hbase-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From git-site-r...@apache.org
Subject [22/27] hbase-site git commit: Published site at .
Date Thu, 18 Jan 2018 15:31:21 GMT
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/cc91ad79/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.Replicator.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.Replicator.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.Replicator.html
index f07ce2e..6cb0b5d 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.Replicator.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/HBaseInterClusterReplicationEndpoint.Replicator.html
@@ -29,488 +29,489 @@
 <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.util.ArrayList;<a name="line.24"></a>
-<span class="sourceLineNo">025</span>import java.util.Collections;<a name="line.25"></a>
-<span class="sourceLineNo">026</span>import java.util.HashMap;<a name="line.26"></a>
-<span class="sourceLineNo">027</span>import java.util.List;<a name="line.27"></a>
-<span class="sourceLineNo">028</span>import java.util.Map;<a name="line.28"></a>
-<span class="sourceLineNo">029</span>import java.util.concurrent.Callable;<a name="line.29"></a>
-<span class="sourceLineNo">030</span>import java.util.concurrent.CompletionService;<a name="line.30"></a>
-<span class="sourceLineNo">031</span>import java.util.concurrent.ExecutionException;<a name="line.31"></a>
-<span class="sourceLineNo">032</span>import java.util.concurrent.ExecutorCompletionService;<a name="line.32"></a>
-<span class="sourceLineNo">033</span>import java.util.concurrent.Future;<a name="line.33"></a>
-<span class="sourceLineNo">034</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.34"></a>
-<span class="sourceLineNo">035</span>import java.util.concurrent.ThreadPoolExecutor;<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.regex.Matcher;<a name="line.37"></a>
-<span class="sourceLineNo">038</span>import java.util.regex.Pattern;<a name="line.38"></a>
-<span class="sourceLineNo">039</span>import org.apache.commons.lang3.StringUtils;<a name="line.39"></a>
-<span class="sourceLineNo">040</span>import org.apache.hadoop.conf.Configuration;<a name="line.40"></a>
-<span class="sourceLineNo">041</span>import org.apache.hadoop.fs.Path;<a name="line.41"></a>
-<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.Abortable;<a name="line.42"></a>
-<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.43"></a>
-<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.HConstants;<a name="line.44"></a>
-<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.TableName;<a name="line.45"></a>
-<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.46"></a>
-<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.47"></a>
-<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.48"></a>
-<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.49"></a>
-<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.ipc.RpcServer;<a name="line.50"></a>
-<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtbufUtil;<a name="line.51"></a>
-<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.52"></a>
-<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.regionserver.ReplicationSinkManager.SinkPeer;<a name="line.53"></a>
-<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.54"></a>
-<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.55"></a>
-<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.56"></a>
-<span class="sourceLineNo">057</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.57"></a>
-<span class="sourceLineNo">058</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.58"></a>
-<span class="sourceLineNo">059</span>import org.slf4j.Logger;<a name="line.59"></a>
-<span class="sourceLineNo">060</span>import org.slf4j.LoggerFactory;<a name="line.60"></a>
-<span class="sourceLineNo">061</span><a name="line.61"></a>
-<span class="sourceLineNo">062</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.62"></a>
-<span class="sourceLineNo">063</span><a name="line.63"></a>
-<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService.BlockingInterface;<a name="line.64"></a>
-<span class="sourceLineNo">065</span><a name="line.65"></a>
-<span class="sourceLineNo">066</span>/**<a name="line.66"></a>
-<span class="sourceLineNo">067</span> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint}<a name="line.67"></a>
-<span class="sourceLineNo">068</span> * implementation for replicating to another HBase cluster.<a name="line.68"></a>
-<span class="sourceLineNo">069</span> * For the slave cluster it selects a random number of peers<a name="line.69"></a>
-<span class="sourceLineNo">070</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.70"></a>
-<span class="sourceLineNo">071</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.71"></a>
-<span class="sourceLineNo">072</span> * &lt;p&gt;<a name="line.72"></a>
-<span class="sourceLineNo">073</span> * A stream is considered down when we cannot contact a region server on the<a name="line.73"></a>
-<span class="sourceLineNo">074</span> * peer cluster for more than 55 seconds by default.<a name="line.74"></a>
-<span class="sourceLineNo">075</span> * &lt;/p&gt;<a name="line.75"></a>
-<span class="sourceLineNo">076</span> */<a name="line.76"></a>
-<span class="sourceLineNo">077</span>@InterfaceAudience.Private<a name="line.77"></a>
-<span class="sourceLineNo">078</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.78"></a>
-<span class="sourceLineNo">079</span>  private static final Logger LOG =<a name="line.79"></a>
-<span class="sourceLineNo">080</span>      LoggerFactory.getLogger(HBaseInterClusterReplicationEndpoint.class);<a name="line.80"></a>
-<span class="sourceLineNo">081</span><a name="line.81"></a>
-<span class="sourceLineNo">082</span>  private static final long DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER = 2;<a name="line.82"></a>
-<span class="sourceLineNo">083</span><a name="line.83"></a>
-<span class="sourceLineNo">084</span>  private ClusterConnection conn;<a name="line.84"></a>
-<span class="sourceLineNo">085</span>  private Configuration localConf;<a name="line.85"></a>
-<span class="sourceLineNo">086</span>  private Configuration conf;<a name="line.86"></a>
-<span class="sourceLineNo">087</span>  // How long should we sleep for each retry<a name="line.87"></a>
-<span class="sourceLineNo">088</span>  private long sleepForRetries;<a name="line.88"></a>
-<span class="sourceLineNo">089</span>  // Maximum number of retries before taking bold actions<a name="line.89"></a>
-<span class="sourceLineNo">090</span>  private int maxRetriesMultiplier;<a name="line.90"></a>
-<span class="sourceLineNo">091</span>  // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.91"></a>
-<span class="sourceLineNo">092</span>  private int socketTimeoutMultiplier;<a name="line.92"></a>
-<span class="sourceLineNo">093</span>  // Amount of time for shutdown to wait for all tasks to complete<a name="line.93"></a>
-<span class="sourceLineNo">094</span>  private long maxTerminationWait;<a name="line.94"></a>
-<span class="sourceLineNo">095</span>  // Size limit for replication RPCs, in bytes<a name="line.95"></a>
-<span class="sourceLineNo">096</span>  private int replicationRpcLimit;<a name="line.96"></a>
-<span class="sourceLineNo">097</span>  //Metrics for this source<a name="line.97"></a>
-<span class="sourceLineNo">098</span>  private MetricsSource metrics;<a name="line.98"></a>
-<span class="sourceLineNo">099</span>  // Handles connecting to peer region servers<a name="line.99"></a>
-<span class="sourceLineNo">100</span>  private ReplicationSinkManager replicationSinkMgr;<a name="line.100"></a>
-<span class="sourceLineNo">101</span>  private boolean peersSelected = false;<a name="line.101"></a>
-<span class="sourceLineNo">102</span>  private String replicationClusterId = "";<a name="line.102"></a>
-<span class="sourceLineNo">103</span>  private ThreadPoolExecutor exec;<a name="line.103"></a>
-<span class="sourceLineNo">104</span>  private int maxThreads;<a name="line.104"></a>
-<span class="sourceLineNo">105</span>  private Path baseNamespaceDir;<a name="line.105"></a>
-<span class="sourceLineNo">106</span>  private Path hfileArchiveDir;<a name="line.106"></a>
-<span class="sourceLineNo">107</span>  private boolean replicationBulkLoadDataEnabled;<a name="line.107"></a>
-<span class="sourceLineNo">108</span>  private Abortable abortable;<a name="line.108"></a>
-<span class="sourceLineNo">109</span>  private boolean dropOnDeletedTables;<a name="line.109"></a>
-<span class="sourceLineNo">110</span><a name="line.110"></a>
-<span class="sourceLineNo">111</span>  @Override<a name="line.111"></a>
-<span class="sourceLineNo">112</span>  public void init(Context context) throws IOException {<a name="line.112"></a>
-<span class="sourceLineNo">113</span>    super.init(context);<a name="line.113"></a>
-<span class="sourceLineNo">114</span>    this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.114"></a>
-<span class="sourceLineNo">115</span>    this.localConf = HBaseConfiguration.create(ctx.getLocalConfiguration());<a name="line.115"></a>
-<span class="sourceLineNo">116</span>    decorateConf();<a name="line.116"></a>
-<span class="sourceLineNo">117</span>    this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.117"></a>
-<span class="sourceLineNo">118</span>    this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.118"></a>
-<span class="sourceLineNo">119</span>        maxRetriesMultiplier);<a name="line.119"></a>
-<span class="sourceLineNo">120</span>    // A Replicator job is bound by the RPC timeout. We will wait this long for all Replicator<a name="line.120"></a>
-<span class="sourceLineNo">121</span>    // tasks to terminate when doStop() is called.<a name="line.121"></a>
-<span class="sourceLineNo">122</span>    long maxTerminationWaitMultiplier = this.conf.getLong(<a name="line.122"></a>
-<span class="sourceLineNo">123</span>        "replication.source.maxterminationmultiplier",<a name="line.123"></a>
-<span class="sourceLineNo">124</span>        DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER);<a name="line.124"></a>
-<span class="sourceLineNo">125</span>    this.maxTerminationWait = maxTerminationWaitMultiplier *<a name="line.125"></a>
-<span class="sourceLineNo">126</span>        this.conf.getLong(HConstants.HBASE_RPC_TIMEOUT_KEY, HConstants.DEFAULT_HBASE_RPC_TIMEOUT);<a name="line.126"></a>
-<span class="sourceLineNo">127</span>    // TODO: This connection is replication specific or we should make it particular to<a name="line.127"></a>
-<span class="sourceLineNo">128</span>    // replication and make replication specific settings such as compression or codec to use<a name="line.128"></a>
-<span class="sourceLineNo">129</span>    // passing Cells.<a name="line.129"></a>
-<span class="sourceLineNo">130</span>    this.conn = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.130"></a>
-<span class="sourceLineNo">131</span>    this.sleepForRetries =<a name="line.131"></a>
-<span class="sourceLineNo">132</span>        this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.132"></a>
-<span class="sourceLineNo">133</span>    this.metrics = context.getMetrics();<a name="line.133"></a>
-<span class="sourceLineNo">134</span>    // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.134"></a>
-<span class="sourceLineNo">135</span>    this.replicationSinkMgr = new ReplicationSinkManager(conn, ctx.getPeerId(), this, this.conf);<a name="line.135"></a>
-<span class="sourceLineNo">136</span>    // per sink thread pool<a name="line.136"></a>
-<span class="sourceLineNo">137</span>    this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.137"></a>
-<span class="sourceLineNo">138</span>      HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.138"></a>
-<span class="sourceLineNo">139</span>    this.exec = new ThreadPoolExecutor(maxThreads, maxThreads, 60, TimeUnit.SECONDS,<a name="line.139"></a>
-<span class="sourceLineNo">140</span>        new LinkedBlockingQueue&lt;&gt;());<a name="line.140"></a>
-<span class="sourceLineNo">141</span>    this.exec.allowCoreThreadTimeOut(true);<a name="line.141"></a>
-<span class="sourceLineNo">142</span>    this.abortable = ctx.getAbortable();<a name="line.142"></a>
-<span class="sourceLineNo">143</span>    // Set the size limit for replication RPCs to 95% of the max request size.<a name="line.143"></a>
-<span class="sourceLineNo">144</span>    // We could do with less slop if we have an accurate estimate of encoded size. Being<a name="line.144"></a>
-<span class="sourceLineNo">145</span>    // conservative for now.<a name="line.145"></a>
-<span class="sourceLineNo">146</span>    this.replicationRpcLimit = (int)(0.95 * conf.getLong(RpcServer.MAX_REQUEST_SIZE,<a name="line.146"></a>
-<span class="sourceLineNo">147</span>      RpcServer.DEFAULT_MAX_REQUEST_SIZE));<a name="line.147"></a>
-<span class="sourceLineNo">148</span>    this.dropOnDeletedTables =<a name="line.148"></a>
-<span class="sourceLineNo">149</span>        this.conf.getBoolean(HConstants.REPLICATION_DROP_ON_DELETED_TABLE_KEY, false);<a name="line.149"></a>
-<span class="sourceLineNo">150</span><a name="line.150"></a>
-<span class="sourceLineNo">151</span>    this.replicationBulkLoadDataEnabled =<a name="line.151"></a>
-<span class="sourceLineNo">152</span>        conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.152"></a>
-<span class="sourceLineNo">153</span>          HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.153"></a>
-<span class="sourceLineNo">154</span>    if (this.replicationBulkLoadDataEnabled) {<a name="line.154"></a>
-<span class="sourceLineNo">155</span>      replicationClusterId = this.conf.get(HConstants.REPLICATION_CLUSTER_ID);<a name="line.155"></a>
-<span class="sourceLineNo">156</span>    }<a name="line.156"></a>
-<span class="sourceLineNo">157</span>    // Construct base namespace directory and hfile archive directory path<a name="line.157"></a>
-<span class="sourceLineNo">158</span>    Path rootDir = FSUtils.getRootDir(conf);<a name="line.158"></a>
-<span class="sourceLineNo">159</span>    Path baseNSDir = new Path(HConstants.BASE_NAMESPACE_DIR);<a name="line.159"></a>
-<span class="sourceLineNo">160</span>    baseNamespaceDir = new Path(rootDir, baseNSDir);<a name="line.160"></a>
-<span class="sourceLineNo">161</span>    hfileArchiveDir = new Path(rootDir, new Path(HConstants.HFILE_ARCHIVE_DIRECTORY, baseNSDir));<a name="line.161"></a>
-<span class="sourceLineNo">162</span>  }<a name="line.162"></a>
-<span class="sourceLineNo">163</span><a name="line.163"></a>
-<span class="sourceLineNo">164</span>  private void decorateConf() {<a name="line.164"></a>
-<span class="sourceLineNo">165</span>    String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.165"></a>
-<span class="sourceLineNo">166</span>    if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.166"></a>
-<span class="sourceLineNo">167</span>      this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.167"></a>
-<span class="sourceLineNo">168</span>    }<a name="line.168"></a>
-<span class="sourceLineNo">169</span>  }<a name="line.169"></a>
-<span class="sourceLineNo">170</span><a name="line.170"></a>
-<span class="sourceLineNo">171</span>  private void connectToPeers() {<a name="line.171"></a>
-<span class="sourceLineNo">172</span>    getRegionServers();<a name="line.172"></a>
-<span class="sourceLineNo">173</span><a name="line.173"></a>
-<span class="sourceLineNo">174</span>    int sleepMultiplier = 1;<a name="line.174"></a>
-<span class="sourceLineNo">175</span><a name="line.175"></a>
-<span class="sourceLineNo">176</span>    // Connect to peer cluster first, unless we have to stop<a name="line.176"></a>
-<span class="sourceLineNo">177</span>    while (this.isRunning() &amp;&amp; replicationSinkMgr.getNumSinks() == 0) {<a name="line.177"></a>
-<span class="sourceLineNo">178</span>      replicationSinkMgr.chooseSinks();<a name="line.178"></a>
-<span class="sourceLineNo">179</span>      if (this.isRunning() &amp;&amp; replicationSinkMgr.getNumSinks() == 0) {<a name="line.179"></a>
-<span class="sourceLineNo">180</span>        if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.180"></a>
-<span class="sourceLineNo">181</span>          sleepMultiplier++;<a name="line.181"></a>
-<span class="sourceLineNo">182</span>        }<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><a name="line.186"></a>
-<span class="sourceLineNo">187</span>  /**<a name="line.187"></a>
-<span class="sourceLineNo">188</span>   * Do the sleeping logic<a name="line.188"></a>
-<span class="sourceLineNo">189</span>   * @param msg Why we sleep<a name="line.189"></a>
-<span class="sourceLineNo">190</span>   * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.190"></a>
-<span class="sourceLineNo">191</span>   * @return True if &lt;code&gt;sleepMultiplier&lt;/code&gt; is &amp;lt; &lt;code&gt;maxRetriesMultiplier&lt;/code&gt;<a name="line.191"></a>
-<span class="sourceLineNo">192</span>   */<a name="line.192"></a>
-<span class="sourceLineNo">193</span>  protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.193"></a>
-<span class="sourceLineNo">194</span>    try {<a name="line.194"></a>
-<span class="sourceLineNo">195</span>      if (LOG.isTraceEnabled()) {<a name="line.195"></a>
-<span class="sourceLineNo">196</span>        LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.196"></a>
-<span class="sourceLineNo">197</span>      }<a name="line.197"></a>
-<span class="sourceLineNo">198</span>      Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.198"></a>
-<span class="sourceLineNo">199</span>    } catch (InterruptedException e) {<a name="line.199"></a>
-<span class="sourceLineNo">200</span>      LOG.debug("Interrupted while sleeping between retries");<a name="line.200"></a>
-<span class="sourceLineNo">201</span>    }<a name="line.201"></a>
-<span class="sourceLineNo">202</span>    return sleepMultiplier &lt; maxRetriesMultiplier;<a name="line.202"></a>
-<span class="sourceLineNo">203</span>  }<a name="line.203"></a>
-<span class="sourceLineNo">204</span><a name="line.204"></a>
-<span class="sourceLineNo">205</span>  private List&lt;List&lt;Entry&gt;&gt; createBatches(final List&lt;Entry&gt; entries) {<a name="line.205"></a>
-<span class="sourceLineNo">206</span>    int numSinks = Math.max(replicationSinkMgr.getNumSinks(), 1);<a name="line.206"></a>
-<span class="sourceLineNo">207</span>    int n = Math.min(Math.min(this.maxThreads, entries.size()/100+1), numSinks);<a name="line.207"></a>
-<span class="sourceLineNo">208</span>    // Maintains the current batch for a given partition index<a name="line.208"></a>
-<span class="sourceLineNo">209</span>    Map&lt;Integer, List&lt;Entry&gt;&gt; entryMap = new HashMap&lt;&gt;(n);<a name="line.209"></a>
-<span class="sourceLineNo">210</span>    List&lt;List&lt;Entry&gt;&gt; entryLists = new ArrayList&lt;&gt;();<a name="line.210"></a>
-<span class="sourceLineNo">211</span>    int[] sizes = new int[n];<a name="line.211"></a>
-<span class="sourceLineNo">212</span><a name="line.212"></a>
-<span class="sourceLineNo">213</span>    for (int i = 0; i &lt; n; i++) {<a name="line.213"></a>
-<span class="sourceLineNo">214</span>      entryMap.put(i, new ArrayList&lt;Entry&gt;(entries.size()/n+1));<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>    for (Entry e: entries) {<a name="line.217"></a>
-<span class="sourceLineNo">218</span>      int index = Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName())%n);<a name="line.218"></a>
-<span class="sourceLineNo">219</span>      int entrySize = (int)e.getKey().estimatedSerializedSizeOf() +<a name="line.219"></a>
-<span class="sourceLineNo">220</span>          (int)e.getEdit().estimatedSerializedSizeOf();<a name="line.220"></a>
-<span class="sourceLineNo">221</span>      // If this batch is oversized, add it to final list and initialize a new empty batch<a name="line.221"></a>
-<span class="sourceLineNo">222</span>      if (sizes[index] &gt; 0 /* must include at least one entry */ &amp;&amp;<a name="line.222"></a>
-<span class="sourceLineNo">223</span>          sizes[index] + entrySize &gt; replicationRpcLimit) {<a name="line.223"></a>
-<span class="sourceLineNo">224</span>        entryLists.add(entryMap.get(index));<a name="line.224"></a>
-<span class="sourceLineNo">225</span>        entryMap.put(index, new ArrayList&lt;Entry&gt;());<a name="line.225"></a>
-<span class="sourceLineNo">226</span>        sizes[index] = 0;<a name="line.226"></a>
-<span class="sourceLineNo">227</span>      }<a name="line.227"></a>
-<span class="sourceLineNo">228</span>      entryMap.get(index).add(e);<a name="line.228"></a>
-<span class="sourceLineNo">229</span>      sizes[index] += entrySize;<a name="line.229"></a>
-<span class="sourceLineNo">230</span>    }<a name="line.230"></a>
-<span class="sourceLineNo">231</span><a name="line.231"></a>
-<span class="sourceLineNo">232</span>    entryLists.addAll(entryMap.values());<a name="line.232"></a>
-<span class="sourceLineNo">233</span>    return entryLists;<a name="line.233"></a>
-<span class="sourceLineNo">234</span>  }<a name="line.234"></a>
-<span class="sourceLineNo">235</span><a name="line.235"></a>
-<span class="sourceLineNo">236</span>  private TableName parseTable(String msg) {<a name="line.236"></a>
-<span class="sourceLineNo">237</span>    // ... TableNotFoundException: '&lt;table&gt;'/n...<a name="line.237"></a>
-<span class="sourceLineNo">238</span>    Pattern p = Pattern.compile("TableNotFoundException: \\'([\\S]*)\\'");<a name="line.238"></a>
-<span class="sourceLineNo">239</span>    Matcher m = p.matcher(msg);<a name="line.239"></a>
-<span class="sourceLineNo">240</span>    if (m.find()) {<a name="line.240"></a>
-<span class="sourceLineNo">241</span>      String table = m.group(1);<a name="line.241"></a>
-<span class="sourceLineNo">242</span>      try {<a name="line.242"></a>
-<span class="sourceLineNo">243</span>        // double check that table is a valid table name<a name="line.243"></a>
-<span class="sourceLineNo">244</span>        TableName.valueOf(TableName.isLegalFullyQualifiedTableName(Bytes.toBytes(table)));<a name="line.244"></a>
-<span class="sourceLineNo">245</span>        return TableName.valueOf(table);<a name="line.245"></a>
-<span class="sourceLineNo">246</span>      } catch (IllegalArgumentException ignore) {<a name="line.246"></a>
-<span class="sourceLineNo">247</span>      }<a name="line.247"></a>
-<span class="sourceLineNo">248</span>    }<a name="line.248"></a>
-<span class="sourceLineNo">249</span>    return null;<a name="line.249"></a>
-<span class="sourceLineNo">250</span>  }<a name="line.250"></a>
-<span class="sourceLineNo">251</span><a name="line.251"></a>
-<span class="sourceLineNo">252</span>  // Filter a set of batches by TableName<a name="line.252"></a>
-<span class="sourceLineNo">253</span>  private List&lt;List&lt;Entry&gt;&gt; filterBatches(final List&lt;List&lt;Entry&gt;&gt; oldEntryList, TableName table) {<a name="line.253"></a>
-<span class="sourceLineNo">254</span>    List&lt;List&lt;Entry&gt;&gt; entryLists = new ArrayList&lt;&gt;();<a name="line.254"></a>
-<span class="sourceLineNo">255</span>    for (List&lt;Entry&gt; entries : oldEntryList) {<a name="line.255"></a>
-<span class="sourceLineNo">256</span>      ArrayList&lt;Entry&gt; thisList = new ArrayList&lt;Entry&gt;(entries.size());<a name="line.256"></a>
-<span class="sourceLineNo">257</span>      entryLists.add(thisList);<a name="line.257"></a>
-<span class="sourceLineNo">258</span>      for (Entry e : entries) {<a name="line.258"></a>
-<span class="sourceLineNo">259</span>        if (!e.getKey().getTablename().equals(table)) {<a name="line.259"></a>
-<span class="sourceLineNo">260</span>          thisList.add(e);<a name="line.260"></a>
-<span class="sourceLineNo">261</span>        }<a name="line.261"></a>
-<span class="sourceLineNo">262</span>      }<a name="line.262"></a>
-<span class="sourceLineNo">263</span>    }<a name="line.263"></a>
-<span class="sourceLineNo">264</span>    return entryLists;<a name="line.264"></a>
-<span class="sourceLineNo">265</span>  }<a name="line.265"></a>
-<span class="sourceLineNo">266</span><a name="line.266"></a>
-<span class="sourceLineNo">267</span>  private void reconnectToPeerCluster() {<a name="line.267"></a>
-<span class="sourceLineNo">268</span>    ClusterConnection connection = null;<a name="line.268"></a>
-<span class="sourceLineNo">269</span>    try {<a name="line.269"></a>
-<span class="sourceLineNo">270</span>      connection = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.270"></a>
-<span class="sourceLineNo">271</span>    } catch (IOException ioe) {<a name="line.271"></a>
-<span class="sourceLineNo">272</span>      LOG.warn("Failed to create connection for peer cluster", ioe);<a name="line.272"></a>
-<span class="sourceLineNo">273</span>    }<a name="line.273"></a>
-<span class="sourceLineNo">274</span>    if (connection != null) {<a name="line.274"></a>
-<span class="sourceLineNo">275</span>      this.conn = connection;<a name="line.275"></a>
-<span class="sourceLineNo">276</span>    }<a name="line.276"></a>
-<span class="sourceLineNo">277</span>  }<a name="line.277"></a>
-<span class="sourceLineNo">278</span><a name="line.278"></a>
-<span class="sourceLineNo">279</span>  /**<a name="line.279"></a>
-<span class="sourceLineNo">280</span>   * Do the shipping logic<a name="line.280"></a>
-<span class="sourceLineNo">281</span>   */<a name="line.281"></a>
-<span class="sourceLineNo">282</span>  @Override<a name="line.282"></a>
-<span class="sourceLineNo">283</span>  public boolean replicate(ReplicateContext replicateContext) {<a name="line.283"></a>
-<span class="sourceLineNo">284</span>    CompletionService&lt;Integer&gt; pool = new ExecutorCompletionService&lt;&gt;(this.exec);<a name="line.284"></a>
-<span class="sourceLineNo">285</span>    List&lt;List&lt;Entry&gt;&gt; batches;<a name="line.285"></a>
-<span class="sourceLineNo">286</span>    String walGroupId = replicateContext.getWalGroupId();<a name="line.286"></a>
-<span class="sourceLineNo">287</span>    int sleepMultiplier = 1;<a name="line.287"></a>
-<span class="sourceLineNo">288</span><a name="line.288"></a>
-<span class="sourceLineNo">289</span>    if (!peersSelected &amp;&amp; this.isRunning()) {<a name="line.289"></a>
-<span class="sourceLineNo">290</span>      connectToPeers();<a name="line.290"></a>
-<span class="sourceLineNo">291</span>      peersSelected = true;<a name="line.291"></a>
-<span class="sourceLineNo">292</span>    }<a name="line.292"></a>
-<span class="sourceLineNo">293</span><a name="line.293"></a>
-<span class="sourceLineNo">294</span>    int numSinks = replicationSinkMgr.getNumSinks();<a name="line.294"></a>
-<span class="sourceLineNo">295</span>    if (numSinks == 0) {<a name="line.295"></a>
-<span class="sourceLineNo">296</span>      LOG.warn("No replication sinks found, returning without replicating. The source should retry"<a name="line.296"></a>
-<span class="sourceLineNo">297</span>          + " with the same set of edits.");<a name="line.297"></a>
-<span class="sourceLineNo">298</span>      return false;<a name="line.298"></a>
-<span class="sourceLineNo">299</span>    }<a name="line.299"></a>
-<span class="sourceLineNo">300</span><a name="line.300"></a>
-<span class="sourceLineNo">301</span>    batches = createBatches(replicateContext.getEntries());<a name="line.301"></a>
-<span class="sourceLineNo">302</span><a name="line.302"></a>
-<span class="sourceLineNo">303</span>    while (this.isRunning() &amp;&amp; !exec.isShutdown()) {<a name="line.303"></a>
-<span class="sourceLineNo">304</span>      if (!isPeerEnabled()) {<a name="line.304"></a>
-<span class="sourceLineNo">305</span>        if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.305"></a>
-<span class="sourceLineNo">306</span>          sleepMultiplier++;<a name="line.306"></a>
-<span class="sourceLineNo">307</span>        }<a name="line.307"></a>
-<span class="sourceLineNo">308</span>        continue;<a name="line.308"></a>
-<span class="sourceLineNo">309</span>      }<a name="line.309"></a>
-<span class="sourceLineNo">310</span>      if (this.conn == null || this.conn.isClosed()) {<a name="line.310"></a>
-<span class="sourceLineNo">311</span>        reconnectToPeerCluster();<a name="line.311"></a>
-<span class="sourceLineNo">312</span>      }<a name="line.312"></a>
-<span class="sourceLineNo">313</span>      try {<a name="line.313"></a>
-<span class="sourceLineNo">314</span>        int futures = 0;<a name="line.314"></a>
-<span class="sourceLineNo">315</span>        for (int i=0; i&lt;batches.size(); i++) {<a name="line.315"></a>
-<span class="sourceLineNo">316</span>          List&lt;Entry&gt; entries = batches.get(i);<a name="line.316"></a>
-<span class="sourceLineNo">317</span>          if (!entries.isEmpty()) {<a name="line.317"></a>
-<span class="sourceLineNo">318</span>            if (LOG.isTraceEnabled()) {<a name="line.318"></a>
-<span class="sourceLineNo">319</span>              LOG.trace("Submitting " + entries.size() +<a name="line.319"></a>
-<span class="sourceLineNo">320</span>                  " entries of total size " + replicateContext.getSize());<a name="line.320"></a>
-<span class="sourceLineNo">321</span>            }<a name="line.321"></a>
-<span class="sourceLineNo">322</span>            // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.322"></a>
-<span class="sourceLineNo">323</span>            pool.submit(createReplicator(entries, i));<a name="line.323"></a>
-<span class="sourceLineNo">324</span>            futures++;<a name="line.324"></a>
-<span class="sourceLineNo">325</span>          }<a name="line.325"></a>
-<span class="sourceLineNo">326</span>        }<a name="line.326"></a>
-<span class="sourceLineNo">327</span>        IOException iox = null;<a name="line.327"></a>
-<span class="sourceLineNo">328</span><a name="line.328"></a>
-<span class="sourceLineNo">329</span>        long lastWriteTime = 0;<a name="line.329"></a>
-<span class="sourceLineNo">330</span>        for (int i=0; i&lt;futures; i++) {<a name="line.330"></a>
-<span class="sourceLineNo">331</span>          try {<a name="line.331"></a>
-<span class="sourceLineNo">332</span>            // wait for all futures, remove successful parts<a name="line.332"></a>
-<span class="sourceLineNo">333</span>            // (only the remaining parts will be retried)<a name="line.333"></a>
-<span class="sourceLineNo">334</span>            Future&lt;Integer&gt; f = pool.take();<a name="line.334"></a>
-<span class="sourceLineNo">335</span>            int index = f.get().intValue();<a name="line.335"></a>
-<span class="sourceLineNo">336</span>            List&lt;Entry&gt; batch = batches.get(index);<a name="line.336"></a>
-<span class="sourceLineNo">337</span>            batches.set(index, Collections.&lt;Entry&gt;emptyList()); // remove successful batch<a name="line.337"></a>
-<span class="sourceLineNo">338</span>            // Find the most recent write time in the batch<a name="line.338"></a>
-<span class="sourceLineNo">339</span>            long writeTime = batch.get(batch.size() - 1).getKey().getWriteTime();<a name="line.339"></a>
-<span class="sourceLineNo">340</span>            if (writeTime &gt; lastWriteTime) {<a name="line.340"></a>
-<span class="sourceLineNo">341</span>              lastWriteTime = writeTime;<a name="line.341"></a>
-<span class="sourceLineNo">342</span>            }<a name="line.342"></a>
-<span class="sourceLineNo">343</span>          } catch (InterruptedException ie) {<a name="line.343"></a>
-<span class="sourceLineNo">344</span>            iox =  new IOException(ie);<a name="line.344"></a>
-<span class="sourceLineNo">345</span>          } catch (ExecutionException ee) {<a name="line.345"></a>
-<span class="sourceLineNo">346</span>            // cause must be an IOException<a name="line.346"></a>
-<span class="sourceLineNo">347</span>            iox = (IOException)ee.getCause();<a name="line.347"></a>
-<span class="sourceLineNo">348</span>          }<a name="line.348"></a>
-<span class="sourceLineNo">349</span>        }<a name="line.349"></a>
-<span class="sourceLineNo">350</span>        if (iox != null) {<a name="line.350"></a>
-<span class="sourceLineNo">351</span>          // if we had any exceptions, try again<a name="line.351"></a>
-<span class="sourceLineNo">352</span>          throw iox;<a name="line.352"></a>
-<span class="sourceLineNo">353</span>        }<a name="line.353"></a>
-<span class="sourceLineNo">354</span>        // update metrics<a name="line.354"></a>
-<span class="sourceLineNo">355</span>        if (lastWriteTime &gt; 0) {<a name="line.355"></a>
-<span class="sourceLineNo">356</span>          this.metrics.setAgeOfLastShippedOp(lastWriteTime, walGroupId);<a name="line.356"></a>
-<span class="sourceLineNo">357</span>        }<a name="line.357"></a>
-<span class="sourceLineNo">358</span>        return true;<a name="line.358"></a>
-<span class="sourceLineNo">359</span><a name="line.359"></a>
-<span class="sourceLineNo">360</span>      } catch (IOException ioe) {<a name="line.360"></a>
-<span class="sourceLineNo">361</span>        // Didn't ship anything, but must still age the last time we did<a name="line.361"></a>
-<span class="sourceLineNo">362</span>        this.metrics.refreshAgeOfLastShippedOp(walGroupId);<a name="line.362"></a>
-<span class="sourceLineNo">363</span>        if (ioe instanceof RemoteException) {<a name="line.363"></a>
-<span class="sourceLineNo">364</span>          ioe = ((RemoteException) ioe).unwrapRemoteException();<a name="line.364"></a>
-<span class="sourceLineNo">365</span>          LOG.warn("Can't replicate because of an error on the remote cluster: ", ioe);<a name="line.365"></a>
-<span class="sourceLineNo">366</span>          if (ioe instanceof TableNotFoundException) {<a name="line.366"></a>
-<span class="sourceLineNo">367</span>            if (dropOnDeletedTables) {<a name="line.367"></a>
-<span class="sourceLineNo">368</span>              // this is a bit fragile, but cannot change how TNFE is serialized<a name="line.368"></a>
-<span class="sourceLineNo">369</span>              // at least check whether the table name is legal<a name="line.369"></a>
-<span class="sourceLineNo">370</span>              TableName table = parseTable(ioe.getMessage());<a name="line.370"></a>
-<span class="sourceLineNo">371</span>              if (table != null) {<a name="line.371"></a>
-<span class="sourceLineNo">372</span>                try (Connection localConn =<a name="line.372"></a>
-<span class="sourceLineNo">373</span>                    ConnectionFactory.createConnection(ctx.getLocalConfiguration())) {<a name="line.373"></a>
-<span class="sourceLineNo">374</span>                  if (!localConn.getAdmin().tableExists(table)) {<a name="line.374"></a>
-<span class="sourceLineNo">375</span>                    // Would potentially be better to retry in one of the outer loops<a name="line.375"></a>
-<span class="sourceLineNo">376</span>                    // and add a table filter there; but that would break the encapsulation,<a name="line.376"></a>
-<span class="sourceLineNo">377</span>                    // so we're doing the filtering here.<a name="line.377"></a>
-<span class="sourceLineNo">378</span>                    LOG.info("Missing table detected at sink, local table also does not exist, filtering edits for '"+table+"'");<a name="line.378"></a>
-<span class="sourceLineNo">379</span>                    batches = filterBatches(batches, table);<a name="line.379"></a>
-<span class="sourceLineNo">380</span>                    continue;<a name="line.380"></a>
-<span class="sourceLineNo">381</span>                  }<a name="line.381"></a>
-<span class="sourceLineNo">382</span>                } catch (IOException iox) {<a name="line.382"></a>
-<span class="sourceLineNo">383</span>                  LOG.warn("Exception checking for local table: ", iox);<a name="line.383"></a>
-<span class="sourceLineNo">384</span>                }<a name="line.384"></a>
-<span class="sourceLineNo">385</span>              }<a name="line.385"></a>
-<span class="sourceLineNo">386</span>            }<a name="line.386"></a>
-<span class="sourceLineNo">387</span>            // fall through and sleep below<a name="line.387"></a>
-<span class="sourceLineNo">388</span>          } else {<a name="line.388"></a>
-<span class="sourceLineNo">389</span>            LOG.warn("Peer encountered RemoteException, rechecking all sinks: ", ioe);<a name="line.389"></a>
-<span class="sourceLineNo">390</span>            replicationSinkMgr.chooseSinks();<a name="line.390"></a>
-<span class="sourceLineNo">391</span>          }<a name="line.391"></a>
-<span class="sourceLineNo">392</span>        } else {<a name="line.392"></a>
-<span class="sourceLineNo">393</span>          if (ioe instanceof SocketTimeoutException) {<a name="line.393"></a>
-<span class="sourceLineNo">394</span>            // This exception means we waited for more than 60s and nothing<a name="line.394"></a>
-<span class="sourceLineNo">395</span>            // happened, the cluster is alive and calling it right away<a name="line.395"></a>
-<span class="sourceLineNo">396</span>            // even for a test just makes things worse.<a name="line.396"></a>
-<span class="sourceLineNo">397</span>            sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.397"></a>
-<span class="sourceLineNo">398</span>              "call to the remote cluster timed out, which is usually " +<a name="line.398"></a>
-<span class="sourceLineNo">399</span>              "caused by a machine failure or a massive slowdown",<a name="line.399"></a>
-<span class="sourceLineNo">400</span>              this.socketTimeoutMultiplier);<a name="line.400"></a>
-<span class="sourceLineNo">401</span>          } else if (ioe instanceof ConnectException) {<a name="line.401"></a>
-<span class="sourceLineNo">402</span>            LOG.warn("Peer is unavailable, rechecking all sinks: ", ioe);<a name="line.402"></a>
-<span class="sourceLineNo">403</span>            replicationSinkMgr.chooseSinks();<a name="line.403"></a>
-<span class="sourceLineNo">404</span>          } else {<a name="line.404"></a>
-<span class="sourceLineNo">405</span>            LOG.warn("Can't replicate because of a local or network error: ", ioe);<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>        if (sleepForRetries("Since we are unable to replicate", sleepMultiplier)) {<a name="line.408"></a>
-<span class="sourceLineNo">409</span>          sleepMultiplier++;<a name="line.409"></a>
-<span class="sourceLineNo">410</span>        }<a name="line.410"></a>
-<span class="sourceLineNo">411</span>      }<a name="line.411"></a>
-<span class="sourceLineNo">412</span>    }<a name="line.412"></a>
-<span class="sourceLineNo">413</span>    return false; // in case we exited before replicating<a name="line.413"></a>
-<span class="sourceLineNo">414</span>  }<a name="line.414"></a>
-<span class="sourceLineNo">415</span><a name="line.415"></a>
-<span class="sourceLineNo">416</span>  protected boolean isPeerEnabled() {<a name="line.416"></a>
-<span class="sourceLineNo">417</span>    return ctx.getReplicationPeer().isPeerEnabled();<a name="line.417"></a>
-<span class="sourceLineNo">418</span>  }<a name="line.418"></a>
-<span class="sourceLineNo">419</span><a name="line.419"></a>
-<span class="sourceLineNo">420</span>  @Override<a name="line.420"></a>
-<span class="sourceLineNo">421</span>  protected void doStop() {<a name="line.421"></a>
-<span class="sourceLineNo">422</span>    disconnect(); //don't call super.doStop()<a name="line.422"></a>
-<span class="sourceLineNo">423</span>    if (this.conn != null) {<a name="line.423"></a>
-<span class="sourceLineNo">424</span>      try {<a name="line.424"></a>
-<span class="sourceLineNo">425</span>        this.conn.close();<a name="line.425"></a>
-<span class="sourceLineNo">426</span>        this.conn = null;<a name="line.426"></a>
-<span class="sourceLineNo">427</span>      } catch (IOException e) {<a name="line.427"></a>
-<span class="sourceLineNo">428</span>        LOG.warn("Failed to close the connection");<a name="line.428"></a>
-<span class="sourceLineNo">429</span>      }<a name="line.429"></a>
-<span class="sourceLineNo">430</span>    }<a name="line.430"></a>
-<span class="sourceLineNo">431</span>    // Allow currently running replication tasks to finish<a name="line.431"></a>
-<span class="sourceLineNo">432</span>    exec.shutdown();<a name="line.432"></a>
-<span class="sourceLineNo">433</span>    try {<a name="line.433"></a>
-<span class="sourceLineNo">434</span>      exec.awaitTermination(maxTerminationWait, TimeUnit.MILLISECONDS);<a name="line.434"></a>
-<span class="sourceLineNo">435</span>    } catch (InterruptedException e) {<a name="line.435"></a>
-<span class="sourceLineNo">436</span>    }<a name="line.436"></a>
-<span class="sourceLineNo">437</span>    // Abort if the tasks did not terminate in time<a name="line.437"></a>
-<span class="sourceLineNo">438</span>    if (!exec.isTerminated()) {<a name="line.438"></a>
-<span class="sourceLineNo">439</span>      String errMsg = "HBaseInterClusterReplicationEndpoint termination failed. The " +<a name="line.439"></a>
-<span class="sourceLineNo">440</span>          "ThreadPoolExecutor failed to finish all tasks within " + maxTerminationWait + "ms. " +<a name="line.440"></a>
-<span class="sourceLineNo">441</span>          "Aborting to prevent Replication from deadlocking. See HBASE-16081.";<a name="line.441"></a>
-<span class="sourceLineNo">442</span>      abortable.abort(errMsg, new IOException(errMsg));<a name="line.442"></a>
-<span class="sourceLineNo">443</span>    }<a name="line.443"></a>
-<span class="sourceLineNo">444</span>    notifyStopped();<a name="line.444"></a>
-<span class="sourceLineNo">445</span>  }<a name="line.445"></a>
-<span class="sourceLineNo">446</span><a name="line.446"></a>
-<span class="sourceLineNo">447</span>  @VisibleForTesting<a name="line.447"></a>
-<span class="sourceLineNo">448</span>  protected Replicator createReplicator(List&lt;Entry&gt; entries, int ordinal) {<a name="line.448"></a>
-<span class="sourceLineNo">449</span>    return new Replicator(entries, ordinal);<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>  @VisibleForTesting<a name="line.452"></a>
-<span class="sourceLineNo">453</span>  protected class Replicator implements Callable&lt;Integer&gt; {<a name="line.453"></a>
-<span class="sourceLineNo">454</span>    private List&lt;Entry&gt; entries;<a name="line.454"></a>
-<span class="sourceLineNo">455</span>    private int ordinal;<a name="line.455"></a>
-<span class="sourceLineNo">456</span>    public Replicator(List&lt;Entry&gt; entries, int ordinal) {<a name="line.456"></a>
-<span class="sourceLineNo">457</span>      this.entries = entries;<a name="line.457"></a>
-<span class="sourceLineNo">458</span>      this.ordinal = ordinal;<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 void replicateEntries(BlockingInterface rrs, final List&lt;Entry&gt; batch,<a name="line.461"></a>
-<span class="sourceLineNo">462</span>        String replicationClusterId, Path baseNamespaceDir, Path hfileArchiveDir)<a name="line.462"></a>
-<span class="sourceLineNo">463</span>        throws IOException {<a name="line.463"></a>
-<span class="sourceLineNo">464</span>      if (LOG.isTraceEnabled()) {<a name="line.464"></a>
-<span class="sourceLineNo">465</span>        long size = 0;<a name="line.465"></a>
-<span class="sourceLineNo">466</span>        for (Entry e: entries) {<a name="line.466"></a>
-<span class="sourceLineNo">467</span>          size += e.getKey().estimatedSerializedSizeOf();<a name="line.467"></a>
-<span class="sourceLineNo">468</span>          size += e.getEdit().estimatedSerializedSizeOf();<a name="line.468"></a>
-<span class="sourceLineNo">469</span>        }<a name="line.469"></a>
-<span class="sourceLineNo">470</span>        LOG.trace("Replicating batch " + System.identityHashCode(entries) + " of " +<a name="line.470"></a>
-<span class="sourceLineNo">471</span>            entries.size() + " entries with total size " + size + " bytes to " +<a name="line.471"></a>
-<span class="sourceLineNo">472</span>            replicationClusterId);<a name="line.472"></a>
-<span class="sourceLineNo">473</span>      }<a name="line.473"></a>
-<span class="sourceLineNo">474</span>      try {<a name="line.474"></a>
-<span class="sourceLineNo">475</span>        ReplicationProtbufUtil.replicateWALEntry(rrs, batch.toArray(new Entry[batch.size()]),<a name="line.475"></a>
-<span class="sourceLineNo">476</span>          replicationClusterId, baseNamespaceDir, hfileArchiveDir);<a name="line.476"></a>
-<span class="sourceLineNo">477</span>        if (LOG.isTraceEnabled()) {<a name="line.477"></a>
-<span class="sourceLineNo">478</span>          LOG.trace("Completed replicating batch " + System.identityHashCode(entries));<a name="line.478"></a>
-<span class="sourceLineNo">479</span>        }<a name="line.479"></a>
-<span class="sourceLineNo">480</span>      } catch (IOException e) {<a name="line.480"></a>
-<span class="sourceLineNo">481</span>        if (LOG.isTraceEnabled()) {<a name="line.481"></a>
-<span class="sourceLineNo">482</span>          LOG.trace("Failed replicating batch " + System.identityHashCode(entries), e);<a name="line.482"></a>
-<span class="sourceLineNo">483</span>        }<a name="line.483"></a>
-<span class="sourceLineNo">484</span>        throw e;<a name="line.484"></a>
-<span class="sourceLineNo">485</span>      }<a name="line.485"></a>
-<span class="sourceLineNo">486</span>    }<a name="line.486"></a>
-<span class="sourceLineNo">487</span><a name="line.487"></a>
-<span class="sourceLineNo">488</span>    @Override<a name="line.488"></a>
-<span class="sourceLineNo">489</span>    public Integer call() throws IOException {<a name="line.489"></a>
-<span class="sourceLineNo">490</span>      SinkPeer sinkPeer = null;<a name="line.490"></a>
-<span class="sourceLineNo">491</span>      try {<a name="line.491"></a>
-<span class="sourceLineNo">492</span>        sinkPeer = replicationSinkMgr.getReplicationSink();<a name="line.492"></a>
-<span class="sourceLineNo">493</span>        BlockingInterface rrs = sinkPeer.getRegionServer();<a name="line.493"></a>
-<span class="sourceLineNo">494</span>        replicateEntries(rrs, entries, replicationClusterId, baseNamespaceDir, hfileArchiveDir);<a name="line.494"></a>
-<span class="sourceLineNo">495</span>        replicationSinkMgr.reportSinkSuccess(sinkPeer);<a name="line.495"></a>
-<span class="sourceLineNo">496</span>        return ordinal;<a name="line.496"></a>
-<span class="sourceLineNo">497</span>      } catch (IOException ioe) {<a name="line.497"></a>
-<span class="sourceLineNo">498</span>        if (sinkPeer != null) {<a name="line.498"></a>
-<span class="sourceLineNo">499</span>          replicationSinkMgr.reportBadSink(sinkPeer);<a name="line.499"></a>
-<span class="sourceLineNo">500</span>        }<a name="line.500"></a>
-<span class="sourceLineNo">501</span>        throw ioe;<a name="line.501"></a>
-<span class="sourceLineNo">502</span>      }<a name="line.502"></a>
-<span class="sourceLineNo">503</span>    }<a name="line.503"></a>
-<span class="sourceLineNo">504</span>  }<a name="line.504"></a>
-<span class="sourceLineNo">505</span>}<a name="line.505"></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.List;<a name="line.28"></a>
+<span class="sourceLineNo">029</span>import java.util.Map;<a name="line.29"></a>
+<span class="sourceLineNo">030</span>import java.util.concurrent.Callable;<a name="line.30"></a>
+<span class="sourceLineNo">031</span>import java.util.concurrent.CompletionService;<a name="line.31"></a>
+<span class="sourceLineNo">032</span>import java.util.concurrent.ExecutionException;<a name="line.32"></a>
+<span class="sourceLineNo">033</span>import java.util.concurrent.ExecutorCompletionService;<a name="line.33"></a>
+<span class="sourceLineNo">034</span>import java.util.concurrent.Future;<a name="line.34"></a>
+<span class="sourceLineNo">035</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.35"></a>
+<span class="sourceLineNo">036</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.36"></a>
+<span class="sourceLineNo">037</span>import java.util.concurrent.TimeUnit;<a name="line.37"></a>
+<span class="sourceLineNo">038</span>import java.util.regex.Matcher;<a name="line.38"></a>
+<span class="sourceLineNo">039</span>import java.util.regex.Pattern;<a name="line.39"></a>
+<span class="sourceLineNo">040</span>import org.apache.commons.lang3.StringUtils;<a name="line.40"></a>
+<span class="sourceLineNo">041</span>import org.apache.hadoop.conf.Configuration;<a name="line.41"></a>
+<span class="sourceLineNo">042</span>import org.apache.hadoop.fs.Path;<a name="line.42"></a>
+<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.Abortable;<a name="line.43"></a>
+<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.44"></a>
+<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.HConstants;<a name="line.45"></a>
+<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.TableName;<a name="line.46"></a>
+<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.47"></a>
+<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.48"></a>
+<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.49"></a>
+<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.50"></a>
+<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.ipc.RpcServer;<a name="line.51"></a>
+<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtbufUtil;<a name="line.52"></a>
+<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.53"></a>
+<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.regionserver.ReplicationSinkManager.SinkPeer;<a name="line.54"></a>
+<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.55"></a>
+<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.56"></a>
+<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.57"></a>
+<span class="sourceLineNo">058</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.58"></a>
+<span class="sourceLineNo">059</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.59"></a>
+<span class="sourceLineNo">060</span>import org.slf4j.Logger;<a name="line.60"></a>
+<span class="sourceLineNo">061</span>import org.slf4j.LoggerFactory;<a name="line.61"></a>
+<span class="sourceLineNo">062</span><a name="line.62"></a>
+<span class="sourceLineNo">063</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.63"></a>
+<span class="sourceLineNo">064</span><a name="line.64"></a>
+<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService.BlockingInterface;<a name="line.65"></a>
+<span class="sourceLineNo">066</span><a name="line.66"></a>
+<span class="sourceLineNo">067</span>/**<a name="line.67"></a>
+<span class="sourceLineNo">068</span> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint}<a name="line.68"></a>
+<span class="sourceLineNo">069</span> * implementation for replicating to another HBase cluster.<a name="line.69"></a>
+<span class="sourceLineNo">070</span> * For the slave cluster it selects a random number of peers<a name="line.70"></a>
+<span class="sourceLineNo">071</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.71"></a>
+<span class="sourceLineNo">072</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.72"></a>
+<span class="sourceLineNo">073</span> * &lt;p&gt;<a name="line.73"></a>
+<span class="sourceLineNo">074</span> * A stream is considered down when we cannot contact a region server on the<a name="line.74"></a>
+<span class="sourceLineNo">075</span> * peer cluster for more than 55 seconds by default.<a name="line.75"></a>
+<span class="sourceLineNo">076</span> * &lt;/p&gt;<a name="line.76"></a>
+<span class="sourceLineNo">077</span> */<a name="line.77"></a>
+<span class="sourceLineNo">078</span>@InterfaceAudience.Private<a name="line.78"></a>
+<span class="sourceLineNo">079</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.79"></a>
+<span class="sourceLineNo">080</span>  private static final Logger LOG =<a name="line.80"></a>
+<span class="sourceLineNo">081</span>      LoggerFactory.getLogger(HBaseInterClusterReplicationEndpoint.class);<a name="line.81"></a>
+<span class="sourceLineNo">082</span><a name="line.82"></a>
+<span class="sourceLineNo">083</span>  private static final long DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER = 2;<a name="line.83"></a>
+<span class="sourceLineNo">084</span><a name="line.84"></a>
+<span class="sourceLineNo">085</span>  private ClusterConnection conn;<a name="line.85"></a>
+<span class="sourceLineNo">086</span>  private Configuration localConf;<a name="line.86"></a>
+<span class="sourceLineNo">087</span>  private Configuration conf;<a name="line.87"></a>
+<span class="sourceLineNo">088</span>  // How long should we sleep for each retry<a name="line.88"></a>
+<span class="sourceLineNo">089</span>  private long sleepForRetries;<a name="line.89"></a>
+<span class="sourceLineNo">090</span>  // Maximum number of retries before taking bold actions<a name="line.90"></a>
+<span class="sourceLineNo">091</span>  private int maxRetriesMultiplier;<a name="line.91"></a>
+<span class="sourceLineNo">092</span>  // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.92"></a>
+<span class="sourceLineNo">093</span>  private int socketTimeoutMultiplier;<a name="line.93"></a>
+<span class="sourceLineNo">094</span>  // Amount of time for shutdown to wait for all tasks to complete<a name="line.94"></a>
+<span class="sourceLineNo">095</span>  private long maxTerminationWait;<a name="line.95"></a>
+<span class="sourceLineNo">096</span>  // Size limit for replication RPCs, in bytes<a name="line.96"></a>
+<span class="sourceLineNo">097</span>  private int replicationRpcLimit;<a name="line.97"></a>
+<span class="sourceLineNo">098</span>  //Metrics for this source<a name="line.98"></a>
+<span class="sourceLineNo">099</span>  private MetricsSource metrics;<a name="line.99"></a>
+<span class="sourceLineNo">100</span>  // Handles connecting to peer region servers<a name="line.100"></a>
+<span class="sourceLineNo">101</span>  private ReplicationSinkManager replicationSinkMgr;<a name="line.101"></a>
+<span class="sourceLineNo">102</span>  private boolean peersSelected = false;<a name="line.102"></a>
+<span class="sourceLineNo">103</span>  private String replicationClusterId = "";<a name="line.103"></a>
+<span class="sourceLineNo">104</span>  private ThreadPoolExecutor exec;<a name="line.104"></a>
+<span class="sourceLineNo">105</span>  private int maxThreads;<a name="line.105"></a>
+<span class="sourceLineNo">106</span>  private Path baseNamespaceDir;<a name="line.106"></a>
+<span class="sourceLineNo">107</span>  private Path hfileArchiveDir;<a name="line.107"></a>
+<span class="sourceLineNo">108</span>  private boolean replicationBulkLoadDataEnabled;<a name="line.108"></a>
+<span class="sourceLineNo">109</span>  private Abortable abortable;<a name="line.109"></a>
+<span class="sourceLineNo">110</span>  private boolean dropOnDeletedTables;<a name="line.110"></a>
+<span class="sourceLineNo">111</span><a name="line.111"></a>
+<span class="sourceLineNo">112</span>  @Override<a name="line.112"></a>
+<span class="sourceLineNo">113</span>  public void init(Context context) throws IOException {<a name="line.113"></a>
+<span class="sourceLineNo">114</span>    super.init(context);<a name="line.114"></a>
+<span class="sourceLineNo">115</span>    this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.115"></a>
+<span class="sourceLineNo">116</span>    this.localConf = HBaseConfiguration.create(ctx.getLocalConfiguration());<a name="line.116"></a>
+<span class="sourceLineNo">117</span>    decorateConf();<a name="line.117"></a>
+<span class="sourceLineNo">118</span>    this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.118"></a>
+<span class="sourceLineNo">119</span>    this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.119"></a>
+<span class="sourceLineNo">120</span>        maxRetriesMultiplier);<a name="line.120"></a>
+<span class="sourceLineNo">121</span>    // A Replicator job is bound by the RPC timeout. We will wait this long for all Replicator<a name="line.121"></a>
+<span class="sourceLineNo">122</span>    // tasks to terminate when doStop() is called.<a name="line.122"></a>
+<span class="sourceLineNo">123</span>    long maxTerminationWaitMultiplier = this.conf.getLong(<a name="line.123"></a>
+<span class="sourceLineNo">124</span>        "replication.source.maxterminationmultiplier",<a name="line.124"></a>
+<span class="sourceLineNo">125</span>        DEFAULT_MAX_TERMINATION_WAIT_MULTIPLIER);<a name="line.125"></a>
+<span class="sourceLineNo">126</span>    this.maxTerminationWait = maxTerminationWaitMultiplier *<a name="line.126"></a>
+<span class="sourceLineNo">127</span>        this.conf.getLong(HConstants.HBASE_RPC_TIMEOUT_KEY, HConstants.DEFAULT_HBASE_RPC_TIMEOUT);<a name="line.127"></a>
+<span class="sourceLineNo">128</span>    // TODO: This connection is replication specific or we should make it particular to<a name="line.128"></a>
+<span class="sourceLineNo">129</span>    // replication and make replication specific settings such as compression or codec to use<a name="line.129"></a>
+<span class="sourceLineNo">130</span>    // passing Cells.<a name="line.130"></a>
+<span class="sourceLineNo">131</span>    this.conn = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.131"></a>
+<span class="sourceLineNo">132</span>    this.sleepForRetries =<a name="line.132"></a>
+<span class="sourceLineNo">133</span>        this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.133"></a>
+<span class="sourceLineNo">134</span>    this.metrics = context.getMetrics();<a name="line.134"></a>
+<span class="sourceLineNo">135</span>    // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.135"></a>
+<span class="sourceLineNo">136</span>    this.replicationSinkMgr = new ReplicationSinkManager(conn, ctx.getPeerId(), this, this.conf);<a name="line.136"></a>
+<span class="sourceLineNo">137</span>    // per sink thread pool<a name="line.137"></a>
+<span class="sourceLineNo">138</span>    this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.138"></a>
+<span class="sourceLineNo">139</span>      HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.139"></a>
+<span class="sourceLineNo">140</span>    this.exec = new ThreadPoolExecutor(maxThreads, maxThreads, 60, TimeUnit.SECONDS,<a name="line.140"></a>
+<span class="sourceLineNo">141</span>        new LinkedBlockingQueue&lt;&gt;());<a name="line.141"></a>
+<span class="sourceLineNo">142</span>    this.exec.allowCoreThreadTimeOut(true);<a name="line.142"></a>
+<span class="sourceLineNo">143</span>    this.abortable = ctx.getAbortable();<a name="line.143"></a>
+<span class="sourceLineNo">144</span>    // Set the size limit for replication RPCs to 95% of the max request size.<a name="line.144"></a>
+<span class="sourceLineNo">145</span>    // We could do with less slop if we have an accurate estimate of encoded size. Being<a name="line.145"></a>
+<span class="sourceLineNo">146</span>    // conservative for now.<a name="line.146"></a>
+<span class="sourceLineNo">147</span>    this.replicationRpcLimit = (int)(0.95 * conf.getLong(RpcServer.MAX_REQUEST_SIZE,<a name="line.147"></a>
+<span class="sourceLineNo">148</span>      RpcServer.DEFAULT_MAX_REQUEST_SIZE));<a name="line.148"></a>
+<span class="sourceLineNo">149</span>    this.dropOnDeletedTables =<a name="line.149"></a>
+<span class="sourceLineNo">150</span>        this.conf.getBoolean(HConstants.REPLICATION_DROP_ON_DELETED_TABLE_KEY, false);<a name="line.150"></a>
+<span class="sourceLineNo">151</span><a name="line.151"></a>
+<span class="sourceLineNo">152</span>    this.replicationBulkLoadDataEnabled =<a name="line.152"></a>
+<span class="sourceLineNo">153</span>        conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.153"></a>
+<span class="sourceLineNo">154</span>          HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.154"></a>
+<span class="sourceLineNo">155</span>    if (this.replicationBulkLoadDataEnabled) {<a name="line.155"></a>
+<span class="sourceLineNo">156</span>      replicationClusterId = this.conf.get(HConstants.REPLICATION_CLUSTER_ID);<a name="line.156"></a>
+<span class="sourceLineNo">157</span>    }<a name="line.157"></a>
+<span class="sourceLineNo">158</span>    // Construct base namespace directory and hfile archive directory path<a name="line.158"></a>
+<span class="sourceLineNo">159</span>    Path rootDir = FSUtils.getRootDir(conf);<a name="line.159"></a>
+<span class="sourceLineNo">160</span>    Path baseNSDir = new Path(HConstants.BASE_NAMESPACE_DIR);<a name="line.160"></a>
+<span class="sourceLineNo">161</span>    baseNamespaceDir = new Path(rootDir, baseNSDir);<a name="line.161"></a>
+<span class="sourceLineNo">162</span>    hfileArchiveDir = new Path(rootDir, new Path(HConstants.HFILE_ARCHIVE_DIRECTORY, baseNSDir));<a name="line.162"></a>
+<span class="sourceLineNo">163</span>  }<a name="line.163"></a>
+<span class="sourceLineNo">164</span><a name="line.164"></a>
+<span class="sourceLineNo">165</span>  private void decorateConf() {<a name="line.165"></a>
+<span class="sourceLineNo">166</span>    String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.166"></a>
+<span class="sourceLineNo">167</span>    if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.167"></a>
+<span class="sourceLineNo">168</span>      this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.168"></a>
+<span class="sourceLineNo">169</span>    }<a name="line.169"></a>
+<span class="sourceLineNo">170</span>  }<a name="line.170"></a>
+<span class="sourceLineNo">171</span><a name="line.171"></a>
+<span class="sourceLineNo">172</span>  private void connectToPeers() {<a name="line.172"></a>
+<span class="sourceLineNo">173</span>    getRegionServers();<a name="line.173"></a>
+<span class="sourceLineNo">174</span><a name="line.174"></a>
+<span class="sourceLineNo">175</span>    int sleepMultiplier = 1;<a name="line.175"></a>
+<span class="sourceLineNo">176</span><a name="line.176"></a>
+<span class="sourceLineNo">177</span>    // Connect to peer cluster first, unless we have to stop<a name="line.177"></a>
+<span class="sourceLineNo">178</span>    while (this.isRunning() &amp;&amp; replicationSinkMgr.getNumSinks() == 0) {<a name="line.178"></a>
+<span class="sourceLineNo">179</span>      replicationSinkMgr.chooseSinks();<a name="line.179"></a>
+<span class="sourceLineNo">180</span>      if (this.isRunning() &amp;&amp; replicationSinkMgr.getNumSinks() == 0) {<a name="line.180"></a>
+<span class="sourceLineNo">181</span>        if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.181"></a>
+<span class="sourceLineNo">182</span>          sleepMultiplier++;<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>  }<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>   * Do the sleeping logic<a name="line.189"></a>
+<span class="sourceLineNo">190</span>   * @param msg Why we sleep<a name="line.190"></a>
+<span class="sourceLineNo">191</span>   * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.191"></a>
+<span class="sourceLineNo">192</span>   * @return True if &lt;code&gt;sleepMultiplier&lt;/code&gt; is &amp;lt; &lt;code&gt;maxRetriesMultiplier&lt;/code&gt;<a name="line.192"></a>
+<span class="sourceLineNo">193</span>   */<a name="line.193"></a>
+<span class="sourceLineNo">194</span>  protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.194"></a>
+<span class="sourceLineNo">195</span>    try {<a name="line.195"></a>
+<span class="sourceLineNo">196</span>      if (LOG.isTraceEnabled()) {<a name="line.196"></a>
+<span class="sourceLineNo">197</span>        LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.197"></a>
+<span class="sourceLineNo">198</span>      }<a name="line.198"></a>
+<span class="sourceLineNo">199</span>      Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.199"></a>
+<span class="sourceLineNo">200</span>    } catch (InterruptedException e) {<a name="line.200"></a>
+<span class="sourceLineNo">201</span>      LOG.debug("Interrupted while sleeping between retries");<a name="line.201"></a>
+<span class="sourceLineNo">202</span>    }<a name="line.202"></a>
+<span class="sourceLineNo">203</span>    return sleepMultiplier &lt; maxRetriesMultiplier;<a name="line.203"></a>
+<span class="sourceLineNo">204</span>  }<a name="line.204"></a>
+<span class="sourceLineNo">205</span><a name="line.205"></a>
+<span class="sourceLineNo">206</span>  private List&lt;List&lt;Entry&gt;&gt; createBatches(final List&lt;Entry&gt; entries) {<a name="line.206"></a>
+<span class="sourceLineNo">207</span>    int numSinks = Math.max(replicationSinkMgr.getNumSinks(), 1);<a name="line.207"></a>
+<span class="sourceLineNo">208</span>    int n = Math.min(Math.min(this.maxThreads, entries.size()/100+1), numSinks);<a name="line.208"></a>
+<span class="sourceLineNo">209</span>    // Maintains the current batch for a given partition index<a name="line.209"></a>
+<span class="sourceLineNo">210</span>    Map&lt;Integer, List&lt;Entry&gt;&gt; entryMap = new HashMap&lt;&gt;(n);<a name="line.210"></a>
+<span class="sourceLineNo">211</span>    List&lt;List&lt;Entry&gt;&gt; entryLists = new ArrayList&lt;&gt;();<a name="line.211"></a>
+<span class="sourceLineNo">212</span>    int[] sizes = new int[n];<a name="line.212"></a>
+<span class="sourceLineNo">213</span><a name="line.213"></a>
+<span class="sourceLineNo">214</span>    for (int i = 0; i &lt; n; i++) {<a name="line.214"></a>
+<span class="sourceLineNo">215</span>      entryMap.put(i, new ArrayList&lt;Entry&gt;(entries.size()/n+1));<a name="line.215"></a>
+<span class="sourceLineNo">216</span>    }<a name="line.216"></a>
+<span class="sourceLineNo">217</span><a name="line.217"></a>
+<span class="sourceLineNo">218</span>    for (Entry e: entries) {<a name="line.218"></a>
+<span class="sourceLineNo">219</span>      int index = Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName())%n);<a name="line.219"></a>
+<span class="sourceLineNo">220</span>      int entrySize = (int)e.getKey().estimatedSerializedSizeOf() +<a name="line.220"></a>
+<span class="sourceLineNo">221</span>          (int)e.getEdit().estimatedSerializedSizeOf();<a name="line.221"></a>
+<span class="sourceLineNo">222</span>      // If this batch is oversized, add it to final list and initialize a new empty batch<a name="line.222"></a>
+<span class="sourceLineNo">223</span>      if (sizes[index] &gt; 0 /* must include at least one entry */ &amp;&amp;<a name="line.223"></a>
+<span class="sourceLineNo">224</span>          sizes[index] + entrySize &gt; replicationRpcLimit) {<a name="line.224"></a>
+<span class="sourceLineNo">225</span>        entryLists.add(entryMap.get(index));<a name="line.225"></a>
+<span class="sourceLineNo">226</span>        entryMap.put(index, new ArrayList&lt;Entry&gt;());<a name="line.226"></a>
+<span class="sourceLineNo">227</span>        sizes[index] = 0;<a name="line.227"></a>
+<span class="sourceLineNo">228</span>      }<a name="line.228"></a>
+<span class="sourceLineNo">229</span>      entryMap.get(index).add(e);<a name="line.229"></a>
+<span class="sourceLineNo">230</span>      sizes[index] += entrySize;<a name="line.230"></a>
+<span class="sourceLineNo">231</span>    }<a name="line.231"></a>
+<span class="sourceLineNo">232</span><a name="line.232"></a>
+<span class="sourceLineNo">233</span>    entryLists.addAll(entryMap.values());<a name="line.233"></a>
+<span class="sourceLineNo">234</span>    return entryLists;<a name="line.234"></a>
+<span class="sourceLineNo">235</span>  }<a name="line.235"></a>
+<span class="sourceLineNo">236</span><a name="line.236"></a>
+<span class="sourceLineNo">237</span>  private TableName parseTable(String msg) {<a name="line.237"></a>
+<span class="sourceLineNo">238</span>    // ... TableNotFoundException: '&lt;table&gt;'/n...<a name="line.238"></a>
+<span class="sourceLineNo">239</span>    Pattern p = Pattern.compile("TableNotFoundException: \\'([\\S]*)\\'");<a name="line.239"></a>
+<span class="sourceLineNo">240</span>    Matcher m = p.matcher(msg);<a name="line.240"></a>
+<span class="sourceLineNo">241</span>    if (m.find()) {<a name="line.241"></a>
+<span class="sourceLineNo">242</span>      String table = m.group(1);<a name="line.242"></a>
+<span class="sourceLineNo">243</span>      try {<a name="line.243"></a>
+<span class="sourceLineNo">244</span>        // double check that table is a valid table name<a name="line.244"></a>
+<span class="sourceLineNo">245</span>        TableName.valueOf(TableName.isLegalFullyQualifiedTableName(Bytes.toBytes(table)));<a name="line.245"></a>
+<span class="sourceLineNo">246</span>        return TableName.valueOf(table);<a name="line.246"></a>
+<span class="sourceLineNo">247</span>      } catch (IllegalArgumentException ignore) {<a name="line.247"></a>
+<span class="sourceLineNo">248</span>      }<a name="line.248"></a>
+<span class="sourceLineNo">249</span>    }<a name="line.249"></a>
+<span class="sourceLineNo">250</span>    return null;<a name="line.250"></a>
+<span class="sourceLineNo">251</span>  }<a name="line.251"></a>
+<span class="sourceLineNo">252</span><a name="line.252"></a>
+<span class="sourceLineNo">253</span>  // Filter a set of batches by TableName<a name="line.253"></a>
+<span class="sourceLineNo">254</span>  private List&lt;List&lt;Entry&gt;&gt; filterBatches(final List&lt;List&lt;Entry&gt;&gt; oldEntryList, TableName table) {<a name="line.254"></a>
+<span class="sourceLineNo">255</span>    List&lt;List&lt;Entry&gt;&gt; entryLists = new ArrayList&lt;&gt;();<a name="line.255"></a>
+<span class="sourceLineNo">256</span>    for (List&lt;Entry&gt; entries : oldEntryList) {<a name="line.256"></a>
+<span class="sourceLineNo">257</span>      ArrayList&lt;Entry&gt; thisList = new ArrayList&lt;Entry&gt;(entries.size());<a name="line.257"></a>
+<span class="sourceLineNo">258</span>      entryLists.add(thisList);<a name="line.258"></a>
+<span class="sourceLineNo">259</span>      for (Entry e : entries) {<a name="line.259"></a>
+<span class="sourceLineNo">260</span>        if (!e.getKey().getTablename().equals(table)) {<a name="line.260"></a>
+<span class="sourceLineNo">261</span>          thisList.add(e);<a name="line.261"></a>
+<span class="sourceLineNo">262</span>        }<a name="line.262"></a>
+<span class="sourceLineNo">263</span>      }<a name="line.263"></a>
+<span class="sourceLineNo">264</span>    }<a name="line.264"></a>
+<span class="sourceLineNo">265</span>    return entryLists;<a name="line.265"></a>
+<span class="sourceLineNo">266</span>  }<a name="line.266"></a>
+<span class="sourceLineNo">267</span><a name="line.267"></a>
+<span class="sourceLineNo">268</span>  private void reconnectToPeerCluster() {<a name="line.268"></a>
+<span class="sourceLineNo">269</span>    ClusterConnection connection = null;<a name="line.269"></a>
+<span class="sourceLineNo">270</span>    try {<a name="line.270"></a>
+<span class="sourceLineNo">271</span>      connection = (ClusterConnection) ConnectionFactory.createConnection(this.conf);<a name="line.271"></a>
+<span class="sourceLineNo">272</span>    } catch (IOException ioe) {<a name="line.272"></a>
+<span class="sourceLineNo">273</span>      LOG.warn("Failed to create connection for peer cluster", ioe);<a name="line.273"></a>
+<span class="sourceLineNo">274</span>    }<a name="line.274"></a>
+<span class="sourceLineNo">275</span>    if (connection != null) {<a name="line.275"></a>
+<span class="sourceLineNo">276</span>      this.conn = connection;<a name="line.276"></a>
+<span class="sourceLineNo">277</span>    }<a name="line.277"></a>
+<span class="sourceLineNo">278</span>  }<a name="line.278"></a>
+<span class="sourceLineNo">279</span><a name="line.279"></a>
+<span class="sourceLineNo">280</span>  /**<a name="line.280"></a>
+<span class="sourceLineNo">281</span>   * Do the shipping logic<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 boolean replicate(ReplicateContext replicateContext) {<a name="line.284"></a>
+<span class="sourceLineNo">285</span>    CompletionService&lt;Integer&gt; pool = new ExecutorCompletionService&lt;&gt;(this.exec);<a name="line.285"></a>
+<span class="sourceLineNo">286</span>    List&lt;List&lt;Entry&gt;&gt; batches;<a name="line.286"></a>
+<span class="sourceLineNo">287</span>    String walGroupId = replicateContext.getWalGroupId();<a name="line.287"></a>
+<span class="sourceLineNo">288</span>    int sleepMultiplier = 1;<a name="line.288"></a>
+<span class="sourceLineNo">289</span><a name="line.289"></a>
+<span class="sourceLineNo">290</span>    if (!peersSelected &amp;&amp; this.isRunning()) {<a name="line.290"></a>
+<span class="sourceLineNo">291</span>      connectToPeers();<a name="line.291"></a>
+<span class="sourceLineNo">292</span>      peersSelected = true;<a name="line.292"></a>
+<span class="sourceLineNo">293</span>    }<a name="line.293"></a>
+<span class="sourceLineNo">294</span><a name="line.294"></a>
+<span class="sourceLineNo">295</span>    int numSinks = replicationSinkMgr.getNumSinks();<a name="line.295"></a>
+<span class="sourceLineNo">296</span>    if (numSinks == 0) {<a name="line.296"></a>
+<span class="sourceLineNo">297</span>      LOG.warn("No replication sinks found, returning without replicating. The source should retry"<a name="line.297"></a>
+<span class="sourceLineNo">298</span>          + " with the same set of edits.");<a name="line.298"></a>
+<span class="sourceLineNo">299</span>      return false;<a name="line.299"></a>
+<span class="sourceLineNo">300</span>    }<a name="line.300"></a>
+<span class="sourceLineNo">301</span><a name="line.301"></a>
+<span class="sourceLineNo">302</span>    batches = createBatches(replicateContext.getEntries());<a name="line.302"></a>
+<span class="sourceLineNo">303</span><a name="line.303"></a>
+<span class="sourceLineNo">304</span>    while (this.isRunning() &amp;&amp; !exec.isShutdown()) {<a name="line.304"></a>
+<span class="sourceLineNo">305</span>      if (!isPeerEnabled()) {<a name="line.305"></a>
+<span class="sourceLineNo">306</span>        if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.306"></a>
+<span class="sourceLineNo">307</span>          sleepMultiplier++;<a name="line.307"></a>
+<span class="sourceLineNo">308</span>        }<a name="line.308"></a>
+<span class="sourceLineNo">309</span>        continue;<a name="line.309"></a>
+<span class="sourceLineNo">310</span>      }<a name="line.310"></a>
+<span class="sourceLineNo">311</span>      if (this.conn == null || this.conn.isClosed()) {<a name="line.311"></a>
+<span class="sourceLineNo">312</span>        reconnectToPeerCluster();<a name="line.312"></a>
+<span class="sourceLineNo">313</span>      }<a name="line.313"></a>
+<span class="sourceLineNo">314</span>      try {<a name="line.314"></a>
+<span class="sourceLineNo">315</span>        int futures = 0;<a name="line.315"></a>
+<span class="sourceLineNo">316</span>        for (int i=0; i&lt;batches.size(); i++) {<a name="line.316"></a>
+<span class="sourceLineNo">317</span>          List&lt;Entry&gt; entries = batches.get(i);<a name="line.317"></a>
+<span class="sourceLineNo">318</span>          if (!entries.isEmpty()) {<a name="line.318"></a>
+<span class="sourceLineNo">319</span>            if (LOG.isTraceEnabled()) {<a name="line.319"></a>
+<span class="sourceLineNo">320</span>              LOG.trace("Submitting " + entries.size() +<a name="line.320"></a>
+<span class="sourceLineNo">321</span>                  " entries of total size " + replicateContext.getSize());<a name="line.321"></a>
+<span class="sourceLineNo">322</span>            }<a name="line.322"></a>
+<span class="sourceLineNo">323</span>            // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.323"></a>
+<span class="sourceLineNo">324</span>            pool.submit(createReplicator(entries, i));<a name="line.324"></a>
+<span class="sourceLineNo">325</span>            futures++;<a name="line.325"></a>
+<span class="sourceLineNo">326</span>          }<a name="line.326"></a>
+<span class="sourceLineNo">327</span>        }<a name="line.327"></a>
+<span class="sourceLineNo">328</span>        IOException iox = null;<a name="line.328"></a>
+<span class="sourceLineNo">329</span><a name="line.329"></a>
+<span class="sourceLineNo">330</span>        long lastWriteTime = 0;<a name="line.330"></a>
+<span class="sourceLineNo">331</span>        for (int i=0; i&lt;futures; i++) {<a name="line.331"></a>
+<span class="sourceLineNo">332</span>          try {<a name="line.332"></a>
+<span class="sourceLineNo">333</span>            // wait for all futures, remove successful parts<a name="line.333"></a>
+<span class="sourceLineNo">334</span>            // (only the remaining parts will be retried)<a name="line.334"></a>
+<span class="sourceLineNo">335</span>            Future&lt;Integer&gt; f = pool.take();<a name="line.335"></a>
+<span class="sourceLineNo">336</span>            int index = f.get().intValue();<a name="line.336"></a>
+<span class="sourceLineNo">337</span>            List&lt;Entry&gt; batch = batches.get(index);<a name="line.337"></a>
+<span class="sourceLineNo">338</span>            batches.set(index, Collections.&lt;Entry&gt;emptyList()); // remove successful batch<a name="line.338"></a>
+<span class="sourceLineNo">339</span>            // Find the most recent write time in the batch<a name="line.339"></a>
+<span class="sourceLineNo">340</span>            long writeTime = batch.get(batch.size() - 1).getKey().getWriteTime();<a name="line.340"></a>
+<span class="sourceLineNo">341</span>            if (writeTime &gt; lastWriteTime) {<a name="line.341"></a>
+<span class="sourceLineNo">342</span>              lastWriteTime = writeTime;<a name="line.342"></a>
+<span class="sourceLineNo">343</span>            }<a name="line.343"></a>
+<span class="sourceLineNo">344</span>          } catch (InterruptedException ie) {<a name="line.344"></a>
+<span class="sourceLineNo">345</span>            iox =  new IOException(ie);<a name="line.345"></a>
+<span class="sourceLineNo">346</span>          } catch (ExecutionException ee) {<a name="line.346"></a>
+<span class="sourceLineNo">347</span>            // cause must be an IOException<a name="line.347"></a>
+<span class="sourceLineNo">348</span>            iox = (IOException)ee.getCause();<a name="line.348"></a>
+<span class="sourceLineNo">349</span>          }<a name="line.349"></a>
+<span class="sourceLineNo">350</span>        }<a name="line.350"></a>
+<span class="sourceLineNo">351</span>        if (iox != null) {<a name="line.351"></a>
+<span class="sourceLineNo">352</span>          // if we had any exceptions, try again<a name="line.352"></a>
+<span class="sourceLineNo">353</span>          throw iox;<a name="line.353"></a>
+<span class="sourceLineNo">354</span>        }<a name="line.354"></a>
+<span class="sourceLineNo">355</span>        // update metrics<a name="line.355"></a>
+<span class="sourceLineNo">356</span>        if (lastWriteTime &gt; 0) {<a name="line.356"></a>
+<span class="sourceLineNo">357</span>          this.metrics.setAgeOfLastShippedOp(lastWriteTime, walGroupId);<a name="line.357"></a>
+<span class="sourceLineNo">358</span>        }<a name="line.358"></a>
+<span class="sourceLineNo">359</span>        return true;<a name="line.359"></a>
+<span class="sourceLineNo">360</span><a name="line.360"></a>
+<span class="sourceLineNo">361</span>      } catch (IOException ioe) {<a name="line.361"></a>
+<span class="sourceLineNo">362</span>        // Didn't ship anything, but must still age the last time we did<a name="line.362"></a>
+<span class="sourceLineNo">363</span>        this.metrics.refreshAgeOfLastShippedOp(walGroupId);<a name="line.363"></a>
+<span class="sourceLineNo">364</span>        if (ioe instanceof RemoteException) {<a name="line.364"></a>
+<span class="sourceLineNo">365</span>          ioe = ((RemoteException) ioe).unwrapRemoteException();<a name="line.365"></a>
+<span class="sourceLineNo">366</span>          LOG.warn("Can't replicate because of an error on the remote cluster: ", ioe);<a name="line.366"></a>
+<span class="sourceLineNo">367</span>          if (ioe instanceof TableNotFoundException) {<a name="line.367"></a>
+<span class="sourceLineNo">368</span>            if (dropOnDeletedTables) {<a name="line.368"></a>
+<span class="sourceLineNo">369</span>              // this is a bit fragile, but cannot change how TNFE is serialized<a name="line.369"></a>
+<span class="sourceLineNo">370</span>              // at least check whether the table name is legal<a name="line.370"></a>
+<span class="sourceLineNo">371</span>              TableName table = parseTable(ioe.getMessage());<a name="line.371"></a>
+<span class="sourceLineNo">372</span>              if (table != null) {<a name="line.372"></a>
+<span class="sourceLineNo">373</span>                try (Connection localConn =<a name="line.373"></a>
+<span class="sourceLineNo">374</span>                    ConnectionFactory.createConnection(ctx.getLocalConfiguration())) {<a name="line.374"></a>
+<span class="sourceLineNo">375</span>                  if (!localConn.getAdmin().tableExists(table)) {<a name="line.375"></a>
+<span class="sourceLineNo">376</span>                    // Would potentially be better to retry in one of the outer loops<a name="line.376"></a>
+<span class="sourceLineNo">377</span>                    // and add a table filter there; but that would break the encapsulation,<a name="line.377"></a>
+<span class="sourceLineNo">378</span>                    // so we're doing the filtering here.<a name="line.378"></a>
+<span class="sourceLineNo">379</span>                    LOG.info("Missing table detected at sink, local table also does not exist, filtering edits for '"+table+"'");<a name="line.379"></a>
+<span class="sourceLineNo">380</span>                    batches = filterBatches(batches, table);<a name="line.380"></a>
+<span class="sourceLineNo">381</span>                    continue;<a name="line.381"></a>
+<span class="sourceLineNo">382</span>                  }<a name="line.382"></a>
+<span class="sourceLineNo">383</span>                } catch (IOException iox) {<a name="line.383"></a>
+<span class="sourceLineNo">384</span>                  LOG.warn("Exception checking for local table: ", iox);<a name="line.384"></a>
+<span class="sourceLineNo">385</span>                }<a name="line.385"></a>
+<span class="sourceLineNo">386</span>              }<a name="line.386"></a>
+<span class="sourceLineNo">387</span>            }<a name="line.387"></a>
+<span class="sourceLineNo">388</span>            // fall through and sleep below<a name="line.388"></a>
+<span class="sourceLineNo">389</span>          } else {<a name="line.389"></a>
+<span class="sourceLineNo">390</span>            LOG.warn("Peer encountered RemoteException, rechecking all sinks: ", ioe);<a name="line.390"></a>
+<span class="sourceLineNo">391</span>            replicationSinkMgr.chooseSinks();<a name="line.391"></a>
+<span class="sourceLineNo">392</span>          }<a name="line.392"></a>
+<span class="sourceLineNo">393</span>        } else {<a name="line.393"></a>
+<span class="sourceLineNo">394</span>          if (ioe instanceof SocketTimeoutException) {<a name="line.394"></a>
+<span class="sourceLineNo">395</span>            // This exception means we waited for more than 60s and nothing<a name="line.395"></a>
+<span class="sourceLineNo">396</span>            // happened, the cluster is alive and calling it right away<a name="line.396"></a>
+<span class="sourceLineNo">397</span>            // even for a test just makes things worse.<a name="line.397"></a>
+<span class="sourceLineNo">398</span>            sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.398"></a>
+<span class="sourceLineNo">399</span>              "call to the remote cluster timed out, which is usually " +<a name="line.399"></a>
+<span class="sourceLineNo">400</span>              "caused by a machine failure or a massive slowdown",<a name="line.400"></a>
+<span class="sourceLineNo">401</span>              this.socketTimeoutMultiplier);<a name="line.401"></a>
+<span class="sourceLineNo">402</span>          } else if (ioe instanceof ConnectExceptio

<TRUNCATED>

Mime
View raw message