hbase-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From mi...@apache.org
Subject [04/51] [partial] hbase git commit: Published site at 86be690b0723e814a655ad0ae8a6577d7111c1f2.
Date Sun, 22 Nov 2015 22:40:14 GMT
http://git-wip-us.apache.org/repos/asf/hbase/blob/155d5cd0/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 0793f65..fb9094c 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
@@ -38,294 +38,302 @@
 <span class="sourceLineNo">030</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.30"></a>
 <span class="sourceLineNo">031</span>import java.util.concurrent.TimeUnit;<a name="line.31"></a>
 <span class="sourceLineNo">032</span><a name="line.32"></a>
-<span class="sourceLineNo">033</span>import org.apache.commons.lang.StringUtils;<a name="line.33"></a>
-<span class="sourceLineNo">034</span>import org.apache.commons.logging.Log;<a name="line.34"></a>
-<span class="sourceLineNo">035</span>import org.apache.commons.logging.LogFactory;<a name="line.35"></a>
-<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.classification.InterfaceAudience;<a name="line.36"></a>
-<span class="sourceLineNo">037</span>import org.apache.hadoop.conf.Configuration;<a name="line.37"></a>
-<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.38"></a>
-<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.HConstants;<a name="line.39"></a>
-<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.40"></a>
-<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.41"></a>
-<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.client.HConnection;<a name="line.42"></a>
-<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtbufUtil;<a name="line.43"></a>
-<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.AdminService.BlockingInterface;<a name="line.44"></a>
-<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.45"></a>
-<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.46"></a>
-<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.47"></a>
-<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.48"></a>
-<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.replication.regionserver.ReplicationSinkManager.SinkPeer;<a name="line.49"></a>
-<span class="sourceLineNo">050</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.50"></a>
-<span class="sourceLineNo">051</span><a name="line.51"></a>
-<span class="sourceLineNo">052</span>/**<a name="line.52"></a>
-<span class="sourceLineNo">053</span> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint} <a name="line.53"></a>
-<span class="sourceLineNo">054</span> * implementation for replicating to another HBase cluster.<a name="line.54"></a>
-<span class="sourceLineNo">055</span> * For the slave cluster it selects a random number of peers<a name="line.55"></a>
-<span class="sourceLineNo">056</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.56"></a>
-<span class="sourceLineNo">057</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.57"></a>
-<span class="sourceLineNo">058</span> * &lt;p&gt;<a name="line.58"></a>
-<span class="sourceLineNo">059</span> * A stream is considered down when we cannot contact a region server on the<a name="line.59"></a>
-<span class="sourceLineNo">060</span> * peer cluster for more than 55 seconds by default.<a name="line.60"></a>
-<span class="sourceLineNo">061</span> * &lt;/p&gt;<a name="line.61"></a>
-<span class="sourceLineNo">062</span> */<a name="line.62"></a>
-<span class="sourceLineNo">063</span>@InterfaceAudience.Private<a name="line.63"></a>
-<span class="sourceLineNo">064</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.64"></a>
-<span class="sourceLineNo">065</span><a name="line.65"></a>
-<span class="sourceLineNo">066</span>  private static final Log LOG = LogFactory.getLog(HBaseInterClusterReplicationEndpoint.class);<a name="line.66"></a>
-<span class="sourceLineNo">067</span>  private HConnection conn;<a name="line.67"></a>
-<span class="sourceLineNo">068</span><a name="line.68"></a>
-<span class="sourceLineNo">069</span>  private Configuration conf;<a name="line.69"></a>
-<span class="sourceLineNo">070</span><a name="line.70"></a>
-<span class="sourceLineNo">071</span>  // How long should we sleep for each retry<a name="line.71"></a>
-<span class="sourceLineNo">072</span>  private long sleepForRetries;<a name="line.72"></a>
-<span class="sourceLineNo">073</span><a name="line.73"></a>
-<span class="sourceLineNo">074</span>  // Maximum number of retries before taking bold actions<a name="line.74"></a>
-<span class="sourceLineNo">075</span>  private int maxRetriesMultiplier;<a name="line.75"></a>
-<span class="sourceLineNo">076</span>  // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.76"></a>
-<span class="sourceLineNo">077</span>  private int socketTimeoutMultiplier;<a name="line.77"></a>
-<span class="sourceLineNo">078</span>  //Metrics for this source<a name="line.78"></a>
-<span class="sourceLineNo">079</span>  private MetricsSource metrics;<a name="line.79"></a>
-<span class="sourceLineNo">080</span>  // Handles connecting to peer region servers<a name="line.80"></a>
-<span class="sourceLineNo">081</span>  private ReplicationSinkManager replicationSinkMgr;<a name="line.81"></a>
-<span class="sourceLineNo">082</span>  private boolean peersSelected = false;<a name="line.82"></a>
-<span class="sourceLineNo">083</span>  private ThreadPoolExecutor exec;<a name="line.83"></a>
-<span class="sourceLineNo">084</span>  private int maxThreads;<a name="line.84"></a>
-<span class="sourceLineNo">085</span><a name="line.85"></a>
-<span class="sourceLineNo">086</span>  @Override<a name="line.86"></a>
-<span class="sourceLineNo">087</span>  public void init(Context context) throws IOException {<a name="line.87"></a>
-<span class="sourceLineNo">088</span>    super.init(context);<a name="line.88"></a>
-<span class="sourceLineNo">089</span>    this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.89"></a>
-<span class="sourceLineNo">090</span>    decorateConf();<a name="line.90"></a>
-<span class="sourceLineNo">091</span>    this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.91"></a>
-<span class="sourceLineNo">092</span>    this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.92"></a>
-<span class="sourceLineNo">093</span>        maxRetriesMultiplier);<a name="line.93"></a>
-<span class="sourceLineNo">094</span>    // TODO: This connection is replication specific or we should make it particular to<a name="line.94"></a>
-<span class="sourceLineNo">095</span>    // replication and make replication specific settings such as compression or codec to use<a name="line.95"></a>
-<span class="sourceLineNo">096</span>    // passing Cells.<a name="line.96"></a>
-<span class="sourceLineNo">097</span>    this.conn = (HConnection) ConnectionFactory.createConnection(this.conf);<a name="line.97"></a>
-<span class="sourceLineNo">098</span>    this.sleepForRetries =<a name="line.98"></a>
-<span class="sourceLineNo">099</span>        this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.99"></a>
-<span class="sourceLineNo">100</span>    this.metrics = context.getMetrics();<a name="line.100"></a>
-<span class="sourceLineNo">101</span>    // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.101"></a>
-<span class="sourceLineNo">102</span>    this.replicationSinkMgr = new ReplicationSinkManager(conn, ctx.getPeerId(), this, this.conf);<a name="line.102"></a>
-<span class="sourceLineNo">103</span>    // per sink thread pool<a name="line.103"></a>
-<span class="sourceLineNo">104</span>    this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.104"></a>
-<span class="sourceLineNo">105</span>      HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.105"></a>
-<span class="sourceLineNo">106</span>    this.exec = new ThreadPoolExecutor(1, maxThreads, 60, TimeUnit.SECONDS,<a name="line.106"></a>
-<span class="sourceLineNo">107</span>      new SynchronousQueue&lt;Runnable&gt;());<a name="line.107"></a>
-<span class="sourceLineNo">108</span>  }<a name="line.108"></a>
-<span class="sourceLineNo">109</span><a name="line.109"></a>
-<span class="sourceLineNo">110</span>  private void decorateConf() {<a name="line.110"></a>
-<span class="sourceLineNo">111</span>    String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.111"></a>
-<span class="sourceLineNo">112</span>    if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.112"></a>
-<span class="sourceLineNo">113</span>      this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.113"></a>
-<span class="sourceLineNo">114</span>    }<a name="line.114"></a>
-<span class="sourceLineNo">115</span>  }<a name="line.115"></a>
-<span class="sourceLineNo">116</span><a name="line.116"></a>
-<span class="sourceLineNo">117</span>  private void connectToPeers() {<a name="line.117"></a>
-<span class="sourceLineNo">118</span>    getRegionServers();<a name="line.118"></a>
-<span class="sourceLineNo">119</span><a name="line.119"></a>
-<span class="sourceLineNo">120</span>    int sleepMultiplier = 1;<a name="line.120"></a>
-<span class="sourceLineNo">121</span><a name="line.121"></a>
-<span class="sourceLineNo">122</span>    // Connect to peer cluster first, unless we have to stop<a name="line.122"></a>
-<span class="sourceLineNo">123</span>    while (this.isRunning() &amp;&amp; replicationSinkMgr.getSinks().size() == 0) {<a name="line.123"></a>
-<span class="sourceLineNo">124</span>      replicationSinkMgr.chooseSinks();<a name="line.124"></a>
-<span class="sourceLineNo">125</span>      if (this.isRunning() &amp;&amp; replicationSinkMgr.getSinks().size() == 0) {<a name="line.125"></a>
-<span class="sourceLineNo">126</span>        if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.126"></a>
-<span class="sourceLineNo">127</span>          sleepMultiplier++;<a name="line.127"></a>
-<span class="sourceLineNo">128</span>        }<a name="line.128"></a>
-<span class="sourceLineNo">129</span>      }<a name="line.129"></a>
-<span class="sourceLineNo">130</span>    }<a name="line.130"></a>
-<span class="sourceLineNo">131</span>  }<a name="line.131"></a>
-<span class="sourceLineNo">132</span><a name="line.132"></a>
-<span class="sourceLineNo">133</span>  /**<a name="line.133"></a>
-<span class="sourceLineNo">134</span>   * Do the sleeping logic<a name="line.134"></a>
-<span class="sourceLineNo">135</span>   * @param msg Why we sleep<a name="line.135"></a>
-<span class="sourceLineNo">136</span>   * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.136"></a>
-<span class="sourceLineNo">137</span>   * @return True if &lt;code&gt;sleepMultiplier&lt;/code&gt; is &amp;lt; &lt;code&gt;maxRetriesMultiplier&lt;/code&gt;<a name="line.137"></a>
-<span class="sourceLineNo">138</span>   */<a name="line.138"></a>
-<span class="sourceLineNo">139</span>  protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.139"></a>
-<span class="sourceLineNo">140</span>    try {<a name="line.140"></a>
-<span class="sourceLineNo">141</span>      if (LOG.isTraceEnabled()) {<a name="line.141"></a>
-<span class="sourceLineNo">142</span>        LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.142"></a>
-<span class="sourceLineNo">143</span>      }<a name="line.143"></a>
-<span class="sourceLineNo">144</span>      Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.144"></a>
-<span class="sourceLineNo">145</span>    } catch (InterruptedException e) {<a name="line.145"></a>
-<span class="sourceLineNo">146</span>      LOG.debug("Interrupted while sleeping between retries");<a name="line.146"></a>
-<span class="sourceLineNo">147</span>    }<a name="line.147"></a>
-<span class="sourceLineNo">148</span>    return sleepMultiplier &lt; maxRetriesMultiplier;<a name="line.148"></a>
-<span class="sourceLineNo">149</span>  }<a name="line.149"></a>
-<span class="sourceLineNo">150</span><a name="line.150"></a>
-<span class="sourceLineNo">151</span>  /**<a name="line.151"></a>
-<span class="sourceLineNo">152</span>   * Do the shipping logic<a name="line.152"></a>
-<span class="sourceLineNo">153</span>   */<a name="line.153"></a>
-<span class="sourceLineNo">154</span>  @Override<a name="line.154"></a>
-<span class="sourceLineNo">155</span>  public boolean replicate(ReplicateContext replicateContext) {<a name="line.155"></a>
-<span class="sourceLineNo">156</span>    List&lt;Entry&gt; entries = replicateContext.getEntries();<a name="line.156"></a>
-<span class="sourceLineNo">157</span>    String walGroupId = replicateContext.getWalGroupId();<a name="line.157"></a>
-<span class="sourceLineNo">158</span>    int sleepMultiplier = 1;<a name="line.158"></a>
-<span class="sourceLineNo">159</span><a name="line.159"></a>
-<span class="sourceLineNo">160</span>    if (!peersSelected &amp;&amp; this.isRunning()) {<a name="line.160"></a>
-<span class="sourceLineNo">161</span>      connectToPeers();<a name="line.161"></a>
-<span class="sourceLineNo">162</span>      peersSelected = true;<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>    if (replicationSinkMgr.getSinks().size() == 0) {<a name="line.165"></a>
-<span class="sourceLineNo">166</span>      return false;<a name="line.166"></a>
-<span class="sourceLineNo">167</span>    }<a name="line.167"></a>
-<span class="sourceLineNo">168</span>    // minimum of: configured threads, number of 100-waledit batches,<a name="line.168"></a>
-<span class="sourceLineNo">169</span>    //  and number of current sinks<a name="line.169"></a>
-<span class="sourceLineNo">170</span>    int n = Math.min(Math.min(this.maxThreads, entries.size()/100+1),<a name="line.170"></a>
-<span class="sourceLineNo">171</span>      replicationSinkMgr.getSinks().size());<a name="line.171"></a>
-<span class="sourceLineNo">172</span>    List&lt;List&lt;Entry&gt;&gt; entryLists = new ArrayList&lt;List&lt;Entry&gt;&gt;(n);<a name="line.172"></a>
-<span class="sourceLineNo">173</span>    if (n == 1) {<a name="line.173"></a>
-<span class="sourceLineNo">174</span>      entryLists.add(entries);<a name="line.174"></a>
-<span class="sourceLineNo">175</span>    } else {<a name="line.175"></a>
-<span class="sourceLineNo">176</span>      for (int i=0; i&lt;n; i++) {<a name="line.176"></a>
-<span class="sourceLineNo">177</span>        entryLists.add(new ArrayList&lt;Entry&gt;(entries.size()/n+1));<a name="line.177"></a>
-<span class="sourceLineNo">178</span>      }<a name="line.178"></a>
-<span class="sourceLineNo">179</span>      // now group by region<a name="line.179"></a>
-<span class="sourceLineNo">180</span>      for (Entry e : entries) {<a name="line.180"></a>
-<span class="sourceLineNo">181</span>        entryLists.get(Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName())%n)).add(e);<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>    while (this.isRunning()) {<a name="line.184"></a>
-<span class="sourceLineNo">185</span>      if (!isPeerEnabled()) {<a name="line.185"></a>
-<span class="sourceLineNo">186</span>        if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.186"></a>
-<span class="sourceLineNo">187</span>          sleepMultiplier++;<a name="line.187"></a>
-<span class="sourceLineNo">188</span>        }<a name="line.188"></a>
-<span class="sourceLineNo">189</span>        continue;<a name="line.189"></a>
-<span class="sourceLineNo">190</span>      }<a name="line.190"></a>
-<span class="sourceLineNo">191</span>      try {<a name="line.191"></a>
-<span class="sourceLineNo">192</span>        if (LOG.isTraceEnabled()) {<a name="line.192"></a>
-<span class="sourceLineNo">193</span>          LOG.trace("Replicating " + entries.size() +<a name="line.193"></a>
-<span class="sourceLineNo">194</span>              " entries of total size " + replicateContext.getSize());<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>        List&lt;Future&lt;Integer&gt;&gt; futures = new ArrayList&lt;Future&lt;Integer&gt;&gt;(entryLists.size());<a name="line.197"></a>
-<span class="sourceLineNo">198</span>        for (int i=0; i&lt;entryLists.size(); i++) {<a name="line.198"></a>
-<span class="sourceLineNo">199</span>          if (!entryLists.get(i).isEmpty()) {<a name="line.199"></a>
-<span class="sourceLineNo">200</span>            if (LOG.isTraceEnabled()) {<a name="line.200"></a>
-<span class="sourceLineNo">201</span>              LOG.trace("Submitting " + entryLists.get(i).size() +<a name="line.201"></a>
-<span class="sourceLineNo">202</span>                  " entries of total size " + replicateContext.getSize());<a name="line.202"></a>
-<span class="sourceLineNo">203</span>            }<a name="line.203"></a>
-<span class="sourceLineNo">204</span>            // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.204"></a>
-<span class="sourceLineNo">205</span>            futures.add(exec.submit(new Replicator(entryLists.get(i), i)));<a name="line.205"></a>
-<span class="sourceLineNo">206</span>          }<a name="line.206"></a>
-<span class="sourceLineNo">207</span>        }<a name="line.207"></a>
-<span class="sourceLineNo">208</span>        IOException iox = null;<a name="line.208"></a>
-<span class="sourceLineNo">209</span>        for (Future&lt;Integer&gt; f : futures) {<a name="line.209"></a>
-<span class="sourceLineNo">210</span>          try {<a name="line.210"></a>
-<span class="sourceLineNo">211</span>            // wait for all futures, remove successful parts<a name="line.211"></a>
-<span class="sourceLineNo">212</span>            // (only the remaining parts will be retried)<a name="line.212"></a>
-<span class="sourceLineNo">213</span>            entryLists.remove(f.get());<a name="line.213"></a>
-<span class="sourceLineNo">214</span>          } catch (InterruptedException ie) {<a name="line.214"></a>
-<span class="sourceLineNo">215</span>            iox =  new IOException(ie);<a name="line.215"></a>
-<span class="sourceLineNo">216</span>          } catch (ExecutionException ee) {<a name="line.216"></a>
-<span class="sourceLineNo">217</span>            // cause must be an IOException<a name="line.217"></a>
-<span class="sourceLineNo">218</span>            iox = (IOException)ee.getCause();<a name="line.218"></a>
-<span class="sourceLineNo">219</span>          }<a name="line.219"></a>
-<span class="sourceLineNo">220</span>        }<a name="line.220"></a>
-<span class="sourceLineNo">221</span>        if (iox != null) {<a name="line.221"></a>
-<span class="sourceLineNo">222</span>          // if we had any exceptions, try again<a name="line.222"></a>
-<span class="sourceLineNo">223</span>          throw iox;<a name="line.223"></a>
-<span class="sourceLineNo">224</span>        }<a name="line.224"></a>
-<span class="sourceLineNo">225</span>        // update metrics<a name="line.225"></a>
-<span class="sourceLineNo">226</span>        this.metrics.setAgeOfLastShippedOp(entries.get(entries.size() - 1).getKey().getWriteTime(),<a name="line.226"></a>
-<span class="sourceLineNo">227</span>          walGroupId);<a name="line.227"></a>
-<span class="sourceLineNo">228</span>        return true;<a name="line.228"></a>
-<span class="sourceLineNo">229</span><a name="line.229"></a>
-<span class="sourceLineNo">230</span>      } catch (IOException ioe) {<a name="line.230"></a>
-<span class="sourceLineNo">231</span>        // Didn't ship anything, but must still age the last time we did<a name="line.231"></a>
-<span class="sourceLineNo">232</span>        this.metrics.refreshAgeOfLastShippedOp(walGroupId);<a name="line.232"></a>
-<span class="sourceLineNo">233</span>        if (ioe instanceof RemoteException) {<a name="line.233"></a>
-<span class="sourceLineNo">234</span>          ioe = ((RemoteException) ioe).unwrapRemoteException();<a name="line.234"></a>
-<span class="sourceLineNo">235</span>          LOG.warn("Can't replicate because of an error on the remote cluster: ", ioe);<a name="line.235"></a>
-<span class="sourceLineNo">236</span>          if (ioe instanceof TableNotFoundException) {<a name="line.236"></a>
-<span class="sourceLineNo">237</span>            if (sleepForRetries("A table is missing in the peer cluster. "<a name="line.237"></a>
-<span class="sourceLineNo">238</span>                + "Replication cannot proceed without losing data.", sleepMultiplier)) {<a name="line.238"></a>
-<span class="sourceLineNo">239</span>              sleepMultiplier++;<a name="line.239"></a>
-<span class="sourceLineNo">240</span>            }<a name="line.240"></a>
-<span class="sourceLineNo">241</span>          }<a name="line.241"></a>
-<span class="sourceLineNo">242</span>        } else {<a name="line.242"></a>
-<span class="sourceLineNo">243</span>          if (ioe instanceof SocketTimeoutException) {<a name="line.243"></a>
-<span class="sourceLineNo">244</span>            // This exception means we waited for more than 60s and nothing<a name="line.244"></a>
-<span class="sourceLineNo">245</span>            // happened, the cluster is alive and calling it right away<a name="line.245"></a>
-<span class="sourceLineNo">246</span>            // even for a test just makes things worse.<a name="line.246"></a>
-<span class="sourceLineNo">247</span>            sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.247"></a>
-<span class="sourceLineNo">248</span>              "call to the remote cluster timed out, which is usually " +<a name="line.248"></a>
-<span class="sourceLineNo">249</span>              "caused by a machine failure or a massive slowdown",<a name="line.249"></a>
-<span class="sourceLineNo">250</span>              this.socketTimeoutMultiplier);<a name="line.250"></a>
-<span class="sourceLineNo">251</span>          } else if (ioe instanceof ConnectException) {<a name="line.251"></a>
-<span class="sourceLineNo">252</span>            LOG.warn("Peer is unavailable, rechecking all sinks: ", ioe);<a name="line.252"></a>
-<span class="sourceLineNo">253</span>            replicationSinkMgr.chooseSinks();<a name="line.253"></a>
-<span class="sourceLineNo">254</span>          } else {<a name="line.254"></a>
-<span class="sourceLineNo">255</span>            LOG.warn("Can't replicate because of a local or network error: ", ioe);<a name="line.255"></a>
-<span class="sourceLineNo">256</span>          }<a name="line.256"></a>
-<span class="sourceLineNo">257</span>        }<a name="line.257"></a>
-<span class="sourceLineNo">258</span>        if (sleepForRetries("Since we are unable to replicate", sleepMultiplier)) {<a name="line.258"></a>
-<span class="sourceLineNo">259</span>          sleepMultiplier++;<a name="line.259"></a>
-<span class="sourceLineNo">260</span>        }<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>    return false; // in case we exited before replicating<a name="line.263"></a>
-<span class="sourceLineNo">264</span>  }<a name="line.264"></a>
-<span class="sourceLineNo">265</span><a name="line.265"></a>
-<span class="sourceLineNo">266</span>  protected boolean isPeerEnabled() {<a name="line.266"></a>
-<span class="sourceLineNo">267</span>    return ctx.getReplicationPeer().getPeerState() == PeerState.ENABLED;<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>  @Override<a name="line.270"></a>
-<span class="sourceLineNo">271</span>  protected void doStop() {<a name="line.271"></a>
-<span class="sourceLineNo">272</span>    disconnect(); //don't call super.doStop()<a name="line.272"></a>
-<span class="sourceLineNo">273</span>    if (this.conn != null) {<a name="line.273"></a>
-<span class="sourceLineNo">274</span>      try {<a name="line.274"></a>
-<span class="sourceLineNo">275</span>        this.conn.close();<a name="line.275"></a>
-<span class="sourceLineNo">276</span>        this.conn = null;<a name="line.276"></a>
-<span class="sourceLineNo">277</span>      } catch (IOException e) {<a name="line.277"></a>
-<span class="sourceLineNo">278</span>        LOG.warn("Failed to close the connection");<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>    exec.shutdownNow();<a name="line.281"></a>
-<span class="sourceLineNo">282</span>    notifyStopped();<a name="line.282"></a>
-<span class="sourceLineNo">283</span>  }<a name="line.283"></a>
-<span class="sourceLineNo">284</span><a name="line.284"></a>
-<span class="sourceLineNo">285</span>  // is this needed? Nobody else will call doStop() otherwise<a name="line.285"></a>
-<span class="sourceLineNo">286</span>  @Override<a name="line.286"></a>
-<span class="sourceLineNo">287</span>  public State stopAndWait() {<a name="line.287"></a>
-<span class="sourceLineNo">288</span>    doStop();<a name="line.288"></a>
-<span class="sourceLineNo">289</span>    return super.stopAndWait();<a name="line.289"></a>
-<span class="sourceLineNo">290</span>  }<a name="line.290"></a>
-<span class="sourceLineNo">291</span><a name="line.291"></a>
-<span class="sourceLineNo">292</span>  private class Replicator implements Callable&lt;Integer&gt; {<a name="line.292"></a>
-<span class="sourceLineNo">293</span>    private List&lt;Entry&gt; entries;<a name="line.293"></a>
-<span class="sourceLineNo">294</span>    private int ordinal;<a name="line.294"></a>
-<span class="sourceLineNo">295</span>    public Replicator(List&lt;Entry&gt; entries, int ordinal) {<a name="line.295"></a>
-<span class="sourceLineNo">296</span>      this.entries = entries;<a name="line.296"></a>
-<span class="sourceLineNo">297</span>      this.ordinal = ordinal;<a name="line.297"></a>
-<span class="sourceLineNo">298</span>    }<a name="line.298"></a>
-<span class="sourceLineNo">299</span><a name="line.299"></a>
-<span class="sourceLineNo">300</span>    @Override<a name="line.300"></a>
-<span class="sourceLineNo">301</span>    public Integer call() throws IOException {<a name="line.301"></a>
-<span class="sourceLineNo">302</span>      SinkPeer sinkPeer = null;<a name="line.302"></a>
-<span class="sourceLineNo">303</span>      try {<a name="line.303"></a>
-<span class="sourceLineNo">304</span>        sinkPeer = replicationSinkMgr.getReplicationSink();<a name="line.304"></a>
-<span class="sourceLineNo">305</span>        BlockingInterface rrs = sinkPeer.getRegionServer();<a name="line.305"></a>
-<span class="sourceLineNo">306</span>        ReplicationProtbufUtil.replicateWALEntry(rrs,<a name="line.306"></a>
-<span class="sourceLineNo">307</span>            entries.toArray(new Entry[entries.size()]));<a name="line.307"></a>
-<span class="sourceLineNo">308</span>        replicationSinkMgr.reportSinkSuccess(sinkPeer);<a name="line.308"></a>
-<span class="sourceLineNo">309</span>        return ordinal;<a name="line.309"></a>
-<span class="sourceLineNo">310</span><a name="line.310"></a>
-<span class="sourceLineNo">311</span>      } catch (IOException ioe) {<a name="line.311"></a>
-<span class="sourceLineNo">312</span>        if (sinkPeer != null) {<a name="line.312"></a>
-<span class="sourceLineNo">313</span>          replicationSinkMgr.reportBadSink(sinkPeer);<a name="line.313"></a>
-<span class="sourceLineNo">314</span>        }<a name="line.314"></a>
-<span class="sourceLineNo">315</span>        throw ioe;<a name="line.315"></a>
-<span class="sourceLineNo">316</span>      }<a name="line.316"></a>
-<span class="sourceLineNo">317</span>    }<a name="line.317"></a>
+<span class="sourceLineNo">033</span>import com.google.common.annotations.VisibleForTesting;<a name="line.33"></a>
+<span class="sourceLineNo">034</span>import org.apache.commons.lang.StringUtils;<a name="line.34"></a>
+<span class="sourceLineNo">035</span>import org.apache.commons.logging.Log;<a name="line.35"></a>
+<span class="sourceLineNo">036</span>import org.apache.commons.logging.LogFactory;<a name="line.36"></a>
+<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.classification.InterfaceAudience;<a name="line.37"></a>
+<span class="sourceLineNo">038</span>import org.apache.hadoop.conf.Configuration;<a name="line.38"></a>
+<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.39"></a>
+<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.HConstants;<a name="line.40"></a>
+<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.41"></a>
+<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.42"></a>
+<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.client.HConnection;<a name="line.43"></a>
+<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.protobuf.ReplicationProtbufUtil;<a name="line.44"></a>
+<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.protobuf.generated.AdminProtos.AdminService.BlockingInterface;<a name="line.45"></a>
+<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.46"></a>
+<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.47"></a>
+<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.48"></a>
+<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.49"></a>
+<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.replication.regionserver.ReplicationSinkManager.SinkPeer;<a name="line.50"></a>
+<span class="sourceLineNo">051</span>import org.apache.hadoop.ipc.RemoteException;<a name="line.51"></a>
+<span class="sourceLineNo">052</span><a name="line.52"></a>
+<span class="sourceLineNo">053</span>/**<a name="line.53"></a>
+<span class="sourceLineNo">054</span> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint} <a name="line.54"></a>
+<span class="sourceLineNo">055</span> * implementation for replicating to another HBase cluster.<a name="line.55"></a>
+<span class="sourceLineNo">056</span> * For the slave cluster it selects a random number of peers<a name="line.56"></a>
+<span class="sourceLineNo">057</span> * using a replication ratio. For example, if replication ration = 0.1<a name="line.57"></a>
+<span class="sourceLineNo">058</span> * and slave cluster has 100 region servers, 10 will be selected.<a name="line.58"></a>
+<span class="sourceLineNo">059</span> * &lt;p&gt;<a name="line.59"></a>
+<span class="sourceLineNo">060</span> * A stream is considered down when we cannot contact a region server on the<a name="line.60"></a>
+<span class="sourceLineNo">061</span> * peer cluster for more than 55 seconds by default.<a name="line.61"></a>
+<span class="sourceLineNo">062</span> * &lt;/p&gt;<a name="line.62"></a>
+<span class="sourceLineNo">063</span> */<a name="line.63"></a>
+<span class="sourceLineNo">064</span>@InterfaceAudience.Private<a name="line.64"></a>
+<span class="sourceLineNo">065</span>public class HBaseInterClusterReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.65"></a>
+<span class="sourceLineNo">066</span><a name="line.66"></a>
+<span class="sourceLineNo">067</span>  private static final Log LOG = LogFactory.getLog(HBaseInterClusterReplicationEndpoint.class);<a name="line.67"></a>
+<span class="sourceLineNo">068</span>  private HConnection conn;<a name="line.68"></a>
+<span class="sourceLineNo">069</span><a name="line.69"></a>
+<span class="sourceLineNo">070</span>  private Configuration conf;<a name="line.70"></a>
+<span class="sourceLineNo">071</span><a name="line.71"></a>
+<span class="sourceLineNo">072</span>  // How long should we sleep for each retry<a name="line.72"></a>
+<span class="sourceLineNo">073</span>  private long sleepForRetries;<a name="line.73"></a>
+<span class="sourceLineNo">074</span><a name="line.74"></a>
+<span class="sourceLineNo">075</span>  // Maximum number of retries before taking bold actions<a name="line.75"></a>
+<span class="sourceLineNo">076</span>  private int maxRetriesMultiplier;<a name="line.76"></a>
+<span class="sourceLineNo">077</span>  // Socket timeouts require even bolder actions since we don't want to DDOS<a name="line.77"></a>
+<span class="sourceLineNo">078</span>  private int socketTimeoutMultiplier;<a name="line.78"></a>
+<span class="sourceLineNo">079</span>  //Metrics for this source<a name="line.79"></a>
+<span class="sourceLineNo">080</span>  private MetricsSource metrics;<a name="line.80"></a>
+<span class="sourceLineNo">081</span>  // Handles connecting to peer region servers<a name="line.81"></a>
+<span class="sourceLineNo">082</span>  private ReplicationSinkManager replicationSinkMgr;<a name="line.82"></a>
+<span class="sourceLineNo">083</span>  private boolean peersSelected = false;<a name="line.83"></a>
+<span class="sourceLineNo">084</span>  private ThreadPoolExecutor exec;<a name="line.84"></a>
+<span class="sourceLineNo">085</span>  private int maxThreads;<a name="line.85"></a>
+<span class="sourceLineNo">086</span><a name="line.86"></a>
+<span class="sourceLineNo">087</span>  @Override<a name="line.87"></a>
+<span class="sourceLineNo">088</span>  public void init(Context context) throws IOException {<a name="line.88"></a>
+<span class="sourceLineNo">089</span>    super.init(context);<a name="line.89"></a>
+<span class="sourceLineNo">090</span>    this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.90"></a>
+<span class="sourceLineNo">091</span>    decorateConf();<a name="line.91"></a>
+<span class="sourceLineNo">092</span>    this.maxRetriesMultiplier = this.conf.getInt("replication.source.maxretriesmultiplier", 300);<a name="line.92"></a>
+<span class="sourceLineNo">093</span>    this.socketTimeoutMultiplier = this.conf.getInt("replication.source.socketTimeoutMultiplier",<a name="line.93"></a>
+<span class="sourceLineNo">094</span>        maxRetriesMultiplier);<a name="line.94"></a>
+<span class="sourceLineNo">095</span>    // TODO: This connection is replication specific or we should make it particular to<a name="line.95"></a>
+<span class="sourceLineNo">096</span>    // replication and make replication specific settings such as compression or codec to use<a name="line.96"></a>
+<span class="sourceLineNo">097</span>    // passing Cells.<a name="line.97"></a>
+<span class="sourceLineNo">098</span>    this.conn = (HConnection) ConnectionFactory.createConnection(this.conf);<a name="line.98"></a>
+<span class="sourceLineNo">099</span>    this.sleepForRetries =<a name="line.99"></a>
+<span class="sourceLineNo">100</span>        this.conf.getLong("replication.source.sleepforretries", 1000);<a name="line.100"></a>
+<span class="sourceLineNo">101</span>    this.metrics = context.getMetrics();<a name="line.101"></a>
+<span class="sourceLineNo">102</span>    // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.102"></a>
+<span class="sourceLineNo">103</span>    this.replicationSinkMgr = new ReplicationSinkManager(conn, ctx.getPeerId(), this, this.conf);<a name="line.103"></a>
+<span class="sourceLineNo">104</span>    // per sink thread pool<a name="line.104"></a>
+<span class="sourceLineNo">105</span>    this.maxThreads = this.conf.getInt(HConstants.REPLICATION_SOURCE_MAXTHREADS_KEY,<a name="line.105"></a>
+<span class="sourceLineNo">106</span>      HConstants.REPLICATION_SOURCE_MAXTHREADS_DEFAULT);<a name="line.106"></a>
+<span class="sourceLineNo">107</span>    this.exec = new ThreadPoolExecutor(1, maxThreads, 60, TimeUnit.SECONDS,<a name="line.107"></a>
+<span class="sourceLineNo">108</span>      new SynchronousQueue&lt;Runnable&gt;());<a name="line.108"></a>
+<span class="sourceLineNo">109</span>  }<a name="line.109"></a>
+<span class="sourceLineNo">110</span><a name="line.110"></a>
+<span class="sourceLineNo">111</span>  private void decorateConf() {<a name="line.111"></a>
+<span class="sourceLineNo">112</span>    String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.112"></a>
+<span class="sourceLineNo">113</span>    if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.113"></a>
+<span class="sourceLineNo">114</span>      this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.114"></a>
+<span class="sourceLineNo">115</span>    }<a name="line.115"></a>
+<span class="sourceLineNo">116</span>  }<a name="line.116"></a>
+<span class="sourceLineNo">117</span><a name="line.117"></a>
+<span class="sourceLineNo">118</span>  private void connectToPeers() {<a name="line.118"></a>
+<span class="sourceLineNo">119</span>    getRegionServers();<a name="line.119"></a>
+<span class="sourceLineNo">120</span><a name="line.120"></a>
+<span class="sourceLineNo">121</span>    int sleepMultiplier = 1;<a name="line.121"></a>
+<span class="sourceLineNo">122</span><a name="line.122"></a>
+<span class="sourceLineNo">123</span>    // Connect to peer cluster first, unless we have to stop<a name="line.123"></a>
+<span class="sourceLineNo">124</span>    while (this.isRunning() &amp;&amp; replicationSinkMgr.getSinks().size() == 0) {<a name="line.124"></a>
+<span class="sourceLineNo">125</span>      replicationSinkMgr.chooseSinks();<a name="line.125"></a>
+<span class="sourceLineNo">126</span>      if (this.isRunning() &amp;&amp; replicationSinkMgr.getSinks().size() == 0) {<a name="line.126"></a>
+<span class="sourceLineNo">127</span>        if (sleepForRetries("Waiting for peers", sleepMultiplier)) {<a name="line.127"></a>
+<span class="sourceLineNo">128</span>          sleepMultiplier++;<a name="line.128"></a>
+<span class="sourceLineNo">129</span>        }<a name="line.129"></a>
+<span class="sourceLineNo">130</span>      }<a name="line.130"></a>
+<span class="sourceLineNo">131</span>    }<a name="line.131"></a>
+<span class="sourceLineNo">132</span>  }<a name="line.132"></a>
+<span class="sourceLineNo">133</span><a name="line.133"></a>
+<span class="sourceLineNo">134</span>  /**<a name="line.134"></a>
+<span class="sourceLineNo">135</span>   * Do the sleeping logic<a name="line.135"></a>
+<span class="sourceLineNo">136</span>   * @param msg Why we sleep<a name="line.136"></a>
+<span class="sourceLineNo">137</span>   * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.137"></a>
+<span class="sourceLineNo">138</span>   * @return True if &lt;code&gt;sleepMultiplier&lt;/code&gt; is &amp;lt; &lt;code&gt;maxRetriesMultiplier&lt;/code&gt;<a name="line.138"></a>
+<span class="sourceLineNo">139</span>   */<a name="line.139"></a>
+<span class="sourceLineNo">140</span>  protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.140"></a>
+<span class="sourceLineNo">141</span>    try {<a name="line.141"></a>
+<span class="sourceLineNo">142</span>      if (LOG.isTraceEnabled()) {<a name="line.142"></a>
+<span class="sourceLineNo">143</span>        LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.143"></a>
+<span class="sourceLineNo">144</span>      }<a name="line.144"></a>
+<span class="sourceLineNo">145</span>      Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.145"></a>
+<span class="sourceLineNo">146</span>    } catch (InterruptedException e) {<a name="line.146"></a>
+<span class="sourceLineNo">147</span>      LOG.debug("Interrupted while sleeping between retries");<a name="line.147"></a>
+<span class="sourceLineNo">148</span>    }<a name="line.148"></a>
+<span class="sourceLineNo">149</span>    return sleepMultiplier &lt; maxRetriesMultiplier;<a name="line.149"></a>
+<span class="sourceLineNo">150</span>  }<a name="line.150"></a>
+<span class="sourceLineNo">151</span><a name="line.151"></a>
+<span class="sourceLineNo">152</span>  /**<a name="line.152"></a>
+<span class="sourceLineNo">153</span>   * Do the shipping logic<a name="line.153"></a>
+<span class="sourceLineNo">154</span>   */<a name="line.154"></a>
+<span class="sourceLineNo">155</span>  @Override<a name="line.155"></a>
+<span class="sourceLineNo">156</span>  public boolean replicate(ReplicateContext replicateContext) {<a name="line.156"></a>
+<span class="sourceLineNo">157</span>    List&lt;Entry&gt; entries = replicateContext.getEntries();<a name="line.157"></a>
+<span class="sourceLineNo">158</span>    String walGroupId = replicateContext.getWalGroupId();<a name="line.158"></a>
+<span class="sourceLineNo">159</span>    int sleepMultiplier = 1;<a name="line.159"></a>
+<span class="sourceLineNo">160</span><a name="line.160"></a>
+<span class="sourceLineNo">161</span>    if (!peersSelected &amp;&amp; this.isRunning()) {<a name="line.161"></a>
+<span class="sourceLineNo">162</span>      connectToPeers();<a name="line.162"></a>
+<span class="sourceLineNo">163</span>      peersSelected = true;<a name="line.163"></a>
+<span class="sourceLineNo">164</span>    }<a name="line.164"></a>
+<span class="sourceLineNo">165</span><a name="line.165"></a>
+<span class="sourceLineNo">166</span>    if (replicationSinkMgr.getSinks().size() == 0) {<a name="line.166"></a>
+<span class="sourceLineNo">167</span>      return false;<a name="line.167"></a>
+<span class="sourceLineNo">168</span>    }<a name="line.168"></a>
+<span class="sourceLineNo">169</span>    // minimum of: configured threads, number of 100-waledit batches,<a name="line.169"></a>
+<span class="sourceLineNo">170</span>    //  and number of current sinks<a name="line.170"></a>
+<span class="sourceLineNo">171</span>    int n = Math.min(Math.min(this.maxThreads, entries.size()/100+1),<a name="line.171"></a>
+<span class="sourceLineNo">172</span>      replicationSinkMgr.getSinks().size());<a name="line.172"></a>
+<span class="sourceLineNo">173</span>    List&lt;List&lt;Entry&gt;&gt; entryLists = new ArrayList&lt;List&lt;Entry&gt;&gt;(n);<a name="line.173"></a>
+<span class="sourceLineNo">174</span>    if (n == 1) {<a name="line.174"></a>
+<span class="sourceLineNo">175</span>      entryLists.add(entries);<a name="line.175"></a>
+<span class="sourceLineNo">176</span>    } else {<a name="line.176"></a>
+<span class="sourceLineNo">177</span>      for (int i=0; i&lt;n; i++) {<a name="line.177"></a>
+<span class="sourceLineNo">178</span>        entryLists.add(new ArrayList&lt;Entry&gt;(entries.size()/n+1));<a name="line.178"></a>
+<span class="sourceLineNo">179</span>      }<a name="line.179"></a>
+<span class="sourceLineNo">180</span>      // now group by region<a name="line.180"></a>
+<span class="sourceLineNo">181</span>      for (Entry e : entries) {<a name="line.181"></a>
+<span class="sourceLineNo">182</span>        entryLists.get(Math.abs(Bytes.hashCode(e.getKey().getEncodedRegionName())%n)).add(e);<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>    while (this.isRunning()) {<a name="line.185"></a>
+<span class="sourceLineNo">186</span>      if (!isPeerEnabled()) {<a name="line.186"></a>
+<span class="sourceLineNo">187</span>        if (sleepForRetries("Replication is disabled", sleepMultiplier)) {<a name="line.187"></a>
+<span class="sourceLineNo">188</span>          sleepMultiplier++;<a name="line.188"></a>
+<span class="sourceLineNo">189</span>        }<a name="line.189"></a>
+<span class="sourceLineNo">190</span>        continue;<a name="line.190"></a>
+<span class="sourceLineNo">191</span>      }<a name="line.191"></a>
+<span class="sourceLineNo">192</span>      try {<a name="line.192"></a>
+<span class="sourceLineNo">193</span>        if (LOG.isTraceEnabled()) {<a name="line.193"></a>
+<span class="sourceLineNo">194</span>          LOG.trace("Replicating " + entries.size() +<a name="line.194"></a>
+<span class="sourceLineNo">195</span>              " entries of total size " + replicateContext.getSize());<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>        List&lt;Future&lt;Integer&gt;&gt; futures = new ArrayList&lt;Future&lt;Integer&gt;&gt;(entryLists.size());<a name="line.198"></a>
+<span class="sourceLineNo">199</span>        for (int i=0; i&lt;entryLists.size(); i++) {<a name="line.199"></a>
+<span class="sourceLineNo">200</span>          if (!entryLists.get(i).isEmpty()) {<a name="line.200"></a>
+<span class="sourceLineNo">201</span>            if (LOG.isTraceEnabled()) {<a name="line.201"></a>
+<span class="sourceLineNo">202</span>              LOG.trace("Submitting " + entryLists.get(i).size() +<a name="line.202"></a>
+<span class="sourceLineNo">203</span>                  " entries of total size " + replicateContext.getSize());<a name="line.203"></a>
+<span class="sourceLineNo">204</span>            }<a name="line.204"></a>
+<span class="sourceLineNo">205</span>            // RuntimeExceptions encountered here bubble up and are handled in ReplicationSource<a name="line.205"></a>
+<span class="sourceLineNo">206</span>            futures.add(exec.submit(createReplicator(entryLists.get(i), i)));<a name="line.206"></a>
+<span class="sourceLineNo">207</span>          }<a name="line.207"></a>
+<span class="sourceLineNo">208</span>        }<a name="line.208"></a>
+<span class="sourceLineNo">209</span>        IOException iox = null;<a name="line.209"></a>
+<span class="sourceLineNo">210</span>        for (int index = futures.size() - 1; index &gt;= 0; index--) {<a name="line.210"></a>
+<span class="sourceLineNo">211</span>          try {<a name="line.211"></a>
+<span class="sourceLineNo">212</span>            // wait for all futures, remove successful parts<a name="line.212"></a>
+<span class="sourceLineNo">213</span>            // (only the remaining parts will be retried)<a name="line.213"></a>
+<span class="sourceLineNo">214</span>            Future&lt;Integer&gt; f = futures.get(index);<a name="line.214"></a>
+<span class="sourceLineNo">215</span>            entryLists.remove(f.get().intValue());<a name="line.215"></a>
+<span class="sourceLineNo">216</span>          } catch (InterruptedException ie) {<a name="line.216"></a>
+<span class="sourceLineNo">217</span>            iox =  new IOException(ie);<a name="line.217"></a>
+<span class="sourceLineNo">218</span>          } catch (ExecutionException ee) {<a name="line.218"></a>
+<span class="sourceLineNo">219</span>            // cause must be an IOException<a name="line.219"></a>
+<span class="sourceLineNo">220</span>            iox = (IOException)ee.getCause();<a name="line.220"></a>
+<span class="sourceLineNo">221</span>          }<a name="line.221"></a>
+<span class="sourceLineNo">222</span>        }<a name="line.222"></a>
+<span class="sourceLineNo">223</span>        if (iox != null) {<a name="line.223"></a>
+<span class="sourceLineNo">224</span>          // if we had any exceptions, try again<a name="line.224"></a>
+<span class="sourceLineNo">225</span>          throw iox;<a name="line.225"></a>
+<span class="sourceLineNo">226</span>        }<a name="line.226"></a>
+<span class="sourceLineNo">227</span>        // update metrics<a name="line.227"></a>
+<span class="sourceLineNo">228</span>        this.metrics.setAgeOfLastShippedOp(entries.get(entries.size() - 1).getKey().getWriteTime(),<a name="line.228"></a>
+<span class="sourceLineNo">229</span>          walGroupId);<a name="line.229"></a>
+<span class="sourceLineNo">230</span>        return true;<a name="line.230"></a>
+<span class="sourceLineNo">231</span><a name="line.231"></a>
+<span class="sourceLineNo">232</span>      } catch (IOException ioe) {<a name="line.232"></a>
+<span class="sourceLineNo">233</span>        // Didn't ship anything, but must still age the last time we did<a name="line.233"></a>
+<span class="sourceLineNo">234</span>        this.metrics.refreshAgeOfLastShippedOp(walGroupId);<a name="line.234"></a>
+<span class="sourceLineNo">235</span>        if (ioe instanceof RemoteException) {<a name="line.235"></a>
+<span class="sourceLineNo">236</span>          ioe = ((RemoteException) ioe).unwrapRemoteException();<a name="line.236"></a>
+<span class="sourceLineNo">237</span>          LOG.warn("Can't replicate because of an error on the remote cluster: ", ioe);<a name="line.237"></a>
+<span class="sourceLineNo">238</span>          if (ioe instanceof TableNotFoundException) {<a name="line.238"></a>
+<span class="sourceLineNo">239</span>            if (sleepForRetries("A table is missing in the peer cluster. "<a name="line.239"></a>
+<span class="sourceLineNo">240</span>                + "Replication cannot proceed without losing data.", sleepMultiplier)) {<a name="line.240"></a>
+<span class="sourceLineNo">241</span>              sleepMultiplier++;<a name="line.241"></a>
+<span class="sourceLineNo">242</span>            }<a name="line.242"></a>
+<span class="sourceLineNo">243</span>          }<a name="line.243"></a>
+<span class="sourceLineNo">244</span>        } else {<a name="line.244"></a>
+<span class="sourceLineNo">245</span>          if (ioe instanceof SocketTimeoutException) {<a name="line.245"></a>
+<span class="sourceLineNo">246</span>            // This exception means we waited for more than 60s and nothing<a name="line.246"></a>
+<span class="sourceLineNo">247</span>            // happened, the cluster is alive and calling it right away<a name="line.247"></a>
+<span class="sourceLineNo">248</span>            // even for a test just makes things worse.<a name="line.248"></a>
+<span class="sourceLineNo">249</span>            sleepForRetries("Encountered a SocketTimeoutException. Since the " +<a name="line.249"></a>
+<span class="sourceLineNo">250</span>              "call to the remote cluster timed out, which is usually " +<a name="line.250"></a>
+<span class="sourceLineNo">251</span>              "caused by a machine failure or a massive slowdown",<a name="line.251"></a>
+<span class="sourceLineNo">252</span>              this.socketTimeoutMultiplier);<a name="line.252"></a>
+<span class="sourceLineNo">253</span>          } else if (ioe instanceof ConnectException) {<a name="line.253"></a>
+<span class="sourceLineNo">254</span>            LOG.warn("Peer is unavailable, rechecking all sinks: ", ioe);<a name="line.254"></a>
+<span class="sourceLineNo">255</span>            replicationSinkMgr.chooseSinks();<a name="line.255"></a>
+<span class="sourceLineNo">256</span>          } else {<a name="line.256"></a>
+<span class="sourceLineNo">257</span>            LOG.warn("Can't replicate because of a local or network error: ", ioe);<a name="line.257"></a>
+<span class="sourceLineNo">258</span>          }<a name="line.258"></a>
+<span class="sourceLineNo">259</span>        }<a name="line.259"></a>
+<span class="sourceLineNo">260</span>        if (sleepForRetries("Since we are unable to replicate", sleepMultiplier)) {<a name="line.260"></a>
+<span class="sourceLineNo">261</span>          sleepMultiplier++;<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 false; // in case we exited before replicating<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>  protected boolean isPeerEnabled() {<a name="line.268"></a>
+<span class="sourceLineNo">269</span>    return ctx.getReplicationPeer().getPeerState() == PeerState.ENABLED;<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>  @Override<a name="line.272"></a>
+<span class="sourceLineNo">273</span>  protected void doStop() {<a name="line.273"></a>
+<span class="sourceLineNo">274</span>    disconnect(); //don't call super.doStop()<a name="line.274"></a>
+<span class="sourceLineNo">275</span>    if (this.conn != null) {<a name="line.275"></a>
+<span class="sourceLineNo">276</span>      try {<a name="line.276"></a>
+<span class="sourceLineNo">277</span>        this.conn.close();<a name="line.277"></a>
+<span class="sourceLineNo">278</span>        this.conn = null;<a name="line.278"></a>
+<span class="sourceLineNo">279</span>      } catch (IOException e) {<a name="line.279"></a>
+<span class="sourceLineNo">280</span>        LOG.warn("Failed to close the connection");<a name="line.280"></a>
+<span class="sourceLineNo">281</span>      }<a name="line.281"></a>
+<span class="sourceLineNo">282</span>    }<a name="line.282"></a>
+<span class="sourceLineNo">283</span>    exec.shutdownNow();<a name="line.283"></a>
+<span class="sourceLineNo">284</span>    notifyStopped();<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>  // is this needed? Nobody else will call doStop() otherwise<a name="line.287"></a>
+<span class="sourceLineNo">288</span>  @Override<a name="line.288"></a>
+<span class="sourceLineNo">289</span>  public State stopAndWait() {<a name="line.289"></a>
+<span class="sourceLineNo">290</span>    doStop();<a name="line.290"></a>
+<span class="sourceLineNo">291</span>    return super.stopAndWait();<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>  @VisibleForTesting<a name="line.294"></a>
+<span class="sourceLineNo">295</span>  protected Replicator createReplicator(List&lt;Entry&gt; entries, int ordinal) {<a name="line.295"></a>
+<span class="sourceLineNo">296</span>    return new Replicator(entries, ordinal);<a name="line.296"></a>
+<span class="sourceLineNo">297</span>  }<a name="line.297"></a>
+<span class="sourceLineNo">298</span><a name="line.298"></a>
+<span class="sourceLineNo">299</span>  @VisibleForTesting<a name="line.299"></a>
+<span class="sourceLineNo">300</span>  protected class Replicator implements Callable&lt;Integer&gt; {<a name="line.300"></a>
+<span class="sourceLineNo">301</span>    private List&lt;Entry&gt; entries;<a name="line.301"></a>
+<span class="sourceLineNo">302</span>    private int ordinal;<a name="line.302"></a>
+<span class="sourceLineNo">303</span>    public Replicator(List&lt;Entry&gt; entries, int ordinal) {<a name="line.303"></a>
+<span class="sourceLineNo">304</span>      this.entries = entries;<a name="line.304"></a>
+<span class="sourceLineNo">305</span>      this.ordinal = ordinal;<a name="line.305"></a>
+<span class="sourceLineNo">306</span>    }<a name="line.306"></a>
+<span class="sourceLineNo">307</span><a name="line.307"></a>
+<span class="sourceLineNo">308</span>    @Override<a name="line.308"></a>
+<span class="sourceLineNo">309</span>    public Integer call() throws IOException {<a name="line.309"></a>
+<span class="sourceLineNo">310</span>      SinkPeer sinkPeer = null;<a name="line.310"></a>
+<span class="sourceLineNo">311</span>      try {<a name="line.311"></a>
+<span class="sourceLineNo">312</span>        sinkPeer = replicationSinkMgr.getReplicationSink();<a name="line.312"></a>
+<span class="sourceLineNo">313</span>        BlockingInterface rrs = sinkPeer.getRegionServer();<a name="line.313"></a>
+<span class="sourceLineNo">314</span>        ReplicationProtbufUtil.replicateWALEntry(rrs,<a name="line.314"></a>
+<span class="sourceLineNo">315</span>            entries.toArray(new Entry[entries.size()]));<a name="line.315"></a>
+<span class="sourceLineNo">316</span>        replicationSinkMgr.reportSinkSuccess(sinkPeer);<a name="line.316"></a>
+<span class="sourceLineNo">317</span>        return ordinal;<a name="line.317"></a>
 <span class="sourceLineNo">318</span><a name="line.318"></a>
-<span class="sourceLineNo">319</span>  }<a name="line.319"></a>
-<span class="sourceLineNo">320</span>}<a name="line.320"></a>
+<span class="sourceLineNo">319</span>      } catch (IOException ioe) {<a name="line.319"></a>
+<span class="sourceLineNo">320</span>        if (sinkPeer != null) {<a name="line.320"></a>
+<span class="sourceLineNo">321</span>          replicationSinkMgr.reportBadSink(sinkPeer);<a name="line.321"></a>
+<span class="sourceLineNo">322</span>        }<a name="line.322"></a>
+<span class="sourceLineNo">323</span>        throw ioe;<a name="line.323"></a>
+<span class="sourceLineNo">324</span>      }<a name="line.324"></a>
+<span class="sourceLineNo">325</span>    }<a name="line.325"></a>
+<span class="sourceLineNo">326</span><a name="line.326"></a>
+<span class="sourceLineNo">327</span>  }<a name="line.327"></a>
+<span class="sourceLineNo">328</span>}<a name="line.328"></a>
 
 
 


Mime
View raw message