You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hbase.apache.org by gi...@apache.org on 2018/09/07 14:52:52 UTC

[09/23] hbase-site git commit: Published site at 24f2893059e745aa6691cdd6cb37fc248880b3b9.

http://git-wip-us.apache.org/repos/asf/hbase-site/blob/c548e10d/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html
index 681e263..7cdf584 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.TimeoutMonitor.html
@@ -182,7 +182,7 @@
 <span class="sourceLineNo">174</span>      }<a name="line.174"></a>
 <span class="sourceLineNo">175</span>      FileStatus[] logfiles = FSUtils.listStatus(fs, logDir, filter);<a name="line.175"></a>
 <span class="sourceLineNo">176</span>      if (logfiles == null || logfiles.length == 0) {<a name="line.176"></a>
-<span class="sourceLineNo">177</span>        LOG.info(logDir + " is empty dir, no logs to split");<a name="line.177"></a>
+<span class="sourceLineNo">177</span>        LOG.info("{} dir is empty, no logs to split.", logDir);<a name="line.177"></a>
 <span class="sourceLineNo">178</span>      } else {<a name="line.178"></a>
 <span class="sourceLineNo">179</span>        Collections.addAll(fileStatus, logfiles);<a name="line.179"></a>
 <span class="sourceLineNo">180</span>      }<a name="line.180"></a>
@@ -243,426 +243,430 @@
 <span class="sourceLineNo">235</span>      PathFilter filter) throws IOException {<a name="line.235"></a>
 <span class="sourceLineNo">236</span>    MonitoredTask status = TaskMonitor.get().createStatus("Doing distributed log split in " +<a name="line.236"></a>
 <span class="sourceLineNo">237</span>      logDirs + " for serverName=" + serverNames);<a name="line.237"></a>
-<span class="sourceLineNo">238</span>    FileStatus[] logfiles = getFileList(logDirs, filter);<a name="line.238"></a>
-<span class="sourceLineNo">239</span>    status.setStatus("Checking directory contents...");<a name="line.239"></a>
-<span class="sourceLineNo">240</span>    SplitLogCounters.tot_mgr_log_split_batch_start.increment();<a name="line.240"></a>
-<span class="sourceLineNo">241</span>    LOG.info("Started splitting " + logfiles.length + " logs in " + logDirs +<a name="line.241"></a>
-<span class="sourceLineNo">242</span>      " for " + serverNames);<a name="line.242"></a>
-<span class="sourceLineNo">243</span>    long t = EnvironmentEdgeManager.currentTime();<a name="line.243"></a>
-<span class="sourceLineNo">244</span>    long totalSize = 0;<a name="line.244"></a>
-<span class="sourceLineNo">245</span>    TaskBatch batch = new TaskBatch();<a name="line.245"></a>
-<span class="sourceLineNo">246</span>    for (FileStatus lf : logfiles) {<a name="line.246"></a>
-<span class="sourceLineNo">247</span>      // TODO If the log file is still being written to - which is most likely<a name="line.247"></a>
-<span class="sourceLineNo">248</span>      // the case for the last log file - then its length will show up here<a name="line.248"></a>
-<span class="sourceLineNo">249</span>      // as zero. The size of such a file can only be retrieved after<a name="line.249"></a>
-<span class="sourceLineNo">250</span>      // recover-lease is done. totalSize will be under in most cases and the<a name="line.250"></a>
-<span class="sourceLineNo">251</span>      // metrics that it drives will also be under-reported.<a name="line.251"></a>
-<span class="sourceLineNo">252</span>      totalSize += lf.getLen();<a name="line.252"></a>
-<span class="sourceLineNo">253</span>      String pathToLog = FSUtils.removeWALRootPath(lf.getPath(), conf);<a name="line.253"></a>
-<span class="sourceLineNo">254</span>      if (!enqueueSplitTask(pathToLog, batch)) {<a name="line.254"></a>
-<span class="sourceLineNo">255</span>        throw new IOException("duplicate log split scheduled for " + lf.getPath());<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>    waitForSplittingCompletion(batch, status);<a name="line.258"></a>
-<span class="sourceLineNo">259</span><a name="line.259"></a>
-<span class="sourceLineNo">260</span>    if (batch.done != batch.installed) {<a name="line.260"></a>
-<span class="sourceLineNo">261</span>      batch.isDead = true;<a name="line.261"></a>
-<span class="sourceLineNo">262</span>      SplitLogCounters.tot_mgr_log_split_batch_err.increment();<a name="line.262"></a>
-<span class="sourceLineNo">263</span>      LOG.warn("error while splitting logs in " + logDirs + " installed = " + batch.installed<a name="line.263"></a>
-<span class="sourceLineNo">264</span>          + " but only " + batch.done + " done");<a name="line.264"></a>
-<span class="sourceLineNo">265</span>      String msg = "error or interrupted while splitting logs in " + logDirs + " Task = " + batch;<a name="line.265"></a>
-<span class="sourceLineNo">266</span>      status.abort(msg);<a name="line.266"></a>
-<span class="sourceLineNo">267</span>      throw new IOException(msg);<a name="line.267"></a>
-<span class="sourceLineNo">268</span>    }<a name="line.268"></a>
-<span class="sourceLineNo">269</span>    for (Path logDir : logDirs) {<a name="line.269"></a>
-<span class="sourceLineNo">270</span>      status.setStatus("Cleaning up log directory...");<a name="line.270"></a>
-<span class="sourceLineNo">271</span>      final FileSystem fs = logDir.getFileSystem(conf);<a name="line.271"></a>
-<span class="sourceLineNo">272</span>      try {<a name="line.272"></a>
-<span class="sourceLineNo">273</span>        if (fs.exists(logDir) &amp;&amp; !fs.delete(logDir, false)) {<a name="line.273"></a>
-<span class="sourceLineNo">274</span>          LOG.warn("Unable to delete log src dir. Ignoring. " + logDir);<a name="line.274"></a>
-<span class="sourceLineNo">275</span>        }<a name="line.275"></a>
-<span class="sourceLineNo">276</span>      } catch (IOException ioe) {<a name="line.276"></a>
-<span class="sourceLineNo">277</span>        FileStatus[] files = fs.listStatus(logDir);<a name="line.277"></a>
-<span class="sourceLineNo">278</span>        if (files != null &amp;&amp; files.length &gt; 0) {<a name="line.278"></a>
-<span class="sourceLineNo">279</span>          LOG.warn("Returning success without actually splitting and "<a name="line.279"></a>
-<span class="sourceLineNo">280</span>              + "deleting all the log files in path " + logDir + ": "<a name="line.280"></a>
-<span class="sourceLineNo">281</span>              + Arrays.toString(files), ioe);<a name="line.281"></a>
-<span class="sourceLineNo">282</span>        } else {<a name="line.282"></a>
-<span class="sourceLineNo">283</span>          LOG.warn("Unable to delete log src dir. Ignoring. " + logDir, ioe);<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>      SplitLogCounters.tot_mgr_log_split_batch_success.increment();<a name="line.286"></a>
-<span class="sourceLineNo">287</span>    }<a name="line.287"></a>
-<span class="sourceLineNo">288</span>    String msg =<a name="line.288"></a>
-<span class="sourceLineNo">289</span>        "finished splitting (more than or equal to) " + totalSize + " bytes in " + batch.installed<a name="line.289"></a>
-<span class="sourceLineNo">290</span>            + " log files in " + logDirs + " in "<a name="line.290"></a>
-<span class="sourceLineNo">291</span>            + (EnvironmentEdgeManager.currentTime() - t) + "ms";<a name="line.291"></a>
-<span class="sourceLineNo">292</span>    status.markComplete(msg);<a name="line.292"></a>
-<span class="sourceLineNo">293</span>    LOG.info(msg);<a name="line.293"></a>
-<span class="sourceLineNo">294</span>    return totalSize;<a name="line.294"></a>
-<span class="sourceLineNo">295</span>  }<a name="line.295"></a>
-<span class="sourceLineNo">296</span><a name="line.296"></a>
-<span class="sourceLineNo">297</span>  /**<a name="line.297"></a>
-<span class="sourceLineNo">298</span>   * Add a task entry to coordination if it is not already there.<a name="line.298"></a>
-<span class="sourceLineNo">299</span>   * @param taskname the path of the log to be split<a name="line.299"></a>
-<span class="sourceLineNo">300</span>   * @param batch the batch this task belongs to<a name="line.300"></a>
-<span class="sourceLineNo">301</span>   * @return true if a new entry is created, false if it is already there.<a name="line.301"></a>
-<span class="sourceLineNo">302</span>   */<a name="line.302"></a>
-<span class="sourceLineNo">303</span>  boolean enqueueSplitTask(String taskname, TaskBatch batch) {<a name="line.303"></a>
-<span class="sourceLineNo">304</span>    lastTaskCreateTime = EnvironmentEdgeManager.currentTime();<a name="line.304"></a>
-<span class="sourceLineNo">305</span>    String task = getSplitLogManagerCoordination().prepareTask(taskname);<a name="line.305"></a>
-<span class="sourceLineNo">306</span>    Task oldtask = createTaskIfAbsent(task, batch);<a name="line.306"></a>
-<span class="sourceLineNo">307</span>    if (oldtask == null) {<a name="line.307"></a>
-<span class="sourceLineNo">308</span>      // publish the task in the coordination engine<a name="line.308"></a>
-<span class="sourceLineNo">309</span>      getSplitLogManagerCoordination().submitTask(task);<a name="line.309"></a>
-<span class="sourceLineNo">310</span>      return true;<a name="line.310"></a>
-<span class="sourceLineNo">311</span>    }<a name="line.311"></a>
-<span class="sourceLineNo">312</span>    return false;<a name="line.312"></a>
-<span class="sourceLineNo">313</span>  }<a name="line.313"></a>
-<span class="sourceLineNo">314</span><a name="line.314"></a>
-<span class="sourceLineNo">315</span>  private void waitForSplittingCompletion(TaskBatch batch, MonitoredTask status) {<a name="line.315"></a>
-<span class="sourceLineNo">316</span>    synchronized (batch) {<a name="line.316"></a>
-<span class="sourceLineNo">317</span>      while ((batch.done + batch.error) != batch.installed) {<a name="line.317"></a>
-<span class="sourceLineNo">318</span>        try {<a name="line.318"></a>
-<span class="sourceLineNo">319</span>          status.setStatus("Waiting for distributed tasks to finish. " + " scheduled="<a name="line.319"></a>
-<span class="sourceLineNo">320</span>              + batch.installed + " done=" + batch.done + " error=" + batch.error);<a name="line.320"></a>
-<span class="sourceLineNo">321</span>          int remaining = batch.installed - (batch.done + batch.error);<a name="line.321"></a>
-<span class="sourceLineNo">322</span>          int actual = activeTasks(batch);<a name="line.322"></a>
-<span class="sourceLineNo">323</span>          if (remaining != actual) {<a name="line.323"></a>
-<span class="sourceLineNo">324</span>            LOG.warn("Expected " + remaining + " active tasks, but actually there are " + actual);<a name="line.324"></a>
-<span class="sourceLineNo">325</span>          }<a name="line.325"></a>
-<span class="sourceLineNo">326</span>          int remainingTasks = getSplitLogManagerCoordination().remainingTasksInCoordination();<a name="line.326"></a>
-<span class="sourceLineNo">327</span>          if (remainingTasks &gt;= 0 &amp;&amp; actual &gt; remainingTasks) {<a name="line.327"></a>
-<span class="sourceLineNo">328</span>            LOG.warn("Expected at least" + actual + " tasks remaining, but actually there are "<a name="line.328"></a>
-<span class="sourceLineNo">329</span>                + remainingTasks);<a name="line.329"></a>
-<span class="sourceLineNo">330</span>          }<a name="line.330"></a>
-<span class="sourceLineNo">331</span>          if (remainingTasks == 0 || actual == 0) {<a name="line.331"></a>
-<span class="sourceLineNo">332</span>            LOG.warn("No more task remaining, splitting "<a name="line.332"></a>
-<span class="sourceLineNo">333</span>                + "should have completed. Remaining tasks is " + remainingTasks<a name="line.333"></a>
-<span class="sourceLineNo">334</span>                + ", active tasks in map " + actual);<a name="line.334"></a>
-<span class="sourceLineNo">335</span>            if (remainingTasks == 0 &amp;&amp; actual == 0) {<a name="line.335"></a>
-<span class="sourceLineNo">336</span>              return;<a name="line.336"></a>
-<span class="sourceLineNo">337</span>            }<a name="line.337"></a>
-<span class="sourceLineNo">338</span>          }<a name="line.338"></a>
-<span class="sourceLineNo">339</span>          batch.wait(100);<a name="line.339"></a>
-<span class="sourceLineNo">340</span>          if (server.isStopped()) {<a name="line.340"></a>
-<span class="sourceLineNo">341</span>            LOG.warn("Stopped while waiting for log splits to be completed");<a name="line.341"></a>
-<span class="sourceLineNo">342</span>            return;<a name="line.342"></a>
-<span class="sourceLineNo">343</span>          }<a name="line.343"></a>
-<span class="sourceLineNo">344</span>        } catch (InterruptedException e) {<a name="line.344"></a>
-<span class="sourceLineNo">345</span>          LOG.warn("Interrupted while waiting for log splits to be completed");<a name="line.345"></a>
-<span class="sourceLineNo">346</span>          Thread.currentThread().interrupt();<a name="line.346"></a>
-<span class="sourceLineNo">347</span>          return;<a name="line.347"></a>
-<span class="sourceLineNo">348</span>        }<a name="line.348"></a>
-<span class="sourceLineNo">349</span>      }<a name="line.349"></a>
-<span class="sourceLineNo">350</span>    }<a name="line.350"></a>
-<span class="sourceLineNo">351</span>  }<a name="line.351"></a>
-<span class="sourceLineNo">352</span><a name="line.352"></a>
-<span class="sourceLineNo">353</span>  @VisibleForTesting<a name="line.353"></a>
-<span class="sourceLineNo">354</span>  ConcurrentMap&lt;String, Task&gt; getTasks() {<a name="line.354"></a>
-<span class="sourceLineNo">355</span>    return tasks;<a name="line.355"></a>
-<span class="sourceLineNo">356</span>  }<a name="line.356"></a>
-<span class="sourceLineNo">357</span><a name="line.357"></a>
-<span class="sourceLineNo">358</span>  private int activeTasks(final TaskBatch batch) {<a name="line.358"></a>
-<span class="sourceLineNo">359</span>    int count = 0;<a name="line.359"></a>
-<span class="sourceLineNo">360</span>    for (Task t : tasks.values()) {<a name="line.360"></a>
-<span class="sourceLineNo">361</span>      if (t.batch == batch &amp;&amp; t.status == TerminationStatus.IN_PROGRESS) {<a name="line.361"></a>
-<span class="sourceLineNo">362</span>        count++;<a name="line.362"></a>
-<span class="sourceLineNo">363</span>      }<a name="line.363"></a>
-<span class="sourceLineNo">364</span>    }<a name="line.364"></a>
-<span class="sourceLineNo">365</span>    return count;<a name="line.365"></a>
-<span class="sourceLineNo">366</span><a name="line.366"></a>
-<span class="sourceLineNo">367</span>  }<a name="line.367"></a>
-<span class="sourceLineNo">368</span><a name="line.368"></a>
-<span class="sourceLineNo">369</span>  /**<a name="line.369"></a>
-<span class="sourceLineNo">370</span>   * @param path<a name="line.370"></a>
-<span class="sourceLineNo">371</span>   * @param batch<a name="line.371"></a>
-<span class="sourceLineNo">372</span>   * @return null on success, existing task on error<a name="line.372"></a>
-<span class="sourceLineNo">373</span>   */<a name="line.373"></a>
-<span class="sourceLineNo">374</span>  private Task createTaskIfAbsent(String path, TaskBatch batch) {<a name="line.374"></a>
-<span class="sourceLineNo">375</span>    Task oldtask;<a name="line.375"></a>
-<span class="sourceLineNo">376</span>    // batch.installed is only changed via this function and<a name="line.376"></a>
-<span class="sourceLineNo">377</span>    // a single thread touches batch.installed.<a name="line.377"></a>
-<span class="sourceLineNo">378</span>    Task newtask = new Task();<a name="line.378"></a>
-<span class="sourceLineNo">379</span>    newtask.batch = batch;<a name="line.379"></a>
-<span class="sourceLineNo">380</span>    oldtask = tasks.putIfAbsent(path, newtask);<a name="line.380"></a>
-<span class="sourceLineNo">381</span>    if (oldtask == null) {<a name="line.381"></a>
-<span class="sourceLineNo">382</span>      batch.installed++;<a name="line.382"></a>
-<span class="sourceLineNo">383</span>      return null;<a name="line.383"></a>
-<span class="sourceLineNo">384</span>    }<a name="line.384"></a>
-<span class="sourceLineNo">385</span>    // new task was not used.<a name="line.385"></a>
-<span class="sourceLineNo">386</span>    synchronized (oldtask) {<a name="line.386"></a>
-<span class="sourceLineNo">387</span>      if (oldtask.isOrphan()) {<a name="line.387"></a>
-<span class="sourceLineNo">388</span>        if (oldtask.status == SUCCESS) {<a name="line.388"></a>
-<span class="sourceLineNo">389</span>          // The task is already done. Do not install the batch for this<a name="line.389"></a>
-<span class="sourceLineNo">390</span>          // task because it might be too late for setDone() to update<a name="line.390"></a>
-<span class="sourceLineNo">391</span>          // batch.done. There is no need for the batch creator to wait for<a name="line.391"></a>
-<span class="sourceLineNo">392</span>          // this task to complete.<a name="line.392"></a>
-<span class="sourceLineNo">393</span>          return (null);<a name="line.393"></a>
-<span class="sourceLineNo">394</span>        }<a name="line.394"></a>
-<span class="sourceLineNo">395</span>        if (oldtask.status == IN_PROGRESS) {<a name="line.395"></a>
-<span class="sourceLineNo">396</span>          oldtask.batch = batch;<a name="line.396"></a>
-<span class="sourceLineNo">397</span>          batch.installed++;<a name="line.397"></a>
-<span class="sourceLineNo">398</span>          LOG.debug("Previously orphan task " + path + " is now being waited upon");<a name="line.398"></a>
-<span class="sourceLineNo">399</span>          return null;<a name="line.399"></a>
-<span class="sourceLineNo">400</span>        }<a name="line.400"></a>
-<span class="sourceLineNo">401</span>        while (oldtask.status == FAILURE) {<a name="line.401"></a>
-<span class="sourceLineNo">402</span>          LOG.debug("wait for status of task " + path + " to change to DELETED");<a name="line.402"></a>
-<span class="sourceLineNo">403</span>          SplitLogCounters.tot_mgr_wait_for_zk_delete.increment();<a name="line.403"></a>
-<span class="sourceLineNo">404</span>          try {<a name="line.404"></a>
-<span class="sourceLineNo">405</span>            oldtask.wait();<a name="line.405"></a>
-<span class="sourceLineNo">406</span>          } catch (InterruptedException e) {<a name="line.406"></a>
-<span class="sourceLineNo">407</span>            Thread.currentThread().interrupt();<a name="line.407"></a>
-<span class="sourceLineNo">408</span>            LOG.warn("Interrupted when waiting for znode delete callback");<a name="line.408"></a>
-<span class="sourceLineNo">409</span>            // fall through to return failure<a name="line.409"></a>
-<span class="sourceLineNo">410</span>            break;<a name="line.410"></a>
-<span class="sourceLineNo">411</span>          }<a name="line.411"></a>
-<span class="sourceLineNo">412</span>        }<a name="line.412"></a>
-<span class="sourceLineNo">413</span>        if (oldtask.status != DELETED) {<a name="line.413"></a>
-<span class="sourceLineNo">414</span>          LOG.warn("Failure because previously failed task"<a name="line.414"></a>
-<span class="sourceLineNo">415</span>              + " state still present. Waiting for znode delete callback" + " path=" + path);<a name="line.415"></a>
-<span class="sourceLineNo">416</span>          return oldtask;<a name="line.416"></a>
-<span class="sourceLineNo">417</span>        }<a name="line.417"></a>
-<span class="sourceLineNo">418</span>        // reinsert the newTask and it must succeed this time<a name="line.418"></a>
-<span class="sourceLineNo">419</span>        Task t = tasks.putIfAbsent(path, newtask);<a name="line.419"></a>
-<span class="sourceLineNo">420</span>        if (t == null) {<a name="line.420"></a>
-<span class="sourceLineNo">421</span>          batch.installed++;<a name="line.421"></a>
-<span class="sourceLineNo">422</span>          return null;<a name="line.422"></a>
-<span class="sourceLineNo">423</span>        }<a name="line.423"></a>
-<span class="sourceLineNo">424</span>        LOG.error(HBaseMarkers.FATAL, "Logic error. Deleted task still present in tasks map");<a name="line.424"></a>
-<span class="sourceLineNo">425</span>        assert false : "Deleted task still present in tasks map";<a name="line.425"></a>
-<span class="sourceLineNo">426</span>        return t;<a name="line.426"></a>
-<span class="sourceLineNo">427</span>      }<a name="line.427"></a>
-<span class="sourceLineNo">428</span>      LOG.warn("Failure because two threads can't wait for the same task; path=" + path);<a name="line.428"></a>
-<span class="sourceLineNo">429</span>      return oldtask;<a name="line.429"></a>
-<span class="sourceLineNo">430</span>    }<a name="line.430"></a>
-<span class="sourceLineNo">431</span>  }<a name="line.431"></a>
-<span class="sourceLineNo">432</span><a name="line.432"></a>
-<span class="sourceLineNo">433</span>  public void stop() {<a name="line.433"></a>
-<span class="sourceLineNo">434</span>    if (choreService != null) {<a name="line.434"></a>
-<span class="sourceLineNo">435</span>      choreService.shutdown();<a name="line.435"></a>
-<span class="sourceLineNo">436</span>    }<a name="line.436"></a>
-<span class="sourceLineNo">437</span>    if (timeoutMonitor != null) {<a name="line.437"></a>
-<span class="sourceLineNo">438</span>      timeoutMonitor.cancel(true);<a name="line.438"></a>
-<span class="sourceLineNo">439</span>    }<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>  void handleDeadWorker(ServerName workerName) {<a name="line.442"></a>
-<span class="sourceLineNo">443</span>    // resubmit the tasks on the TimeoutMonitor thread. Makes it easier<a name="line.443"></a>
-<span class="sourceLineNo">444</span>    // to reason about concurrency. Makes it easier to retry.<a name="line.444"></a>
-<span class="sourceLineNo">445</span>    synchronized (deadWorkersLock) {<a name="line.445"></a>
-<span class="sourceLineNo">446</span>      if (deadWorkers == null) {<a name="line.446"></a>
-<span class="sourceLineNo">447</span>        deadWorkers = new HashSet&lt;&gt;(100);<a name="line.447"></a>
-<span class="sourceLineNo">448</span>      }<a name="line.448"></a>
-<span class="sourceLineNo">449</span>      deadWorkers.add(workerName);<a name="line.449"></a>
-<span class="sourceLineNo">450</span>    }<a name="line.450"></a>
-<span class="sourceLineNo">451</span>    LOG.info("dead splitlog worker " + workerName);<a name="line.451"></a>
-<span class="sourceLineNo">452</span>  }<a name="line.452"></a>
-<span class="sourceLineNo">453</span><a name="line.453"></a>
-<span class="sourceLineNo">454</span>  void handleDeadWorkers(Set&lt;ServerName&gt; serverNames) {<a name="line.454"></a>
-<span class="sourceLineNo">455</span>    synchronized (deadWorkersLock) {<a name="line.455"></a>
-<span class="sourceLineNo">456</span>      if (deadWorkers == null) {<a name="line.456"></a>
-<span class="sourceLineNo">457</span>        deadWorkers = new HashSet&lt;&gt;(100);<a name="line.457"></a>
-<span class="sourceLineNo">458</span>      }<a name="line.458"></a>
-<span class="sourceLineNo">459</span>      deadWorkers.addAll(serverNames);<a name="line.459"></a>
-<span class="sourceLineNo">460</span>    }<a name="line.460"></a>
-<span class="sourceLineNo">461</span>    LOG.info("dead splitlog workers " + serverNames);<a name="line.461"></a>
-<span class="sourceLineNo">462</span>  }<a name="line.462"></a>
-<span class="sourceLineNo">463</span><a name="line.463"></a>
-<span class="sourceLineNo">464</span>  /**<a name="line.464"></a>
-<span class="sourceLineNo">465</span>   * Keeps track of the batch of tasks submitted together by a caller in splitLogDistributed().<a name="line.465"></a>
-<span class="sourceLineNo">466</span>   * Clients threads use this object to wait for all their tasks to be done.<a name="line.466"></a>
-<span class="sourceLineNo">467</span>   * &lt;p&gt;<a name="line.467"></a>
-<span class="sourceLineNo">468</span>   * All access is synchronized.<a name="line.468"></a>
-<span class="sourceLineNo">469</span>   */<a name="line.469"></a>
-<span class="sourceLineNo">470</span>  @InterfaceAudience.Private<a name="line.470"></a>
-<span class="sourceLineNo">471</span>  public static class TaskBatch {<a name="line.471"></a>
-<span class="sourceLineNo">472</span>    public int installed = 0;<a name="line.472"></a>
-<span class="sourceLineNo">473</span>    public int done = 0;<a name="line.473"></a>
-<span class="sourceLineNo">474</span>    public int error = 0;<a name="line.474"></a>
-<span class="sourceLineNo">475</span>    public volatile boolean isDead = false;<a name="line.475"></a>
-<span class="sourceLineNo">476</span><a name="line.476"></a>
-<span class="sourceLineNo">477</span>    @Override<a name="line.477"></a>
-<span class="sourceLineNo">478</span>    public String toString() {<a name="line.478"></a>
-<span class="sourceLineNo">479</span>      return ("installed = " + installed + " done = " + done + " error = " + error);<a name="line.479"></a>
-<span class="sourceLineNo">480</span>    }<a name="line.480"></a>
-<span class="sourceLineNo">481</span>  }<a name="line.481"></a>
-<span class="sourceLineNo">482</span><a name="line.482"></a>
-<span class="sourceLineNo">483</span>  /**<a name="line.483"></a>
-<span class="sourceLineNo">484</span>   * in memory state of an active task.<a name="line.484"></a>
-<span class="sourceLineNo">485</span>   */<a name="line.485"></a>
-<span class="sourceLineNo">486</span>  @InterfaceAudience.Private<a name="line.486"></a>
-<span class="sourceLineNo">487</span>  public static class Task {<a name="line.487"></a>
-<span class="sourceLineNo">488</span>    public volatile long last_update;<a name="line.488"></a>
-<span class="sourceLineNo">489</span>    public volatile int last_version;<a name="line.489"></a>
-<span class="sourceLineNo">490</span>    public volatile ServerName cur_worker_name;<a name="line.490"></a>
-<span class="sourceLineNo">491</span>    public volatile TaskBatch batch;<a name="line.491"></a>
-<span class="sourceLineNo">492</span>    public volatile TerminationStatus status;<a name="line.492"></a>
-<span class="sourceLineNo">493</span>    public volatile AtomicInteger incarnation = new AtomicInteger(0);<a name="line.493"></a>
-<span class="sourceLineNo">494</span>    public final AtomicInteger unforcedResubmits = new AtomicInteger();<a name="line.494"></a>
-<span class="sourceLineNo">495</span>    public volatile boolean resubmitThresholdReached;<a name="line.495"></a>
-<span class="sourceLineNo">496</span><a name="line.496"></a>
-<span class="sourceLineNo">497</span>    @Override<a name="line.497"></a>
-<span class="sourceLineNo">498</span>    public String toString() {<a name="line.498"></a>
-<span class="sourceLineNo">499</span>      return ("last_update = " + last_update + " last_version = " + last_version<a name="line.499"></a>
-<span class="sourceLineNo">500</span>          + " cur_worker_name = " + cur_worker_name + " status = " + status + " incarnation = "<a name="line.500"></a>
-<span class="sourceLineNo">501</span>          + incarnation + " resubmits = " + unforcedResubmits.get() + " batch = " + batch);<a name="line.501"></a>
-<span class="sourceLineNo">502</span>    }<a name="line.502"></a>
-<span class="sourceLineNo">503</span><a name="line.503"></a>
-<span class="sourceLineNo">504</span>    public Task() {<a name="line.504"></a>
-<span class="sourceLineNo">505</span>      last_version = -1;<a name="line.505"></a>
-<span class="sourceLineNo">506</span>      status = IN_PROGRESS;<a name="line.506"></a>
-<span class="sourceLineNo">507</span>      setUnassigned();<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>    public boolean isOrphan() {<a name="line.510"></a>
-<span class="sourceLineNo">511</span>      return (batch == null || batch.isDead);<a name="line.511"></a>
+<span class="sourceLineNo">238</span>    long totalSize = 0;<a name="line.238"></a>
+<span class="sourceLineNo">239</span>    TaskBatch batch = null;<a name="line.239"></a>
+<span class="sourceLineNo">240</span>    long startTime = 0;<a name="line.240"></a>
+<span class="sourceLineNo">241</span>    FileStatus[] logfiles = getFileList(logDirs, filter);<a name="line.241"></a>
+<span class="sourceLineNo">242</span>    if (logfiles.length != 0) {<a name="line.242"></a>
+<span class="sourceLineNo">243</span>      status.setStatus("Checking directory contents...");<a name="line.243"></a>
+<span class="sourceLineNo">244</span>      SplitLogCounters.tot_mgr_log_split_batch_start.increment();<a name="line.244"></a>
+<span class="sourceLineNo">245</span>      LOG.info("Started splitting " + logfiles.length + " logs in " + logDirs +<a name="line.245"></a>
+<span class="sourceLineNo">246</span>          " for " + serverNames);<a name="line.246"></a>
+<span class="sourceLineNo">247</span>      startTime = EnvironmentEdgeManager.currentTime();<a name="line.247"></a>
+<span class="sourceLineNo">248</span>      batch = new TaskBatch();<a name="line.248"></a>
+<span class="sourceLineNo">249</span>      for (FileStatus lf : logfiles) {<a name="line.249"></a>
+<span class="sourceLineNo">250</span>        // TODO If the log file is still being written to - which is most likely<a name="line.250"></a>
+<span class="sourceLineNo">251</span>        // the case for the last log file - then its length will show up here<a name="line.251"></a>
+<span class="sourceLineNo">252</span>        // as zero. The size of such a file can only be retrieved after<a name="line.252"></a>
+<span class="sourceLineNo">253</span>        // recover-lease is done. totalSize will be under in most cases and the<a name="line.253"></a>
+<span class="sourceLineNo">254</span>        // metrics that it drives will also be under-reported.<a name="line.254"></a>
+<span class="sourceLineNo">255</span>        totalSize += lf.getLen();<a name="line.255"></a>
+<span class="sourceLineNo">256</span>        String pathToLog = FSUtils.removeWALRootPath(lf.getPath(), conf);<a name="line.256"></a>
+<span class="sourceLineNo">257</span>        if (!enqueueSplitTask(pathToLog, batch)) {<a name="line.257"></a>
+<span class="sourceLineNo">258</span>          throw new IOException("duplicate log split scheduled for " + lf.getPath());<a name="line.258"></a>
+<span class="sourceLineNo">259</span>        }<a name="line.259"></a>
+<span class="sourceLineNo">260</span>      }<a name="line.260"></a>
+<span class="sourceLineNo">261</span>      waitForSplittingCompletion(batch, status);<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>    if (batch != null &amp;&amp; batch.done != batch.installed) {<a name="line.264"></a>
+<span class="sourceLineNo">265</span>      batch.isDead = true;<a name="line.265"></a>
+<span class="sourceLineNo">266</span>      SplitLogCounters.tot_mgr_log_split_batch_err.increment();<a name="line.266"></a>
+<span class="sourceLineNo">267</span>      LOG.warn("error while splitting logs in " + logDirs + " installed = " + batch.installed<a name="line.267"></a>
+<span class="sourceLineNo">268</span>          + " but only " + batch.done + " done");<a name="line.268"></a>
+<span class="sourceLineNo">269</span>      String msg = "error or interrupted while splitting logs in " + logDirs + " Task = " + batch;<a name="line.269"></a>
+<span class="sourceLineNo">270</span>      status.abort(msg);<a name="line.270"></a>
+<span class="sourceLineNo">271</span>      throw new IOException(msg);<a name="line.271"></a>
+<span class="sourceLineNo">272</span>    }<a name="line.272"></a>
+<span class="sourceLineNo">273</span>    for (Path logDir : logDirs) {<a name="line.273"></a>
+<span class="sourceLineNo">274</span>      status.setStatus("Cleaning up log directory...");<a name="line.274"></a>
+<span class="sourceLineNo">275</span>      final FileSystem fs = logDir.getFileSystem(conf);<a name="line.275"></a>
+<span class="sourceLineNo">276</span>      try {<a name="line.276"></a>
+<span class="sourceLineNo">277</span>        if (fs.exists(logDir) &amp;&amp; !fs.delete(logDir, false)) {<a name="line.277"></a>
+<span class="sourceLineNo">278</span>          LOG.warn("Unable to delete log src dir. Ignoring. " + logDir);<a name="line.278"></a>
+<span class="sourceLineNo">279</span>        }<a name="line.279"></a>
+<span class="sourceLineNo">280</span>      } catch (IOException ioe) {<a name="line.280"></a>
+<span class="sourceLineNo">281</span>        FileStatus[] files = fs.listStatus(logDir);<a name="line.281"></a>
+<span class="sourceLineNo">282</span>        if (files != null &amp;&amp; files.length &gt; 0) {<a name="line.282"></a>
+<span class="sourceLineNo">283</span>          LOG.warn("Returning success without actually splitting and "<a name="line.283"></a>
+<span class="sourceLineNo">284</span>              + "deleting all the log files in path " + logDir + ": "<a name="line.284"></a>
+<span class="sourceLineNo">285</span>              + Arrays.toString(files), ioe);<a name="line.285"></a>
+<span class="sourceLineNo">286</span>        } else {<a name="line.286"></a>
+<span class="sourceLineNo">287</span>          LOG.warn("Unable to delete log src dir. Ignoring. " + logDir, ioe);<a name="line.287"></a>
+<span class="sourceLineNo">288</span>        }<a name="line.288"></a>
+<span class="sourceLineNo">289</span>      }<a name="line.289"></a>
+<span class="sourceLineNo">290</span>      SplitLogCounters.tot_mgr_log_split_batch_success.increment();<a name="line.290"></a>
+<span class="sourceLineNo">291</span>    }<a name="line.291"></a>
+<span class="sourceLineNo">292</span>    String msg = "Finished splitting (more than or equal to) " + totalSize +<a name="line.292"></a>
+<span class="sourceLineNo">293</span>        " bytes in " + ((batch == null)? 0: batch.installed) +<a name="line.293"></a>
+<span class="sourceLineNo">294</span>        " log files in " + logDirs + " in " +<a name="line.294"></a>
+<span class="sourceLineNo">295</span>        ((startTime == 0)? startTime: (EnvironmentEdgeManager.currentTime() - startTime)) + "ms";<a name="line.295"></a>
+<span class="sourceLineNo">296</span>    status.markComplete(msg);<a name="line.296"></a>
+<span class="sourceLineNo">297</span>    LOG.info(msg);<a name="line.297"></a>
+<span class="sourceLineNo">298</span>    return totalSize;<a name="line.298"></a>
+<span class="sourceLineNo">299</span>  }<a name="line.299"></a>
+<span class="sourceLineNo">300</span><a name="line.300"></a>
+<span class="sourceLineNo">301</span>  /**<a name="line.301"></a>
+<span class="sourceLineNo">302</span>   * Add a task entry to coordination if it is not already there.<a name="line.302"></a>
+<span class="sourceLineNo">303</span>   * @param taskname the path of the log to be split<a name="line.303"></a>
+<span class="sourceLineNo">304</span>   * @param batch the batch this task belongs to<a name="line.304"></a>
+<span class="sourceLineNo">305</span>   * @return true if a new entry is created, false if it is already there.<a name="line.305"></a>
+<span class="sourceLineNo">306</span>   */<a name="line.306"></a>
+<span class="sourceLineNo">307</span>  boolean enqueueSplitTask(String taskname, TaskBatch batch) {<a name="line.307"></a>
+<span class="sourceLineNo">308</span>    lastTaskCreateTime = EnvironmentEdgeManager.currentTime();<a name="line.308"></a>
+<span class="sourceLineNo">309</span>    String task = getSplitLogManagerCoordination().prepareTask(taskname);<a name="line.309"></a>
+<span class="sourceLineNo">310</span>    Task oldtask = createTaskIfAbsent(task, batch);<a name="line.310"></a>
+<span class="sourceLineNo">311</span>    if (oldtask == null) {<a name="line.311"></a>
+<span class="sourceLineNo">312</span>      // publish the task in the coordination engine<a name="line.312"></a>
+<span class="sourceLineNo">313</span>      getSplitLogManagerCoordination().submitTask(task);<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>    return false;<a name="line.316"></a>
+<span class="sourceLineNo">317</span>  }<a name="line.317"></a>
+<span class="sourceLineNo">318</span><a name="line.318"></a>
+<span class="sourceLineNo">319</span>  private void waitForSplittingCompletion(TaskBatch batch, MonitoredTask status) {<a name="line.319"></a>
+<span class="sourceLineNo">320</span>    synchronized (batch) {<a name="line.320"></a>
+<span class="sourceLineNo">321</span>      while ((batch.done + batch.error) != batch.installed) {<a name="line.321"></a>
+<span class="sourceLineNo">322</span>        try {<a name="line.322"></a>
+<span class="sourceLineNo">323</span>          status.setStatus("Waiting for distributed tasks to finish. " + " scheduled="<a name="line.323"></a>
+<span class="sourceLineNo">324</span>              + batch.installed + " done=" + batch.done + " error=" + batch.error);<a name="line.324"></a>
+<span class="sourceLineNo">325</span>          int remaining = batch.installed - (batch.done + batch.error);<a name="line.325"></a>
+<span class="sourceLineNo">326</span>          int actual = activeTasks(batch);<a name="line.326"></a>
+<span class="sourceLineNo">327</span>          if (remaining != actual) {<a name="line.327"></a>
+<span class="sourceLineNo">328</span>            LOG.warn("Expected " + remaining + " active tasks, but actually there are " + actual);<a name="line.328"></a>
+<span class="sourceLineNo">329</span>          }<a name="line.329"></a>
+<span class="sourceLineNo">330</span>          int remainingTasks = getSplitLogManagerCoordination().remainingTasksInCoordination();<a name="line.330"></a>
+<span class="sourceLineNo">331</span>          if (remainingTasks &gt;= 0 &amp;&amp; actual &gt; remainingTasks) {<a name="line.331"></a>
+<span class="sourceLineNo">332</span>            LOG.warn("Expected at least" + actual + " tasks remaining, but actually there are "<a name="line.332"></a>
+<span class="sourceLineNo">333</span>                + remainingTasks);<a name="line.333"></a>
+<span class="sourceLineNo">334</span>          }<a name="line.334"></a>
+<span class="sourceLineNo">335</span>          if (remainingTasks == 0 || actual == 0) {<a name="line.335"></a>
+<span class="sourceLineNo">336</span>            LOG.warn("No more task remaining, splitting "<a name="line.336"></a>
+<span class="sourceLineNo">337</span>                + "should have completed. Remaining tasks is " + remainingTasks<a name="line.337"></a>
+<span class="sourceLineNo">338</span>                + ", active tasks in map " + actual);<a name="line.338"></a>
+<span class="sourceLineNo">339</span>            if (remainingTasks == 0 &amp;&amp; actual == 0) {<a name="line.339"></a>
+<span class="sourceLineNo">340</span>              return;<a name="line.340"></a>
+<span class="sourceLineNo">341</span>            }<a name="line.341"></a>
+<span class="sourceLineNo">342</span>          }<a name="line.342"></a>
+<span class="sourceLineNo">343</span>          batch.wait(100);<a name="line.343"></a>
+<span class="sourceLineNo">344</span>          if (server.isStopped()) {<a name="line.344"></a>
+<span class="sourceLineNo">345</span>            LOG.warn("Stopped while waiting for log splits to be completed");<a name="line.345"></a>
+<span class="sourceLineNo">346</span>            return;<a name="line.346"></a>
+<span class="sourceLineNo">347</span>          }<a name="line.347"></a>
+<span class="sourceLineNo">348</span>        } catch (InterruptedException e) {<a name="line.348"></a>
+<span class="sourceLineNo">349</span>          LOG.warn("Interrupted while waiting for log splits to be completed");<a name="line.349"></a>
+<span class="sourceLineNo">350</span>          Thread.currentThread().interrupt();<a name="line.350"></a>
+<span class="sourceLineNo">351</span>          return;<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>    }<a name="line.354"></a>
+<span class="sourceLineNo">355</span>  }<a name="line.355"></a>
+<span class="sourceLineNo">356</span><a name="line.356"></a>
+<span class="sourceLineNo">357</span>  @VisibleForTesting<a name="line.357"></a>
+<span class="sourceLineNo">358</span>  ConcurrentMap&lt;String, Task&gt; getTasks() {<a name="line.358"></a>
+<span class="sourceLineNo">359</span>    return tasks;<a name="line.359"></a>
+<span class="sourceLineNo">360</span>  }<a name="line.360"></a>
+<span class="sourceLineNo">361</span><a name="line.361"></a>
+<span class="sourceLineNo">362</span>  private int activeTasks(final TaskBatch batch) {<a name="line.362"></a>
+<span class="sourceLineNo">363</span>    int count = 0;<a name="line.363"></a>
+<span class="sourceLineNo">364</span>    for (Task t : tasks.values()) {<a name="line.364"></a>
+<span class="sourceLineNo">365</span>      if (t.batch == batch &amp;&amp; t.status == TerminationStatus.IN_PROGRESS) {<a name="line.365"></a>
+<span class="sourceLineNo">366</span>        count++;<a name="line.366"></a>
+<span class="sourceLineNo">367</span>      }<a name="line.367"></a>
+<span class="sourceLineNo">368</span>    }<a name="line.368"></a>
+<span class="sourceLineNo">369</span>    return count;<a name="line.369"></a>
+<span class="sourceLineNo">370</span><a name="line.370"></a>
+<span class="sourceLineNo">371</span>  }<a name="line.371"></a>
+<span class="sourceLineNo">372</span><a name="line.372"></a>
+<span class="sourceLineNo">373</span>  /**<a name="line.373"></a>
+<span class="sourceLineNo">374</span>   * @param path<a name="line.374"></a>
+<span class="sourceLineNo">375</span>   * @param batch<a name="line.375"></a>
+<span class="sourceLineNo">376</span>   * @return null on success, existing task on error<a name="line.376"></a>
+<span class="sourceLineNo">377</span>   */<a name="line.377"></a>
+<span class="sourceLineNo">378</span>  private Task createTaskIfAbsent(String path, TaskBatch batch) {<a name="line.378"></a>
+<span class="sourceLineNo">379</span>    Task oldtask;<a name="line.379"></a>
+<span class="sourceLineNo">380</span>    // batch.installed is only changed via this function and<a name="line.380"></a>
+<span class="sourceLineNo">381</span>    // a single thread touches batch.installed.<a name="line.381"></a>
+<span class="sourceLineNo">382</span>    Task newtask = new Task();<a name="line.382"></a>
+<span class="sourceLineNo">383</span>    newtask.batch = batch;<a name="line.383"></a>
+<span class="sourceLineNo">384</span>    oldtask = tasks.putIfAbsent(path, newtask);<a name="line.384"></a>
+<span class="sourceLineNo">385</span>    if (oldtask == null) {<a name="line.385"></a>
+<span class="sourceLineNo">386</span>      batch.installed++;<a name="line.386"></a>
+<span class="sourceLineNo">387</span>      return null;<a name="line.387"></a>
+<span class="sourceLineNo">388</span>    }<a name="line.388"></a>
+<span class="sourceLineNo">389</span>    // new task was not used.<a name="line.389"></a>
+<span class="sourceLineNo">390</span>    synchronized (oldtask) {<a name="line.390"></a>
+<span class="sourceLineNo">391</span>      if (oldtask.isOrphan()) {<a name="line.391"></a>
+<span class="sourceLineNo">392</span>        if (oldtask.status == SUCCESS) {<a name="line.392"></a>
+<span class="sourceLineNo">393</span>          // The task is already done. Do not install the batch for this<a name="line.393"></a>
+<span class="sourceLineNo">394</span>          // task because it might be too late for setDone() to update<a name="line.394"></a>
+<span class="sourceLineNo">395</span>          // batch.done. There is no need for the batch creator to wait for<a name="line.395"></a>
+<span class="sourceLineNo">396</span>          // this task to complete.<a name="line.396"></a>
+<span class="sourceLineNo">397</span>          return (null);<a name="line.397"></a>
+<span class="sourceLineNo">398</span>        }<a name="line.398"></a>
+<span class="sourceLineNo">399</span>        if (oldtask.status == IN_PROGRESS) {<a name="line.399"></a>
+<span class="sourceLineNo">400</span>          oldtask.batch = batch;<a name="line.400"></a>
+<span class="sourceLineNo">401</span>          batch.installed++;<a name="line.401"></a>
+<span class="sourceLineNo">402</span>          LOG.debug("Previously orphan task " + path + " is now being waited upon");<a name="line.402"></a>
+<span class="sourceLineNo">403</span>          return null;<a name="line.403"></a>
+<span class="sourceLineNo">404</span>        }<a name="line.404"></a>
+<span class="sourceLineNo">405</span>        while (oldtask.status == FAILURE) {<a name="line.405"></a>
+<span class="sourceLineNo">406</span>          LOG.debug("wait for status of task " + path + " to change to DELETED");<a name="line.406"></a>
+<span class="sourceLineNo">407</span>          SplitLogCounters.tot_mgr_wait_for_zk_delete.increment();<a name="line.407"></a>
+<span class="sourceLineNo">408</span>          try {<a name="line.408"></a>
+<span class="sourceLineNo">409</span>            oldtask.wait();<a name="line.409"></a>
+<span class="sourceLineNo">410</span>          } catch (InterruptedException e) {<a name="line.410"></a>
+<span class="sourceLineNo">411</span>            Thread.currentThread().interrupt();<a name="line.411"></a>
+<span class="sourceLineNo">412</span>            LOG.warn("Interrupted when waiting for znode delete callback");<a name="line.412"></a>
+<span class="sourceLineNo">413</span>            // fall through to return failure<a name="line.413"></a>
+<span class="sourceLineNo">414</span>            break;<a name="line.414"></a>
+<span class="sourceLineNo">415</span>          }<a name="line.415"></a>
+<span class="sourceLineNo">416</span>        }<a name="line.416"></a>
+<span class="sourceLineNo">417</span>        if (oldtask.status != DELETED) {<a name="line.417"></a>
+<span class="sourceLineNo">418</span>          LOG.warn("Failure because previously failed task"<a name="line.418"></a>
+<span class="sourceLineNo">419</span>              + " state still present. Waiting for znode delete callback" + " path=" + path);<a name="line.419"></a>
+<span class="sourceLineNo">420</span>          return oldtask;<a name="line.420"></a>
+<span class="sourceLineNo">421</span>        }<a name="line.421"></a>
+<span class="sourceLineNo">422</span>        // reinsert the newTask and it must succeed this time<a name="line.422"></a>
+<span class="sourceLineNo">423</span>        Task t = tasks.putIfAbsent(path, newtask);<a name="line.423"></a>
+<span class="sourceLineNo">424</span>        if (t == null) {<a name="line.424"></a>
+<span class="sourceLineNo">425</span>          batch.installed++;<a name="line.425"></a>
+<span class="sourceLineNo">426</span>          return null;<a name="line.426"></a>
+<span class="sourceLineNo">427</span>        }<a name="line.427"></a>
+<span class="sourceLineNo">428</span>        LOG.error(HBaseMarkers.FATAL, "Logic error. Deleted task still present in tasks map");<a name="line.428"></a>
+<span class="sourceLineNo">429</span>        assert false : "Deleted task still present in tasks map";<a name="line.429"></a>
+<span class="sourceLineNo">430</span>        return t;<a name="line.430"></a>
+<span class="sourceLineNo">431</span>      }<a name="line.431"></a>
+<span class="sourceLineNo">432</span>      LOG.warn("Failure because two threads can't wait for the same task; path=" + path);<a name="line.432"></a>
+<span class="sourceLineNo">433</span>      return oldtask;<a name="line.433"></a>
+<span class="sourceLineNo">434</span>    }<a name="line.434"></a>
+<span class="sourceLineNo">435</span>  }<a name="line.435"></a>
+<span class="sourceLineNo">436</span><a name="line.436"></a>
+<span class="sourceLineNo">437</span>  public void stop() {<a name="line.437"></a>
+<span class="sourceLineNo">438</span>    if (choreService != null) {<a name="line.438"></a>
+<span class="sourceLineNo">439</span>      choreService.shutdown();<a name="line.439"></a>
+<span class="sourceLineNo">440</span>    }<a name="line.440"></a>
+<span class="sourceLineNo">441</span>    if (timeoutMonitor != null) {<a name="line.441"></a>
+<span class="sourceLineNo">442</span>      timeoutMonitor.cancel(true);<a name="line.442"></a>
+<span class="sourceLineNo">443</span>    }<a name="line.443"></a>
+<span class="sourceLineNo">444</span>  }<a name="line.444"></a>
+<span class="sourceLineNo">445</span><a name="line.445"></a>
+<span class="sourceLineNo">446</span>  void handleDeadWorker(ServerName workerName) {<a name="line.446"></a>
+<span class="sourceLineNo">447</span>    // resubmit the tasks on the TimeoutMonitor thread. Makes it easier<a name="line.447"></a>
+<span class="sourceLineNo">448</span>    // to reason about concurrency. Makes it easier to retry.<a name="line.448"></a>
+<span class="sourceLineNo">449</span>    synchronized (deadWorkersLock) {<a name="line.449"></a>
+<span class="sourceLineNo">450</span>      if (deadWorkers == null) {<a name="line.450"></a>
+<span class="sourceLineNo">451</span>        deadWorkers = new HashSet&lt;&gt;(100);<a name="line.451"></a>
+<span class="sourceLineNo">452</span>      }<a name="line.452"></a>
+<span class="sourceLineNo">453</span>      deadWorkers.add(workerName);<a name="line.453"></a>
+<span class="sourceLineNo">454</span>    }<a name="line.454"></a>
+<span class="sourceLineNo">455</span>    LOG.info("Dead splitlog worker {}", workerName);<a name="line.455"></a>
+<span class="sourceLineNo">456</span>  }<a name="line.456"></a>
+<span class="sourceLineNo">457</span><a name="line.457"></a>
+<span class="sourceLineNo">458</span>  void handleDeadWorkers(Set&lt;ServerName&gt; serverNames) {<a name="line.458"></a>
+<span class="sourceLineNo">459</span>    synchronized (deadWorkersLock) {<a name="line.459"></a>
+<span class="sourceLineNo">460</span>      if (deadWorkers == null) {<a name="line.460"></a>
+<span class="sourceLineNo">461</span>        deadWorkers = new HashSet&lt;&gt;(100);<a name="line.461"></a>
+<span class="sourceLineNo">462</span>      }<a name="line.462"></a>
+<span class="sourceLineNo">463</span>      deadWorkers.addAll(serverNames);<a name="line.463"></a>
+<span class="sourceLineNo">464</span>    }<a name="line.464"></a>
+<span class="sourceLineNo">465</span>    LOG.info("dead splitlog workers " + serverNames);<a name="line.465"></a>
+<span class="sourceLineNo">466</span>  }<a name="line.466"></a>
+<span class="sourceLineNo">467</span><a name="line.467"></a>
+<span class="sourceLineNo">468</span>  /**<a name="line.468"></a>
+<span class="sourceLineNo">469</span>   * Keeps track of the batch of tasks submitted together by a caller in splitLogDistributed().<a name="line.469"></a>
+<span class="sourceLineNo">470</span>   * Clients threads use this object to wait for all their tasks to be done.<a name="line.470"></a>
+<span class="sourceLineNo">471</span>   * &lt;p&gt;<a name="line.471"></a>
+<span class="sourceLineNo">472</span>   * All access is synchronized.<a name="line.472"></a>
+<span class="sourceLineNo">473</span>   */<a name="line.473"></a>
+<span class="sourceLineNo">474</span>  @InterfaceAudience.Private<a name="line.474"></a>
+<span class="sourceLineNo">475</span>  public static class TaskBatch {<a name="line.475"></a>
+<span class="sourceLineNo">476</span>    public int installed = 0;<a name="line.476"></a>
+<span class="sourceLineNo">477</span>    public int done = 0;<a name="line.477"></a>
+<span class="sourceLineNo">478</span>    public int error = 0;<a name="line.478"></a>
+<span class="sourceLineNo">479</span>    public volatile boolean isDead = false;<a name="line.479"></a>
+<span class="sourceLineNo">480</span><a name="line.480"></a>
+<span class="sourceLineNo">481</span>    @Override<a name="line.481"></a>
+<span class="sourceLineNo">482</span>    public String toString() {<a name="line.482"></a>
+<span class="sourceLineNo">483</span>      return ("installed = " + installed + " done = " + done + " error = " + error);<a name="line.483"></a>
+<span class="sourceLineNo">484</span>    }<a name="line.484"></a>
+<span class="sourceLineNo">485</span>  }<a name="line.485"></a>
+<span class="sourceLineNo">486</span><a name="line.486"></a>
+<span class="sourceLineNo">487</span>  /**<a name="line.487"></a>
+<span class="sourceLineNo">488</span>   * in memory state of an active task.<a name="line.488"></a>
+<span class="sourceLineNo">489</span>   */<a name="line.489"></a>
+<span class="sourceLineNo">490</span>  @InterfaceAudience.Private<a name="line.490"></a>
+<span class="sourceLineNo">491</span>  public static class Task {<a name="line.491"></a>
+<span class="sourceLineNo">492</span>    public volatile long last_update;<a name="line.492"></a>
+<span class="sourceLineNo">493</span>    public volatile int last_version;<a name="line.493"></a>
+<span class="sourceLineNo">494</span>    public volatile ServerName cur_worker_name;<a name="line.494"></a>
+<span class="sourceLineNo">495</span>    public volatile TaskBatch batch;<a name="line.495"></a>
+<span class="sourceLineNo">496</span>    public volatile TerminationStatus status;<a name="line.496"></a>
+<span class="sourceLineNo">497</span>    public volatile AtomicInteger incarnation = new AtomicInteger(0);<a name="line.497"></a>
+<span class="sourceLineNo">498</span>    public final AtomicInteger unforcedResubmits = new AtomicInteger();<a name="line.498"></a>
+<span class="sourceLineNo">499</span>    public volatile boolean resubmitThresholdReached;<a name="line.499"></a>
+<span class="sourceLineNo">500</span><a name="line.500"></a>
+<span class="sourceLineNo">501</span>    @Override<a name="line.501"></a>
+<span class="sourceLineNo">502</span>    public String toString() {<a name="line.502"></a>
+<span class="sourceLineNo">503</span>      return ("last_update = " + last_update + " last_version = " + last_version<a name="line.503"></a>
+<span class="sourceLineNo">504</span>          + " cur_worker_name = " + cur_worker_name + " status = " + status + " incarnation = "<a name="line.504"></a>
+<span class="sourceLineNo">505</span>          + incarnation + " resubmits = " + unforcedResubmits.get() + " batch = " + batch);<a name="line.505"></a>
+<span class="sourceLineNo">506</span>    }<a name="line.506"></a>
+<span class="sourceLineNo">507</span><a name="line.507"></a>
+<span class="sourceLineNo">508</span>    public Task() {<a name="line.508"></a>
+<span class="sourceLineNo">509</span>      last_version = -1;<a name="line.509"></a>
+<span class="sourceLineNo">510</span>      status = IN_PROGRESS;<a name="line.510"></a>
+<span class="sourceLineNo">511</span>      setUnassigned();<a name="line.511"></a>
 <span class="sourceLineNo">512</span>    }<a name="line.512"></a>
 <span class="sourceLineNo">513</span><a name="line.513"></a>
-<span class="sourceLineNo">514</span>    public boolean isUnassigned() {<a name="line.514"></a>
-<span class="sourceLineNo">515</span>      return (cur_worker_name == null);<a name="line.515"></a>
+<span class="sourceLineNo">514</span>    public boolean isOrphan() {<a name="line.514"></a>
+<span class="sourceLineNo">515</span>      return (batch == null || batch.isDead);<a name="line.515"></a>
 <span class="sourceLineNo">516</span>    }<a name="line.516"></a>
 <span class="sourceLineNo">517</span><a name="line.517"></a>
-<span class="sourceLineNo">518</span>    public void heartbeatNoDetails(long time) {<a name="line.518"></a>
-<span class="sourceLineNo">519</span>      last_update = time;<a name="line.519"></a>
+<span class="sourceLineNo">518</span>    public boolean isUnassigned() {<a name="line.518"></a>
+<span class="sourceLineNo">519</span>      return (cur_worker_name == null);<a name="line.519"></a>
 <span class="sourceLineNo">520</span>    }<a name="line.520"></a>
 <span class="sourceLineNo">521</span><a name="line.521"></a>
-<span class="sourceLineNo">522</span>    public void heartbeat(long time, int version, ServerName worker) {<a name="line.522"></a>
-<span class="sourceLineNo">523</span>      last_version = version;<a name="line.523"></a>
-<span class="sourceLineNo">524</span>      last_update = time;<a name="line.524"></a>
-<span class="sourceLineNo">525</span>      cur_worker_name = worker;<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>    public void setUnassigned() {<a name="line.528"></a>
-<span class="sourceLineNo">529</span>      cur_worker_name = null;<a name="line.529"></a>
-<span class="sourceLineNo">530</span>      last_update = -1;<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><a name="line.533"></a>
-<span class="sourceLineNo">534</span>  /**<a name="line.534"></a>
-<span class="sourceLineNo">535</span>   * Periodically checks all active tasks and resubmits the ones that have timed out<a name="line.535"></a>
-<span class="sourceLineNo">536</span>   */<a name="line.536"></a>
-<span class="sourceLineNo">537</span>  private class TimeoutMonitor extends ScheduledChore {<a name="line.537"></a>
-<span class="sourceLineNo">538</span>    private long lastLog = 0;<a name="line.538"></a>
-<span class="sourceLineNo">539</span><a name="line.539"></a>
-<span class="sourceLineNo">540</span>    public TimeoutMonitor(final int period, Stoppable stopper) {<a name="line.540"></a>
-<span class="sourceLineNo">541</span>      super("SplitLogManager Timeout Monitor", stopper, period);<a name="line.541"></a>
-<span class="sourceLineNo">542</span>    }<a name="line.542"></a>
+<span class="sourceLineNo">522</span>    public void heartbeatNoDetails(long time) {<a name="line.522"></a>
+<span class="sourceLineNo">523</span>      last_update = time;<a name="line.523"></a>
+<span class="sourceLineNo">524</span>    }<a name="line.524"></a>
+<span class="sourceLineNo">525</span><a name="line.525"></a>
+<span class="sourceLineNo">526</span>    public void heartbeat(long time, int version, ServerName worker) {<a name="line.526"></a>
+<span class="sourceLineNo">527</span>      last_version = version;<a name="line.527"></a>
+<span class="sourceLineNo">528</span>      last_update = time;<a name="line.528"></a>
+<span class="sourceLineNo">529</span>      cur_worker_name = worker;<a name="line.529"></a>
+<span class="sourceLineNo">530</span>    }<a name="line.530"></a>
+<span class="sourceLineNo">531</span><a name="line.531"></a>
+<span class="sourceLineNo">532</span>    public void setUnassigned() {<a name="line.532"></a>
+<span class="sourceLineNo">533</span>      cur_worker_name = null;<a name="line.533"></a>
+<span class="sourceLineNo">534</span>      last_update = -1;<a name="line.534"></a>
+<span class="sourceLineNo">535</span>    }<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>  /**<a name="line.538"></a>
+<span class="sourceLineNo">539</span>   * Periodically checks all active tasks and resubmits the ones that have timed out<a name="line.539"></a>
+<span class="sourceLineNo">540</span>   */<a name="line.540"></a>
+<span class="sourceLineNo">541</span>  private class TimeoutMonitor extends ScheduledChore {<a name="line.541"></a>
+<span class="sourceLineNo">542</span>    private long lastLog = 0;<a name="line.542"></a>
 <span class="sourceLineNo">543</span><a name="line.543"></a>
-<span class="sourceLineNo">544</span>    @Override<a name="line.544"></a>
-<span class="sourceLineNo">545</span>    protected void chore() {<a name="line.545"></a>
-<span class="sourceLineNo">546</span>      if (server.getCoordinatedStateManager() == null) return;<a name="line.546"></a>
+<span class="sourceLineNo">544</span>    public TimeoutMonitor(final int period, Stoppable stopper) {<a name="line.544"></a>
+<span class="sourceLineNo">545</span>      super("SplitLogManager Timeout Monitor", stopper, period);<a name="line.545"></a>
+<span class="sourceLineNo">546</span>    }<a name="line.546"></a>
 <span class="sourceLineNo">547</span><a name="line.547"></a>
-<span class="sourceLineNo">548</span>      int resubmitted = 0;<a name="line.548"></a>
-<span class="sourceLineNo">549</span>      int unassigned = 0;<a name="line.549"></a>
-<span class="sourceLineNo">550</span>      int tot = 0;<a name="line.550"></a>
-<span class="sourceLineNo">551</span>      boolean found_assigned_task = false;<a name="line.551"></a>
-<span class="sourceLineNo">552</span>      Set&lt;ServerName&gt; localDeadWorkers;<a name="line.552"></a>
-<span class="sourceLineNo">553</span><a name="line.553"></a>
-<span class="sourceLineNo">554</span>      synchronized (deadWorkersLock) {<a name="line.554"></a>
-<span class="sourceLineNo">555</span>        localDeadWorkers = deadWorkers;<a name="line.555"></a>
-<span class="sourceLineNo">556</span>        deadWorkers = null;<a name="line.556"></a>
-<span class="sourceLineNo">557</span>      }<a name="line.557"></a>
-<span class="sourceLineNo">558</span><a name="line.558"></a>
-<span class="sourceLineNo">559</span>      for (Map.Entry&lt;String, Task&gt; e : tasks.entrySet()) {<a name="line.559"></a>
-<span class="sourceLineNo">560</span>        String path = e.getKey();<a name="line.560"></a>
-<span class="sourceLineNo">561</span>        Task task = e.getValue();<a name="line.561"></a>
-<span class="sourceLineNo">562</span>        ServerName cur_worker = task.cur_worker_name;<a name="line.562"></a>
-<span class="sourceLineNo">563</span>        tot++;<a name="line.563"></a>
-<span class="sourceLineNo">564</span>        // don't easily resubmit a task which hasn't been picked up yet. It<a name="line.564"></a>
-<span class="sourceLineNo">565</span>        // might be a long while before a SplitLogWorker is free to pick up a<a name="line.565"></a>
-<span class="sourceLineNo">566</span>        // task. This is because a SplitLogWorker picks up a task one at a<a name="line.566"></a>
-<span class="sourceLineNo">567</span>        // time. If we want progress when there are no region servers then we<a name="line.567"></a>
-<span class="sourceLineNo">568</span>        // will have to run a SplitLogWorker thread in the Master.<a name="line.568"></a>
-<span class="sourceLineNo">569</span>        if (task.isUnassigned()) {<a name="line.569"></a>
-<span class="sourceLineNo">570</span>          unassigned++;<a name="line.570"></a>
-<span class="sourceLineNo">571</span>          continue;<a name="line.571"></a>
-<span class="sourceLineNo">572</span>        }<a name="line.572"></a>
-<span class="sourceLineNo">573</span>        found_assigned_task = true;<a name="line.573"></a>
-<span class="sourceLineNo">574</span>        if (localDeadWorkers != null &amp;&amp; localDeadWorkers.contains(cur_worker)) {<a name="line.574"></a>
-<span class="sourceLineNo">575</span>          SplitLogCounters.tot_mgr_resubmit_dead_server_task.increment();<a name="line.575"></a>
-<span class="sourceLineNo">576</span>          if (getSplitLogManagerCoordination().resubmitTask(path, task, FORCE)) {<a name="line.576"></a>
-<span class="sourceLineNo">577</span>            resubmitted++;<a name="line.577"></a>
-<span class="sourceLineNo">578</span>          } else {<a name="line.578"></a>
-<span class="sourceLineNo">579</span>            handleDeadWorker(cur_worker);<a name="line.579"></a>
-<span class="sourceLineNo">580</span>            LOG.warn("Failed to resubmit task " + path + " owned by dead " + cur_worker<a name="line.580"></a>
-<span class="sourceLineNo">581</span>                + ", will retry.");<a name="line.581"></a>
-<span class="sourceLineNo">582</span>          }<a name="line.582"></a>
-<span class="sourceLineNo">583</span>        } else if (getSplitLogManagerCoordination().resubmitTask(path, task, CHECK)) {<a name="line.583"></a>
-<span class="sourceLineNo">584</span>          resubmitted++;<a name="line.584"></a>
-<span class="sourceLineNo">585</span>        }<a name="line.585"></a>
-<span class="sourceLineNo">586</span>      }<a name="line.586"></a>
-<span class="sourceLineNo">587</span>      if (tot &gt; 0) {<a name="line.587"></a>
-<span class="sourceLineNo">588</span>        long now = EnvironmentEdgeManager.currentTime();<a name="line.588"></a>
-<span class="sourceLineNo">589</span>        if (now &gt; lastLog + 5000) {<a name="line.589"></a>
-<span class="sourceLineNo">590</span>          lastLog = now;<a name="line.590"></a>
-<span class="sourceLineNo">591</span>          LOG.info("total=" + tot + ", unassigned=" + unassigned + ", tasks=" + tasks);<a name="line.591"></a>
-<span class="sourceLineNo">592</span>        }<a name="line.592"></a>
-<span class="sourceLineNo">593</span>      }<a name="line.593"></a>
-<span class="sourceLineNo">594</span>      if (resubmitted &gt; 0) {<a name="line.594"></a>
-<span class="sourceLineNo">595</span>        LOG.info("resubmitted " + resubmitted + " out of " + tot + " tasks");<a name="line.595"></a>
-<span class="sourceLineNo">596</span>      }<a name="line.596"></a>
-<span class="sourceLineNo">597</span>      // If there are pending tasks and all of them have been unassigned for<a name="line.597"></a>
-<span class="sourceLineNo">598</span>      // some time then put up a RESCAN node to ping the workers.<a name="line.598"></a>
-<span class="sourceLineNo">599</span>      // ZKSplitlog.DEFAULT_UNASSIGNED_TIMEOUT is of the order of minutes<a name="line.599"></a>
-<span class="sourceLineNo">600</span>      // because a. it is very unlikely that every worker had a<a name="line.600"></a>
-<span class="sourceLineNo">601</span>      // transient error when trying to grab the task b. if there are no<a name="line.601"></a>
-<span class="sourceLineNo">602</span>      // workers then all tasks wills stay unassigned indefinitely and the<a name="line.602"></a>
-<span class="sourceLineNo">603</span>      // manager will be indefinitely creating RESCAN nodes. TODO may be the<a name="line.603"></a>
-<span class="sourceLineNo">604</span>      // master should spawn both a manager and a worker thread to guarantee<a name="line.604"></a>
-<span class="sourceLineNo">605</span>      // that there is always one worker in the system<a name="line.605"></a>
-<span class="sourceLineNo">606</span>      if (tot &gt; 0<a name="line.606"></a>
-<span class="sourceLineNo">607</span>          &amp;&amp; !found_assigned_task<a name="line.607"></a>
-<span class="sourceLineNo">608</span>          &amp;&amp; ((EnvironmentEdgeManager.currentTime() - lastTaskCreateTime) &gt; unassignedTimeout)) {<a name="line.608"></a>
-<span class="sourceLineNo">609</span>        for (Map.Entry&lt;String, Task&gt; e : tasks.entrySet()) {<a name="line.609"></a>
-<span class="sourceLineNo">610</span>          String key = e.getKey();<a name="line.610"></a>
-<span class="sourceLineNo">611</span>          Task task = e.getValue();<a name="line.611"></a>
-<span class="sourceLineNo">612</span>          // we have to do task.isUnassigned() check again because tasks might<a name="line.612"></a>
-<span class="sourceLineNo">613</span>          // have been asynchronously assigned. There is no locking required<a name="line.613"></a>
-<span class="sourceLineNo">614</span>          // for these checks ... it is OK even if tryGetDataSetWatch() is<a name="line.614"></a>
-<span class="sourceLineNo">615</span>          // called unnecessarily for a taskpath<a name="line.615"></a>
-<span class="sourceLineNo">616</span>          if (task.isUnassigned() &amp;&amp; (task.status != FAILURE)) {<a name="line.616"></a>
-<span class="sourceLineNo">617</span>            // We just touch the znode to make sure its still there<a name="line.617"></a>
-<span class="sourceLineNo">618</span>            getSplitLogManagerCoordination().checkTaskStillAvailable(key);<a name="line.618"></a>
-<span class="sourceLineNo">619</span>          }<a name="line.619"></a>
-<span class="sourceLineNo">620</span>        }<a name="line.620"></a>
-<span class="sourceLineNo">621</span>        getSplitLogManagerCoordination().checkTasks();<a name="line.621"></a>
-<span class="sourceLineNo">622</span>        SplitLogCounters.tot_mgr_resubmit_unassigned.increment();<a name="line.622"></a>
-<span class="sourceLineNo">623</span>        LOG.debug("resubmitting unassigned task(s) after timeout");<a name="line.623"></a>
-<span class="sourceLineNo">624</span>      }<a name="line.624"></a>
-<span class="sourceLineNo">625</span>      Set&lt;String&gt; failedDeletions =<a name="line.625"></a>
-<span class="sourceLineNo">626</span>        getSplitLogManagerCoordination().getDetails().getFailedDeletions();<a name="line.626"></a>
-<span class="sourceLineNo">627</span>      // Retry previously failed deletes<a name="line.627"></a>
-<span class="sourceLineNo">628</span>      if (failedDeletions.size() &gt; 0) {<a name="line.628"></a>
-<span class="sourceLineNo">629</span>        List&lt;String&gt; tmpPaths = new ArrayList&lt;&gt;(failedDeletions);<a name="line.629"></a>
-<span class="sourceLineNo">630</span>        for (String tmpPath : tmpPaths) {<a name="line.630"></a>
-<span class="sourceLineNo">631</span>          // deleteNode is an async call<a name="line.631"></a>
-<span class="sourceLineNo">632</span>          getSplitLogManagerCoordination().deleteTask(tmpPath);<a name="line.632"></a>
-<span class="sourceLineNo">633</span>        }<a name="line.633"></a>
-<span class="sourceLineNo">634</span>        failedDeletions.removeAll(tmpPaths);<a name="line.634"></a>
-<span class="sourceLineNo">635</span>      }<a name="line.635"></a>
-<span class="sourceLineNo">636</span>    }<a name="line.636"></a>
-<span class="sourceLineNo">637</span>  }<a name="line.637"></a>
-<span class="sourceLineNo">638</span><a name="line.638"></a>
-<span class="sourceLineNo">639</span>  public enum ResubmitDirective {<a name="line.639"></a>
-<span class="sourceLineNo">640</span>    CHECK(), FORCE()<a name="line.640"></a>
+<span class="sourceLineNo">548</span>    @Override<a name="line.548"></a>
+<span class="sourceLineNo">549</span>    protected void chore() {<a name="line.549"></a>
+<span class="sourceLineNo">550</span>      if (server.getCoordinatedStateManager() == null) return;<a name="line.550"></a>
+<span class="sourceLineNo">551</span><a name="line.551"></a>
+<span class="sourceLineNo">552</span>      int resubmitted = 0;<a name="line.552"></a>
+<span class="sourceLineNo">553</span>      int unassigned = 0;<a name="line.553"></a>
+<span class="sourceLineNo">554</span>      int tot = 0;<a name="line.554"></a>
+<span class="sourceLineNo">555</span>      boolean found_assigned_task = false;<a name="line.555"></a>
+<span class="sourceLineNo">556</span>      Set&lt;ServerName&gt; localDeadWorkers;<a name="line.556"></a>
+<span class="sourceLineNo">557</span><a name="line.557"></a>
+<span class="sourceLineNo">558</span>      synchronized (deadWorkersLock) {<a name="line.558"></a>
+<span class="sourceLineNo">559</span>        localDeadWorkers = deadWorkers;<a name="line.559"></a>
+<span class="sourceLineNo">560</span>        deadWorkers = null;<a name="line.560"></a>
+<span class="sourceLineNo">561</span>      }<a name="line.561"></a>
+<span class="sourceLineNo">562</span><a name="line.562"></a>
+<span class="sourceLineNo">563</span>      for (Map.Entry&lt;String, Task&gt; e : tasks.entrySet()) {<a name="line.563"></a>
+<span class="sourceLineNo">564</span>        String path = e.getKey();<a name="line.564"></a>
+<span class="sourceLineNo">565</span>        Task task = e.getValue();<a name="line.565"></a>
+<span class="sourceLineNo">566</span>        ServerName cur_worker = task.cur_worker_name;<a name="line.566"></a>
+<span class="sourceLineNo">567</span>        tot++;<a name="line.567"></a>
+<span class="sourceLineNo">568</span>        // don't easily resubmit a task which hasn't been picked up yet. It<a name="line.568"></a>
+<span class="sourceLineNo">569</span>        // might be a long while before a SplitLogWorker is free to pick up a<a name="line.569"></a>
+<span class="sourceLineNo">570</span>        // task. This is because a SplitLogWorker picks up a task one at a<a name="line.570"></a>
+<span class="sourceLineNo">571</span>        // time. If we want progress when there are no region servers then we<a name="line.571"></a>
+<span class="sourceLineNo">572</span>        // will have to run a SplitLogWorker thread in the Master.<a name="line.572"></a>
+<span class="sourceLineNo">573</span>        if (task.isUnassigned()) {<a name="line.573"></a>
+<span class="sourceLineNo">574</span>          unassigned++;<a name="line.574"></a>
+<span class="sourceLineNo">575</span>          continue;<a name="line.575"></a>
+<span class="sourceLineNo">576</span>        }<a name="line.576"></a>
+<span class="sourceLineNo">577</span>        found_assigned_task = true;<a name="line.577"></a>
+<span class="sourceLineNo">578</span>        if (localDeadWorkers != null &amp;&amp; localDeadWorkers.contains(cur_worker)) {<a name="line.578"></a>
+<span class="sourceLineNo">579</span>          SplitLogCounters.tot_mgr_resubmit_dead_server_task.increment();<a name="line.579"></a>
+<span class="sourceLineNo">580</span>          if (getSplitLogManagerCoordination().resubmitTask(path, task, FORCE)) {<a name="line.580"></a>
+<span class="sourceLineNo">581</span>            resubmitted++;<a name="line.581"></a>
+<span class="sourceLineNo">582</span>          } else {<a name="line.582"></a>
+<span class="sourceLineNo">583</span>            handleDeadWorker(cur_worker);<a name="line.583"></a>
+<span class="sourceLineNo">584</span>            LOG.warn("Failed to resubmit task " + path + " owned by dead " + cur_worker<a name="line.584"></a>
+<span class="sourceLineNo">585</span>                + ", will retry.");<a name="line.585"></a>
+<span class="sourceLineNo">586</span>          }<a name="line.586"></a>
+<span class="sourceLineNo">587</span>        } else if (getSplitLogManagerCoordination().resubmitTask(path, task, CHECK)) {<a name="line.587"></a>
+<span class="sourceLineNo">588</span>          resubmitted++;<a name="line.588"></a>
+<span class="sourceLineNo">589</span>        }<a name="line.589"></a>
+<span class="sourceLineNo">590</span>      }<a name="line.590"></a>
+<span class="sourceLineNo">591</span>      if (tot &gt; 0) {<a name="line.591"></a>
+<span class="sourceLineNo">592</span>        long now = EnvironmentEdgeManager.currentTime();<a name="line.592"></a>
+<span class="sourceLineNo">593</span>        if (now &gt; lastLog + 5000) {<a name="line.593"></a>
+<span class="sourceLineNo">594</span>          lastLog = now;<a name="line.594"></a>
+<span class="sourceLineNo">595</span>          LOG.info("total=" + tot + ", unassigned=" + unassigned + ", tasks=" + tasks);<a name="line.595"></a>
+<span class="sourceLineNo">596</span>        }<a name="line.596"></a>
+<span class="sourceLineNo">597</span>      }<a name="line.597"></a>
+<span class="sourceLineNo">598</span>      if (resubmitted &gt; 0) {<a name="line.598"></a>
+<span class="sourceLineNo">599</span>        LOG.info("resubmitted " + resubmitted + " out of " + tot + " tasks");<a name="line.599"></a>
+<span class="sourceLineNo">600</span>      }<a name="line.600"></a>
+<span class="sourceLineNo">601</span>      // If there are pending tasks and all of them have been unassigned for<a name="line.601"></a>
+<span class="sourceLineNo">602</span>      // some time then put up a RESCAN node to ping the workers.<a name="line.602"></a>
+<span class="sourceLineNo">603</span>      // ZKSplitlog.DEFAULT_UNASSIGNED_TIMEOUT is of the order of minutes<a name="line.603"></a>
+<span class="sourceLineNo">604</span>      // because a. it is very unlikely that every worker had a<a name="line.604"></a>
+<span class="sourceLineNo">605</span>      // transient error when trying to grab the task b. if there are no<a name="line.605"></a>
+<span class="sourceLineNo">606</span>      // workers then all tasks wills stay unassigned indefinitely and the<a name="line.606"></a>
+<span class="sourceLineNo">607</span>      // manager will be indefinitely creating RESCAN nodes. TODO may be the<a name="line.607"></a>
+<span class="sourceLineNo">608</span>      // master should spawn both a manager and a worker thread to guarantee<a name="line.608"></a>
+<span class="sourceLineNo">609</span>      // that there is always one worker in the system<a name="line.609"></a>
+<span class="sourceLineNo">610</span>      if (tot &gt; 0<a name="line.610"></a>
+<span class="sourceLineNo">611</span>          &amp;&amp; !found_assigned_task<a name="line.611"></a>
+<span class="sourceLineNo">612</span>          &amp;&amp; ((EnvironmentEdgeManager.currentTime() - lastTaskCreateTime) &gt; unassignedTimeout)) {<a name="line.612"></a>
+<span class="sourceLineNo">613</span>        for (Map.Entry&lt;String, Task&gt; e : tasks.entrySet()) {<a name="line.613"></a>
+<span class="sourceLineNo">614</span>          String key = e.getKey();<a name="line.614"></a>
+<span class="sourceLineNo">615</span>          Task task = e.getValue();<a name="line.615"></a>
+<span class="sourceLineNo">616</span>          // we have to do task.isUnassigned() check again because tasks might<a name="line.616"></a>
+<span class="sourceLineNo">617</span>          // have been asynchronously assigned. There is no locking required<a name="line.617"></a>
+<span class="sourceLineNo">618</span>          // for these checks ... it is OK even if tryGetDataSetWatch() is<a name="line.618"></a>
+<span class="sourceLineNo">619</span>          // called unnecessarily for a taskpath<a name="line.619"></a>
+<span class="sourceLineNo">620</span>          if (task.isUnassigned() &amp;&amp; (task.status != FAILURE)) {<a name="line.620"></a>
+<span class="sourceLineNo">621</span>            // We just touch the znode to make sure its still there<a name="line.621"></a>
+<span class="sourceLineNo">622</span>            getSplitLogManagerCoordination().checkTaskStillAvailable(key);<a name="line.622"></a>
+<span class="sourceLineNo">623</span>          }<a name="line.623"></a>
+<span class="sourceLineNo">624</span>        }<a name="line.624"></a>
+<span class="sourceLineNo">625</span>        getSplitLogManagerCoordination().checkTasks();<a name="line.625"></a>
+<span class="sourceLineNo">626</span>        SplitLogCounters.tot_mgr_resubmit_unassigned.increment();<a name="line.626"></a>
+<span class="sourceLineNo">627</span>        LOG.debug("resubmitting unassigned task(s) after timeout");<a name="line.627"></a>
+<span class="sourceLineNo">628</span>      }<a name="line.628"></a>
+<span class="sourceLineNo">629</span>      Set&lt;String&gt; failedDeletions =<a name="line.629"></a>
+<span class="sourceLineNo">630</span>        getSplitLogManagerCoordination().getDetails().getFailedDeletions();<a name="line.630"></a>
+<span class="sourceLineNo">631</span>      // Retry previously failed deletes<a name="line.631"></a>
+<span class="sourceLineNo">632</span>      if (failedDeletions.size() &gt; 0) {<a name="line.632"></a>
+<span class="sourceLineNo">633</span>        List&lt;String&gt; tmpPaths = new ArrayList&lt;&gt;(failedDeletions);<a name="line.633"></a>
+<span class="sourceLineNo">634</span>        for (String tmpPath : tmpPaths) {<a name="line.634"></a>
+<span class="sourceLineNo">635</span>          // deleteNode is an async call<a name="line.635"></a>
+<span class="sourceLineNo">636</span>          getSplitLogManagerCoordination().deleteTask(tmpPath);<a name="line.636"></a>
+<span class="sourceLineNo">637</span>        }<a name="line.637"></a>
+<span class="sourceLineNo">638</span>        failedDeletions.removeAll(tmpPaths);<a name="line.638"></a>
+<span class="sourceLineNo">639</span>      }<a name="line.639"></a>
+<span class="sourceLineNo">640</span>    }<a name="line.640"></a>
 <span class="sourceLineNo">641</span>  }<a name="line.641"></a>
 <span class="sourceLineNo">642</span><a name="line.642"></a>
-<span class="sourceLineNo">643</span>  public enum TerminationStatus {<a name="line.643"></a>
-<span class="sourceLineNo">644</span>    IN_PROGRESS("in_progress"), SUCCESS("success"), FAILURE("failure"), DELETED("deleted");<a name="line.644"></a>
-<span class="sourceLineNo">645</span><a name="line.645"></a>
-<span class="sourceLineNo">646</span>    final String statusMsg;<a name="line.646"></a>
-<span class="sourceLineNo">647</span><a name="line.647"></a>
-<span class="sourceLineNo">648</span>    TerminationStatus(String msg) {<a name="line.648"></a>
-<span class="sourceLineNo">649</span>      statusMsg = msg;<a name="line.649"></a>
-<span class="sourceLineNo">650</span>    }<a name="line.650"></a>
+<span class="sourceLineNo">643</span>  public enum ResubmitDirective {<a name="line.643"></a>
+<span class="sourceLineNo">644</span>    CHECK(), FORCE()<a name="line.644"></a>
+<span class="sourceLineNo">645</span>  }<a name="line.645"></a>
+<span class="sourceLineNo">646</span><a name="line.646"></a>
+<span class="sourceLineNo">647</span>  public enum TerminationStatus {<a name="line.647"></a>
+<span class="sourceLineNo">648</span>    IN_PROGRESS("in_progress"), SUCCESS("success"), FAILURE("failure"), DELETED("deleted");<a name="line.648"></a>
+<span class="sourceLineNo">649</span><a name="line.649"></a>
+<span class="sourceLineNo">650</span>    final String statusMsg;<a name="line.650"></a>
 <span class="sourceLineNo">651</span><a name="line.651"></a>
-<span class="sourceLineNo">652</span>    @Override<a name="line.652"></a>
-<span class="sourceLineNo">653</span>    public String toString() {<a name="line.653"></a>
-<span class="sourceLineNo">654</span>      return statusMsg;<a name="line.654"></a>
-<span class="sourceLineNo">655</span>    }<a name="line.655"></a>
-<span class="sourceLineNo">656</span>  }<a name="line.656"></a>
-<span class="sourceLineNo">657</span>}<a name="line.657"></a>
+<span class="sourceLineNo">652</span>    TerminationStatus(String msg) {<a name="line.652"></a>
+<span class="sourceLineNo">653</span>      statusMsg = msg;<a name="line.653"></a>
+<span class="sourceLineNo">654</span>    }<a name="line.654"></a>
+<span class="sourceLineNo">655</span><a name="line.655"></a>
+<span class="sourceLineNo">656</span>    @Override<a name="line.656"></a>
+<span class="sourceLineNo">657</span>    public String toString() {<a name="line.657"></a>
+<span class="sourceLineNo">658</span>      return statusMsg;<a name="line.658"></a>
+<span class="sourceLineNo">659</span>    }<a name="line.659"></a>
+<span class="sourceLineNo">660</span>  }<a name="line.660"></a>
+<span class="sourceLineNo">661</span>}<a name="line.661"></a>