blob: 4ea983efb05f6a71767292749992f6800635e87e [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><a name="line.18"></a>
<span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.replication.regionserver;<a name="line.19"></a>
<span class="sourceLineNo">020</span><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.net.ConnectException;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.net.SocketTimeoutException;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.net.UnknownHostException;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.ArrayList;<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.HashMap;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import java.util.HashSet;<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.concurrent.Callable;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import java.util.concurrent.CompletionService;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import java.util.concurrent.ExecutionException;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import java.util.concurrent.ExecutorCompletionService;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import java.util.concurrent.Future;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import java.util.concurrent.TimeUnit;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import java.util.stream.Collectors;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import java.util.stream.Stream;<a name="line.41"></a>
<span class="sourceLineNo">042</span><a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.commons.lang3.StringUtils;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.fs.Path;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.Abortable;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.CellUtil;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.HBaseConfiguration;<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.TableName;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.client.Admin;<a name="line.51"></a>
<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.client.AsyncClusterConnection;<a name="line.52"></a>
<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.client.AsyncRegionServerAdmin;<a name="line.53"></a>
<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.client.ClusterConnectionFactory;<a name="line.54"></a>
<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.55"></a>
<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.56"></a>
<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.ipc.RpcServer;<a name="line.57"></a>
<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtobufUtil;<a name="line.58"></a>
<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.regionserver.NoSuchColumnFamilyException;<a name="line.59"></a>
<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.regionserver.wal.WALUtil;<a name="line.60"></a>
<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.61"></a>
<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.security.User;<a name="line.62"></a>
<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.63"></a>
<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.util.CommonFSUtils;<a name="line.64"></a>
<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.65"></a>
<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.66"></a>
<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.wal.WALEdit;<a name="line.67"></a>
<span class="sourceLineNo">068</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.68"></a>
<span class="sourceLineNo">069</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.69"></a>
<span class="sourceLineNo">070</span>import org.slf4j.Logger;<a name="line.70"></a>
<span class="sourceLineNo">071</span>import org.slf4j.LoggerFactory;<a name="line.71"></a>
<span class="sourceLineNo">072</span><a name="line.72"></a>
<span class="sourceLineNo">073</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<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> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint}<a name="line.77"></a>
<span class="sourceLineNo">078</span> * implementation for replicating to another HBase cluster.<a name="line.78"></a>
<span class="sourceLineNo">079</span> * For the slave cluster it selects a random number of peers<a name="line.79"></a>
<span class="sourceLineNo">080</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.80"></a>
<span class="sourceLineNo">081</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.81"></a>
<span class="sourceLineNo">082</span> * &lt;p&gt;<a name="line.82"></a>
<span class="sourceLineNo">083</span> * A stream is considered down when we cannot contact a region server on the<a name="line.83"></a>
<span class="sourceLineNo">084</span> * peer cluster for more than 55 seconds by default.<a name="line.84"></a>
<span class="sourceLineNo">085</span> * &lt;/p&gt;<a name="line.85"></a>
<span class="sourceLineNo">086</span> */<a name="line.86"></a>
<span class="sourceLineNo">087</span>@InterfaceAudience.Private<a name="line.87"></a>
<span class="sourceLineNo">088</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.88"></a>
<span class="sourceLineNo">089</span> private static final Logger LOG =<a name="line.89"></a>
<span class="sourceLineNo">090</span> LoggerFactory.getLogger(HBaseInterClusterReplicationEndpoint.class);<a name="line.90"></a>
<span class="sourceLineNo">091</span><a name="line.91"></a>
<span class="sourceLineNo">092</span> private static final long DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER = 2;<a name="line.92"></a>
<span class="sourceLineNo">093</span><a name="line.93"></a>
<span class="sourceLineNo">094</span> /** Drop edits for tables that been deleted from the replication source and target */<a name="line.94"></a>
<span class="sourceLineNo">095</span> public static final String REPLICATION_DROP_ON_DELETED_TABLE_KEY =<a name="line.95"></a>
<span class="sourceLineNo">096</span> "hbase.replication.drop.on.deleted.table";<a name="line.96"></a>
<span class="sourceLineNo">097</span> /** Drop edits for CFs that been deleted from the replication source and target */<a name="line.97"></a>
<span class="sourceLineNo">098</span> public static final String REPLICATION_DROP_ON_DELETED_COLUMN_FAMILY_KEY =<a name="line.98"></a>
<span class="sourceLineNo">099</span> "hbase.replication.drop.on.deleted.columnfamily";<a name="line.99"></a>
<span class="sourceLineNo">100</span><a name="line.100"></a>
<span class="sourceLineNo">101</span> // How long should we sleep for each retry<a name="line.101"></a>
<span class="sourceLineNo">102</span> private long sleepForRetries;<a name="line.102"></a>
<span class="sourceLineNo">103</span> // Maximum number of retries before taking bold actions<a name="line.103"></a>
<span class="sourceLineNo">104</span> private int maxRetriesMultiplier;<a name="line.104"></a>
<span class="sourceLineNo">105</span> // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.105"></a>
<span class="sourceLineNo">106</span> private int socketTimeoutMultiplier;<a name="line.106"></a>
<span class="sourceLineNo">107</span> // Amount of time for shutdown to wait for all tasks to complete<a name="line.107"></a>
<span class="sourceLineNo">108</span> private long maxTerminationWait;<a name="line.108"></a>
<span class="sourceLineNo">109</span> // Size limit for replication RPCs, in bytes<a name="line.109"></a>
<span class="sourceLineNo">110</span> private int replicationRpcLimit;<a name="line.110"></a>
<span class="sourceLineNo">111</span> //Metrics for this source<a name="line.111"></a>
<span class="sourceLineNo">112</span> private MetricsSource metrics;<a name="line.112"></a>
<span class="sourceLineNo">113</span> private boolean peersSelected = false;<a name="line.113"></a>
<span class="sourceLineNo">114</span> private String replicationClusterId = "";<a name="line.114"></a>
<span class="sourceLineNo">115</span> private ThreadPoolExecutor exec;<a name="line.115"></a>
<span class="sourceLineNo">116</span> private int maxThreads;<a name="line.116"></a>
<span class="sourceLineNo">117</span> private Path baseNamespaceDir;<a name="line.117"></a>
<span class="sourceLineNo">118</span> private Path hfileArchiveDir;<a name="line.118"></a>
<span class="sourceLineNo">119</span> private boolean replicationBulkLoadDataEnabled;<a name="line.119"></a>
<span class="sourceLineNo">120</span> private Abortable abortable;<a name="line.120"></a>
<span class="sourceLineNo">121</span> private boolean dropOnDeletedTables;<a name="line.121"></a>
<span class="sourceLineNo">122</span> private boolean dropOnDeletedColumnFamilies;<a name="line.122"></a>
<span class="sourceLineNo">123</span> private boolean isSerial = false;<a name="line.123"></a>
<span class="sourceLineNo">124</span> //Initialising as 0 to guarantee at least one logging message<a name="line.124"></a>
<span class="sourceLineNo">125</span> private long lastSinkFetchTime = 0;<a name="line.125"></a>
<span class="sourceLineNo">126</span><a name="line.126"></a>
<span class="sourceLineNo">127</span> @Override<a name="line.127"></a>
<span class="sourceLineNo">128</span> public void init(Context context) throws IOException {<a name="line.128"></a>
<span class="sourceLineNo">129</span> super.init(context);<a name="line.129"></a>
<span class="sourceLineNo">130</span> this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.130"></a>
<span class="sourceLineNo">131</span> decorateConf();<a name="line.131"></a>
<span class="sourceLineNo">132</span> this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.132"></a>
<span class="sourceLineNo">133</span> this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.133"></a>
<span class="sourceLineNo">134</span> maxRetriesMultiplier);<a name="line.134"></a>
<span class="sourceLineNo">135</span> // A Replicator job is bound by the RPC timeout. We will wait this long for all Replicator<a name="line.135"></a>
<span class="sourceLineNo">136</span> // tasks to terminate when doStop() is called.<a name="line.136"></a>
<span class="sourceLineNo">137</span> long maxTerminationWaitMultiplier = this.conf.getLong(<a name="line.137"></a>
<span class="sourceLineNo">138</span> "replication.source.maxterminationmultiplier",<a name="line.138"></a>
<span class="sourceLineNo">139</span> DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER);<a name="line.139"></a>
<span class="sourceLineNo">140</span> this.maxTerminationWait = maxTerminationWaitMultiplier *<a name="line.140"></a>
<span class="sourceLineNo">141</span> this.conf.getLong(HConstants.HBASE_RPC_TIMEOUT_KEY, HConstants.DEFAULT_HBASE_RPC_TIMEOUT);<a name="line.141"></a>
<span class="sourceLineNo">142</span> // TODO: This connection is replication specific or we should make it particular to<a name="line.142"></a>
<span class="sourceLineNo">143</span> // replication and make replication specific settings such as compression or codec to use<a name="line.143"></a>
<span class="sourceLineNo">144</span> // passing Cells.<a name="line.144"></a>
<span class="sourceLineNo">145</span> this.conn = createConnection(this.conf);<a name="line.145"></a>
<span class="sourceLineNo">146</span> this.sleepForRetries =<a name="line.146"></a>
<span class="sourceLineNo">147</span> this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.147"></a>
<span class="sourceLineNo">148</span> this.metrics = context.getMetrics();<a name="line.148"></a>
<span class="sourceLineNo">149</span> // per sink thread pool<a name="line.149"></a>
<span class="sourceLineNo">150</span> this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.150"></a>
<span class="sourceLineNo">151</span> HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.151"></a>
<span class="sourceLineNo">152</span> this.exec = Threads.getBoundedCachedThreadPool(maxThreads, 60, TimeUnit.SECONDS,<a name="line.152"></a>
<span class="sourceLineNo">153</span> new ThreadFactoryBuilder().setDaemon(true).setNameFormat("SinkThread-%d").build());<a name="line.153"></a>
<span class="sourceLineNo">154</span> this.abortable = ctx.getAbortable();<a name="line.154"></a>
<span class="sourceLineNo">155</span> // Set the size limit for replication RPCs to 95% of the max request size.<a name="line.155"></a>
<span class="sourceLineNo">156</span> // We could do with less slop if we have an accurate estimate of encoded size. Being<a name="line.156"></a>
<span class="sourceLineNo">157</span> // conservative for now.<a name="line.157"></a>
<span class="sourceLineNo">158</span> this.replicationRpcLimit = (int)(0.95 * conf.getLong(RpcServer.MAX_REQUEST_SIZE,<a name="line.158"></a>
<span class="sourceLineNo">159</span> RpcServer.DEFAULT_MAX_REQUEST_SIZE));<a name="line.159"></a>
<span class="sourceLineNo">160</span> this.dropOnDeletedTables =<a name="line.160"></a>
<span class="sourceLineNo">161</span> this.conf.getBoolean(REPLICATION_DROP_ON_DELETED_TABLE_KEY, false);<a name="line.161"></a>
<span class="sourceLineNo">162</span> this.dropOnDeletedColumnFamilies = this.conf<a name="line.162"></a>
<span class="sourceLineNo">163</span> .getBoolean(REPLICATION_DROP_ON_DELETED_COLUMN_FAMILY_KEY, false);<a name="line.163"></a>
<span class="sourceLineNo">164</span><a name="line.164"></a>
<span class="sourceLineNo">165</span> this.replicationBulkLoadDataEnabled =<a name="line.165"></a>
<span class="sourceLineNo">166</span> conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.166"></a>
<span class="sourceLineNo">167</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.167"></a>
<span class="sourceLineNo">168</span> if (this.replicationBulkLoadDataEnabled) {<a name="line.168"></a>
<span class="sourceLineNo">169</span> replicationClusterId = this.conf.get(HConstants.REPLICATION_CLUSTER_ID);<a name="line.169"></a>
<span class="sourceLineNo">170</span> }<a name="line.170"></a>
<span class="sourceLineNo">171</span> // Construct base namespace directory and hfile archive directory path<a name="line.171"></a>
<span class="sourceLineNo">172</span> Path rootDir = CommonFSUtils.getRootDir(conf);<a name="line.172"></a>
<span class="sourceLineNo">173</span> Path baseNSDir = new Path(HConstants.BASE_NAMESPACE_DIR);<a name="line.173"></a>
<span class="sourceLineNo">174</span> baseNamespaceDir = new Path(rootDir, baseNSDir);<a name="line.174"></a>
<span class="sourceLineNo">175</span> hfileArchiveDir = new Path(rootDir, new Path(HConstants.HFILE_ARCHIVE_DIRECTORY, baseNSDir));<a name="line.175"></a>
<span class="sourceLineNo">176</span> isSerial = context.getPeerConfig().isSerial();<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> private void decorateConf() {<a name="line.179"></a>
<span class="sourceLineNo">180</span> String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.180"></a>
<span class="sourceLineNo">181</span> if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.181"></a>
<span class="sourceLineNo">182</span> this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.182"></a>
<span class="sourceLineNo">183</span> }<a name="line.183"></a>
<span class="sourceLineNo">184</span> }<a name="line.184"></a>
<span class="sourceLineNo">185</span><a name="line.185"></a>
<span class="sourceLineNo">186</span> private void connectToPeers() {<a name="line.186"></a>
<span class="sourceLineNo">187</span> int sleepMultiplier = 1;<a name="line.187"></a>
<span class="sourceLineNo">188</span> // Connect to peer cluster first, unless we have to stop<a name="line.188"></a>
<span class="sourceLineNo">189</span> while (this.isRunning() &amp;&amp; getNumSinks() == 0) {<a name="line.189"></a>
<span class="sourceLineNo">190</span> chooseSinks();<a name="line.190"></a>
<span class="sourceLineNo">191</span> if (this.isRunning() &amp;&amp; getNumSinks() == 0) {<a name="line.191"></a>
<span class="sourceLineNo">192</span> if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.192"></a>
<span class="sourceLineNo">193</span> sleepMultiplier++;<a name="line.193"></a>
<span class="sourceLineNo">194</span> }<a name="line.194"></a>
<span class="sourceLineNo">195</span> }<a name="line.195"></a>
<span class="sourceLineNo">196</span> }<a name="line.196"></a>
<span class="sourceLineNo">197</span> }<a name="line.197"></a>
<span class="sourceLineNo">198</span><a name="line.198"></a>
<span class="sourceLineNo">199</span> /**<a name="line.199"></a>
<span class="sourceLineNo">200</span> * Do the sleeping logic<a name="line.200"></a>
<span class="sourceLineNo">201</span> * @param msg Why we sleep<a name="line.201"></a>
<span class="sourceLineNo">202</span> * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.202"></a>
<span class="sourceLineNo">203</span> * @return True if &lt;code&gt;sleepMultiplier&lt;/code&gt; is &amp;lt; &lt;code&gt;maxRetriesMultiplier&lt;/code&gt;<a name="line.203"></a>
<span class="sourceLineNo">204</span> */<a name="line.204"></a>
<span class="sourceLineNo">205</span> protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.205"></a>
<span class="sourceLineNo">206</span> try {<a name="line.206"></a>
<span class="sourceLineNo">207</span> if (LOG.isTraceEnabled()) {<a name="line.207"></a>
<span class="sourceLineNo">208</span> LOG.trace("{} {}, sleeping {} times {}",<a name="line.208"></a>
<span class="sourceLineNo">209</span> logPeerId(), msg, sleepForRetries, sleepMultiplier);<a name="line.209"></a>
<span class="sourceLineNo">210</span> }<a name="line.210"></a>
<span class="sourceLineNo">211</span> Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.211"></a>
<span class="sourceLineNo">212</span> } catch (InterruptedException e) {<a name="line.212"></a>
<span class="sourceLineNo">213</span> if (LOG.isDebugEnabled()) {<a name="line.213"></a>
<span class="sourceLineNo">214</span> LOG.debug("{} Interrupted while sleeping between retries", logPeerId());<a name="line.214"></a>
<span class="sourceLineNo">215</span> }<a name="line.215"></a>
<span class="sourceLineNo">216</span> }<a name="line.216"></a>
<span class="sourceLineNo">217</span> return sleepMultiplier &lt; maxRetriesMultiplier;<a name="line.217"></a>
<span class="sourceLineNo">218</span> }<a name="line.218"></a>
<span class="sourceLineNo">219</span><a name="line.219"></a>
<span class="sourceLineNo">220</span> private int getEstimatedEntrySize(Entry e) {<a name="line.220"></a>
<span class="sourceLineNo">221</span> long size = e.getKey().estimatedSerializedSizeOf() + e.getEdit().estimatedSerializedSizeOf();<a name="line.221"></a>
<span class="sourceLineNo">222</span> return (int) size;<a name="line.222"></a>
<span class="sourceLineNo">223</span> }<a name="line.223"></a>
<span class="sourceLineNo">224</span><a name="line.224"></a>
<span class="sourceLineNo">225</span> private List&lt;List&lt;Entry&gt;&gt; createParallelBatches(final List&lt;Entry&gt; entries) {<a name="line.225"></a>
<span class="sourceLineNo">226</span> int numSinks = Math.max(getNumSinks(), 1);<a name="line.226"></a>
<span class="sourceLineNo">227</span> int n = Math.min(Math.min(this.maxThreads, entries.size() / 100 + 1), numSinks);<a name="line.227"></a>
<span class="sourceLineNo">228</span> List&lt;List&lt;Entry&gt;&gt; entryLists =<a name="line.228"></a>
<span class="sourceLineNo">229</span> Stream.generate(ArrayList&lt;Entry&gt;::new).limit(n).collect(Collectors.toList());<a name="line.229"></a>
<span class="sourceLineNo">230</span> int[] sizes = new int[n];<a name="line.230"></a>
<span class="sourceLineNo">231</span> for (Entry e : entries) {<a name="line.231"></a>
<span class="sourceLineNo">232</span> int index = Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName()) % n);<a name="line.232"></a>
<span class="sourceLineNo">233</span> int entrySize = getEstimatedEntrySize(e);<a name="line.233"></a>
<span class="sourceLineNo">234</span> // If this batch has at least one entry and is over sized, move it to the tail of list and<a name="line.234"></a>
<span class="sourceLineNo">235</span> // initialize the entryLists[index] to be a empty list.<a name="line.235"></a>
<span class="sourceLineNo">236</span> if (sizes[index] &gt; 0 &amp;&amp; sizes[index] + entrySize &gt; replicationRpcLimit) {<a name="line.236"></a>
<span class="sourceLineNo">237</span> entryLists.add(entryLists.get(index));<a name="line.237"></a>
<span class="sourceLineNo">238</span> entryLists.set(index, new ArrayList&lt;&gt;());<a name="line.238"></a>
<span class="sourceLineNo">239</span> sizes[index] = 0;<a name="line.239"></a>
<span class="sourceLineNo">240</span> }<a name="line.240"></a>
<span class="sourceLineNo">241</span> entryLists.get(index).add(e);<a name="line.241"></a>
<span class="sourceLineNo">242</span> sizes[index] += entrySize;<a name="line.242"></a>
<span class="sourceLineNo">243</span> }<a name="line.243"></a>
<span class="sourceLineNo">244</span> return entryLists;<a name="line.244"></a>
<span class="sourceLineNo">245</span> }<a name="line.245"></a>
<span class="sourceLineNo">246</span><a name="line.246"></a>
<span class="sourceLineNo">247</span> private List&lt;List&lt;Entry&gt;&gt; createSerialBatches(final List&lt;Entry&gt; entries) {<a name="line.247"></a>
<span class="sourceLineNo">248</span> Map&lt;byte[], List&lt;Entry&gt;&gt; regionEntries = new TreeMap&lt;&gt;(Bytes.BYTES_COMPARATOR);<a name="line.248"></a>
<span class="sourceLineNo">249</span> for (Entry e : entries) {<a name="line.249"></a>
<span class="sourceLineNo">250</span> regionEntries.computeIfAbsent(e.getKey().getEncodedRegionName(), key -&gt; new ArrayList&lt;&gt;())<a name="line.250"></a>
<span class="sourceLineNo">251</span> .add(e);<a name="line.251"></a>
<span class="sourceLineNo">252</span> }<a name="line.252"></a>
<span class="sourceLineNo">253</span> return new ArrayList&lt;&gt;(regionEntries.values());<a name="line.253"></a>
<span class="sourceLineNo">254</span> }<a name="line.254"></a>
<span class="sourceLineNo">255</span><a name="line.255"></a>
<span class="sourceLineNo">256</span> /**<a name="line.256"></a>
<span class="sourceLineNo">257</span> * Divide the entries into multiple batches, so that we can replicate each batch in a thread pool<a name="line.257"></a>
<span class="sourceLineNo">258</span> * concurrently. Note that, for serial replication, we need to make sure that entries from the<a name="line.258"></a>
<span class="sourceLineNo">259</span> * same region to be replicated serially, so entries from the same region consist of a batch, and<a name="line.259"></a>
<span class="sourceLineNo">260</span> * we will divide a batch into several batches by replicationRpcLimit in method<a name="line.260"></a>
<span class="sourceLineNo">261</span> * serialReplicateRegionEntries()<a name="line.261"></a>
<span class="sourceLineNo">262</span> */<a name="line.262"></a>
<span class="sourceLineNo">263</span> private List&lt;List&lt;Entry&gt;&gt; createBatches(final List&lt;Entry&gt; entries) {<a name="line.263"></a>
<span class="sourceLineNo">264</span> if (isSerial) {<a name="line.264"></a>
<span class="sourceLineNo">265</span> return createSerialBatches(entries);<a name="line.265"></a>
<span class="sourceLineNo">266</span> } else {<a name="line.266"></a>
<span class="sourceLineNo">267</span> return createParallelBatches(entries);<a name="line.267"></a>
<span class="sourceLineNo">268</span> }<a name="line.268"></a>
<span class="sourceLineNo">269</span> }<a name="line.269"></a>
<span class="sourceLineNo">270</span><a name="line.270"></a>
<span class="sourceLineNo">271</span> /**<a name="line.271"></a>
<span class="sourceLineNo">272</span> * Check if there's an {@link TableNotFoundException} in the caused by stacktrace.<a name="line.272"></a>
<span class="sourceLineNo">273</span> */<a name="line.273"></a>
<span class="sourceLineNo">274</span> @VisibleForTesting<a name="line.274"></a>
<span class="sourceLineNo">275</span> public static boolean isTableNotFoundException(Throwable io) {<a name="line.275"></a>
<span class="sourceLineNo">276</span> if (io instanceof RemoteException) {<a name="line.276"></a>
<span class="sourceLineNo">277</span> io = ((RemoteException) io).unwrapRemoteException();<a name="line.277"></a>
<span class="sourceLineNo">278</span> }<a name="line.278"></a>
<span class="sourceLineNo">279</span> if (io != null &amp;&amp; io.getMessage().contains("TableNotFoundException")) {<a name="line.279"></a>
<span class="sourceLineNo">280</span> return true;<a name="line.280"></a>
<span class="sourceLineNo">281</span> }<a name="line.281"></a>
<span class="sourceLineNo">282</span> for (; io != null; io = io.getCause()) {<a name="line.282"></a>
<span class="sourceLineNo">283</span> if (io instanceof TableNotFoundException) {<a name="line.283"></a>
<span class="sourceLineNo">284</span> return true;<a name="line.284"></a>
<span class="sourceLineNo">285</span> }<a name="line.285"></a>
<span class="sourceLineNo">286</span> }<a name="line.286"></a>
<span class="sourceLineNo">287</span> return false;<a name="line.287"></a>
<span class="sourceLineNo">288</span> }<a name="line.288"></a>
<span class="sourceLineNo">289</span><a name="line.289"></a>
<span class="sourceLineNo">290</span> /**<a name="line.290"></a>
<span class="sourceLineNo">291</span> * Check if there's an {@link NoSuchColumnFamilyException} in the caused by stacktrace.<a name="line.291"></a>
<span class="sourceLineNo">292</span> */<a name="line.292"></a>
<span class="sourceLineNo">293</span> @VisibleForTesting<a name="line.293"></a>
<span class="sourceLineNo">294</span> public static boolean isNoSuchColumnFamilyException(Throwable io) {<a name="line.294"></a>
<span class="sourceLineNo">295</span> if (io instanceof RemoteException) {<a name="line.295"></a>
<span class="sourceLineNo">296</span> io = ((RemoteException) io).unwrapRemoteException();<a name="line.296"></a>
<span class="sourceLineNo">297</span> }<a name="line.297"></a>
<span class="sourceLineNo">298</span> if (io != null &amp;&amp; io.getMessage().contains("NoSuchColumnFamilyException")) {<a name="line.298"></a>
<span class="sourceLineNo">299</span> return true;<a name="line.299"></a>
<span class="sourceLineNo">300</span> }<a name="line.300"></a>
<span class="sourceLineNo">301</span> for (; io != null; io = io.getCause()) {<a name="line.301"></a>
<span class="sourceLineNo">302</span> if (io instanceof NoSuchColumnFamilyException) {<a name="line.302"></a>
<span class="sourceLineNo">303</span> return true;<a name="line.303"></a>
<span class="sourceLineNo">304</span> }<a name="line.304"></a>
<span class="sourceLineNo">305</span> }<a name="line.305"></a>
<span class="sourceLineNo">306</span> return false;<a name="line.306"></a>
<span class="sourceLineNo">307</span> }<a name="line.307"></a>
<span class="sourceLineNo">308</span><a name="line.308"></a>
<span class="sourceLineNo">309</span> @VisibleForTesting<a name="line.309"></a>
<span class="sourceLineNo">310</span> List&lt;List&lt;Entry&gt;&gt; filterNotExistTableEdits(final List&lt;List&lt;Entry&gt;&gt; oldEntryList) {<a name="line.310"></a>
<span class="sourceLineNo">311</span> List&lt;List&lt;Entry&gt;&gt; entryList = new ArrayList&lt;&gt;();<a name="line.311"></a>
<span class="sourceLineNo">312</span> Map&lt;TableName, Boolean&gt; existMap = new HashMap&lt;&gt;();<a name="line.312"></a>
<span class="sourceLineNo">313</span> try (Connection localConn = ConnectionFactory.createConnection(ctx.getLocalConfiguration());<a name="line.313"></a>
<span class="sourceLineNo">314</span> Admin localAdmin = localConn.getAdmin()) {<a name="line.314"></a>
<span class="sourceLineNo">315</span> for (List&lt;Entry&gt; oldEntries : oldEntryList) {<a name="line.315"></a>
<span class="sourceLineNo">316</span> List&lt;Entry&gt; entries = new ArrayList&lt;&gt;();<a name="line.316"></a>
<span class="sourceLineNo">317</span> for (Entry e : oldEntries) {<a name="line.317"></a>
<span class="sourceLineNo">318</span> TableName tableName = e.getKey().getTableName();<a name="line.318"></a>
<span class="sourceLineNo">319</span> boolean exist = true;<a name="line.319"></a>
<span class="sourceLineNo">320</span> if (existMap.containsKey(tableName)) {<a name="line.320"></a>
<span class="sourceLineNo">321</span> exist = existMap.get(tableName);<a name="line.321"></a>
<span class="sourceLineNo">322</span> } else {<a name="line.322"></a>
<span class="sourceLineNo">323</span> try {<a name="line.323"></a>
<span class="sourceLineNo">324</span> exist = localAdmin.tableExists(tableName);<a name="line.324"></a>
<span class="sourceLineNo">325</span> existMap.put(tableName, exist);<a name="line.325"></a>
<span class="sourceLineNo">326</span> } catch (IOException iox) {<a name="line.326"></a>
<span class="sourceLineNo">327</span> LOG.warn("Exception checking for local table " + tableName, iox);<a name="line.327"></a>
<span class="sourceLineNo">328</span> // we can't drop edits without full assurance, so we assume table exists.<a name="line.328"></a>
<span class="sourceLineNo">329</span> exist = true;<a name="line.329"></a>
<span class="sourceLineNo">330</span> }<a name="line.330"></a>
<span class="sourceLineNo">331</span> }<a name="line.331"></a>
<span class="sourceLineNo">332</span> if (exist) {<a name="line.332"></a>
<span class="sourceLineNo">333</span> entries.add(e);<a name="line.333"></a>
<span class="sourceLineNo">334</span> } else {<a name="line.334"></a>
<span class="sourceLineNo">335</span> // Would potentially be better to retry in one of the outer loops<a name="line.335"></a>
<span class="sourceLineNo">336</span> // and add a table filter there; but that would break the encapsulation,<a name="line.336"></a>
<span class="sourceLineNo">337</span> // so we're doing the filtering here.<a name="line.337"></a>
<span class="sourceLineNo">338</span> LOG.warn("Missing table detected at sink, local table also does not exist, "<a name="line.338"></a>
<span class="sourceLineNo">339</span> + "filtering edits for table '{}'", tableName);<a name="line.339"></a>
<span class="sourceLineNo">340</span> }<a name="line.340"></a>
<span class="sourceLineNo">341</span> }<a name="line.341"></a>
<span class="sourceLineNo">342</span> if (!entries.isEmpty()) {<a name="line.342"></a>
<span class="sourceLineNo">343</span> entryList.add(entries);<a name="line.343"></a>
<span class="sourceLineNo">344</span> }<a name="line.344"></a>
<span class="sourceLineNo">345</span> }<a name="line.345"></a>
<span class="sourceLineNo">346</span> } catch (IOException iox) {<a name="line.346"></a>
<span class="sourceLineNo">347</span> LOG.warn("Exception when creating connection to check local table", iox);<a name="line.347"></a>
<span class="sourceLineNo">348</span> return oldEntryList;<a name="line.348"></a>
<span class="sourceLineNo">349</span> }<a name="line.349"></a>
<span class="sourceLineNo">350</span> return entryList;<a name="line.350"></a>
<span class="sourceLineNo">351</span> }<a name="line.351"></a>
<span class="sourceLineNo">352</span><a name="line.352"></a>
<span class="sourceLineNo">353</span> @VisibleForTesting<a name="line.353"></a>
<span class="sourceLineNo">354</span> List&lt;List&lt;Entry&gt;&gt; filterNotExistColumnFamilyEdits(final List&lt;List&lt;Entry&gt;&gt; oldEntryList) {<a name="line.354"></a>
<span class="sourceLineNo">355</span> List&lt;List&lt;Entry&gt;&gt; entryList = new ArrayList&lt;&gt;();<a name="line.355"></a>
<span class="sourceLineNo">356</span> Map&lt;TableName, Set&lt;String&gt;&gt; existColumnFamilyMap = new HashMap&lt;&gt;();<a name="line.356"></a>
<span class="sourceLineNo">357</span> try (Connection localConn = ConnectionFactory.createConnection(ctx.getLocalConfiguration());<a name="line.357"></a>
<span class="sourceLineNo">358</span> Admin localAdmin = localConn.getAdmin()) {<a name="line.358"></a>
<span class="sourceLineNo">359</span> for (List&lt;Entry&gt; oldEntries : oldEntryList) {<a name="line.359"></a>
<span class="sourceLineNo">360</span> List&lt;Entry&gt; entries = new ArrayList&lt;&gt;();<a name="line.360"></a>
<span class="sourceLineNo">361</span> for (Entry e : oldEntries) {<a name="line.361"></a>
<span class="sourceLineNo">362</span> TableName tableName = e.getKey().getTableName();<a name="line.362"></a>
<span class="sourceLineNo">363</span> if (!existColumnFamilyMap.containsKey(tableName)) {<a name="line.363"></a>
<span class="sourceLineNo">364</span> try {<a name="line.364"></a>
<span class="sourceLineNo">365</span> Set&lt;String&gt; cfs = localAdmin.getDescriptor(tableName).getColumnFamilyNames().stream()<a name="line.365"></a>
<span class="sourceLineNo">366</span> .map(Bytes::toString).collect(Collectors.toSet());<a name="line.366"></a>
<span class="sourceLineNo">367</span> existColumnFamilyMap.put(tableName, cfs);<a name="line.367"></a>
<span class="sourceLineNo">368</span> } catch (Exception ex) {<a name="line.368"></a>
<span class="sourceLineNo">369</span> LOG.warn("Exception getting cf names for local table {}", tableName, ex);<a name="line.369"></a>
<span class="sourceLineNo">370</span> // if catch any exception, we are not sure about table's description,<a name="line.370"></a>
<span class="sourceLineNo">371</span> // so replicate raw entry<a name="line.371"></a>
<span class="sourceLineNo">372</span> entries.add(e);<a name="line.372"></a>
<span class="sourceLineNo">373</span> continue;<a name="line.373"></a>
<span class="sourceLineNo">374</span> }<a name="line.374"></a>
<span class="sourceLineNo">375</span> }<a name="line.375"></a>
<span class="sourceLineNo">376</span><a name="line.376"></a>
<span class="sourceLineNo">377</span> Set&lt;String&gt; existColumnFamilies = existColumnFamilyMap.get(tableName);<a name="line.377"></a>
<span class="sourceLineNo">378</span> Set&lt;String&gt; missingCFs = new HashSet&lt;&gt;();<a name="line.378"></a>
<span class="sourceLineNo">379</span> WALEdit walEdit = new WALEdit();<a name="line.379"></a>
<span class="sourceLineNo">380</span> walEdit.getCells().addAll(e.getEdit().getCells());<a name="line.380"></a>
<span class="sourceLineNo">381</span> WALUtil.filterCells(walEdit, cell -&gt; {<a name="line.381"></a>
<span class="sourceLineNo">382</span> String cf = Bytes.toString(CellUtil.cloneFamily(cell));<a name="line.382"></a>
<span class="sourceLineNo">383</span> if (existColumnFamilies.contains(cf)) {<a name="line.383"></a>
<span class="sourceLineNo">384</span> return cell;<a name="line.384"></a>
<span class="sourceLineNo">385</span> } else {<a name="line.385"></a>
<span class="sourceLineNo">386</span> missingCFs.add(cf);<a name="line.386"></a>
<span class="sourceLineNo">387</span> return null;<a name="line.387"></a>
<span class="sourceLineNo">388</span> }<a name="line.388"></a>
<span class="sourceLineNo">389</span> });<a name="line.389"></a>
<span class="sourceLineNo">390</span> if (!walEdit.isEmpty()) {<a name="line.390"></a>
<span class="sourceLineNo">391</span> Entry newEntry = new Entry(e.getKey(), walEdit);<a name="line.391"></a>
<span class="sourceLineNo">392</span> entries.add(newEntry);<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> if (!missingCFs.isEmpty()) {<a name="line.395"></a>
<span class="sourceLineNo">396</span> // Would potentially be better to retry in one of the outer loops<a name="line.396"></a>
<span class="sourceLineNo">397</span> // and add a table filter there; but that would break the encapsulation,<a name="line.397"></a>
<span class="sourceLineNo">398</span> // so we're doing the filtering here.<a name="line.398"></a>
<span class="sourceLineNo">399</span> LOG.warn(<a name="line.399"></a>
<span class="sourceLineNo">400</span> "Missing column family detected at sink, local column family also does not exist,"<a name="line.400"></a>
<span class="sourceLineNo">401</span> + " filtering edits for table '{}',column family '{}'", tableName, missingCFs);<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> if (!entries.isEmpty()) {<a name="line.404"></a>
<span class="sourceLineNo">405</span> entryList.add(entries);<a name="line.405"></a>
<span class="sourceLineNo">406</span> }<a name="line.406"></a>
<span class="sourceLineNo">407</span> }<a name="line.407"></a>
<span class="sourceLineNo">408</span> } catch (IOException iox) {<a name="line.408"></a>
<span class="sourceLineNo">409</span> LOG.warn("Exception when creating connection to check local table", iox);<a name="line.409"></a>
<span class="sourceLineNo">410</span> return oldEntryList;<a name="line.410"></a>
<span class="sourceLineNo">411</span> }<a name="line.411"></a>
<span class="sourceLineNo">412</span> return entryList;<a name="line.412"></a>
<span class="sourceLineNo">413</span> }<a name="line.413"></a>
<span class="sourceLineNo">414</span><a name="line.414"></a>
<span class="sourceLineNo">415</span> private void reconnectToPeerCluster() {<a name="line.415"></a>
<span class="sourceLineNo">416</span> AsyncClusterConnection connection = null;<a name="line.416"></a>
<span class="sourceLineNo">417</span> try {<a name="line.417"></a>
<span class="sourceLineNo">418</span> connection =<a name="line.418"></a>
<span class="sourceLineNo">419</span> ClusterConnectionFactory.createAsyncClusterConnection(conf, null, User.getCurrent());<a name="line.419"></a>
<span class="sourceLineNo">420</span> } catch (IOException ioe) {<a name="line.420"></a>
<span class="sourceLineNo">421</span> LOG.warn("{} Failed to create connection for peer cluster", logPeerId(), ioe);<a name="line.421"></a>
<span class="sourceLineNo">422</span> }<a name="line.422"></a>
<span class="sourceLineNo">423</span> if (connection != null) {<a name="line.423"></a>
<span class="sourceLineNo">424</span> this.conn = connection;<a name="line.424"></a>
<span class="sourceLineNo">425</span> }<a name="line.425"></a>
<span class="sourceLineNo">426</span> }<a name="line.426"></a>
<span class="sourceLineNo">427</span><a name="line.427"></a>
<span class="sourceLineNo">428</span> private long parallelReplicate(CompletionService&lt;Integer&gt; pool, ReplicateContext replicateContext,<a name="line.428"></a>
<span class="sourceLineNo">429</span> List&lt;List&lt;Entry&gt;&gt; batches) throws IOException {<a name="line.429"></a>
<span class="sourceLineNo">430</span> int futures = 0;<a name="line.430"></a>
<span class="sourceLineNo">431</span> for (int i = 0; i &lt; batches.size(); i++) {<a name="line.431"></a>
<span class="sourceLineNo">432</span> List&lt;Entry&gt; entries = batches.get(i);<a name="line.432"></a>
<span class="sourceLineNo">433</span> if (!entries.isEmpty()) {<a name="line.433"></a>
<span class="sourceLineNo">434</span> if (LOG.isTraceEnabled()) {<a name="line.434"></a>
<span class="sourceLineNo">435</span> LOG.trace("{} Submitting {} entries of total size {}", logPeerId(), entries.size(),<a name="line.435"></a>
<span class="sourceLineNo">436</span> replicateContext.getSize());<a name="line.436"></a>
<span class="sourceLineNo">437</span> }<a name="line.437"></a>
<span class="sourceLineNo">438</span> // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.438"></a>
<span class="sourceLineNo">439</span> pool.submit(createReplicator(entries, i, replicateContext.getTimeout()));<a name="line.439"></a>
<span class="sourceLineNo">440</span> futures++;<a name="line.440"></a>
<span class="sourceLineNo">441</span> }<a name="line.441"></a>
<span class="sourceLineNo">442</span> }<a name="line.442"></a>
<span class="sourceLineNo">443</span><a name="line.443"></a>
<span class="sourceLineNo">444</span> IOException iox = null;<a name="line.444"></a>
<span class="sourceLineNo">445</span> long lastWriteTime = 0;<a name="line.445"></a>
<span class="sourceLineNo">446</span> for (int i = 0; i &lt; futures; i++) {<a name="line.446"></a>
<span class="sourceLineNo">447</span> try {<a name="line.447"></a>
<span class="sourceLineNo">448</span> // wait for all futures, remove successful parts<a name="line.448"></a>
<span class="sourceLineNo">449</span> // (only the remaining parts will be retried)<a name="line.449"></a>
<span class="sourceLineNo">450</span> Future&lt;Integer&gt; f = pool.take();<a name="line.450"></a>
<span class="sourceLineNo">451</span> int index = f.get();<a name="line.451"></a>
<span class="sourceLineNo">452</span> List&lt;Entry&gt; batch = batches.get(index);<a name="line.452"></a>
<span class="sourceLineNo">453</span> batches.set(index, Collections.emptyList()); // remove successful batch<a name="line.453"></a>
<span class="sourceLineNo">454</span> // Find the most recent write time in the batch<a name="line.454"></a>
<span class="sourceLineNo">455</span> long writeTime = batch.get(batch.size() - 1).getKey().getWriteTime();<a name="line.455"></a>
<span class="sourceLineNo">456</span> if (writeTime &gt; lastWriteTime) {<a name="line.456"></a>
<span class="sourceLineNo">457</span> lastWriteTime = writeTime;<a name="line.457"></a>
<span class="sourceLineNo">458</span> }<a name="line.458"></a>
<span class="sourceLineNo">459</span> } catch (InterruptedException ie) {<a name="line.459"></a>
<span class="sourceLineNo">460</span> iox = new IOException(ie);<a name="line.460"></a>
<span class="sourceLineNo">461</span> } catch (ExecutionException ee) {<a name="line.461"></a>
<span class="sourceLineNo">462</span> iox = ee.getCause() instanceof IOException?<a name="line.462"></a>
<span class="sourceLineNo">463</span> (IOException)ee.getCause(): new IOException(ee.getCause());<a name="line.463"></a>
<span class="sourceLineNo">464</span> }<a name="line.464"></a>
<span class="sourceLineNo">465</span> }<a name="line.465"></a>
<span class="sourceLineNo">466</span> if (iox != null) {<a name="line.466"></a>
<span class="sourceLineNo">467</span> // if we had any exceptions, try again<a name="line.467"></a>
<span class="sourceLineNo">468</span> throw iox;<a name="line.468"></a>
<span class="sourceLineNo">469</span> }<a name="line.469"></a>
<span class="sourceLineNo">470</span> return lastWriteTime;<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> * Do the shipping logic<a name="line.474"></a>
<span class="sourceLineNo">475</span> */<a name="line.475"></a>
<span class="sourceLineNo">476</span> @Override<a name="line.476"></a>
<span class="sourceLineNo">477</span> public boolean replicate(ReplicateContext replicateContext) {<a name="line.477"></a>
<span class="sourceLineNo">478</span> CompletionService&lt;Integer&gt; pool = new ExecutorCompletionService&lt;&gt;(this.exec);<a name="line.478"></a>
<span class="sourceLineNo">479</span> int sleepMultiplier = 1;<a name="line.479"></a>
<span class="sourceLineNo">480</span><a name="line.480"></a>
<span class="sourceLineNo">481</span> if (!peersSelected &amp;&amp; this.isRunning()) {<a name="line.481"></a>
<span class="sourceLineNo">482</span> connectToPeers();<a name="line.482"></a>
<span class="sourceLineNo">483</span> peersSelected = true;<a name="line.483"></a>
<span class="sourceLineNo">484</span> }<a name="line.484"></a>
<span class="sourceLineNo">485</span><a name="line.485"></a>
<span class="sourceLineNo">486</span> int numSinks = getNumSinks();<a name="line.486"></a>
<span class="sourceLineNo">487</span> if (numSinks == 0) {<a name="line.487"></a>
<span class="sourceLineNo">488</span> if((System.currentTimeMillis() - lastSinkFetchTime) &gt;= (maxRetriesMultiplier*1000)) {<a name="line.488"></a>
<span class="sourceLineNo">489</span> LOG.warn(<a name="line.489"></a>
<span class="sourceLineNo">490</span> "No replication sinks found, returning without replicating. "<a name="line.490"></a>
<span class="sourceLineNo">491</span> + "The source should retry with the same set of edits. Not logging this again for "<a name="line.491"></a>
<span class="sourceLineNo">492</span> + "the next {} seconds.", maxRetriesMultiplier);<a name="line.492"></a>
<span class="sourceLineNo">493</span> lastSinkFetchTime = System.currentTimeMillis();<a name="line.493"></a>
<span class="sourceLineNo">494</span> }<a name="line.494"></a>
<span class="sourceLineNo">495</span> sleepForRetries("No sinks available at peer", sleepMultiplier);<a name="line.495"></a>
<span class="sourceLineNo">496</span> return false;<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> List&lt;List&lt;Entry&gt;&gt; batches = createBatches(replicateContext.getEntries());<a name="line.499"></a>
<span class="sourceLineNo">500</span> while (this.isRunning() &amp;&amp; !exec.isShutdown()) {<a name="line.500"></a>
<span class="sourceLineNo">501</span> if (!isPeerEnabled()) {<a name="line.501"></a>
<span class="sourceLineNo">502</span> if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.502"></a>
<span class="sourceLineNo">503</span> sleepMultiplier++;<a name="line.503"></a>
<span class="sourceLineNo">504</span> }<a name="line.504"></a>
<span class="sourceLineNo">505</span> continue;<a name="line.505"></a>
<span class="sourceLineNo">506</span> }<a name="line.506"></a>
<span class="sourceLineNo">507</span> if (this.conn == null) {<a name="line.507"></a>
<span class="sourceLineNo">508</span> reconnectToPeerCluster();<a name="line.508"></a>
<span class="sourceLineNo">509</span> }<a name="line.509"></a>
<span class="sourceLineNo">510</span> try {<a name="line.510"></a>
<span class="sourceLineNo">511</span> // replicate the batches to sink side.<a name="line.511"></a>
<span class="sourceLineNo">512</span> parallelReplicate(pool, replicateContext, batches);<a name="line.512"></a>
<span class="sourceLineNo">513</span> return true;<a name="line.513"></a>
<span class="sourceLineNo">514</span> } catch (IOException ioe) {<a name="line.514"></a>
<span class="sourceLineNo">515</span> if (ioe instanceof RemoteException) {<a name="line.515"></a>
<span class="sourceLineNo">516</span> if (dropOnDeletedTables &amp;&amp; isTableNotFoundException(ioe)) {<a name="line.516"></a>
<span class="sourceLineNo">517</span> // Only filter the edits to replicate and don't change the entries in replicateContext<a name="line.517"></a>
<span class="sourceLineNo">518</span> // as the upper layer rely on it.<a name="line.518"></a>
<span class="sourceLineNo">519</span> batches = filterNotExistTableEdits(batches);<a name="line.519"></a>
<span class="sourceLineNo">520</span> if (batches.isEmpty()) {<a name="line.520"></a>
<span class="sourceLineNo">521</span> LOG.warn("After filter not exist table's edits, 0 edits to replicate, just return");<a name="line.521"></a>
<span class="sourceLineNo">522</span> return true;<a name="line.522"></a>
<span class="sourceLineNo">523</span> }<a name="line.523"></a>
<span class="sourceLineNo">524</span> } else if (dropOnDeletedColumnFamilies &amp;&amp; isNoSuchColumnFamilyException(ioe)) {<a name="line.524"></a>
<span class="sourceLineNo">525</span> batches = filterNotExistColumnFamilyEdits(batches);<a name="line.525"></a>
<span class="sourceLineNo">526</span> if (batches.isEmpty()) {<a name="line.526"></a>
<span class="sourceLineNo">527</span> LOG.warn("After filter not exist column family's edits, 0 edits to replicate, " +<a name="line.527"></a>
<span class="sourceLineNo">528</span> "just return");<a name="line.528"></a>
<span class="sourceLineNo">529</span> return true;<a name="line.529"></a>
<span class="sourceLineNo">530</span> }<a name="line.530"></a>
<span class="sourceLineNo">531</span> } else {<a name="line.531"></a>
<span class="sourceLineNo">532</span> LOG.warn("{} Peer encountered RemoteException, rechecking all sinks: ", logPeerId(),<a name="line.532"></a>
<span class="sourceLineNo">533</span> ioe);<a name="line.533"></a>
<span class="sourceLineNo">534</span> chooseSinks();<a name="line.534"></a>
<span class="sourceLineNo">535</span> }<a name="line.535"></a>
<span class="sourceLineNo">536</span> } else {<a name="line.536"></a>
<span class="sourceLineNo">537</span> if (ioe instanceof SocketTimeoutException) {<a name="line.537"></a>
<span class="sourceLineNo">538</span> // This exception means we waited for more than 60s and nothing<a name="line.538"></a>
<span class="sourceLineNo">539</span> // happened, the cluster is alive and calling it right away<a name="line.539"></a>
<span class="sourceLineNo">540</span> // even for a test just makes things worse.<a name="line.540"></a>
<span class="sourceLineNo">541</span> sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.541"></a>
<span class="sourceLineNo">542</span> "call to the remote cluster timed out, which is usually " +<a name="line.542"></a>
<span class="sourceLineNo">543</span> "caused by a machine failure or a massive slowdown",<a name="line.543"></a>
<span class="sourceLineNo">544</span> this.socketTimeoutMultiplier);<a name="line.544"></a>
<span class="sourceLineNo">545</span> } else if (ioe instanceof ConnectException || ioe instanceof UnknownHostException) {<a name="line.545"></a>
<span class="sourceLineNo">546</span> LOG.warn("{} Peer is unavailable, rechecking all sinks: ", logPeerId(), ioe);<a name="line.546"></a>
<span class="sourceLineNo">547</span> chooseSinks();<a name="line.547"></a>
<span class="sourceLineNo">548</span> } else {<a name="line.548"></a>
<span class="sourceLineNo">549</span> LOG.warn("{} Can't replicate because of a local or network error: ", logPeerId(), ioe);<a name="line.549"></a>
<span class="sourceLineNo">550</span> }<a name="line.550"></a>
<span class="sourceLineNo">551</span> }<a name="line.551"></a>
<span class="sourceLineNo">552</span> if (sleepForRetries("Since we are unable to replicate", sleepMultiplier)) {<a name="line.552"></a>
<span class="sourceLineNo">553</span> sleepMultiplier++;<a name="line.553"></a>
<span class="sourceLineNo">554</span> }<a name="line.554"></a>
<span class="sourceLineNo">555</span> }<a name="line.555"></a>
<span class="sourceLineNo">556</span> }<a name="line.556"></a>
<span class="sourceLineNo">557</span> return false; // in case we exited before replicating<a name="line.557"></a>
<span class="sourceLineNo">558</span> }<a name="line.558"></a>
<span class="sourceLineNo">559</span><a name="line.559"></a>
<span class="sourceLineNo">560</span> protected boolean isPeerEnabled() {<a name="line.560"></a>
<span class="sourceLineNo">561</span> return ctx.getReplicationPeer().isPeerEnabled();<a name="line.561"></a>
<span class="sourceLineNo">562</span> }<a name="line.562"></a>
<span class="sourceLineNo">563</span><a name="line.563"></a>
<span class="sourceLineNo">564</span> @Override<a name="line.564"></a>
<span class="sourceLineNo">565</span> protected void doStop() {<a name="line.565"></a>
<span class="sourceLineNo">566</span> disconnect(); // don't call super.doStop()<a name="line.566"></a>
<span class="sourceLineNo">567</span> if (this.conn != null) {<a name="line.567"></a>
<span class="sourceLineNo">568</span> try {<a name="line.568"></a>
<span class="sourceLineNo">569</span> this.conn.close();<a name="line.569"></a>
<span class="sourceLineNo">570</span> this.conn = null;<a name="line.570"></a>
<span class="sourceLineNo">571</span> } catch (IOException e) {<a name="line.571"></a>
<span class="sourceLineNo">572</span> LOG.warn("{} Failed to close the connection", logPeerId());<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> // Allow currently running replication tasks to finish<a name="line.575"></a>
<span class="sourceLineNo">576</span> exec.shutdown();<a name="line.576"></a>
<span class="sourceLineNo">577</span> try {<a name="line.577"></a>
<span class="sourceLineNo">578</span> exec.awaitTermination(maxTerminationWait, TimeUnit.MILLISECONDS);<a name="line.578"></a>
<span class="sourceLineNo">579</span> } catch (InterruptedException e) {<a name="line.579"></a>
<span class="sourceLineNo">580</span> }<a name="line.580"></a>
<span class="sourceLineNo">581</span> // Abort if the tasks did not terminate in time<a name="line.581"></a>
<span class="sourceLineNo">582</span> if (!exec.isTerminated()) {<a name="line.582"></a>
<span class="sourceLineNo">583</span> String errMsg = "HBaseInterClusterReplicationEndpoint termination failed. The " +<a name="line.583"></a>
<span class="sourceLineNo">584</span> "ThreadPoolExecutor failed to finish all tasks within " + maxTerminationWait + "ms. " +<a name="line.584"></a>
<span class="sourceLineNo">585</span> "Aborting to prevent Replication from deadlocking. See HBASE-16081.";<a name="line.585"></a>
<span class="sourceLineNo">586</span> abortable.abort(errMsg, new IOException(errMsg));<a name="line.586"></a>
<span class="sourceLineNo">587</span> }<a name="line.587"></a>
<span class="sourceLineNo">588</span> notifyStopped();<a name="line.588"></a>
<span class="sourceLineNo">589</span> }<a name="line.589"></a>
<span class="sourceLineNo">590</span><a name="line.590"></a>
<span class="sourceLineNo">591</span> @VisibleForTesting<a name="line.591"></a>
<span class="sourceLineNo">592</span> protected int replicateEntries(List&lt;Entry&gt; entries, int batchIndex, int timeout)<a name="line.592"></a>
<span class="sourceLineNo">593</span> throws IOException {<a name="line.593"></a>
<span class="sourceLineNo">594</span> SinkPeer sinkPeer = null;<a name="line.594"></a>
<span class="sourceLineNo">595</span> try {<a name="line.595"></a>
<span class="sourceLineNo">596</span> int entriesHashCode = System.identityHashCode(entries);<a name="line.596"></a>
<span class="sourceLineNo">597</span> if (LOG.isTraceEnabled()) {<a name="line.597"></a>
<span class="sourceLineNo">598</span> long size = entries.stream().mapToLong(this::getEstimatedEntrySize).sum();<a name="line.598"></a>
<span class="sourceLineNo">599</span> LOG.trace("{} Replicating batch {} of {} entries with total size {} bytes to {}",<a name="line.599"></a>
<span class="sourceLineNo">600</span> logPeerId(), entriesHashCode, entries.size(), size, replicationClusterId);<a name="line.600"></a>
<span class="sourceLineNo">601</span> }<a name="line.601"></a>
<span class="sourceLineNo">602</span> sinkPeer = getReplicationSink();<a name="line.602"></a>
<span class="sourceLineNo">603</span> AsyncRegionServerAdmin rsAdmin = sinkPeer.getRegionServer();<a name="line.603"></a>
<span class="sourceLineNo">604</span> try {<a name="line.604"></a>
<span class="sourceLineNo">605</span> ReplicationProtobufUtil.replicateWALEntry(rsAdmin,<a name="line.605"></a>
<span class="sourceLineNo">606</span> entries.toArray(new Entry[entries.size()]), replicationClusterId, baseNamespaceDir,<a name="line.606"></a>
<span class="sourceLineNo">607</span> hfileArchiveDir, timeout);<a name="line.607"></a>
<span class="sourceLineNo">608</span> if (LOG.isTraceEnabled()) {<a name="line.608"></a>
<span class="sourceLineNo">609</span> LOG.trace("{} Completed replicating batch {}", logPeerId(), entriesHashCode);<a name="line.609"></a>
<span class="sourceLineNo">610</span> }<a name="line.610"></a>
<span class="sourceLineNo">611</span> } catch (IOException e) {<a name="line.611"></a>
<span class="sourceLineNo">612</span> if (LOG.isTraceEnabled()) {<a name="line.612"></a>
<span class="sourceLineNo">613</span> LOG.trace("{} Failed replicating batch {}", logPeerId(), entriesHashCode, e);<a name="line.613"></a>
<span class="sourceLineNo">614</span> }<a name="line.614"></a>
<span class="sourceLineNo">615</span> throw e;<a name="line.615"></a>
<span class="sourceLineNo">616</span> }<a name="line.616"></a>
<span class="sourceLineNo">617</span> reportSinkSuccess(sinkPeer);<a name="line.617"></a>
<span class="sourceLineNo">618</span> } catch (IOException ioe) {<a name="line.618"></a>
<span class="sourceLineNo">619</span> if (sinkPeer != null) {<a name="line.619"></a>
<span class="sourceLineNo">620</span> reportBadSink(sinkPeer);<a name="line.620"></a>
<span class="sourceLineNo">621</span> }<a name="line.621"></a>
<span class="sourceLineNo">622</span> throw ioe;<a name="line.622"></a>
<span class="sourceLineNo">623</span> }<a name="line.623"></a>
<span class="sourceLineNo">624</span> return batchIndex;<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> private int serialReplicateRegionEntries(List&lt;Entry&gt; entries, int batchIndex, int timeout)<a name="line.627"></a>
<span class="sourceLineNo">628</span> throws IOException {<a name="line.628"></a>
<span class="sourceLineNo">629</span> int batchSize = 0, index = 0;<a name="line.629"></a>
<span class="sourceLineNo">630</span> List&lt;Entry&gt; batch = new ArrayList&lt;&gt;();<a name="line.630"></a>
<span class="sourceLineNo">631</span> for (Entry entry : entries) {<a name="line.631"></a>
<span class="sourceLineNo">632</span> int entrySize = getEstimatedEntrySize(entry);<a name="line.632"></a>
<span class="sourceLineNo">633</span> if (batchSize &gt; 0 &amp;&amp; batchSize + entrySize &gt; replicationRpcLimit) {<a name="line.633"></a>
<span class="sourceLineNo">634</span> replicateEntries(batch, index++, timeout);<a name="line.634"></a>
<span class="sourceLineNo">635</span> batch.clear();<a name="line.635"></a>
<span class="sourceLineNo">636</span> batchSize = 0;<a name="line.636"></a>
<span class="sourceLineNo">637</span> }<a name="line.637"></a>
<span class="sourceLineNo">638</span> batch.add(entry);<a name="line.638"></a>
<span class="sourceLineNo">639</span> batchSize += entrySize;<a name="line.639"></a>
<span class="sourceLineNo">640</span> }<a name="line.640"></a>
<span class="sourceLineNo">641</span> if (batchSize &gt; 0) {<a name="line.641"></a>
<span class="sourceLineNo">642</span> replicateEntries(batch, index, timeout);<a name="line.642"></a>
<span class="sourceLineNo">643</span> }<a name="line.643"></a>
<span class="sourceLineNo">644</span> return batchIndex;<a name="line.644"></a>
<span class="sourceLineNo">645</span> }<a name="line.645"></a>
<span class="sourceLineNo">646</span><a name="line.646"></a>
<span class="sourceLineNo">647</span> @VisibleForTesting<a name="line.647"></a>
<span class="sourceLineNo">648</span> protected Callable&lt;Integer&gt; createReplicator(List&lt;Entry&gt; entries, int batchIndex, int timeout) {<a name="line.648"></a>
<span class="sourceLineNo">649</span> return isSerial ? () -&gt; serialReplicateRegionEntries(entries, batchIndex, timeout)<a name="line.649"></a>
<span class="sourceLineNo">650</span> : () -&gt; replicateEntries(entries, batchIndex, timeout);<a name="line.650"></a>
<span class="sourceLineNo">651</span> }<a name="line.651"></a>
<span class="sourceLineNo">652</span><a name="line.652"></a>
<span class="sourceLineNo">653</span> private String logPeerId(){<a name="line.653"></a>
<span class="sourceLineNo">654</span> return "[Source for peer " + this.ctx.getPeerId() + "]:";<a name="line.654"></a>
<span class="sourceLineNo">655</span> }<a name="line.655"></a>
<span class="sourceLineNo">656</span>}<a name="line.656"></a>
</pre>
</div>
</body>
</html>