hbase-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From git-site-r...@apache.org
Subject [36/51] [partial] hbase-site git commit: Published site at 82d554e3783372cc6b05489452c815b57c06f6cd.
Date Wed, 19 Jul 2017 22:07:52 GMT
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/9eba7fcf/devapidocs/src-html/org/apache/hadoop/hbase/coordination/ZkSplitLogWorkerCoordination.GetDataAsyncCallback.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/coordination/ZkSplitLogWorkerCoordination.GetDataAsyncCallback.html b/devapidocs/src-html/org/apache/hadoop/hbase/coordination/ZkSplitLogWorkerCoordination.GetDataAsyncCallback.html
index 5eefe01..59063f5 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/coordination/ZkSplitLogWorkerCoordination.GetDataAsyncCallback.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/coordination/ZkSplitLogWorkerCoordination.GetDataAsyncCallback.html
@@ -58,606 +58,607 @@
 <span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.util.CancelableProgressable;<a name="line.50"></a>
 <span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.51"></a>
 <span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.52"></a>
-<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.zookeeper.ZKSplitLog;<a name="line.53"></a>
-<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.54"></a>
-<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.zookeeper.ZooKeeperListener;<a name="line.55"></a>
-<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher;<a name="line.56"></a>
-<span class="sourceLineNo">057</span>import org.apache.hadoop.util.StringUtils;<a name="line.57"></a>
-<span class="sourceLineNo">058</span>import org.apache.zookeeper.AsyncCallback;<a name="line.58"></a>
-<span class="sourceLineNo">059</span>import org.apache.zookeeper.KeeperException;<a name="line.59"></a>
-<span class="sourceLineNo">060</span>import org.apache.zookeeper.data.Stat;<a name="line.60"></a>
-<span class="sourceLineNo">061</span><a name="line.61"></a>
-<span class="sourceLineNo">062</span>/**<a name="line.62"></a>
-<span class="sourceLineNo">063</span> * ZooKeeper based implementation of {@link SplitLogWorkerCoordination}<a name="line.63"></a>
-<span class="sourceLineNo">064</span> * It listen for changes in ZooKeeper and<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>@InterfaceAudience.Private<a name="line.67"></a>
-<span class="sourceLineNo">068</span>public class ZkSplitLogWorkerCoordination extends ZooKeeperListener implements<a name="line.68"></a>
-<span class="sourceLineNo">069</span>    SplitLogWorkerCoordination {<a name="line.69"></a>
-<span class="sourceLineNo">070</span><a name="line.70"></a>
-<span class="sourceLineNo">071</span>  private static final Log LOG = LogFactory.getLog(ZkSplitLogWorkerCoordination.class);<a name="line.71"></a>
-<span class="sourceLineNo">072</span><a name="line.72"></a>
-<span class="sourceLineNo">073</span>  private static final int checkInterval = 5000; // 5 seconds<a name="line.73"></a>
-<span class="sourceLineNo">074</span>  private static final int FAILED_TO_OWN_TASK = -1;<a name="line.74"></a>
-<span class="sourceLineNo">075</span><a name="line.75"></a>
-<span class="sourceLineNo">076</span>  private  SplitLogWorker worker;<a name="line.76"></a>
-<span class="sourceLineNo">077</span><a name="line.77"></a>
-<span class="sourceLineNo">078</span>  private TaskExecutor splitTaskExecutor;<a name="line.78"></a>
-<span class="sourceLineNo">079</span><a name="line.79"></a>
-<span class="sourceLineNo">080</span>  private final Object taskReadyLock = new Object();<a name="line.80"></a>
-<span class="sourceLineNo">081</span>  private AtomicInteger taskReadySeq = new AtomicInteger(0);<a name="line.81"></a>
-<span class="sourceLineNo">082</span>  private volatile String currentTask = null;<a name="line.82"></a>
-<span class="sourceLineNo">083</span>  private int currentVersion;<a name="line.83"></a>
-<span class="sourceLineNo">084</span>  private volatile boolean shouldStop = false;<a name="line.84"></a>
-<span class="sourceLineNo">085</span>  private final Object grabTaskLock = new Object();<a name="line.85"></a>
-<span class="sourceLineNo">086</span>  private boolean workerInGrabTask = false;<a name="line.86"></a>
-<span class="sourceLineNo">087</span>  private int reportPeriod;<a name="line.87"></a>
-<span class="sourceLineNo">088</span>  private RegionServerServices server = null;<a name="line.88"></a>
-<span class="sourceLineNo">089</span>  protected final AtomicInteger tasksInProgress = new AtomicInteger(0);<a name="line.89"></a>
-<span class="sourceLineNo">090</span>  private int maxConcurrentTasks = 0;<a name="line.90"></a>
-<span class="sourceLineNo">091</span><a name="line.91"></a>
-<span class="sourceLineNo">092</span>  private final ZkCoordinatedStateManager manager;<a name="line.92"></a>
-<span class="sourceLineNo">093</span><a name="line.93"></a>
-<span class="sourceLineNo">094</span>  public ZkSplitLogWorkerCoordination(ZkCoordinatedStateManager zkCoordinatedStateManager,<a name="line.94"></a>
-<span class="sourceLineNo">095</span>      ZooKeeperWatcher watcher) {<a name="line.95"></a>
-<span class="sourceLineNo">096</span>    super(watcher);<a name="line.96"></a>
-<span class="sourceLineNo">097</span>    manager = zkCoordinatedStateManager;<a name="line.97"></a>
-<span class="sourceLineNo">098</span><a name="line.98"></a>
-<span class="sourceLineNo">099</span>  }<a name="line.99"></a>
-<span class="sourceLineNo">100</span><a name="line.100"></a>
-<span class="sourceLineNo">101</span>  /**<a name="line.101"></a>
-<span class="sourceLineNo">102</span>   * Override handler from {@link ZooKeeperListener}<a name="line.102"></a>
-<span class="sourceLineNo">103</span>   */<a name="line.103"></a>
-<span class="sourceLineNo">104</span>  @Override<a name="line.104"></a>
-<span class="sourceLineNo">105</span>  public void nodeChildrenChanged(String path) {<a name="line.105"></a>
-<span class="sourceLineNo">106</span>    if (path.equals(watcher.znodePaths.splitLogZNode)) {<a name="line.106"></a>
-<span class="sourceLineNo">107</span>      if (LOG.isTraceEnabled()) LOG.trace("tasks arrived or departed on " + path);<a name="line.107"></a>
-<span class="sourceLineNo">108</span>      synchronized (taskReadyLock) {<a name="line.108"></a>
-<span class="sourceLineNo">109</span>        this.taskReadySeq.incrementAndGet();<a name="line.109"></a>
-<span class="sourceLineNo">110</span>        taskReadyLock.notify();<a name="line.110"></a>
-<span class="sourceLineNo">111</span>      }<a name="line.111"></a>
-<span class="sourceLineNo">112</span>    }<a name="line.112"></a>
-<span class="sourceLineNo">113</span>  }<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>   * Override handler from {@link ZooKeeperListener}<a name="line.116"></a>
-<span class="sourceLineNo">117</span>   */<a name="line.117"></a>
-<span class="sourceLineNo">118</span>  @Override<a name="line.118"></a>
-<span class="sourceLineNo">119</span>  public void nodeDataChanged(String path) {<a name="line.119"></a>
-<span class="sourceLineNo">120</span>    // there will be a self generated dataChanged event every time attemptToOwnTask()<a name="line.120"></a>
-<span class="sourceLineNo">121</span>    // heartbeats the task znode by upping its version<a name="line.121"></a>
-<span class="sourceLineNo">122</span>    synchronized (grabTaskLock) {<a name="line.122"></a>
-<span class="sourceLineNo">123</span>      if (workerInGrabTask) {<a name="line.123"></a>
-<span class="sourceLineNo">124</span>        // currentTask can change<a name="line.124"></a>
-<span class="sourceLineNo">125</span>        String taskpath = currentTask;<a name="line.125"></a>
-<span class="sourceLineNo">126</span>        if (taskpath != null &amp;&amp; taskpath.equals(path)) {<a name="line.126"></a>
-<span class="sourceLineNo">127</span>          getDataSetWatchAsync();<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>   * Override setter from {@link SplitLogWorkerCoordination}<a name="line.134"></a>
-<span class="sourceLineNo">135</span>   */<a name="line.135"></a>
-<span class="sourceLineNo">136</span>  @Override<a name="line.136"></a>
-<span class="sourceLineNo">137</span>  public void init(RegionServerServices server, Configuration conf,<a name="line.137"></a>
-<span class="sourceLineNo">138</span>      TaskExecutor splitExecutor, SplitLogWorker worker) {<a name="line.138"></a>
-<span class="sourceLineNo">139</span>    this.server = server;<a name="line.139"></a>
-<span class="sourceLineNo">140</span>    this.worker = worker;<a name="line.140"></a>
-<span class="sourceLineNo">141</span>    this.splitTaskExecutor = splitExecutor;<a name="line.141"></a>
-<span class="sourceLineNo">142</span>    maxConcurrentTasks = conf.getInt("hbase.regionserver.wal.max.splitters", DEFAULT_MAX_SPLITTERS);<a name="line.142"></a>
-<span class="sourceLineNo">143</span>    reportPeriod =<a name="line.143"></a>
-<span class="sourceLineNo">144</span>        conf.getInt("hbase.splitlog.report.period",<a name="line.144"></a>
-<span class="sourceLineNo">145</span>          conf.getInt(HConstants.HBASE_SPLITLOG_MANAGER_TIMEOUT,<a name="line.145"></a>
-<span class="sourceLineNo">146</span>            ZKSplitLogManagerCoordination.DEFAULT_TIMEOUT) / 3);<a name="line.146"></a>
-<span class="sourceLineNo">147</span>  }<a name="line.147"></a>
-<span class="sourceLineNo">148</span><a name="line.148"></a>
-<span class="sourceLineNo">149</span>  /* Support functions for ZooKeeper async callback */<a name="line.149"></a>
-<span class="sourceLineNo">150</span><a name="line.150"></a>
-<span class="sourceLineNo">151</span>  void getDataSetWatchFailure(String path) {<a name="line.151"></a>
-<span class="sourceLineNo">152</span>    synchronized (grabTaskLock) {<a name="line.152"></a>
-<span class="sourceLineNo">153</span>      if (workerInGrabTask) {<a name="line.153"></a>
-<span class="sourceLineNo">154</span>        // currentTask can change but that's ok<a name="line.154"></a>
-<span class="sourceLineNo">155</span>        String taskpath = currentTask;<a name="line.155"></a>
-<span class="sourceLineNo">156</span>        if (taskpath != null &amp;&amp; taskpath.equals(path)) {<a name="line.156"></a>
-<span class="sourceLineNo">157</span>          LOG.info("retrying data watch on " + path);<a name="line.157"></a>
-<span class="sourceLineNo">158</span>          SplitLogCounters.tot_wkr_get_data_retry.incrementAndGet();<a name="line.158"></a>
-<span class="sourceLineNo">159</span>          getDataSetWatchAsync();<a name="line.159"></a>
-<span class="sourceLineNo">160</span>        } else {<a name="line.160"></a>
-<span class="sourceLineNo">161</span>          // no point setting a watch on the task which this worker is not<a name="line.161"></a>
-<span class="sourceLineNo">162</span>          // working upon anymore<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>    }<a name="line.165"></a>
-<span class="sourceLineNo">166</span>  }<a name="line.166"></a>
-<span class="sourceLineNo">167</span><a name="line.167"></a>
-<span class="sourceLineNo">168</span>  public void getDataSetWatchAsync() {<a name="line.168"></a>
-<span class="sourceLineNo">169</span>    watcher.getRecoverableZooKeeper().getZooKeeper()<a name="line.169"></a>
-<span class="sourceLineNo">170</span>        .getData(currentTask, watcher, new GetDataAsyncCallback(), null);<a name="line.170"></a>
-<span class="sourceLineNo">171</span>    SplitLogCounters.tot_wkr_get_data_queued.incrementAndGet();<a name="line.171"></a>
-<span class="sourceLineNo">172</span>  }<a name="line.172"></a>
-<span class="sourceLineNo">173</span><a name="line.173"></a>
-<span class="sourceLineNo">174</span>  void getDataSetWatchSuccess(String path, byte[] data) {<a name="line.174"></a>
-<span class="sourceLineNo">175</span>    SplitLogTask slt;<a name="line.175"></a>
-<span class="sourceLineNo">176</span>    try {<a name="line.176"></a>
-<span class="sourceLineNo">177</span>      slt = SplitLogTask.parseFrom(data);<a name="line.177"></a>
-<span class="sourceLineNo">178</span>    } catch (DeserializationException e) {<a name="line.178"></a>
-<span class="sourceLineNo">179</span>      LOG.warn("Failed parse", e);<a name="line.179"></a>
-<span class="sourceLineNo">180</span>      return;<a name="line.180"></a>
-<span class="sourceLineNo">181</span>    }<a name="line.181"></a>
-<span class="sourceLineNo">182</span>    synchronized (grabTaskLock) {<a name="line.182"></a>
-<span class="sourceLineNo">183</span>      if (workerInGrabTask) {<a name="line.183"></a>
-<span class="sourceLineNo">184</span>        // currentTask can change but that's ok<a name="line.184"></a>
-<span class="sourceLineNo">185</span>        String taskpath = currentTask;<a name="line.185"></a>
-<span class="sourceLineNo">186</span>        if (taskpath != null &amp;&amp; taskpath.equals(path)) {<a name="line.186"></a>
-<span class="sourceLineNo">187</span>          ServerName serverName = manager.getServer().getServerName();<a name="line.187"></a>
-<span class="sourceLineNo">188</span>          // have to compare data. cannot compare version because then there<a name="line.188"></a>
-<span class="sourceLineNo">189</span>          // will be race with attemptToOwnTask()<a name="line.189"></a>
-<span class="sourceLineNo">190</span>          // cannot just check whether the node has been transitioned to<a name="line.190"></a>
-<span class="sourceLineNo">191</span>          // UNASSIGNED because by the time this worker sets the data watch<a name="line.191"></a>
-<span class="sourceLineNo">192</span>          // the node might have made two transitions - from owned by this<a name="line.192"></a>
-<span class="sourceLineNo">193</span>          // worker to unassigned to owned by another worker<a name="line.193"></a>
-<span class="sourceLineNo">194</span>          if (!slt.isOwned(serverName) &amp;&amp; !slt.isDone(serverName) &amp;&amp; !slt.isErr(serverName)<a name="line.194"></a>
-<span class="sourceLineNo">195</span>              &amp;&amp; !slt.isResigned(serverName)) {<a name="line.195"></a>
-<span class="sourceLineNo">196</span>            LOG.info("task " + taskpath + " preempted from " + serverName<a name="line.196"></a>
-<span class="sourceLineNo">197</span>                + ", current task state and owner=" + slt.toString());<a name="line.197"></a>
-<span class="sourceLineNo">198</span>            worker.stopTask();<a name="line.198"></a>
-<span class="sourceLineNo">199</span>          }<a name="line.199"></a>
-<span class="sourceLineNo">200</span>        }<a name="line.200"></a>
-<span class="sourceLineNo">201</span>      }<a name="line.201"></a>
-<span class="sourceLineNo">202</span>    }<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>  /**<a name="line.205"></a>
-<span class="sourceLineNo">206</span>   * try to grab a 'lock' on the task zk node to own and execute the task.<a name="line.206"></a>
-<span class="sourceLineNo">207</span>   * &lt;p&gt;<a name="line.207"></a>
-<span class="sourceLineNo">208</span>   * @param path zk node for the task<a name="line.208"></a>
-<span class="sourceLineNo">209</span>   */<a name="line.209"></a>
-<span class="sourceLineNo">210</span>  private void grabTask(String path) {<a name="line.210"></a>
-<span class="sourceLineNo">211</span>    Stat stat = new Stat();<a name="line.211"></a>
-<span class="sourceLineNo">212</span>    byte[] data;<a name="line.212"></a>
-<span class="sourceLineNo">213</span>    synchronized (grabTaskLock) {<a name="line.213"></a>
-<span class="sourceLineNo">214</span>      currentTask = path;<a name="line.214"></a>
-<span class="sourceLineNo">215</span>      workerInGrabTask = true;<a name="line.215"></a>
-<span class="sourceLineNo">216</span>      if (Thread.interrupted()) {<a name="line.216"></a>
-<span class="sourceLineNo">217</span>        return;<a name="line.217"></a>
-<span class="sourceLineNo">218</span>      }<a name="line.218"></a>
-<span class="sourceLineNo">219</span>    }<a name="line.219"></a>
-<span class="sourceLineNo">220</span>    try {<a name="line.220"></a>
-<span class="sourceLineNo">221</span>      try {<a name="line.221"></a>
-<span class="sourceLineNo">222</span>        if ((data = ZKUtil.getDataNoWatch(watcher, path, stat)) == null) {<a name="line.222"></a>
-<span class="sourceLineNo">223</span>          SplitLogCounters.tot_wkr_failed_to_grab_task_no_data.incrementAndGet();<a name="line.223"></a>
-<span class="sourceLineNo">224</span>          return;<a name="line.224"></a>
-<span class="sourceLineNo">225</span>        }<a name="line.225"></a>
-<span class="sourceLineNo">226</span>      } catch (KeeperException e) {<a name="line.226"></a>
-<span class="sourceLineNo">227</span>        LOG.warn("Failed to get data for znode " + path, e);<a name="line.227"></a>
-<span class="sourceLineNo">228</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_exception.incrementAndGet();<a name="line.228"></a>
-<span class="sourceLineNo">229</span>        return;<a name="line.229"></a>
-<span class="sourceLineNo">230</span>      }<a name="line.230"></a>
-<span class="sourceLineNo">231</span>      SplitLogTask slt;<a name="line.231"></a>
-<span class="sourceLineNo">232</span>      try {<a name="line.232"></a>
-<span class="sourceLineNo">233</span>        slt = SplitLogTask.parseFrom(data);<a name="line.233"></a>
-<span class="sourceLineNo">234</span>      } catch (DeserializationException e) {<a name="line.234"></a>
-<span class="sourceLineNo">235</span>        LOG.warn("Failed parse data for znode " + path, e);<a name="line.235"></a>
-<span class="sourceLineNo">236</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_exception.incrementAndGet();<a name="line.236"></a>
-<span class="sourceLineNo">237</span>        return;<a name="line.237"></a>
-<span class="sourceLineNo">238</span>      }<a name="line.238"></a>
-<span class="sourceLineNo">239</span>      if (!slt.isUnassigned()) {<a name="line.239"></a>
-<span class="sourceLineNo">240</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_owned.incrementAndGet();<a name="line.240"></a>
-<span class="sourceLineNo">241</span>        return;<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>      currentVersion =<a name="line.244"></a>
-<span class="sourceLineNo">245</span>          attemptToOwnTask(true, watcher, server.getServerName(), path,<a name="line.245"></a>
-<span class="sourceLineNo">246</span>            slt.getMode(), stat.getVersion());<a name="line.246"></a>
-<span class="sourceLineNo">247</span>      if (currentVersion &lt; 0) {<a name="line.247"></a>
-<span class="sourceLineNo">248</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_lost_race.incrementAndGet();<a name="line.248"></a>
-<span class="sourceLineNo">249</span>        return;<a name="line.249"></a>
-<span class="sourceLineNo">250</span>      }<a name="line.250"></a>
-<span class="sourceLineNo">251</span><a name="line.251"></a>
-<span class="sourceLineNo">252</span>      if (ZKSplitLog.isRescanNode(watcher, currentTask)) {<a name="line.252"></a>
-<span class="sourceLineNo">253</span>        ZkSplitLogWorkerCoordination.ZkSplitTaskDetails splitTaskDetails =<a name="line.253"></a>
-<span class="sourceLineNo">254</span>            new ZkSplitLogWorkerCoordination.ZkSplitTaskDetails();<a name="line.254"></a>
-<span class="sourceLineNo">255</span>        splitTaskDetails.setTaskNode(currentTask);<a name="line.255"></a>
-<span class="sourceLineNo">256</span>        splitTaskDetails.setCurTaskZKVersion(new MutableInt(currentVersion));<a name="line.256"></a>
-<span class="sourceLineNo">257</span><a name="line.257"></a>
-<span class="sourceLineNo">258</span>        endTask(new SplitLogTask.Done(server.getServerName(), slt.getMode()),<a name="line.258"></a>
-<span class="sourceLineNo">259</span>          SplitLogCounters.tot_wkr_task_acquired_rescan, splitTaskDetails);<a name="line.259"></a>
-<span class="sourceLineNo">260</span>        return;<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>      LOG.info("worker " + server.getServerName() + " acquired task " + path);<a name="line.263"></a>
-<span class="sourceLineNo">264</span>      SplitLogCounters.tot_wkr_task_acquired.incrementAndGet();<a name="line.264"></a>
-<span class="sourceLineNo">265</span>      getDataSetWatchAsync();<a name="line.265"></a>
-<span class="sourceLineNo">266</span><a name="line.266"></a>
-<span class="sourceLineNo">267</span>      submitTask(path, slt.getMode(), currentVersion, reportPeriod);<a name="line.267"></a>
-<span class="sourceLineNo">268</span><a name="line.268"></a>
-<span class="sourceLineNo">269</span>      // after a successful submit, sleep a little bit to allow other RSs to grab the rest tasks<a name="line.269"></a>
-<span class="sourceLineNo">270</span>      try {<a name="line.270"></a>
-<span class="sourceLineNo">271</span>        int sleepTime = RandomUtils.nextInt(500) + 500;<a name="line.271"></a>
-<span class="sourceLineNo">272</span>        Thread.sleep(sleepTime);<a name="line.272"></a>
-<span class="sourceLineNo">273</span>      } catch (InterruptedException e) {<a name="line.273"></a>
-<span class="sourceLineNo">274</span>        LOG.warn("Interrupted while yielding for other region servers", e);<a name="line.274"></a>
-<span class="sourceLineNo">275</span>        Thread.currentThread().interrupt();<a name="line.275"></a>
-<span class="sourceLineNo">276</span>      }<a name="line.276"></a>
-<span class="sourceLineNo">277</span>    } finally {<a name="line.277"></a>
-<span class="sourceLineNo">278</span>      synchronized (grabTaskLock) {<a name="line.278"></a>
-<span class="sourceLineNo">279</span>        workerInGrabTask = false;<a name="line.279"></a>
-<span class="sourceLineNo">280</span>        // clear the interrupt from stopTask() otherwise the next task will<a name="line.280"></a>
-<span class="sourceLineNo">281</span>        // suffer<a name="line.281"></a>
-<span class="sourceLineNo">282</span>        Thread.interrupted();<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>  }<a name="line.285"></a>
-<span class="sourceLineNo">286</span><a name="line.286"></a>
-<span class="sourceLineNo">287</span>  /**<a name="line.287"></a>
-<span class="sourceLineNo">288</span>   * Submit a log split task to executor service<a name="line.288"></a>
-<span class="sourceLineNo">289</span>   * @param curTask task to submit<a name="line.289"></a>
-<span class="sourceLineNo">290</span>   * @param curTaskZKVersion current version of task<a name="line.290"></a>
-<span class="sourceLineNo">291</span>   */<a name="line.291"></a>
-<span class="sourceLineNo">292</span>  void submitTask(final String curTask, final RecoveryMode mode, final int curTaskZKVersion,<a name="line.292"></a>
-<span class="sourceLineNo">293</span>      final int reportPeriod) {<a name="line.293"></a>
-<span class="sourceLineNo">294</span>    final MutableInt zkVersion = new MutableInt(curTaskZKVersion);<a name="line.294"></a>
-<span class="sourceLineNo">295</span><a name="line.295"></a>
-<span class="sourceLineNo">296</span>    CancelableProgressable reporter = new CancelableProgressable() {<a name="line.296"></a>
-<span class="sourceLineNo">297</span>      private long last_report_at = 0;<a name="line.297"></a>
-<span class="sourceLineNo">298</span><a name="line.298"></a>
-<span class="sourceLineNo">299</span>      @Override<a name="line.299"></a>
-<span class="sourceLineNo">300</span>      public boolean progress() {<a name="line.300"></a>
-<span class="sourceLineNo">301</span>        long t = EnvironmentEdgeManager.currentTime();<a name="line.301"></a>
-<span class="sourceLineNo">302</span>        if ((t - last_report_at) &gt; reportPeriod) {<a name="line.302"></a>
-<span class="sourceLineNo">303</span>          last_report_at = t;<a name="line.303"></a>
-<span class="sourceLineNo">304</span>          int latestZKVersion =<a name="line.304"></a>
-<span class="sourceLineNo">305</span>              attemptToOwnTask(false, watcher, server.getServerName(), curTask,<a name="line.305"></a>
-<span class="sourceLineNo">306</span>                mode, zkVersion.intValue());<a name="line.306"></a>
-<span class="sourceLineNo">307</span>          if (latestZKVersion &lt; 0) {<a name="line.307"></a>
-<span class="sourceLineNo">308</span>            LOG.warn("Failed to heartbeat the task" + curTask);<a name="line.308"></a>
-<span class="sourceLineNo">309</span>            return false;<a name="line.309"></a>
-<span class="sourceLineNo">310</span>          }<a name="line.310"></a>
-<span class="sourceLineNo">311</span>          zkVersion.setValue(latestZKVersion);<a name="line.311"></a>
-<span class="sourceLineNo">312</span>        }<a name="line.312"></a>
-<span class="sourceLineNo">313</span>        return true;<a name="line.313"></a>
-<span class="sourceLineNo">314</span>      }<a name="line.314"></a>
-<span class="sourceLineNo">315</span>    };<a name="line.315"></a>
-<span class="sourceLineNo">316</span>    ZkSplitLogWorkerCoordination.ZkSplitTaskDetails splitTaskDetails =<a name="line.316"></a>
-<span class="sourceLineNo">317</span>        new ZkSplitLogWorkerCoordination.ZkSplitTaskDetails();<a name="line.317"></a>
-<span class="sourceLineNo">318</span>    splitTaskDetails.setTaskNode(curTask);<a name="line.318"></a>
-<span class="sourceLineNo">319</span>    splitTaskDetails.setCurTaskZKVersion(zkVersion);<a name="line.319"></a>
-<span class="sourceLineNo">320</span><a name="line.320"></a>
-<span class="sourceLineNo">321</span>    WALSplitterHandler hsh =<a name="line.321"></a>
-<span class="sourceLineNo">322</span>        new WALSplitterHandler(server, this, splitTaskDetails, reporter,<a name="line.322"></a>
-<span class="sourceLineNo">323</span>            this.tasksInProgress, splitTaskExecutor, mode);<a name="line.323"></a>
-<span class="sourceLineNo">324</span>    server.getExecutorService().submit(hsh);<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>   * This function calculates how many splitters it could create based on expected average tasks per<a name="line.328"></a>
-<span class="sourceLineNo">329</span>   * RS and the hard limit upper bound(maxConcurrentTasks) set by configuration. &lt;br&gt;<a name="line.329"></a>
-<span class="sourceLineNo">330</span>   * At any given time, a RS allows spawn MIN(Expected Tasks/RS, Hard Upper Bound)<a name="line.330"></a>
-<span class="sourceLineNo">331</span>   * @param numTasks current total number of available tasks<a name="line.331"></a>
-<span class="sourceLineNo">332</span>   */<a name="line.332"></a>
-<span class="sourceLineNo">333</span>  private int calculateAvailableSplitters(int numTasks) {<a name="line.333"></a>
-<span class="sourceLineNo">334</span>    // at lease one RS(itself) available<a name="line.334"></a>
-<span class="sourceLineNo">335</span>    int availableRSs = 1;<a name="line.335"></a>
-<span class="sourceLineNo">336</span>    try {<a name="line.336"></a>
-<span class="sourceLineNo">337</span>      List&lt;String&gt; regionServers =<a name="line.337"></a>
-<span class="sourceLineNo">338</span>          ZKUtil.listChildrenNoWatch(watcher, watcher.znodePaths.rsZNode);<a name="line.338"></a>
-<span class="sourceLineNo">339</span>      availableRSs = Math.max(availableRSs, (regionServers == null) ? 0 : regionServers.size());<a name="line.339"></a>
-<span class="sourceLineNo">340</span>    } catch (KeeperException e) {<a name="line.340"></a>
-<span class="sourceLineNo">341</span>      // do nothing<a name="line.341"></a>
-<span class="sourceLineNo">342</span>      LOG.debug("getAvailableRegionServers got ZooKeeper exception", e);<a name="line.342"></a>
-<span class="sourceLineNo">343</span>    }<a name="line.343"></a>
-<span class="sourceLineNo">344</span><a name="line.344"></a>
-<span class="sourceLineNo">345</span>    int expectedTasksPerRS = (numTasks / availableRSs) + ((numTasks % availableRSs == 0) ? 0 : 1);<a name="line.345"></a>
-<span class="sourceLineNo">346</span>    expectedTasksPerRS = Math.max(1, expectedTasksPerRS); // at least be one<a name="line.346"></a>
-<span class="sourceLineNo">347</span>    // calculate how many more splitters we could spawn<a name="line.347"></a>
-<span class="sourceLineNo">348</span>    return Math.min(expectedTasksPerRS, maxConcurrentTasks)<a name="line.348"></a>
-<span class="sourceLineNo">349</span>        - this.tasksInProgress.get();<a name="line.349"></a>
-<span class="sourceLineNo">350</span>  }<a name="line.350"></a>
-<span class="sourceLineNo">351</span><a name="line.351"></a>
-<span class="sourceLineNo">352</span>  /**<a name="line.352"></a>
-<span class="sourceLineNo">353</span>   * Try to own the task by transitioning the zk node data from UNASSIGNED to OWNED.<a name="line.353"></a>
-<span class="sourceLineNo">354</span>   * &lt;p&gt;<a name="line.354"></a>
-<span class="sourceLineNo">355</span>   * This method is also used to periodically heartbeat the task progress by transitioning the node<a name="line.355"></a>
-<span class="sourceLineNo">356</span>   * from OWNED to OWNED.<a name="line.356"></a>
-<span class="sourceLineNo">357</span>   * &lt;p&gt;<a name="line.357"></a>
-<span class="sourceLineNo">358</span>   * @param isFirstTime shows whther it's the first attempt.<a name="line.358"></a>
-<span class="sourceLineNo">359</span>   * @param zkw zk wathcer<a name="line.359"></a>
-<span class="sourceLineNo">360</span>   * @param server name<a name="line.360"></a>
-<span class="sourceLineNo">361</span>   * @param task to own<a name="line.361"></a>
-<span class="sourceLineNo">362</span>   * @param taskZKVersion version of the task in zk<a name="line.362"></a>
-<span class="sourceLineNo">363</span>   * @return non-negative integer value when task can be owned by current region server otherwise -1<a name="line.363"></a>
-<span class="sourceLineNo">364</span>   */<a name="line.364"></a>
-<span class="sourceLineNo">365</span>  protected static int attemptToOwnTask(boolean isFirstTime, ZooKeeperWatcher zkw,<a name="line.365"></a>
-<span class="sourceLineNo">366</span>      ServerName server, String task, RecoveryMode mode, int taskZKVersion) {<a name="line.366"></a>
-<span class="sourceLineNo">367</span>    int latestZKVersion = FAILED_TO_OWN_TASK;<a name="line.367"></a>
-<span class="sourceLineNo">368</span>    try {<a name="line.368"></a>
-<span class="sourceLineNo">369</span>      SplitLogTask slt = new SplitLogTask.Owned(server, mode);<a name="line.369"></a>
-<span class="sourceLineNo">370</span>      Stat stat = zkw.getRecoverableZooKeeper().setData(task, slt.toByteArray(), taskZKVersion);<a name="line.370"></a>
-<span class="sourceLineNo">371</span>      if (stat == null) {<a name="line.371"></a>
-<span class="sourceLineNo">372</span>        LOG.warn("zk.setData() returned null for path " + task);<a name="line.372"></a>
-<span class="sourceLineNo">373</span>        SplitLogCounters.tot_wkr_task_heartbeat_failed.incrementAndGet();<a name="line.373"></a>
-<span class="sourceLineNo">374</span>        return FAILED_TO_OWN_TASK;<a name="line.374"></a>
-<span class="sourceLineNo">375</span>      }<a name="line.375"></a>
-<span class="sourceLineNo">376</span>      latestZKVersion = stat.getVersion();<a name="line.376"></a>
-<span class="sourceLineNo">377</span>      SplitLogCounters.tot_wkr_task_heartbeat.incrementAndGet();<a name="line.377"></a>
-<span class="sourceLineNo">378</span>      return latestZKVersion;<a name="line.378"></a>
-<span class="sourceLineNo">379</span>    } catch (KeeperException e) {<a name="line.379"></a>
-<span class="sourceLineNo">380</span>      if (!isFirstTime) {<a name="line.380"></a>
-<span class="sourceLineNo">381</span>        if (e.code().equals(KeeperException.Code.NONODE)) {<a name="line.381"></a>
-<span class="sourceLineNo">382</span>          LOG.warn("NONODE failed to assert ownership for " + task, e);<a name="line.382"></a>
-<span class="sourceLineNo">383</span>        } else if (e.code().equals(KeeperException.Code.BADVERSION)) {<a name="line.383"></a>
-<span class="sourceLineNo">384</span>          LOG.warn("BADVERSION failed to assert ownership for " + task, e);<a name="line.384"></a>
-<span class="sourceLineNo">385</span>        } else {<a name="line.385"></a>
-<span class="sourceLineNo">386</span>          LOG.warn("failed to assert ownership for " + task, e);<a name="line.386"></a>
-<span class="sourceLineNo">387</span>        }<a name="line.387"></a>
-<span class="sourceLineNo">388</span>      }<a name="line.388"></a>
-<span class="sourceLineNo">389</span>    } catch (InterruptedException e1) {<a name="line.389"></a>
-<span class="sourceLineNo">390</span>      LOG.warn("Interrupted while trying to assert ownership of " + task + " "<a name="line.390"></a>
-<span class="sourceLineNo">391</span>          + StringUtils.stringifyException(e1));<a name="line.391"></a>
-<span class="sourceLineNo">392</span>      Thread.currentThread().interrupt();<a name="line.392"></a>
-<span class="sourceLineNo">393</span>    }<a name="line.393"></a>
-<span class="sourceLineNo">394</span>    SplitLogCounters.tot_wkr_task_heartbeat_failed.incrementAndGet();<a name="line.394"></a>
-<span class="sourceLineNo">395</span>    return FAILED_TO_OWN_TASK;<a name="line.395"></a>
-<span class="sourceLineNo">396</span>  }<a name="line.396"></a>
-<span class="sourceLineNo">397</span><a name="line.397"></a>
-<span class="sourceLineNo">398</span>  /**<a name="line.398"></a>
-<span class="sourceLineNo">399</span>   * Wait for tasks to become available at /hbase/splitlog zknode. Grab a task one at a time. This<a name="line.399"></a>
-<span class="sourceLineNo">400</span>   * policy puts an upper-limit on the number of simultaneous log splitting that could be happening<a name="line.400"></a>
-<span class="sourceLineNo">401</span>   * in a cluster.<a name="line.401"></a>
-<span class="sourceLineNo">402</span>   * &lt;p&gt;<a name="line.402"></a>
-<span class="sourceLineNo">403</span>   * Synchronization using &lt;code&gt;taskReadyLock&lt;/code&gt; ensures that it will try to grab every task<a name="line.403"></a>
-<span class="sourceLineNo">404</span>   * that has been put up<a name="line.404"></a>
-<span class="sourceLineNo">405</span>   * @throws InterruptedException<a name="line.405"></a>
-<span class="sourceLineNo">406</span>   */<a name="line.406"></a>
-<span class="sourceLineNo">407</span>  @Override<a name="line.407"></a>
-<span class="sourceLineNo">408</span>  public void taskLoop() throws InterruptedException {<a name="line.408"></a>
-<span class="sourceLineNo">409</span>    while (!shouldStop) {<a name="line.409"></a>
-<span class="sourceLineNo">410</span>      int seq_start = taskReadySeq.get();<a name="line.410"></a>
-<span class="sourceLineNo">411</span>      List&lt;String&gt; paths = null;<a name="line.411"></a>
-<span class="sourceLineNo">412</span>      paths = getTaskList();<a name="line.412"></a>
-<span class="sourceLineNo">413</span>      if (paths == null) {<a name="line.413"></a>
-<span class="sourceLineNo">414</span>        LOG.warn("Could not get tasks, did someone remove " + watcher.znodePaths.splitLogZNode<a name="line.414"></a>
-<span class="sourceLineNo">415</span>            + " ... worker thread exiting.");<a name="line.415"></a>
-<span class="sourceLineNo">416</span>        return;<a name="line.416"></a>
-<span class="sourceLineNo">417</span>      }<a name="line.417"></a>
-<span class="sourceLineNo">418</span>      // pick meta wal firstly<a name="line.418"></a>
-<span class="sourceLineNo">419</span>      int offset = (int) (Math.random() * paths.size());<a name="line.419"></a>
-<span class="sourceLineNo">420</span>      for (int i = 0; i &lt; paths.size(); i++) {<a name="line.420"></a>
-<span class="sourceLineNo">421</span>        if (AbstractFSWALProvider.isMetaFile(paths.get(i))) {<a name="line.421"></a>
-<span class="sourceLineNo">422</span>          offset = i;<a name="line.422"></a>
-<span class="sourceLineNo">423</span>          break;<a name="line.423"></a>
-<span class="sourceLineNo">424</span>        }<a name="line.424"></a>
-<span class="sourceLineNo">425</span>      }<a name="line.425"></a>
-<span class="sourceLineNo">426</span>      int numTasks = paths.size();<a name="line.426"></a>
-<span class="sourceLineNo">427</span>      for (int i = 0; i &lt; numTasks; i++) {<a name="line.427"></a>
-<span class="sourceLineNo">428</span>        int idx = (i + offset) % paths.size();<a name="line.428"></a>
-<span class="sourceLineNo">429</span>        // don't call ZKSplitLog.getNodeName() because that will lead to<a name="line.429"></a>
-<span class="sourceLineNo">430</span>        // double encoding of the path name<a name="line.430"></a>
-<span class="sourceLineNo">431</span>        if (this.calculateAvailableSplitters(numTasks) &gt; 0) {<a name="line.431"></a>
-<span class="sourceLineNo">432</span>          grabTask(ZKUtil.joinZNode(watcher.znodePaths.splitLogZNode, paths.get(idx)));<a name="line.432"></a>
-<span class="sourceLineNo">433</span>        } else {<a name="line.433"></a>
-<span class="sourceLineNo">434</span>          LOG.debug("Current region server " + server.getServerName() + " has "<a name="line.434"></a>
-<span class="sourceLineNo">435</span>              + this.tasksInProgress.get() + " tasks in progress and can't take more.");<a name="line.435"></a>
-<span class="sourceLineNo">436</span>          break;<a name="line.436"></a>
-<span class="sourceLineNo">437</span>        }<a name="line.437"></a>
-<span class="sourceLineNo">438</span>        if (shouldStop) {<a name="line.438"></a>
-<span class="sourceLineNo">439</span>          return;<a name="line.439"></a>
-<span class="sourceLineNo">440</span>        }<a name="line.440"></a>
-<span class="sourceLineNo">441</span>      }<a name="line.441"></a>
-<span class="sourceLineNo">442</span>      SplitLogCounters.tot_wkr_task_grabing.incrementAndGet();<a name="line.442"></a>
-<span class="sourceLineNo">443</span>      synchronized (taskReadyLock) {<a name="line.443"></a>
-<span class="sourceLineNo">444</span>        while (seq_start == taskReadySeq.get()) {<a name="line.444"></a>
-<span class="sourceLineNo">445</span>          taskReadyLock.wait(checkInterval);<a name="line.445"></a>
-<span class="sourceLineNo">446</span>          if (server != null) {<a name="line.446"></a>
-<span class="sourceLineNo">447</span>            // check to see if we have stale recovering regions in our internal memory state<a name="line.447"></a>
-<span class="sourceLineNo">448</span>            Map&lt;String, Region&gt; recoveringRegions = server.getRecoveringRegions();<a name="line.448"></a>
-<span class="sourceLineNo">449</span>            if (!recoveringRegions.isEmpty()) {<a name="line.449"></a>
-<span class="sourceLineNo">450</span>              // Make a local copy to prevent ConcurrentModificationException when other threads<a name="line.450"></a>
-<span class="sourceLineNo">451</span>              // modify recoveringRegions<a name="line.451"></a>
-<span class="sourceLineNo">452</span>              List&lt;String&gt; tmpCopy = new ArrayList&lt;&gt;(recoveringRegions.keySet());<a name="line.452"></a>
-<span class="sourceLineNo">453</span>              int listSize = tmpCopy.size();<a name="line.453"></a>
-<span class="sourceLineNo">454</span>              for (int i = 0; i &lt; listSize; i++) {<a name="line.454"></a>
-<span class="sourceLineNo">455</span>                String region = tmpCopy.get(i);<a name="line.455"></a>
-<span class="sourceLineNo">456</span>                String nodePath = ZKUtil.joinZNode(watcher.znodePaths.recoveringRegionsZNode,<a name="line.456"></a>
-<span class="sourceLineNo">457</span>                  region);<a name="line.457"></a>
-<span class="sourceLineNo">458</span>                try {<a name="line.458"></a>
-<span class="sourceLineNo">459</span>                  if (ZKUtil.checkExists(watcher, nodePath) == -1) {<a name="line.459"></a>
-<span class="sourceLineNo">460</span>                    server.getExecutorService().submit(<a name="line.460"></a>
-<span class="sourceLineNo">461</span>                      new FinishRegionRecoveringHandler(server, region, nodePath));<a name="line.461"></a>
-<span class="sourceLineNo">462</span>                  } else {<a name="line.462"></a>
-<span class="sourceLineNo">463</span>                    // current check is a defensive(or redundant) mechanism to prevent us from<a name="line.463"></a>
-<span class="sourceLineNo">464</span>                    // having stale recovering regions in our internal RS memory state while<a name="line.464"></a>
-<span class="sourceLineNo">465</span>                    // zookeeper(source of truth) says differently. We stop at the first good one<a name="line.465"></a>
-<span class="sourceLineNo">466</span>                    // because we should not have a single instance such as this in normal case so<a name="line.466"></a>
-<span class="sourceLineNo">467</span>                    // check the first one is good enough.<a name="line.467"></a>
-<span class="sourceLineNo">468</span>                    break;<a name="line.468"></a>
-<span class="sourceLineNo">469</span>                  }<a name="line.469"></a>
-<span class="sourceLineNo">470</span>                } catch (KeeperException e) {<a name="line.470"></a>
-<span class="sourceLineNo">471</span>                  // ignore zookeeper error<a name="line.471"></a>
-<span class="sourceLineNo">472</span>                  LOG.debug("Got a zookeeper when trying to open a recovering region", e);<a name="line.472"></a>
-<span class="sourceLineNo">473</span>                  break;<a name="line.473"></a>
-<span class="sourceLineNo">474</span>                }<a name="line.474"></a>
-<span class="sourceLineNo">475</span>              }<a name="line.475"></a>
-<span class="sourceLineNo">476</span>            }<a name="line.476"></a>
-<span class="sourceLineNo">477</span>          }<a name="line.477"></a>
-<span class="sourceLineNo">478</span>        }<a name="line.478"></a>
-<span class="sourceLineNo">479</span>      }<a name="line.479"></a>
-<span class="sourceLineNo">480</span>    }<a name="line.480"></a>
-<span class="sourceLineNo">481</span>  }<a name="line.481"></a>
-<span class="sourceLineNo">482</span><a name="line.482"></a>
-<span class="sourceLineNo">483</span>  private List&lt;String&gt; getTaskList() throws InterruptedException {<a name="line.483"></a>
-<span class="sourceLineNo">484</span>    List&lt;String&gt; childrenPaths = null;<a name="line.484"></a>
-<span class="sourceLineNo">485</span>    long sleepTime = 1000;<a name="line.485"></a>
-<span class="sourceLineNo">486</span>    // It will be in loop till it gets the list of children or<a name="line.486"></a>
-<span class="sourceLineNo">487</span>    // it will come out if worker thread exited.<a name="line.487"></a>
-<span class="sourceLineNo">488</span>    while (!shouldStop) {<a name="line.488"></a>
-<span class="sourceLineNo">489</span>      try {<a name="line.489"></a>
-<span class="sourceLineNo">490</span>        childrenPaths = ZKUtil.listChildrenAndWatchForNewChildren(watcher,<a name="line.490"></a>
-<span class="sourceLineNo">491</span>          watcher.znodePaths.splitLogZNode);<a name="line.491"></a>
-<span class="sourceLineNo">492</span>        if (childrenPaths != null) {<a name="line.492"></a>
-<span class="sourceLineNo">493</span>          return childrenPaths;<a name="line.493"></a>
-<span class="sourceLineNo">494</span>        }<a name="line.494"></a>
-<span class="sourceLineNo">495</span>      } catch (KeeperException e) {<a name="line.495"></a>
-<span class="sourceLineNo">496</span>        LOG.warn("Could not get children of znode " + watcher.znodePaths.splitLogZNode, e);<a name="line.496"></a>
-<span class="sourceLineNo">497</span>      }<a name="line.497"></a>
-<span class="sourceLineNo">498</span>      LOG.debug("Retry listChildren of znode " + watcher.znodePaths.splitLogZNode<a name="line.498"></a>
-<span class="sourceLineNo">499</span>          + " after sleep for " + sleepTime + "ms!");<a name="line.499"></a>
-<span class="sourceLineNo">500</span>      Thread.sleep(sleepTime);<a name="line.500"></a>
-<span class="sourceLineNo">501</span>    }<a name="line.501"></a>
-<span class="sourceLineNo">502</span>    return childrenPaths;<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>  @Override<a name="line.505"></a>
-<span class="sourceLineNo">506</span>  public void markCorrupted(Path rootDir, String name, FileSystem fs) {<a name="line.506"></a>
-<span class="sourceLineNo">507</span>    ZKSplitLog.markCorrupted(rootDir, name, fs);<a name="line.507"></a>
-<span class="sourceLineNo">508</span>  }<a name="line.508"></a>
-<span class="sourceLineNo">509</span><a name="line.509"></a>
-<span class="sourceLineNo">510</span>  @Override<a name="line.510"></a>
-<span class="sourceLineNo">511</span>  public boolean isReady() throws InterruptedException {<a name="line.511"></a>
-<span class="sourceLineNo">512</span>    int result = -1;<a name="line.512"></a>
-<span class="sourceLineNo">513</span>    try {<a name="line.513"></a>
-<span class="sourceLineNo">514</span>      result = ZKUtil.checkExists(watcher, watcher.znodePaths.splitLogZNode);<a name="line.514"></a>
-<span class="sourceLineNo">515</span>    } catch (KeeperException e) {<a name="line.515"></a>
-<span class="sourceLineNo">516</span>      // ignore<a name="line.516"></a>
-<span class="sourceLineNo">517</span>      LOG.warn("Exception when checking for " + watcher.znodePaths.splitLogZNode<a name="line.517"></a>
-<span class="sourceLineNo">518</span>          + " ... retrying", e);<a name="line.518"></a>
-<span class="sourceLineNo">519</span>    }<a name="line.519"></a>
-<span class="sourceLineNo">520</span>    if (result == -1) {<a name="line.520"></a>
-<span class="sourceLineNo">521</span>      LOG.info(watcher.znodePaths.splitLogZNode<a name="line.521"></a>
-<span class="sourceLineNo">522</span>          + " znode does not exist, waiting for master to create");<a name="line.522"></a>
-<span class="sourceLineNo">523</span>      Thread.sleep(1000);<a name="line.523"></a>
-<span class="sourceLineNo">524</span>    }<a name="line.524"></a>
-<span class="sourceLineNo">525</span>    return (result != -1);<a name="line.525"></a>
-<span class="sourceLineNo">526</span>  }<a name="line.526"></a>
-<span class="sourceLineNo">527</span><a name="line.527"></a>
-<span class="sourceLineNo">528</span>  @Override<a name="line.528"></a>
-<span class="sourceLineNo">529</span>  public int getTaskReadySeq() {<a name="line.529"></a>
-<span class="sourceLineNo">530</span>    return taskReadySeq.get();<a name="line.530"></a>
-<span class="sourceLineNo">531</span>  }<a name="line.531"></a>
-<span class="sourceLineNo">532</span><a name="line.532"></a>
-<span class="sourceLineNo">533</span>  @Override<a name="line.533"></a>
-<span class="sourceLineNo">534</span>  public void registerListener() {<a name="line.534"></a>
-<span class="sourceLineNo">535</span>    watcher.registerListener(this);<a name="line.535"></a>
-<span class="sourceLineNo">536</span>  }<a name="line.536"></a>
-<span class="sourceLineNo">537</span><a name="line.537"></a>
-<span class="sourceLineNo">538</span>  @Override<a name="line.538"></a>
-<span class="sourceLineNo">539</span>  public void removeListener() {<a name="line.539"></a>
-<span class="sourceLineNo">540</span>    watcher.unregisterListener(this);<a name="line.540"></a>
-<span class="sourceLineNo">541</span>  }<a name="line.541"></a>
-<span class="sourceLineNo">542</span><a name="line.542"></a>
+<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.zookeeper.RecoverableZooKeeper;<a name="line.53"></a>
+<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.zookeeper.ZKSplitLog;<a name="line.54"></a>
+<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.55"></a>
+<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.zookeeper.ZooKeeperListener;<a name="line.56"></a>
+<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher;<a name="line.57"></a>
+<span class="sourceLineNo">058</span>import org.apache.hadoop.util.StringUtils;<a name="line.58"></a>
+<span class="sourceLineNo">059</span>import org.apache.zookeeper.AsyncCallback;<a name="line.59"></a>
+<span class="sourceLineNo">060</span>import org.apache.zookeeper.KeeperException;<a name="line.60"></a>
+<span class="sourceLineNo">061</span>import org.apache.zookeeper.data.Stat;<a name="line.61"></a>
+<span class="sourceLineNo">062</span><a name="line.62"></a>
+<span class="sourceLineNo">063</span>/**<a name="line.63"></a>
+<span class="sourceLineNo">064</span> * ZooKeeper based implementation of {@link SplitLogWorkerCoordination}<a name="line.64"></a>
+<span class="sourceLineNo">065</span> * It listen for changes in ZooKeeper and<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>@InterfaceAudience.Private<a name="line.68"></a>
+<span class="sourceLineNo">069</span>public class ZkSplitLogWorkerCoordination extends ZooKeeperListener implements<a name="line.69"></a>
+<span class="sourceLineNo">070</span>    SplitLogWorkerCoordination {<a name="line.70"></a>
+<span class="sourceLineNo">071</span><a name="line.71"></a>
+<span class="sourceLineNo">072</span>  private static final Log LOG = LogFactory.getLog(ZkSplitLogWorkerCoordination.class);<a name="line.72"></a>
+<span class="sourceLineNo">073</span><a name="line.73"></a>
+<span class="sourceLineNo">074</span>  private static final int checkInterval = 5000; // 5 seconds<a name="line.74"></a>
+<span class="sourceLineNo">075</span>  private static final int FAILED_TO_OWN_TASK = -1;<a name="line.75"></a>
+<span class="sourceLineNo">076</span><a name="line.76"></a>
+<span class="sourceLineNo">077</span>  private  SplitLogWorker worker;<a name="line.77"></a>
+<span class="sourceLineNo">078</span><a name="line.78"></a>
+<span class="sourceLineNo">079</span>  private TaskExecutor splitTaskExecutor;<a name="line.79"></a>
+<span class="sourceLineNo">080</span><a name="line.80"></a>
+<span class="sourceLineNo">081</span>  private final Object taskReadyLock = new Object();<a name="line.81"></a>
+<span class="sourceLineNo">082</span>  private AtomicInteger taskReadySeq = new AtomicInteger(0);<a name="line.82"></a>
+<span class="sourceLineNo">083</span>  private volatile String currentTask = null;<a name="line.83"></a>
+<span class="sourceLineNo">084</span>  private int currentVersion;<a name="line.84"></a>
+<span class="sourceLineNo">085</span>  private volatile boolean shouldStop = false;<a name="line.85"></a>
+<span class="sourceLineNo">086</span>  private final Object grabTaskLock = new Object();<a name="line.86"></a>
+<span class="sourceLineNo">087</span>  private boolean workerInGrabTask = false;<a name="line.87"></a>
+<span class="sourceLineNo">088</span>  private int reportPeriod;<a name="line.88"></a>
+<span class="sourceLineNo">089</span>  private RegionServerServices server = null;<a name="line.89"></a>
+<span class="sourceLineNo">090</span>  protected final AtomicInteger tasksInProgress = new AtomicInteger(0);<a name="line.90"></a>
+<span class="sourceLineNo">091</span>  private int maxConcurrentTasks = 0;<a name="line.91"></a>
+<span class="sourceLineNo">092</span><a name="line.92"></a>
+<span class="sourceLineNo">093</span>  private final ZkCoordinatedStateManager manager;<a name="line.93"></a>
+<span class="sourceLineNo">094</span><a name="line.94"></a>
+<span class="sourceLineNo">095</span>  public ZkSplitLogWorkerCoordination(ZkCoordinatedStateManager zkCoordinatedStateManager,<a name="line.95"></a>
+<span class="sourceLineNo">096</span>      ZooKeeperWatcher watcher) {<a name="line.96"></a>
+<span class="sourceLineNo">097</span>    super(watcher);<a name="line.97"></a>
+<span class="sourceLineNo">098</span>    manager = zkCoordinatedStateManager;<a name="line.98"></a>
+<span class="sourceLineNo">099</span><a name="line.99"></a>
+<span class="sourceLineNo">100</span>  }<a name="line.100"></a>
+<span class="sourceLineNo">101</span><a name="line.101"></a>
+<span class="sourceLineNo">102</span>  /**<a name="line.102"></a>
+<span class="sourceLineNo">103</span>   * Override handler from {@link ZooKeeperListener}<a name="line.103"></a>
+<span class="sourceLineNo">104</span>   */<a name="line.104"></a>
+<span class="sourceLineNo">105</span>  @Override<a name="line.105"></a>
+<span class="sourceLineNo">106</span>  public void nodeChildrenChanged(String path) {<a name="line.106"></a>
+<span class="sourceLineNo">107</span>    if (path.equals(watcher.znodePaths.splitLogZNode)) {<a name="line.107"></a>
+<span class="sourceLineNo">108</span>      if (LOG.isTraceEnabled()) LOG.trace("tasks arrived or departed on " + path);<a name="line.108"></a>
+<span class="sourceLineNo">109</span>      synchronized (taskReadyLock) {<a name="line.109"></a>
+<span class="sourceLineNo">110</span>        this.taskReadySeq.incrementAndGet();<a name="line.110"></a>
+<span class="sourceLineNo">111</span>        taskReadyLock.notify();<a name="line.111"></a>
+<span class="sourceLineNo">112</span>      }<a name="line.112"></a>
+<span class="sourceLineNo">113</span>    }<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>   * Override handler from {@link ZooKeeperListener}<a name="line.117"></a>
+<span class="sourceLineNo">118</span>   */<a name="line.118"></a>
+<span class="sourceLineNo">119</span>  @Override<a name="line.119"></a>
+<span class="sourceLineNo">120</span>  public void nodeDataChanged(String path) {<a name="line.120"></a>
+<span class="sourceLineNo">121</span>    // there will be a self generated dataChanged event every time attemptToOwnTask()<a name="line.121"></a>
+<span class="sourceLineNo">122</span>    // heartbeats the task znode by upping its version<a name="line.122"></a>
+<span class="sourceLineNo">123</span>    synchronized (grabTaskLock) {<a name="line.123"></a>
+<span class="sourceLineNo">124</span>      if (workerInGrabTask) {<a name="line.124"></a>
+<span class="sourceLineNo">125</span>        // currentTask can change<a name="line.125"></a>
+<span class="sourceLineNo">126</span>        String taskpath = currentTask;<a name="line.126"></a>
+<span class="sourceLineNo">127</span>        if (taskpath != null &amp;&amp; taskpath.equals(path)) {<a name="line.127"></a>
+<span class="sourceLineNo">128</span>          getDataSetWatchAsync();<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>   * Override setter from {@link SplitLogWorkerCoordination}<a name="line.135"></a>
+<span class="sourceLineNo">136</span>   */<a name="line.136"></a>
+<span class="sourceLineNo">137</span>  @Override<a name="line.137"></a>
+<span class="sourceLineNo">138</span>  public void init(RegionServerServices server, Configuration conf,<a name="line.138"></a>
+<span class="sourceLineNo">139</span>      TaskExecutor splitExecutor, SplitLogWorker worker) {<a name="line.139"></a>
+<span class="sourceLineNo">140</span>    this.server = server;<a name="line.140"></a>
+<span class="sourceLineNo">141</span>    this.worker = worker;<a name="line.141"></a>
+<span class="sourceLineNo">142</span>    this.splitTaskExecutor = splitExecutor;<a name="line.142"></a>
+<span class="sourceLineNo">143</span>    maxConcurrentTasks = conf.getInt("hbase.regionserver.wal.max.splitters", DEFAULT_MAX_SPLITTERS);<a name="line.143"></a>
+<span class="sourceLineNo">144</span>    reportPeriod =<a name="line.144"></a>
+<span class="sourceLineNo">145</span>        conf.getInt("hbase.splitlog.report.period",<a name="line.145"></a>
+<span class="sourceLineNo">146</span>          conf.getInt(HConstants.HBASE_SPLITLOG_MANAGER_TIMEOUT,<a name="line.146"></a>
+<span class="sourceLineNo">147</span>            ZKSplitLogManagerCoordination.DEFAULT_TIMEOUT) / 3);<a name="line.147"></a>
+<span class="sourceLineNo">148</span>  }<a name="line.148"></a>
+<span class="sourceLineNo">149</span><a name="line.149"></a>
+<span class="sourceLineNo">150</span>  /* Support functions for ZooKeeper async callback */<a name="line.150"></a>
+<span class="sourceLineNo">151</span><a name="line.151"></a>
+<span class="sourceLineNo">152</span>  void getDataSetWatchFailure(String path) {<a name="line.152"></a>
+<span class="sourceLineNo">153</span>    synchronized (grabTaskLock) {<a name="line.153"></a>
+<span class="sourceLineNo">154</span>      if (workerInGrabTask) {<a name="line.154"></a>
+<span class="sourceLineNo">155</span>        // currentTask can change but that's ok<a name="line.155"></a>
+<span class="sourceLineNo">156</span>        String taskpath = currentTask;<a name="line.156"></a>
+<span class="sourceLineNo">157</span>        if (taskpath != null &amp;&amp; taskpath.equals(path)) {<a name="line.157"></a>
+<span class="sourceLineNo">158</span>          LOG.info("retrying data watch on " + path);<a name="line.158"></a>
+<span class="sourceLineNo">159</span>          SplitLogCounters.tot_wkr_get_data_retry.incrementAndGet();<a name="line.159"></a>
+<span class="sourceLineNo">160</span>          getDataSetWatchAsync();<a name="line.160"></a>
+<span class="sourceLineNo">161</span>        } else {<a name="line.161"></a>
+<span class="sourceLineNo">162</span>          // no point setting a watch on the task which this worker is not<a name="line.162"></a>
+<span class="sourceLineNo">163</span>          // working upon anymore<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>    }<a name="line.166"></a>
+<span class="sourceLineNo">167</span>  }<a name="line.167"></a>
+<span class="sourceLineNo">168</span><a name="line.168"></a>
+<span class="sourceLineNo">169</span>  public void getDataSetWatchAsync() {<a name="line.169"></a>
+<span class="sourceLineNo">170</span>    watcher.getRecoverableZooKeeper().getZooKeeper()<a name="line.170"></a>
+<span class="sourceLineNo">171</span>        .getData(currentTask, watcher, new GetDataAsyncCallback(), null);<a name="line.171"></a>
+<span class="sourceLineNo">172</span>    SplitLogCounters.tot_wkr_get_data_queued.incrementAndGet();<a name="line.172"></a>
+<span class="sourceLineNo">173</span>  }<a name="line.173"></a>
+<span class="sourceLineNo">174</span><a name="line.174"></a>
+<span class="sourceLineNo">175</span>  void getDataSetWatchSuccess(String path, byte[] data) {<a name="line.175"></a>
+<span class="sourceLineNo">176</span>    SplitLogTask slt;<a name="line.176"></a>
+<span class="sourceLineNo">177</span>    try {<a name="line.177"></a>
+<span class="sourceLineNo">178</span>      slt = SplitLogTask.parseFrom(data);<a name="line.178"></a>
+<span class="sourceLineNo">179</span>    } catch (DeserializationException e) {<a name="line.179"></a>
+<span class="sourceLineNo">180</span>      LOG.warn("Failed parse", e);<a name="line.180"></a>
+<span class="sourceLineNo">181</span>      return;<a name="line.181"></a>
+<span class="sourceLineNo">182</span>    }<a name="line.182"></a>
+<span class="sourceLineNo">183</span>    synchronized (grabTaskLock) {<a name="line.183"></a>
+<span class="sourceLineNo">184</span>      if (workerInGrabTask) {<a name="line.184"></a>
+<span class="sourceLineNo">185</span>        // currentTask can change but that's ok<a name="line.185"></a>
+<span class="sourceLineNo">186</span>        String taskpath = currentTask;<a name="line.186"></a>
+<span class="sourceLineNo">187</span>        if (taskpath != null &amp;&amp; taskpath.equals(path)) {<a name="line.187"></a>
+<span class="sourceLineNo">188</span>          ServerName serverName = manager.getServer().getServerName();<a name="line.188"></a>
+<span class="sourceLineNo">189</span>          // have to compare data. cannot compare version because then there<a name="line.189"></a>
+<span class="sourceLineNo">190</span>          // will be race with attemptToOwnTask()<a name="line.190"></a>
+<span class="sourceLineNo">191</span>          // cannot just check whether the node has been transitioned to<a name="line.191"></a>
+<span class="sourceLineNo">192</span>          // UNASSIGNED because by the time this worker sets the data watch<a name="line.192"></a>
+<span class="sourceLineNo">193</span>          // the node might have made two transitions - from owned by this<a name="line.193"></a>
+<span class="sourceLineNo">194</span>          // worker to unassigned to owned by another worker<a name="line.194"></a>
+<span class="sourceLineNo">195</span>          if (!slt.isOwned(serverName) &amp;&amp; !slt.isDone(serverName) &amp;&amp; !slt.isErr(serverName)<a name="line.195"></a>
+<span class="sourceLineNo">196</span>              &amp;&amp; !slt.isResigned(serverName)) {<a name="line.196"></a>
+<span class="sourceLineNo">197</span>            LOG.info("task " + taskpath + " preempted from " + serverName<a name="line.197"></a>
+<span class="sourceLineNo">198</span>                + ", current task state and owner=" + slt.toString());<a name="line.198"></a>
+<span class="sourceLineNo">199</span>            worker.stopTask();<a name="line.199"></a>
+<span class="sourceLineNo">200</span>          }<a name="line.200"></a>
+<span class="sourceLineNo">201</span>        }<a name="line.201"></a>
+<span class="sourceLineNo">202</span>      }<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><a name="line.205"></a>
+<span class="sourceLineNo">206</span>  /**<a name="line.206"></a>
+<span class="sourceLineNo">207</span>   * try to grab a 'lock' on the task zk node to own and execute the task.<a name="line.207"></a>
+<span class="sourceLineNo">208</span>   * &lt;p&gt;<a name="line.208"></a>
+<span class="sourceLineNo">209</span>   * @param path zk node for the task<a name="line.209"></a>
+<span class="sourceLineNo">210</span>   */<a name="line.210"></a>
+<span class="sourceLineNo">211</span>  private void grabTask(String path) {<a name="line.211"></a>
+<span class="sourceLineNo">212</span>    Stat stat = new Stat();<a name="line.212"></a>
+<span class="sourceLineNo">213</span>    byte[] data;<a name="line.213"></a>
+<span class="sourceLineNo">214</span>    synchronized (grabTaskLock) {<a name="line.214"></a>
+<span class="sourceLineNo">215</span>      currentTask = path;<a name="line.215"></a>
+<span class="sourceLineNo">216</span>      workerInGrabTask = true;<a name="line.216"></a>
+<span class="sourceLineNo">217</span>      if (Thread.interrupted()) {<a name="line.217"></a>
+<span class="sourceLineNo">218</span>        return;<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>    try {<a name="line.221"></a>
+<span class="sourceLineNo">222</span>      try {<a name="line.222"></a>
+<span class="sourceLineNo">223</span>        if ((data = ZKUtil.getDataNoWatch(watcher, path, stat)) == null) {<a name="line.223"></a>
+<span class="sourceLineNo">224</span>          SplitLogCounters.tot_wkr_failed_to_grab_task_no_data.incrementAndGet();<a name="line.224"></a>
+<span class="sourceLineNo">225</span>          return;<a name="line.225"></a>
+<span class="sourceLineNo">226</span>        }<a name="line.226"></a>
+<span class="sourceLineNo">227</span>      } catch (KeeperException e) {<a name="line.227"></a>
+<span class="sourceLineNo">228</span>        LOG.warn("Failed to get data for znode " + path, e);<a name="line.228"></a>
+<span class="sourceLineNo">229</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_exception.incrementAndGet();<a name="line.229"></a>
+<span class="sourceLineNo">230</span>        return;<a name="line.230"></a>
+<span class="sourceLineNo">231</span>      }<a name="line.231"></a>
+<span class="sourceLineNo">232</span>      SplitLogTask slt;<a name="line.232"></a>
+<span class="sourceLineNo">233</span>      try {<a name="line.233"></a>
+<span class="sourceLineNo">234</span>        slt = SplitLogTask.parseFrom(data);<a name="line.234"></a>
+<span class="sourceLineNo">235</span>      } catch (DeserializationException e) {<a name="line.235"></a>
+<span class="sourceLineNo">236</span>        LOG.warn("Failed parse data for znode " + path, e);<a name="line.236"></a>
+<span class="sourceLineNo">237</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_exception.incrementAndGet();<a name="line.237"></a>
+<span class="sourceLineNo">238</span>        return;<a name="line.238"></a>
+<span class="sourceLineNo">239</span>      }<a name="line.239"></a>
+<span class="sourceLineNo">240</span>      if (!slt.isUnassigned()) {<a name="line.240"></a>
+<span class="sourceLineNo">241</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_owned.incrementAndGet();<a name="line.241"></a>
+<span class="sourceLineNo">242</span>        return;<a name="line.242"></a>
+<span class="sourceLineNo">243</span>      }<a name="line.243"></a>
+<span class="sourceLineNo">244</span><a name="line.244"></a>
+<span class="sourceLineNo">245</span>      currentVersion =<a name="line.245"></a>
+<span class="sourceLineNo">246</span>          attemptToOwnTask(true, watcher, server.getServerName(), path,<a name="line.246"></a>
+<span class="sourceLineNo">247</span>            slt.getMode(), stat.getVersion());<a name="line.247"></a>
+<span class="sourceLineNo">248</span>      if (currentVersion &lt; 0) {<a name="line.248"></a>
+<span class="sourceLineNo">249</span>        SplitLogCounters.tot_wkr_failed_to_grab_task_lost_race.incrementAndGet();<a name="line.249"></a>
+<span class="sourceLineNo">250</span>        return;<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>      if (ZKSplitLog.isRescanNode(watcher, currentTask)) {<a name="line.253"></a>
+<span class="sourceLineNo">254</span>        ZkSplitLogWorkerCoordination.ZkSplitTaskDetails splitTaskDetails =<a name="line.254"></a>
+<span class="sourceLineNo">255</span>            new ZkSplitLogWorkerCoordination.ZkSplitTaskDetails();<a name="line.255"></a>
+<span class="sourceLineNo">256</span>        splitTaskDetails.setTaskNode(currentTask);<a name="line.256"></a>
+<span class="sourceLineNo">257</span>        splitTaskDetails.setCurTaskZKVersion(new MutableInt(currentVersion));<a name="line.257"></a>
+<span class="sourceLineNo">258</span><a name="line.258"></a>
+<span class="sourceLineNo">259</span>        endTask(new SplitLogTask.Done(server.getServerName(), slt.getMode()),<a name="line.259"></a>
+<span class="sourceLineNo">260</span>          SplitLogCounters.tot_wkr_task_acquired_rescan, splitTaskDetails);<a name="line.260"></a>
+<span class="sourceLineNo">261</span>        return;<a name="line.261"></a>
+<span class="sourceLineNo">262</span>      }<a name="line.262"></a>
+<span class="sourceLineNo">263</span><a name="line.263"></a>
+<span class="sourceLineNo">264</span>      LOG.info("worker " + server.getServerName() + " acquired task " + path);<a name="line.264"></a>
+<span class="sourceLineNo">265</span>      SplitLogCounters.tot_wkr_task_acquired.incrementAndGet();<a name="line.265"></a>
+<span class="sourceLineNo">266</span>      getDataSetWatchAsync();<a name="line.266"></a>
+<span class="sourceLineNo">267</span><a name="line.267"></a>
+<span class="sourceLineNo">268</span>      submitTask(path, slt.getMode(), currentVersion, reportPeriod);<a name="line.268"></a>
+<span class="sourceLineNo">269</span><a name="line.269"></a>
+<span class="sourceLineNo">270</span>      // after a successful submit, sleep a little bit to allow other RSs to grab the rest tasks<a name="line.270"></a>
+<span class="sourceLineNo">271</span>      try {<a name="line.271"></a>
+<span class="sourceLineNo">272</span>        int sleepTime = RandomUtils.nextInt(500) + 500;<a name="line.272"></a>
+<span class="sourceLineNo">273</span>        Thread.sleep(sleepTime);<a name="line.273"></a>
+<span class="sourceLineNo">274</span>      } catch (InterruptedException e) {<a name="line.274"></a>
+<span class="sourceLineNo">275</span>        LOG.warn("Interrupted while yielding for other region servers", e);<a name="line.275"></a>
+<span class="sourceLineNo">276</span>        Thread.currentThread().interrupt();<a name="line.276"></a>
+<span class="sourceLineNo">277</span>      }<a name="line.277"></a>
+<span class="sourceLineNo">278</span>    } finally {<a name="line.278"></a>
+<span class="sourceLineNo">279</span>      synchronized (grabTaskLock) {<a name="line.279"></a>
+<span class="sourceLineNo">280</span>        workerInGrabTask = false;<a name="line.280"></a>
+<span class="sourceLineNo">281</span>        // clear the interrupt from stopTask() otherwise the next task will<a name="line.281"></a>
+<span class="sourceLineNo">282</span>        // suffer<a name="line.282"></a>
+<span class="sourceLineNo">283</span>        Thread.interrupted();<a name="line.283"></a>
+<span class="sourceLineNo">284</span>      }<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><a name="line.287"></a>
+<span class="sourceLineNo">288</span>  /**<a name="line.288"></a>
+<span class="sourceLineNo">289</span>   * Submit a log split task to executor service<a name="line.289"></a>
+<span class="sourceLineNo">290</span>   * @param curTask task to submit<a name="line.290"></a>
+<span class="sourceLineNo">291</span>   * @param curTaskZKVersion current version of task<a name="line.291"></a>
+<span class="sourceLineNo">292</span>   */<a name="line.292"></a>
+<span class="sourceLineNo">293</span>  void submitTask(final String curTask, final RecoveryMode mode, final int curTaskZKVersion,<a name="line.293"></a>
+<span class="sourceLineNo">294</span>      final int reportPeriod) {<a name="line.294"></a>
+<span class="sourceLineNo">295</span>    final MutableInt zkVersion = new MutableInt(curTaskZKVersion);<a name="line.295"></a>
+<span class="sourceLineNo">296</span><a name="line.296"></a>
+<span class="sourceLineNo">297</span>    CancelableProgressable reporter = new CancelableProgressable() {<a name="line.297"></a>
+<span class="sourceLineNo">298</span>      private long last_report_at = 0;<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 boolean progress() {<a name="line.301"></a>
+<span class="sourceLineNo">302</span>        long t = EnvironmentEdgeManager.currentTime();<a name="line.302"></a>
+<span class="sourceLineNo">303</span>        if ((t - last_report_at) &gt; reportPeriod) {<a name="line.303"></a>
+<span class="sourceLineNo">304</span>          last_report_at = t;<a name="line.304"></a>
+<span class="sourceLineNo">305</span>          int latestZKVersion =<a name="line.305"></a>
+<span class="sourceLineNo">306</span>              attemptToOwnTask(false, watcher, server.getServerName(), curTask,<a name="line.306"></a>
+<span class="sourceLineNo">307</span>                mode, zkVersion.intValue());<a name="line.307"></a>
+<span class="sourceLineNo">308</span>          if (latestZKVersion &lt; 0) {<a name="line.308"></a>
+<span class="sourceLineNo">309</span>            LOG.warn("Failed to heartbeat the task" + curTask);<a name="line.309"></a>
+<span class="sourceLineNo">310</span>            return false;<a name="line.310"></a>
+<span class="sourceLineNo">311</span>          }<a name="line.311"></a>
+<span class="sourceLineNo">312</span>          zkVersion.setValue(latestZKVersion);<a name="line.312"></a>
+<span class="sourceLineNo">313</span>        }<a name="line.313"></a>
+<span class="sourceLineNo">314</span>        return true;<a name="line.314"></a>
+<span class="sourceLineNo">315</span>      }<a name="line.315"></a>
+<span class="sourceLineNo">316</span>    };<a name="line.316"></a>
+<span class="sourceLineNo">317</span>    ZkSplitLogWorkerCoordination.ZkSplitTaskDetails splitTaskDetails =<a name="line.317"></a>
+<span class="sourceLineNo">318</span>        new ZkSplitLogWorkerCoordination.ZkSplitTaskDetails();<a name="line.318"></a>
+<span class="sourceLineNo">319</span>    splitTaskDetails.setTaskNode(curTask);<a name="line.319"></a>
+<span class="sourceLineNo">320</span>    splitTaskDetails.setCurTaskZKVersion(zkVersion);<a name="line.320"></a>
+<span class="sourceLineNo">321</span><a name="line.321"></a>
+<span class="sourceLineNo">322</span>    WALSplitterHandler hsh =<a name="line.322"></a>
+<span class="sourceLineNo">323</span>        new WALSplitterHandler(server, this, splitTaskDetails, reporter,<a name="line.323"></a>
+<span class="sourceLineNo">324</span>            this.tasksInProgress, splitTaskExecutor, mode);<a name="line.324"></a>
+<span class="sourceLineNo">325</span>    server.getExecutorService().submit(hsh);<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>
+<span class="sourceLineNo">329</span>   * This function calculates how many splitters it could create based on expected average tasks per<a name="line.329"></a>
+<span class="sourceLineNo">330</span>   * RS and the hard limit upper bound(maxConcurrentTasks) set by configuration. &lt;br&gt;<a name="line.330"></a>
+<span class="sourceLineNo">331</span>   * At any given time, a RS allows spawn MIN(Expected Tasks/RS, Hard Upper Bound)<a name="line.331"></a>
+<span class="sourceLineNo">332</span>   * @param numTasks current total number of available tasks<a name="line.332"></a>
+<span class="sourceLineNo">333</span>   */<a name="line.333"></a>
+<span class="sourceLineNo">334</span>  private int calculateAvailableSplitters(int numTasks) {<a name="line.334"></a>
+<span class="sourceLineNo">335</span>    // at lease one RS(itself) available<a name="line.335"></a>
+<span class="sourceLineNo">336</span>    int availableRSs = 1;<a name="line.336"></a>
+<span class="sourceLineNo">337</span>    try {<a name="line.337"></a>
+<span class="sourceLineNo">338</span>      List&lt;String&gt; regionServers =<a name="line.338"></a>
+<span class="sourceLineNo">339</span>          ZKUtil.listChildrenNoWatch(watcher, watcher.znodePaths.rsZNode);<a name="line.339"></a>
+<span class="sourceLineNo">340</span>      availableRSs = Math.max(availableRSs, (regionServers == null) ? 0 : regionServers.size());<a name="line.340"></a>
+<span class="sourceLineNo">341</span>    } catch (KeeperException e) {<a name="line.341"></a>
+<span class="sourceLineNo">342</span>      // do nothing<a name="line.342"></a>
+<span class="sourceLineNo">343</span>      LOG.debug("getAvailableRegionServers got ZooKeeper exception", e);<a name="line.343"></a>
+<span class="sourceLineNo">344</span>    }<a name="line.344"></a>
+<span class="sourceLineNo">345</span><a name="line.345"></a>
+<span class="sourceLineNo">346</span>    int expectedTasksPerRS = (numTasks / availableRSs) + ((numTasks % availableRSs == 0) ? 0 : 1);<a name="line.346"></a>
+<span class="sourceLineNo">347</span>    expectedTasksPerRS = Math.max(1, expectedTasksPerRS); // at least be one<a name="line.347"></a>
+<span class="sourceLineNo">348</span>    // calculate how many more splitters we could spawn<a name="line.348"></a>
+<span class="sourceLineNo">349</span>    return Math.min(expectedTasksPerRS, maxConcurrentTasks)<a name="line.349"></a>
+<span class="sourceLineNo">350</span>        - this.tasksInProgress.get();<a name="line.350"></a>
+<span class="sourceLineNo">351</span>  }<a name="line.351"></a>
+<span class="sourceLineNo">352</span><a name="line.352"></a>
+<span class="sourceLineNo">353</span>  /**<a name="line.353"></a>
+<span class="sourceLineNo">354</span>   * Try to own the task by transitioning the zk node data from UNASSIGNED to OWNED.<a name="line.354"></a>
+<span class="sourceLineNo">355</span>   * &lt;p&gt;<a name="line.355"></a>
+<span class="sourceLineNo">356</span>   * This method is also used to periodically heartbeat the task progress by transitioning the node<a name="line.356"></a>
+<span class="sourceLineNo">357</span>   * from OWNED to OWNED.<a name="line.357"></a>
+<span class="sourceLineNo">358</span>   * &lt;p&gt;<a name="line.358"></a>
+<span class="sourceLineNo">359</span>   * @param isFirstTime shows whther it's the first attempt.<a name="line.359"></a>
+<span class="sourceLineNo">360</span>   * @param zkw zk wathcer<a name="line.360"></a>
+<span class="sourceLineNo">361</span>   * @param server name<a name="line.361"></a>
+<span class="sourceLineNo">362</span>   * @param task to own<a name="line.362"></a>
+<span class="sourceLineNo">363</span>   * @param taskZKVersion version of the task in zk<a name="line.363"></a>
+<span class="sourceLineNo">364</span>   * @return non-negative integer value when task can be owned by current region server otherwise -1<a name="line.364"></a>
+<span class="sourceLineNo">365</span>   */<a name="line.365"></a>
+<span class="sourceLineNo">366</span>  protected static int attemptToOwnTask(boolean isFirstTime, ZooKeeperWatcher zkw,<a name="line.366"></a>
+<span class="sourceLineNo">367</span>      ServerName server, String task, RecoveryMode mode, int taskZKVersion) {<a name="line.367"></a>
+<span class="sourceLineNo">368</span>    int latestZKVersion = FAILED_TO_OWN_TASK;<a name="line.368"></a>
+<span class="sourceLineNo">369</span>    try {<a name="line.369"></a>
+<span class="sourceLineNo">370</span>      SplitLogTask slt = new SplitLogTask.Owned(server, mode);<a name="line.370"></a>
+<span class="sourceLineNo">371</span>      Stat stat = zkw.getRecoverableZooKeeper().setData(task, slt.toByteArray(), taskZKVersion);<a name="line.371"></a>
+<span class="sourceLineNo">372</span>      if (stat == null) {<a name="line.372"></a>
+<span class="sourceLineNo">373</span>        LOG.warn("zk.setData() returned null for path " + task);<a name="line.373"></a>
+<span class="sourceLineNo">374</span>        SplitLogCounters.tot_wkr_task_heartbeat_failed.incrementAndGet();<a name="line.374"></a>
+<span class="sourceLineNo">375</span>        return FAILED_TO_OWN_TASK;<a name="line.375"></a>
+<span class="sourceLineNo">376</span>      }<a name="line.376"></a>
+<span class="sourceLineNo">377</span>      latestZKVersion = stat.getVersion();<a name="line.377"></a>
+<span class="sourceLineNo">378</span>      SplitLogCounters.tot_wkr_task_heartbeat.incrementAndGet();<a name="line.378"></a>
+<span class="sourceLineNo">379</span>      return latestZKVersion;<a name="line.379"></a>
+<span class="sourceLineNo">380</span>    } catch (KeeperException e) {<a name="line.380"></a>
+<span class="sourceLineNo">381</span>      if (!isFirstTime) {<a name="line.381"></a>
+<span class="sourceLineNo">382</span>        if (e.code().equals(KeeperException.Code.NONODE)) {<a name="line.382"></a>
+<span class="sourceLineNo">383</span>          LOG.warn("NONODE failed to assert ownership for " + task, e);<a name="line.383"></a>
+<span class="sourceLineNo">384</span>        } else if (e.code().equals(KeeperException.Code.BADVERSION)) {<a name="line.384"></a>
+<span class="sourceLineNo">385</span>          LOG.warn("BADVERSION failed to assert ownership for " + task, e);<a name="line.385"></a>
+<span class="sourceLineNo">386</span>        } else {<a name="line.386"></a>
+<span class="sourceLineNo">387</span>          LOG.warn("failed to assert ownership for " + task, e);<a name="line.387"></a>
+<span class="sourceLineNo">388</span>        }<a name="line.388"></a>
+<span class="sourceLineNo">389</span>      }<a name="line.389"></a>
+<span class="sourceLineNo">390</span>    } catch (InterruptedException e1) {<a name="line.390"></a>
+<span class="sourceLineNo">391</span>      LOG.warn("Interrupted while trying to assert ownership of " + task + " "<a name="line.391"></a>
+<span class="sourceLineNo">392</span>          + StringUtils.stringifyException(e1));<a name="line.392"></a>
+<span class="sourceLineNo">393</span>      Thread.currentThread().interrupt();<a name="line.393"></a>
+<span class="sourceLineNo">394</span>    }<a name="line.394"></a>
+<span class="sourceLineNo">395</span>    SplitLogCounters.tot_wkr_task_heartbeat_failed.incrementAndGet();<a name="line.395"></a>
+<span class="sourceLineNo">396</span>    return FAILED_TO_OWN_TASK;<a name="line.396"></a>
+<span class="sourceLineNo">397</span>  }<a name="line.397"></a>
+<span class="sourceLineNo">398</span><a name="line.398"></a>
+<span class="sourceLineNo">399</span>  /**<a name="line.399"></a>
+<span class="sourceLineNo">400</span>   * Wait for tasks to become available at /hbase/splitlog zknode. Grab a task one at a time. This<a name="line.400"></a>
+<span class="sourceLineNo">401</span>   * policy puts an upper-limit on the number of simultaneous log splitting that could be happening<a name="line.401"></a>
+<span class="sourceLineNo">402</span>   * in a cluster.<a name="line.402"></a>
+<span class="sourceLineNo">403</span>   * &lt;p&gt;<a name="line.403"></a>
+<span class="sourceLineNo">404</span>   * Synchronization using &lt;code&gt;taskReadyLock&lt;/code&gt; ensures that it will try to grab every task<a name="line.404"></a>
+<span class="sourceLineNo">405</span>   * that has been put up<a name="line.405"></a>
+<span class="sourceLineNo">406</span>   * @throws InterruptedException<a name="line.406"></a>
+<span class="sourceLineNo">407</span>   */<a name="line.407"></a>
+<span class="sourceLineNo">408</span>  @Override<a name="line.408"></a>
+<span class="sourceLineNo">409</span>  public void taskLoop() throws InterruptedException {<a name="line.409"></a>
+<span class="sourceLineNo">410</span>    while (!shouldStop) {<a name="line.410"></a>
+<span class="sourceLineNo">411</span>      int seq_start = taskReadySeq.get();<a name="line.411"></a>
+<span class="sourceLineNo">412</span>      List&lt;String&gt; paths = null;<a name="line.412"></a>
+<span class="sourceLineNo">413</span>      paths = getTaskList();<a name="line.413"></a>
+<span class="sourceLineNo">414</span>      if (paths == null) {<a name="line.414"></a>
+<span class="sourceLineNo">415</span>        LOG.warn("Could not get tasks, did someone remove " + watcher.znodePaths.splitLogZNode<a name="line.415"></a>
+<span class="sourceLineNo">416</span>            + " ... worker thread exiting.");<a name="line.416"></a>
+<span class="sourceLineNo">417</span>        return;<a name="line.417"></a>
+<span class="sourceLineNo">418</span>      }<a name="line.418"></a>
+<span class="sourceLineNo">419</span>      // pick meta wal firstly<a name="line.419"></a>
+<span class="sourceLineNo">420</span>      int offset = (int) (Math.random() * paths.size());<a name="line.420"></a>
+<span class="sourceLineNo">421</span>      for (int i = 0; i &lt; paths.size(); i++) {<a name="line.421"></a>
+<span class="sourceLineNo">422</span>        if (AbstractFSWALProvider.isMetaFile(paths.get(i))) {<a name="line.422"></a>
+<span class="sourceLineNo">423</span>          offset = i;<a name="line.423"></a>
+<span class="sourceLineNo">424</span>          break;<a name="line.424"></a>
+<span class="sourceLineNo">425</span>        }<a name="line.425"></a>
+<span class="sourceLineNo">426</span>      }<a name="line.426"></a>
+<span class="sourceLineNo">427</span>      int numTasks = paths.size();<a name="line.427"></a>
+<span class="sourceLineNo">428</span>      for (int i = 0; i &lt; numTasks; i++) {<a name="line.428"></a>
+<span class="sourceLineNo">429</span>        int idx = (i + offset) % paths.size();<a name="line.429"></a>
+<span class="sourceLineNo">430</span>        // don't call ZKSplitLog.getNodeName() because that will lead to<a name="line.430"></a>
+<span class="sourceLineNo">431</span>        // double encoding of the path name<a name="line.431"></a>
+<span class="sourceLineNo">432</span>        if (this.calculateAvailableSplitters(numTasks) &gt; 0) {<a name="line.432"></a>
+<span class="sourceLineNo">433</span>          grabTask(ZKUtil.joinZNode(watcher.znodePaths.splitLogZNode, paths.get(idx)));<a name="line.433"></a>
+<span class="sourceLineNo">434</span>        } else {<a name="line.434"></a>
+<span class="sourceLineNo">435</span>          LOG.debug("Current region server " + server.getServerName() + " has "<a name="line.435"></a>
+<span class="sourceLineNo">436</span>              + this.tasksInProgress.get() + " tasks in progress and can't take more.");<a name="line.436"></a>
+<span class="sourceLineNo">437</span>          break;<a name="line.437"></a>
+<span class="sourceLineNo">438</span>        }<a name="line.438"></a>
+<span class="sourceLineNo">439</span>        if (shouldStop) {<a name="line.439"></a>
+<span class="sourceLineNo">440</span>          return;<a name="line.440"></a>
+<span class="sourceLineNo">441</span>        }<a name="line.441"></a>
+<span class="sourceLineNo">442</span>      }<a name="line.442"></a>
+<span class="sourceLineNo">443</span>      SplitLogCounters.tot_wkr_task_grabing.incrementAndGet();<a name="line.443"></a>
+<span class="sourceLineNo">444</span>      synchronized (taskReadyLock) {<a name="line.444"></a>
+<span class="sourceLineNo">445</span>        while (seq_start == taskReadySeq.get()) {<a name="line.445"></a>
+<span class="sourceLineNo">446</span>          taskReadyLock.wait(checkInterval);<a name="line.446"></a>
+<span class="sourceLineNo">447</span>          if (server != null) {<a name="line.447"></a>
+<span class="sourceLineNo">448</span>            // check to see if we have stale recovering regions in our internal memory state<a name="line.448"></a>
+<span class="sourceLineNo">449</span>            Map&lt;String, Region&gt; recoveringRegions = server.getRecoveringRegions();<a name="line.449"></a>
+<span class="sourceLineNo">450</span>            if (!recoveringRegions.isEmpty()) {<a name="line.450"></a>
+<span class="sourceLineNo">451</span>              // Make a local copy to prevent ConcurrentModificationException when other threads<a name="line.451"></a>
+<span class="sourceLineNo">452</span>              // modify recoveringRegions<a name="line.452"></a>
+<span class="sourceLineNo">453</span>              List&lt;String&gt; tmpCopy = new ArrayList&lt;&gt;(recoveringRegions.keySet());<a name="line.453"></a>
+<span class="sourceLineNo">454</span>              int listSize = tmpCopy.size();<a name="line.454"></a>
+<span class="sourceLineNo">455</span>              for (int i = 0; i &lt; listSize; i++) {<a name="line.455"

<TRUNCATED>

Mime
View raw message