name="line.854"></a>
<span class="sourceLineNo">855</span> private String logPeerId(){<a name="line.855"></a>
<span class="sourceLineNo">856</span> return "peerId=" + this.getPeerId() + ",";<a name="line.856"></a>
<span class="sourceLineNo">857</span> }<a name="line.857"></a>
<span class="sourceLineNo">858</span>}<a name="line.858"></a>
</pre>
</div>
</body>
</html>