hbase-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From git-site-r...@apache.org
Subject [16/39] hbase-site git commit: Published site at 66469733ec9bffb236c143b858e5748182ad71b3.
Date Thu, 25 Oct 2018 14:54:53 GMT
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/5f45c993/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.CompletedProcedureCleaner.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.CompletedProcedureCleaner.html b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.CompletedProcedureCleaner.html
index 37bf508..0af8acd 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.CompletedProcedureCleaner.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.CompletedProcedureCleaner.html
@@ -660,1578 +660,1600 @@
 <span class="sourceLineNo">652</span>      if (!p.hasParent()) {<a name="line.652"></a>
 <span class="sourceLineNo">653</span>        sendProcedureLoadedNotification(p.getProcId());<a name="line.653"></a>
 <span class="sourceLineNo">654</span>      }<a name="line.654"></a>
-<span class="sourceLineNo">655</span>      scheduler.addBack(p);<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">658</span><a name="line.658"></a>
-<span class="sourceLineNo">659</span>  /**<a name="line.659"></a>
-<span class="sourceLineNo">660</span>   * Initialize the procedure executor, but do not start workers. We will start them later.<a name="line.660"></a>
-<span class="sourceLineNo">661</span>   * &lt;p/&gt;<a name="line.661"></a>
-<span class="sourceLineNo">662</span>   * It calls ProcedureStore.recoverLease() and ProcedureStore.load() to recover the lease, and<a name="line.662"></a>
-<span class="sourceLineNo">663</span>   * ensure a single executor, and start the procedure replay to resume and recover the previous<a name="line.663"></a>
-<span class="sourceLineNo">664</span>   * pending and in-progress procedures.<a name="line.664"></a>
-<span class="sourceLineNo">665</span>   * @param numThreads number of threads available for procedure execution.<a name="line.665"></a>
-<span class="sourceLineNo">666</span>   * @param abortOnCorruption true if you want to abort your service in case a corrupted procedure<a name="line.666"></a>
-<span class="sourceLineNo">667</span>   *          is found on replay. otherwise false.<a name="line.667"></a>
-<span class="sourceLineNo">668</span>   */<a name="line.668"></a>
-<span class="sourceLineNo">669</span>  public void init(int numThreads, boolean abortOnCorruption) throws IOException {<a name="line.669"></a>
-<span class="sourceLineNo">670</span>    // We have numThreads executor + one timer thread used for timing out<a name="line.670"></a>
-<span class="sourceLineNo">671</span>    // procedures and triggering periodic procedures.<a name="line.671"></a>
-<span class="sourceLineNo">672</span>    this.corePoolSize = numThreads;<a name="line.672"></a>
-<span class="sourceLineNo">673</span>    this.maxPoolSize = 10 * numThreads;<a name="line.673"></a>
-<span class="sourceLineNo">674</span>    LOG.info("Starting {} core workers (bigger of cpus/4 or 16) with max (burst) worker count={}",<a name="line.674"></a>
-<span class="sourceLineNo">675</span>        corePoolSize, maxPoolSize);<a name="line.675"></a>
-<span class="sourceLineNo">676</span><a name="line.676"></a>
-<span class="sourceLineNo">677</span>    this.threadGroup = new ThreadGroup("PEWorkerGroup");<a name="line.677"></a>
-<span class="sourceLineNo">678</span>    this.timeoutExecutor = new TimeoutExecutorThread&lt;&gt;(this, threadGroup);<a name="line.678"></a>
-<span class="sourceLineNo">679</span><a name="line.679"></a>
-<span class="sourceLineNo">680</span>    // Create the workers<a name="line.680"></a>
-<span class="sourceLineNo">681</span>    workerId.set(0);<a name="line.681"></a>
-<span class="sourceLineNo">682</span>    workerThreads = new CopyOnWriteArrayList&lt;&gt;();<a name="line.682"></a>
-<span class="sourceLineNo">683</span>    for (int i = 0; i &lt; corePoolSize; ++i) {<a name="line.683"></a>
-<span class="sourceLineNo">684</span>      workerThreads.add(new WorkerThread(threadGroup));<a name="line.684"></a>
-<span class="sourceLineNo">685</span>    }<a name="line.685"></a>
-<span class="sourceLineNo">686</span><a name="line.686"></a>
-<span class="sourceLineNo">687</span>    long st, et;<a name="line.687"></a>
-<span class="sourceLineNo">688</span><a name="line.688"></a>
-<span class="sourceLineNo">689</span>    // Acquire the store lease.<a name="line.689"></a>
-<span class="sourceLineNo">690</span>    st = System.nanoTime();<a name="line.690"></a>
-<span class="sourceLineNo">691</span>    store.recoverLease();<a name="line.691"></a>
-<span class="sourceLineNo">692</span>    et = System.nanoTime();<a name="line.692"></a>
-<span class="sourceLineNo">693</span>    LOG.info("Recovered {} lease in {}", store.getClass().getSimpleName(),<a name="line.693"></a>
-<span class="sourceLineNo">694</span>      StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.694"></a>
-<span class="sourceLineNo">695</span><a name="line.695"></a>
-<span class="sourceLineNo">696</span>    // start the procedure scheduler<a name="line.696"></a>
-<span class="sourceLineNo">697</span>    scheduler.start();<a name="line.697"></a>
+<span class="sourceLineNo">655</span>      // If the procedure holds the lock, put the procedure in front<a name="line.655"></a>
+<span class="sourceLineNo">656</span>      // If its parent holds the lock, put the procedure in front<a name="line.656"></a>
+<span class="sourceLineNo">657</span>      // TODO. Is that possible that its ancestor holds the lock?<a name="line.657"></a>
+<span class="sourceLineNo">658</span>      // For now, the deepest procedure hierarchy is:<a name="line.658"></a>
+<span class="sourceLineNo">659</span>      // ModifyTableProcedure -&gt; ReopenTableProcedure -&gt;<a name="line.659"></a>
+<span class="sourceLineNo">660</span>      // MoveTableProcedure -&gt; Unassign/AssignProcedure<a name="line.660"></a>
+<span class="sourceLineNo">661</span>      // But ModifyTableProcedure and ReopenTableProcedure won't hold the lock<a name="line.661"></a>
+<span class="sourceLineNo">662</span>      // So, check parent lock is enough(a tricky case is resovled by HBASE-21384).<a name="line.662"></a>
+<span class="sourceLineNo">663</span>      // If some one change or add new procedures making 'grandpa' procedure<a name="line.663"></a>
+<span class="sourceLineNo">664</span>      // holds the lock, but parent procedure don't hold the lock, there will<a name="line.664"></a>
+<span class="sourceLineNo">665</span>      // be a problem here. We have to check one procedure's ancestors.<a name="line.665"></a>
+<span class="sourceLineNo">666</span>      // And we need to change LockAndQueue.hasParentLock(Procedure&lt;?&gt; proc) method<a name="line.666"></a>
+<span class="sourceLineNo">667</span>      // to check all ancestors too.<a name="line.667"></a>
+<span class="sourceLineNo">668</span>      if (p.isLockedWhenLoading() || (p.hasParent() &amp;&amp; procedures<a name="line.668"></a>
+<span class="sourceLineNo">669</span>          .get(p.getParentProcId()).isLockedWhenLoading())) {<a name="line.669"></a>
+<span class="sourceLineNo">670</span>        scheduler.addFront(p, false);<a name="line.670"></a>
+<span class="sourceLineNo">671</span>      } else {<a name="line.671"></a>
+<span class="sourceLineNo">672</span>        // if it was not, it can wait.<a name="line.672"></a>
+<span class="sourceLineNo">673</span>        scheduler.addBack(p, false);<a name="line.673"></a>
+<span class="sourceLineNo">674</span>      }<a name="line.674"></a>
+<span class="sourceLineNo">675</span>    });<a name="line.675"></a>
+<span class="sourceLineNo">676</span>    // After all procedures put into the queue, signal the worker threads.<a name="line.676"></a>
+<span class="sourceLineNo">677</span>    // Otherwise, there is a race condition. See HBASE-21364.<a name="line.677"></a>
+<span class="sourceLineNo">678</span>    scheduler.signalAll();<a name="line.678"></a>
+<span class="sourceLineNo">679</span>  }<a name="line.679"></a>
+<span class="sourceLineNo">680</span><a name="line.680"></a>
+<span class="sourceLineNo">681</span>  /**<a name="line.681"></a>
+<span class="sourceLineNo">682</span>   * Initialize the procedure executor, but do not start workers. We will start them later.<a name="line.682"></a>
+<span class="sourceLineNo">683</span>   * &lt;p/&gt;<a name="line.683"></a>
+<span class="sourceLineNo">684</span>   * It calls ProcedureStore.recoverLease() and ProcedureStore.load() to recover the lease, and<a name="line.684"></a>
+<span class="sourceLineNo">685</span>   * ensure a single executor, and start the procedure replay to resume and recover the previous<a name="line.685"></a>
+<span class="sourceLineNo">686</span>   * pending and in-progress procedures.<a name="line.686"></a>
+<span class="sourceLineNo">687</span>   * @param numThreads number of threads available for procedure execution.<a name="line.687"></a>
+<span class="sourceLineNo">688</span>   * @param abortOnCorruption true if you want to abort your service in case a corrupted procedure<a name="line.688"></a>
+<span class="sourceLineNo">689</span>   *          is found on replay. otherwise false.<a name="line.689"></a>
+<span class="sourceLineNo">690</span>   */<a name="line.690"></a>
+<span class="sourceLineNo">691</span>  public void init(int numThreads, boolean abortOnCorruption) throws IOException {<a name="line.691"></a>
+<span class="sourceLineNo">692</span>    // We have numThreads executor + one timer thread used for timing out<a name="line.692"></a>
+<span class="sourceLineNo">693</span>    // procedures and triggering periodic procedures.<a name="line.693"></a>
+<span class="sourceLineNo">694</span>    this.corePoolSize = numThreads;<a name="line.694"></a>
+<span class="sourceLineNo">695</span>    this.maxPoolSize = 10 * numThreads;<a name="line.695"></a>
+<span class="sourceLineNo">696</span>    LOG.info("Starting {} core workers (bigger of cpus/4 or 16) with max (burst) worker count={}",<a name="line.696"></a>
+<span class="sourceLineNo">697</span>        corePoolSize, maxPoolSize);<a name="line.697"></a>
 <span class="sourceLineNo">698</span><a name="line.698"></a>
-<span class="sourceLineNo">699</span>    // TODO: Split in two steps.<a name="line.699"></a>
-<span class="sourceLineNo">700</span>    // TODO: Handle corrupted procedures (currently just a warn)<a name="line.700"></a>
-<span class="sourceLineNo">701</span>    // The first one will make sure that we have the latest id,<a name="line.701"></a>
-<span class="sourceLineNo">702</span>    // so we can start the threads and accept new procedures.<a name="line.702"></a>
-<span class="sourceLineNo">703</span>    // The second step will do the actual load of old procedures.<a name="line.703"></a>
-<span class="sourceLineNo">704</span>    st = System.nanoTime();<a name="line.704"></a>
-<span class="sourceLineNo">705</span>    load(abortOnCorruption);<a name="line.705"></a>
-<span class="sourceLineNo">706</span>    et = System.nanoTime();<a name="line.706"></a>
-<span class="sourceLineNo">707</span>    LOG.info("Loaded {} in {}", store.getClass().getSimpleName(),<a name="line.707"></a>
-<span class="sourceLineNo">708</span>      StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.708"></a>
-<span class="sourceLineNo">709</span>  }<a name="line.709"></a>
+<span class="sourceLineNo">699</span>    this.threadGroup = new ThreadGroup("PEWorkerGroup");<a name="line.699"></a>
+<span class="sourceLineNo">700</span>    this.timeoutExecutor = new TimeoutExecutorThread&lt;&gt;(this, threadGroup);<a name="line.700"></a>
+<span class="sourceLineNo">701</span><a name="line.701"></a>
+<span class="sourceLineNo">702</span>    // Create the workers<a name="line.702"></a>
+<span class="sourceLineNo">703</span>    workerId.set(0);<a name="line.703"></a>
+<span class="sourceLineNo">704</span>    workerThreads = new CopyOnWriteArrayList&lt;&gt;();<a name="line.704"></a>
+<span class="sourceLineNo">705</span>    for (int i = 0; i &lt; corePoolSize; ++i) {<a name="line.705"></a>
+<span class="sourceLineNo">706</span>      workerThreads.add(new WorkerThread(threadGroup));<a name="line.706"></a>
+<span class="sourceLineNo">707</span>    }<a name="line.707"></a>
+<span class="sourceLineNo">708</span><a name="line.708"></a>
+<span class="sourceLineNo">709</span>    long st, et;<a name="line.709"></a>
 <span class="sourceLineNo">710</span><a name="line.710"></a>
-<span class="sourceLineNo">711</span>  /**<a name="line.711"></a>
-<span class="sourceLineNo">712</span>   * Start the workers.<a name="line.712"></a>
-<span class="sourceLineNo">713</span>   */<a name="line.713"></a>
-<span class="sourceLineNo">714</span>  public void startWorkers() throws IOException {<a name="line.714"></a>
-<span class="sourceLineNo">715</span>    if (!running.compareAndSet(false, true)) {<a name="line.715"></a>
-<span class="sourceLineNo">716</span>      LOG.warn("Already running");<a name="line.716"></a>
-<span class="sourceLineNo">717</span>      return;<a name="line.717"></a>
-<span class="sourceLineNo">718</span>    }<a name="line.718"></a>
-<span class="sourceLineNo">719</span>    // Start the executors. Here we must have the lastProcId set.<a name="line.719"></a>
-<span class="sourceLineNo">720</span>    LOG.trace("Start workers {}", workerThreads.size());<a name="line.720"></a>
-<span class="sourceLineNo">721</span>    timeoutExecutor.start();<a name="line.721"></a>
-<span class="sourceLineNo">722</span>    for (WorkerThread worker: workerThreads) {<a name="line.722"></a>
-<span class="sourceLineNo">723</span>      worker.start();<a name="line.723"></a>
-<span class="sourceLineNo">724</span>    }<a name="line.724"></a>
-<span class="sourceLineNo">725</span><a name="line.725"></a>
-<span class="sourceLineNo">726</span>    // Internal chores<a name="line.726"></a>
-<span class="sourceLineNo">727</span>    timeoutExecutor.add(new WorkerMonitor());<a name="line.727"></a>
-<span class="sourceLineNo">728</span><a name="line.728"></a>
-<span class="sourceLineNo">729</span>    // Add completed cleaner chore<a name="line.729"></a>
-<span class="sourceLineNo">730</span>    addChore(new CompletedProcedureCleaner&lt;&gt;(conf, store, completed, nonceKeysToProcIdsMap));<a name="line.730"></a>
+<span class="sourceLineNo">711</span>    // Acquire the store lease.<a name="line.711"></a>
+<span class="sourceLineNo">712</span>    st = System.nanoTime();<a name="line.712"></a>
+<span class="sourceLineNo">713</span>    store.recoverLease();<a name="line.713"></a>
+<span class="sourceLineNo">714</span>    et = System.nanoTime();<a name="line.714"></a>
+<span class="sourceLineNo">715</span>    LOG.info("Recovered {} lease in {}", store.getClass().getSimpleName(),<a name="line.715"></a>
+<span class="sourceLineNo">716</span>      StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.716"></a>
+<span class="sourceLineNo">717</span><a name="line.717"></a>
+<span class="sourceLineNo">718</span>    // start the procedure scheduler<a name="line.718"></a>
+<span class="sourceLineNo">719</span>    scheduler.start();<a name="line.719"></a>
+<span class="sourceLineNo">720</span><a name="line.720"></a>
+<span class="sourceLineNo">721</span>    // TODO: Split in two steps.<a name="line.721"></a>
+<span class="sourceLineNo">722</span>    // TODO: Handle corrupted procedures (currently just a warn)<a name="line.722"></a>
+<span class="sourceLineNo">723</span>    // The first one will make sure that we have the latest id,<a name="line.723"></a>
+<span class="sourceLineNo">724</span>    // so we can start the threads and accept new procedures.<a name="line.724"></a>
+<span class="sourceLineNo">725</span>    // The second step will do the actual load of old procedures.<a name="line.725"></a>
+<span class="sourceLineNo">726</span>    st = System.nanoTime();<a name="line.726"></a>
+<span class="sourceLineNo">727</span>    load(abortOnCorruption);<a name="line.727"></a>
+<span class="sourceLineNo">728</span>    et = System.nanoTime();<a name="line.728"></a>
+<span class="sourceLineNo">729</span>    LOG.info("Loaded {} in {}", store.getClass().getSimpleName(),<a name="line.729"></a>
+<span class="sourceLineNo">730</span>      StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.730"></a>
 <span class="sourceLineNo">731</span>  }<a name="line.731"></a>
 <span class="sourceLineNo">732</span><a name="line.732"></a>
-<span class="sourceLineNo">733</span>  public void stop() {<a name="line.733"></a>
-<span class="sourceLineNo">734</span>    if (!running.getAndSet(false)) {<a name="line.734"></a>
-<span class="sourceLineNo">735</span>      return;<a name="line.735"></a>
-<span class="sourceLineNo">736</span>    }<a name="line.736"></a>
-<span class="sourceLineNo">737</span><a name="line.737"></a>
-<span class="sourceLineNo">738</span>    LOG.info("Stopping");<a name="line.738"></a>
-<span class="sourceLineNo">739</span>    scheduler.stop();<a name="line.739"></a>
-<span class="sourceLineNo">740</span>    timeoutExecutor.sendStopSignal();<a name="line.740"></a>
-<span class="sourceLineNo">741</span>  }<a name="line.741"></a>
-<span class="sourceLineNo">742</span><a name="line.742"></a>
-<span class="sourceLineNo">743</span>  @VisibleForTesting<a name="line.743"></a>
-<span class="sourceLineNo">744</span>  public void join() {<a name="line.744"></a>
-<span class="sourceLineNo">745</span>    assert !isRunning() : "expected not running";<a name="line.745"></a>
-<span class="sourceLineNo">746</span><a name="line.746"></a>
-<span class="sourceLineNo">747</span>    // stop the timeout executor<a name="line.747"></a>
-<span class="sourceLineNo">748</span>    timeoutExecutor.awaitTermination();<a name="line.748"></a>
-<span class="sourceLineNo">749</span><a name="line.749"></a>
-<span class="sourceLineNo">750</span>    // stop the worker threads<a name="line.750"></a>
-<span class="sourceLineNo">751</span>    for (WorkerThread worker: workerThreads) {<a name="line.751"></a>
-<span class="sourceLineNo">752</span>      worker.awaitTermination();<a name="line.752"></a>
-<span class="sourceLineNo">753</span>    }<a name="line.753"></a>
+<span class="sourceLineNo">733</span>  /**<a name="line.733"></a>
+<span class="sourceLineNo">734</span>   * Start the workers.<a name="line.734"></a>
+<span class="sourceLineNo">735</span>   */<a name="line.735"></a>
+<span class="sourceLineNo">736</span>  public void startWorkers() throws IOException {<a name="line.736"></a>
+<span class="sourceLineNo">737</span>    if (!running.compareAndSet(false, true)) {<a name="line.737"></a>
+<span class="sourceLineNo">738</span>      LOG.warn("Already running");<a name="line.738"></a>
+<span class="sourceLineNo">739</span>      return;<a name="line.739"></a>
+<span class="sourceLineNo">740</span>    }<a name="line.740"></a>
+<span class="sourceLineNo">741</span>    // Start the executors. Here we must have the lastProcId set.<a name="line.741"></a>
+<span class="sourceLineNo">742</span>    LOG.trace("Start workers {}", workerThreads.size());<a name="line.742"></a>
+<span class="sourceLineNo">743</span>    timeoutExecutor.start();<a name="line.743"></a>
+<span class="sourceLineNo">744</span>    for (WorkerThread worker: workerThreads) {<a name="line.744"></a>
+<span class="sourceLineNo">745</span>      worker.start();<a name="line.745"></a>
+<span class="sourceLineNo">746</span>    }<a name="line.746"></a>
+<span class="sourceLineNo">747</span><a name="line.747"></a>
+<span class="sourceLineNo">748</span>    // Internal chores<a name="line.748"></a>
+<span class="sourceLineNo">749</span>    timeoutExecutor.add(new WorkerMonitor());<a name="line.749"></a>
+<span class="sourceLineNo">750</span><a name="line.750"></a>
+<span class="sourceLineNo">751</span>    // Add completed cleaner chore<a name="line.751"></a>
+<span class="sourceLineNo">752</span>    addChore(new CompletedProcedureCleaner&lt;&gt;(conf, store, completed, nonceKeysToProcIdsMap));<a name="line.752"></a>
+<span class="sourceLineNo">753</span>  }<a name="line.753"></a>
 <span class="sourceLineNo">754</span><a name="line.754"></a>
-<span class="sourceLineNo">755</span>    // Destroy the Thread Group for the executors<a name="line.755"></a>
-<span class="sourceLineNo">756</span>    // TODO: Fix. #join is not place to destroy resources.<a name="line.756"></a>
-<span class="sourceLineNo">757</span>    try {<a name="line.757"></a>
-<span class="sourceLineNo">758</span>      threadGroup.destroy();<a name="line.758"></a>
-<span class="sourceLineNo">759</span>    } catch (IllegalThreadStateException e) {<a name="line.759"></a>
-<span class="sourceLineNo">760</span>      LOG.error("ThreadGroup {} contains running threads; {}: See STDOUT",<a name="line.760"></a>
-<span class="sourceLineNo">761</span>          this.threadGroup, e.getMessage());<a name="line.761"></a>
-<span class="sourceLineNo">762</span>      // This dumps list of threads on STDOUT.<a name="line.762"></a>
-<span class="sourceLineNo">763</span>      this.threadGroup.list();<a name="line.763"></a>
-<span class="sourceLineNo">764</span>    }<a name="line.764"></a>
-<span class="sourceLineNo">765</span><a name="line.765"></a>
-<span class="sourceLineNo">766</span>    // reset the in-memory state for testing<a name="line.766"></a>
-<span class="sourceLineNo">767</span>    completed.clear();<a name="line.767"></a>
-<span class="sourceLineNo">768</span>    rollbackStack.clear();<a name="line.768"></a>
-<span class="sourceLineNo">769</span>    procedures.clear();<a name="line.769"></a>
-<span class="sourceLineNo">770</span>    nonceKeysToProcIdsMap.clear();<a name="line.770"></a>
-<span class="sourceLineNo">771</span>    scheduler.clear();<a name="line.771"></a>
-<span class="sourceLineNo">772</span>    lastProcId.set(-1);<a name="line.772"></a>
-<span class="sourceLineNo">773</span>  }<a name="line.773"></a>
-<span class="sourceLineNo">774</span><a name="line.774"></a>
-<span class="sourceLineNo">775</span>  public void refreshConfiguration(final Configuration conf) {<a name="line.775"></a>
-<span class="sourceLineNo">776</span>    this.conf = conf;<a name="line.776"></a>
-<span class="sourceLineNo">777</span>    setKeepAliveTime(conf.getLong(WORKER_KEEP_ALIVE_TIME_CONF_KEY,<a name="line.777"></a>
-<span class="sourceLineNo">778</span>        DEFAULT_WORKER_KEEP_ALIVE_TIME), TimeUnit.MILLISECONDS);<a name="line.778"></a>
-<span class="sourceLineNo">779</span>  }<a name="line.779"></a>
-<span class="sourceLineNo">780</span><a name="line.780"></a>
-<span class="sourceLineNo">781</span>  // ==========================================================================<a name="line.781"></a>
-<span class="sourceLineNo">782</span>  //  Accessors<a name="line.782"></a>
-<span class="sourceLineNo">783</span>  // ==========================================================================<a name="line.783"></a>
-<span class="sourceLineNo">784</span>  public boolean isRunning() {<a name="line.784"></a>
-<span class="sourceLineNo">785</span>    return running.get();<a name="line.785"></a>
-<span class="sourceLineNo">786</span>  }<a name="line.786"></a>
+<span class="sourceLineNo">755</span>  public void stop() {<a name="line.755"></a>
+<span class="sourceLineNo">756</span>    if (!running.getAndSet(false)) {<a name="line.756"></a>
+<span class="sourceLineNo">757</span>      return;<a name="line.757"></a>
+<span class="sourceLineNo">758</span>    }<a name="line.758"></a>
+<span class="sourceLineNo">759</span><a name="line.759"></a>
+<span class="sourceLineNo">760</span>    LOG.info("Stopping");<a name="line.760"></a>
+<span class="sourceLineNo">761</span>    scheduler.stop();<a name="line.761"></a>
+<span class="sourceLineNo">762</span>    timeoutExecutor.sendStopSignal();<a name="line.762"></a>
+<span class="sourceLineNo">763</span>  }<a name="line.763"></a>
+<span class="sourceLineNo">764</span><a name="line.764"></a>
+<span class="sourceLineNo">765</span>  @VisibleForTesting<a name="line.765"></a>
+<span class="sourceLineNo">766</span>  public void join() {<a name="line.766"></a>
+<span class="sourceLineNo">767</span>    assert !isRunning() : "expected not running";<a name="line.767"></a>
+<span class="sourceLineNo">768</span><a name="line.768"></a>
+<span class="sourceLineNo">769</span>    // stop the timeout executor<a name="line.769"></a>
+<span class="sourceLineNo">770</span>    timeoutExecutor.awaitTermination();<a name="line.770"></a>
+<span class="sourceLineNo">771</span><a name="line.771"></a>
+<span class="sourceLineNo">772</span>    // stop the worker threads<a name="line.772"></a>
+<span class="sourceLineNo">773</span>    for (WorkerThread worker: workerThreads) {<a name="line.773"></a>
+<span class="sourceLineNo">774</span>      worker.awaitTermination();<a name="line.774"></a>
+<span class="sourceLineNo">775</span>    }<a name="line.775"></a>
+<span class="sourceLineNo">776</span><a name="line.776"></a>
+<span class="sourceLineNo">777</span>    // Destroy the Thread Group for the executors<a name="line.777"></a>
+<span class="sourceLineNo">778</span>    // TODO: Fix. #join is not place to destroy resources.<a name="line.778"></a>
+<span class="sourceLineNo">779</span>    try {<a name="line.779"></a>
+<span class="sourceLineNo">780</span>      threadGroup.destroy();<a name="line.780"></a>
+<span class="sourceLineNo">781</span>    } catch (IllegalThreadStateException e) {<a name="line.781"></a>
+<span class="sourceLineNo">782</span>      LOG.error("ThreadGroup {} contains running threads; {}: See STDOUT",<a name="line.782"></a>
+<span class="sourceLineNo">783</span>          this.threadGroup, e.getMessage());<a name="line.783"></a>
+<span class="sourceLineNo">784</span>      // This dumps list of threads on STDOUT.<a name="line.784"></a>
+<span class="sourceLineNo">785</span>      this.threadGroup.list();<a name="line.785"></a>
+<span class="sourceLineNo">786</span>    }<a name="line.786"></a>
 <span class="sourceLineNo">787</span><a name="line.787"></a>
-<span class="sourceLineNo">788</span>  /**<a name="line.788"></a>
-<span class="sourceLineNo">789</span>   * @return the current number of worker threads.<a name="line.789"></a>
-<span class="sourceLineNo">790</span>   */<a name="line.790"></a>
-<span class="sourceLineNo">791</span>  public int getWorkerThreadCount() {<a name="line.791"></a>
-<span class="sourceLineNo">792</span>    return workerThreads.size();<a name="line.792"></a>
-<span class="sourceLineNo">793</span>  }<a name="line.793"></a>
-<span class="sourceLineNo">794</span><a name="line.794"></a>
-<span class="sourceLineNo">795</span>  /**<a name="line.795"></a>
-<span class="sourceLineNo">796</span>   * @return the core pool size settings.<a name="line.796"></a>
-<span class="sourceLineNo">797</span>   */<a name="line.797"></a>
-<span class="sourceLineNo">798</span>  public int getCorePoolSize() {<a name="line.798"></a>
-<span class="sourceLineNo">799</span>    return corePoolSize;<a name="line.799"></a>
-<span class="sourceLineNo">800</span>  }<a name="line.800"></a>
-<span class="sourceLineNo">801</span><a name="line.801"></a>
-<span class="sourceLineNo">802</span>  public int getActiveExecutorCount() {<a name="line.802"></a>
-<span class="sourceLineNo">803</span>    return activeExecutorCount.get();<a name="line.803"></a>
-<span class="sourceLineNo">804</span>  }<a name="line.804"></a>
-<span class="sourceLineNo">805</span><a name="line.805"></a>
-<span class="sourceLineNo">806</span>  public TEnvironment getEnvironment() {<a name="line.806"></a>
-<span class="sourceLineNo">807</span>    return this.environment;<a name="line.807"></a>
+<span class="sourceLineNo">788</span>    // reset the in-memory state for testing<a name="line.788"></a>
+<span class="sourceLineNo">789</span>    completed.clear();<a name="line.789"></a>
+<span class="sourceLineNo">790</span>    rollbackStack.clear();<a name="line.790"></a>
+<span class="sourceLineNo">791</span>    procedures.clear();<a name="line.791"></a>
+<span class="sourceLineNo">792</span>    nonceKeysToProcIdsMap.clear();<a name="line.792"></a>
+<span class="sourceLineNo">793</span>    scheduler.clear();<a name="line.793"></a>
+<span class="sourceLineNo">794</span>    lastProcId.set(-1);<a name="line.794"></a>
+<span class="sourceLineNo">795</span>  }<a name="line.795"></a>
+<span class="sourceLineNo">796</span><a name="line.796"></a>
+<span class="sourceLineNo">797</span>  public void refreshConfiguration(final Configuration conf) {<a name="line.797"></a>
+<span class="sourceLineNo">798</span>    this.conf = conf;<a name="line.798"></a>
+<span class="sourceLineNo">799</span>    setKeepAliveTime(conf.getLong(WORKER_KEEP_ALIVE_TIME_CONF_KEY,<a name="line.799"></a>
+<span class="sourceLineNo">800</span>        DEFAULT_WORKER_KEEP_ALIVE_TIME), TimeUnit.MILLISECONDS);<a name="line.800"></a>
+<span class="sourceLineNo">801</span>  }<a name="line.801"></a>
+<span class="sourceLineNo">802</span><a name="line.802"></a>
+<span class="sourceLineNo">803</span>  // ==========================================================================<a name="line.803"></a>
+<span class="sourceLineNo">804</span>  //  Accessors<a name="line.804"></a>
+<span class="sourceLineNo">805</span>  // ==========================================================================<a name="line.805"></a>
+<span class="sourceLineNo">806</span>  public boolean isRunning() {<a name="line.806"></a>
+<span class="sourceLineNo">807</span>    return running.get();<a name="line.807"></a>
 <span class="sourceLineNo">808</span>  }<a name="line.808"></a>
 <span class="sourceLineNo">809</span><a name="line.809"></a>
-<span class="sourceLineNo">810</span>  public ProcedureStore getStore() {<a name="line.810"></a>
-<span class="sourceLineNo">811</span>    return this.store;<a name="line.811"></a>
-<span class="sourceLineNo">812</span>  }<a name="line.812"></a>
-<span class="sourceLineNo">813</span><a name="line.813"></a>
-<span class="sourceLineNo">814</span>  ProcedureScheduler getScheduler() {<a name="line.814"></a>
-<span class="sourceLineNo">815</span>    return scheduler;<a name="line.815"></a>
-<span class="sourceLineNo">816</span>  }<a name="line.816"></a>
-<span class="sourceLineNo">817</span><a name="line.817"></a>
-<span class="sourceLineNo">818</span>  public void setKeepAliveTime(final long keepAliveTime, final TimeUnit timeUnit) {<a name="line.818"></a>
-<span class="sourceLineNo">819</span>    this.keepAliveTime = timeUnit.toMillis(keepAliveTime);<a name="line.819"></a>
-<span class="sourceLineNo">820</span>    this.scheduler.signalAll();<a name="line.820"></a>
-<span class="sourceLineNo">821</span>  }<a name="line.821"></a>
-<span class="sourceLineNo">822</span><a name="line.822"></a>
-<span class="sourceLineNo">823</span>  public long getKeepAliveTime(final TimeUnit timeUnit) {<a name="line.823"></a>
-<span class="sourceLineNo">824</span>    return timeUnit.convert(keepAliveTime, TimeUnit.MILLISECONDS);<a name="line.824"></a>
-<span class="sourceLineNo">825</span>  }<a name="line.825"></a>
-<span class="sourceLineNo">826</span><a name="line.826"></a>
-<span class="sourceLineNo">827</span>  // ==========================================================================<a name="line.827"></a>
-<span class="sourceLineNo">828</span>  //  Submit/Remove Chores<a name="line.828"></a>
-<span class="sourceLineNo">829</span>  // ==========================================================================<a name="line.829"></a>
-<span class="sourceLineNo">830</span><a name="line.830"></a>
-<span class="sourceLineNo">831</span>  /**<a name="line.831"></a>
-<span class="sourceLineNo">832</span>   * Add a chore procedure to the executor<a name="line.832"></a>
-<span class="sourceLineNo">833</span>   * @param chore the chore to add<a name="line.833"></a>
-<span class="sourceLineNo">834</span>   */<a name="line.834"></a>
-<span class="sourceLineNo">835</span>  public void addChore(ProcedureInMemoryChore&lt;TEnvironment&gt; chore) {<a name="line.835"></a>
-<span class="sourceLineNo">836</span>    chore.setState(ProcedureState.WAITING_TIMEOUT);<a name="line.836"></a>
-<span class="sourceLineNo">837</span>    timeoutExecutor.add(chore);<a name="line.837"></a>
+<span class="sourceLineNo">810</span>  /**<a name="line.810"></a>
+<span class="sourceLineNo">811</span>   * @return the current number of worker threads.<a name="line.811"></a>
+<span class="sourceLineNo">812</span>   */<a name="line.812"></a>
+<span class="sourceLineNo">813</span>  public int getWorkerThreadCount() {<a name="line.813"></a>
+<span class="sourceLineNo">814</span>    return workerThreads.size();<a name="line.814"></a>
+<span class="sourceLineNo">815</span>  }<a name="line.815"></a>
+<span class="sourceLineNo">816</span><a name="line.816"></a>
+<span class="sourceLineNo">817</span>  /**<a name="line.817"></a>
+<span class="sourceLineNo">818</span>   * @return the core pool size settings.<a name="line.818"></a>
+<span class="sourceLineNo">819</span>   */<a name="line.819"></a>
+<span class="sourceLineNo">820</span>  public int getCorePoolSize() {<a name="line.820"></a>
+<span class="sourceLineNo">821</span>    return corePoolSize;<a name="line.821"></a>
+<span class="sourceLineNo">822</span>  }<a name="line.822"></a>
+<span class="sourceLineNo">823</span><a name="line.823"></a>
+<span class="sourceLineNo">824</span>  public int getActiveExecutorCount() {<a name="line.824"></a>
+<span class="sourceLineNo">825</span>    return activeExecutorCount.get();<a name="line.825"></a>
+<span class="sourceLineNo">826</span>  }<a name="line.826"></a>
+<span class="sourceLineNo">827</span><a name="line.827"></a>
+<span class="sourceLineNo">828</span>  public TEnvironment getEnvironment() {<a name="line.828"></a>
+<span class="sourceLineNo">829</span>    return this.environment;<a name="line.829"></a>
+<span class="sourceLineNo">830</span>  }<a name="line.830"></a>
+<span class="sourceLineNo">831</span><a name="line.831"></a>
+<span class="sourceLineNo">832</span>  public ProcedureStore getStore() {<a name="line.832"></a>
+<span class="sourceLineNo">833</span>    return this.store;<a name="line.833"></a>
+<span class="sourceLineNo">834</span>  }<a name="line.834"></a>
+<span class="sourceLineNo">835</span><a name="line.835"></a>
+<span class="sourceLineNo">836</span>  ProcedureScheduler getScheduler() {<a name="line.836"></a>
+<span class="sourceLineNo">837</span>    return scheduler;<a name="line.837"></a>
 <span class="sourceLineNo">838</span>  }<a name="line.838"></a>
 <span class="sourceLineNo">839</span><a name="line.839"></a>
-<span class="sourceLineNo">840</span>  /**<a name="line.840"></a>
-<span class="sourceLineNo">841</span>   * Remove a chore procedure from the executor<a name="line.841"></a>
-<span class="sourceLineNo">842</span>   * @param chore the chore to remove<a name="line.842"></a>
-<span class="sourceLineNo">843</span>   * @return whether the chore is removed, or it will be removed later<a name="line.843"></a>
-<span class="sourceLineNo">844</span>   */<a name="line.844"></a>
-<span class="sourceLineNo">845</span>  public boolean removeChore(ProcedureInMemoryChore&lt;TEnvironment&gt; chore) {<a name="line.845"></a>
-<span class="sourceLineNo">846</span>    chore.setState(ProcedureState.SUCCESS);<a name="line.846"></a>
-<span class="sourceLineNo">847</span>    return timeoutExecutor.remove(chore);<a name="line.847"></a>
-<span class="sourceLineNo">848</span>  }<a name="line.848"></a>
-<span class="sourceLineNo">849</span><a name="line.849"></a>
-<span class="sourceLineNo">850</span>  // ==========================================================================<a name="line.850"></a>
-<span class="sourceLineNo">851</span>  //  Nonce Procedure helpers<a name="line.851"></a>
-<span class="sourceLineNo">852</span>  // ==========================================================================<a name="line.852"></a>
+<span class="sourceLineNo">840</span>  public void setKeepAliveTime(final long keepAliveTime, final TimeUnit timeUnit) {<a name="line.840"></a>
+<span class="sourceLineNo">841</span>    this.keepAliveTime = timeUnit.toMillis(keepAliveTime);<a name="line.841"></a>
+<span class="sourceLineNo">842</span>    this.scheduler.signalAll();<a name="line.842"></a>
+<span class="sourceLineNo">843</span>  }<a name="line.843"></a>
+<span class="sourceLineNo">844</span><a name="line.844"></a>
+<span class="sourceLineNo">845</span>  public long getKeepAliveTime(final TimeUnit timeUnit) {<a name="line.845"></a>
+<span class="sourceLineNo">846</span>    return timeUnit.convert(keepAliveTime, TimeUnit.MILLISECONDS);<a name="line.846"></a>
+<span class="sourceLineNo">847</span>  }<a name="line.847"></a>
+<span class="sourceLineNo">848</span><a name="line.848"></a>
+<span class="sourceLineNo">849</span>  // ==========================================================================<a name="line.849"></a>
+<span class="sourceLineNo">850</span>  //  Submit/Remove Chores<a name="line.850"></a>
+<span class="sourceLineNo">851</span>  // ==========================================================================<a name="line.851"></a>
+<span class="sourceLineNo">852</span><a name="line.852"></a>
 <span class="sourceLineNo">853</span>  /**<a name="line.853"></a>
-<span class="sourceLineNo">854</span>   * Create a NoneKey from the specified nonceGroup and nonce.<a name="line.854"></a>
-<span class="sourceLineNo">855</span>   * @param nonceGroup<a name="line.855"></a>
-<span class="sourceLineNo">856</span>   * @param nonce<a name="line.856"></a>
-<span class="sourceLineNo">857</span>   * @return the generated NonceKey<a name="line.857"></a>
-<span class="sourceLineNo">858</span>   */<a name="line.858"></a>
-<span class="sourceLineNo">859</span>  public NonceKey createNonceKey(final long nonceGroup, final long nonce) {<a name="line.859"></a>
-<span class="sourceLineNo">860</span>    return (nonce == HConstants.NO_NONCE) ? null : new NonceKey(nonceGroup, nonce);<a name="line.860"></a>
-<span class="sourceLineNo">861</span>  }<a name="line.861"></a>
-<span class="sourceLineNo">862</span><a name="line.862"></a>
-<span class="sourceLineNo">863</span>  /**<a name="line.863"></a>
-<span class="sourceLineNo">864</span>   * Register a nonce for a procedure that is going to be submitted.<a name="line.864"></a>
-<span class="sourceLineNo">865</span>   * A procId will be reserved and on submitProcedure(),<a name="line.865"></a>
-<span class="sourceLineNo">866</span>   * the procedure with the specified nonce will take the reserved ProcId.<a name="line.866"></a>
-<span class="sourceLineNo">867</span>   * If someone already reserved the nonce, this method will return the procId reserved,<a name="line.867"></a>
-<span class="sourceLineNo">868</span>   * otherwise an invalid procId will be returned. and the caller should procede<a name="line.868"></a>
-<span class="sourceLineNo">869</span>   * and submit the procedure.<a name="line.869"></a>
-<span class="sourceLineNo">870</span>   *<a name="line.870"></a>
-<span class="sourceLineNo">871</span>   * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.871"></a>
-<span class="sourceLineNo">872</span>   * @return the procId associated with the nonce, if any otherwise an invalid procId.<a name="line.872"></a>
-<span class="sourceLineNo">873</span>   */<a name="line.873"></a>
-<span class="sourceLineNo">874</span>  public long registerNonce(final NonceKey nonceKey) {<a name="line.874"></a>
-<span class="sourceLineNo">875</span>    if (nonceKey == null) return -1;<a name="line.875"></a>
-<span class="sourceLineNo">876</span><a name="line.876"></a>
-<span class="sourceLineNo">877</span>    // check if we have already a Reserved ID for the nonce<a name="line.877"></a>
-<span class="sourceLineNo">878</span>    Long oldProcId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.878"></a>
-<span class="sourceLineNo">879</span>    if (oldProcId == null) {<a name="line.879"></a>
-<span class="sourceLineNo">880</span>      // reserve a new Procedure ID, this will be associated with the nonce<a name="line.880"></a>
-<span class="sourceLineNo">881</span>      // and the procedure submitted with the specified nonce will use this ID.<a name="line.881"></a>
-<span class="sourceLineNo">882</span>      final long newProcId = nextProcId();<a name="line.882"></a>
-<span class="sourceLineNo">883</span>      oldProcId = nonceKeysToProcIdsMap.putIfAbsent(nonceKey, newProcId);<a name="line.883"></a>
-<span class="sourceLineNo">884</span>      if (oldProcId == null) return -1;<a name="line.884"></a>
-<span class="sourceLineNo">885</span>    }<a name="line.885"></a>
-<span class="sourceLineNo">886</span><a name="line.886"></a>
-<span class="sourceLineNo">887</span>    // we found a registered nonce, but the procedure may not have been submitted yet.<a name="line.887"></a>
-<span class="sourceLineNo">888</span>    // since the client expect the procedure to be submitted, spin here until it is.<a name="line.888"></a>
-<span class="sourceLineNo">889</span>    final boolean traceEnabled = LOG.isTraceEnabled();<a name="line.889"></a>
-<span class="sourceLineNo">890</span>    while (isRunning() &amp;&amp;<a name="line.890"></a>
-<span class="sourceLineNo">891</span>           !(procedures.containsKey(oldProcId) || completed.containsKey(oldProcId)) &amp;&amp;<a name="line.891"></a>
-<span class="sourceLineNo">892</span>           nonceKeysToProcIdsMap.containsKey(nonceKey)) {<a name="line.892"></a>
-<span class="sourceLineNo">893</span>      if (traceEnabled) {<a name="line.893"></a>
-<span class="sourceLineNo">894</span>        LOG.trace("Waiting for pid=" + oldProcId.longValue() + " to be submitted");<a name="line.894"></a>
-<span class="sourceLineNo">895</span>      }<a name="line.895"></a>
-<span class="sourceLineNo">896</span>      Threads.sleep(100);<a name="line.896"></a>
-<span class="sourceLineNo">897</span>    }<a name="line.897"></a>
-<span class="sourceLineNo">898</span>    return oldProcId.longValue();<a name="line.898"></a>
-<span class="sourceLineNo">899</span>  }<a name="line.899"></a>
-<span class="sourceLineNo">900</span><a name="line.900"></a>
-<span class="sourceLineNo">901</span>  /**<a name="line.901"></a>
-<span class="sourceLineNo">902</span>   * Remove the NonceKey if the procedure was not submitted to the executor.<a name="line.902"></a>
-<span class="sourceLineNo">903</span>   * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.903"></a>
-<span class="sourceLineNo">904</span>   */<a name="line.904"></a>
-<span class="sourceLineNo">905</span>  public void unregisterNonceIfProcedureWasNotSubmitted(final NonceKey nonceKey) {<a name="line.905"></a>
-<span class="sourceLineNo">906</span>    if (nonceKey == null) return;<a name="line.906"></a>
-<span class="sourceLineNo">907</span><a name="line.907"></a>
-<span class="sourceLineNo">908</span>    final Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.908"></a>
-<span class="sourceLineNo">909</span>    if (procId == null) return;<a name="line.909"></a>
-<span class="sourceLineNo">910</span><a name="line.910"></a>
-<span class="sourceLineNo">911</span>    // if the procedure was not submitted, remove the nonce<a name="line.911"></a>
-<span class="sourceLineNo">912</span>    if (!(procedures.containsKey(procId) || completed.containsKey(procId))) {<a name="line.912"></a>
-<span class="sourceLineNo">913</span>      nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.913"></a>
-<span class="sourceLineNo">914</span>    }<a name="line.914"></a>
-<span class="sourceLineNo">915</span>  }<a name="line.915"></a>
-<span class="sourceLineNo">916</span><a name="line.916"></a>
-<span class="sourceLineNo">917</span>  public static class FailedProcedure&lt;TEnvironment&gt; extends Procedure&lt;TEnvironment&gt; {<a name="line.917"></a>
-<span class="sourceLineNo">918</span>    private String procName;<a name="line.918"></a>
-<span class="sourceLineNo">919</span><a name="line.919"></a>
-<span class="sourceLineNo">920</span>    public FailedProcedure() {<a name="line.920"></a>
-<span class="sourceLineNo">921</span>    }<a name="line.921"></a>
+<span class="sourceLineNo">854</span>   * Add a chore procedure to the executor<a name="line.854"></a>
+<span class="sourceLineNo">855</span>   * @param chore the chore to add<a name="line.855"></a>
+<span class="sourceLineNo">856</span>   */<a name="line.856"></a>
+<span class="sourceLineNo">857</span>  public void addChore(ProcedureInMemoryChore&lt;TEnvironment&gt; chore) {<a name="line.857"></a>
+<span class="sourceLineNo">858</span>    chore.setState(ProcedureState.WAITING_TIMEOUT);<a name="line.858"></a>
+<span class="sourceLineNo">859</span>    timeoutExecutor.add(chore);<a name="line.859"></a>
+<span class="sourceLineNo">860</span>  }<a name="line.860"></a>
+<span class="sourceLineNo">861</span><a name="line.861"></a>
+<span class="sourceLineNo">862</span>  /**<a name="line.862"></a>
+<span class="sourceLineNo">863</span>   * Remove a chore procedure from the executor<a name="line.863"></a>
+<span class="sourceLineNo">864</span>   * @param chore the chore to remove<a name="line.864"></a>
+<span class="sourceLineNo">865</span>   * @return whether the chore is removed, or it will be removed later<a name="line.865"></a>
+<span class="sourceLineNo">866</span>   */<a name="line.866"></a>
+<span class="sourceLineNo">867</span>  public boolean removeChore(ProcedureInMemoryChore&lt;TEnvironment&gt; chore) {<a name="line.867"></a>
+<span class="sourceLineNo">868</span>    chore.setState(ProcedureState.SUCCESS);<a name="line.868"></a>
+<span class="sourceLineNo">869</span>    return timeoutExecutor.remove(chore);<a name="line.869"></a>
+<span class="sourceLineNo">870</span>  }<a name="line.870"></a>
+<span class="sourceLineNo">871</span><a name="line.871"></a>
+<span class="sourceLineNo">872</span>  // ==========================================================================<a name="line.872"></a>
+<span class="sourceLineNo">873</span>  //  Nonce Procedure helpers<a name="line.873"></a>
+<span class="sourceLineNo">874</span>  // ==========================================================================<a name="line.874"></a>
+<span class="sourceLineNo">875</span>  /**<a name="line.875"></a>
+<span class="sourceLineNo">876</span>   * Create a NoneKey from the specified nonceGroup and nonce.<a name="line.876"></a>
+<span class="sourceLineNo">877</span>   * @param nonceGroup<a name="line.877"></a>
+<span class="sourceLineNo">878</span>   * @param nonce<a name="line.878"></a>
+<span class="sourceLineNo">879</span>   * @return the generated NonceKey<a name="line.879"></a>
+<span class="sourceLineNo">880</span>   */<a name="line.880"></a>
+<span class="sourceLineNo">881</span>  public NonceKey createNonceKey(final long nonceGroup, final long nonce) {<a name="line.881"></a>
+<span class="sourceLineNo">882</span>    return (nonce == HConstants.NO_NONCE) ? null : new NonceKey(nonceGroup, nonce);<a name="line.882"></a>
+<span class="sourceLineNo">883</span>  }<a name="line.883"></a>
+<span class="sourceLineNo">884</span><a name="line.884"></a>
+<span class="sourceLineNo">885</span>  /**<a name="line.885"></a>
+<span class="sourceLineNo">886</span>   * Register a nonce for a procedure that is going to be submitted.<a name="line.886"></a>
+<span class="sourceLineNo">887</span>   * A procId will be reserved and on submitProcedure(),<a name="line.887"></a>
+<span class="sourceLineNo">888</span>   * the procedure with the specified nonce will take the reserved ProcId.<a name="line.888"></a>
+<span class="sourceLineNo">889</span>   * If someone already reserved the nonce, this method will return the procId reserved,<a name="line.889"></a>
+<span class="sourceLineNo">890</span>   * otherwise an invalid procId will be returned. and the caller should procede<a name="line.890"></a>
+<span class="sourceLineNo">891</span>   * and submit the procedure.<a name="line.891"></a>
+<span class="sourceLineNo">892</span>   *<a name="line.892"></a>
+<span class="sourceLineNo">893</span>   * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.893"></a>
+<span class="sourceLineNo">894</span>   * @return the procId associated with the nonce, if any otherwise an invalid procId.<a name="line.894"></a>
+<span class="sourceLineNo">895</span>   */<a name="line.895"></a>
+<span class="sourceLineNo">896</span>  public long registerNonce(final NonceKey nonceKey) {<a name="line.896"></a>
+<span class="sourceLineNo">897</span>    if (nonceKey == null) return -1;<a name="line.897"></a>
+<span class="sourceLineNo">898</span><a name="line.898"></a>
+<span class="sourceLineNo">899</span>    // check if we have already a Reserved ID for the nonce<a name="line.899"></a>
+<span class="sourceLineNo">900</span>    Long oldProcId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.900"></a>
+<span class="sourceLineNo">901</span>    if (oldProcId == null) {<a name="line.901"></a>
+<span class="sourceLineNo">902</span>      // reserve a new Procedure ID, this will be associated with the nonce<a name="line.902"></a>
+<span class="sourceLineNo">903</span>      // and the procedure submitted with the specified nonce will use this ID.<a name="line.903"></a>
+<span class="sourceLineNo">904</span>      final long newProcId = nextProcId();<a name="line.904"></a>
+<span class="sourceLineNo">905</span>      oldProcId = nonceKeysToProcIdsMap.putIfAbsent(nonceKey, newProcId);<a name="line.905"></a>
+<span class="sourceLineNo">906</span>      if (oldProcId == null) return -1;<a name="line.906"></a>
+<span class="sourceLineNo">907</span>    }<a name="line.907"></a>
+<span class="sourceLineNo">908</span><a name="line.908"></a>
+<span class="sourceLineNo">909</span>    // we found a registered nonce, but the procedure may not have been submitted yet.<a name="line.909"></a>
+<span class="sourceLineNo">910</span>    // since the client expect the procedure to be submitted, spin here until it is.<a name="line.910"></a>
+<span class="sourceLineNo">911</span>    final boolean traceEnabled = LOG.isTraceEnabled();<a name="line.911"></a>
+<span class="sourceLineNo">912</span>    while (isRunning() &amp;&amp;<a name="line.912"></a>
+<span class="sourceLineNo">913</span>           !(procedures.containsKey(oldProcId) || completed.containsKey(oldProcId)) &amp;&amp;<a name="line.913"></a>
+<span class="sourceLineNo">914</span>           nonceKeysToProcIdsMap.containsKey(nonceKey)) {<a name="line.914"></a>
+<span class="sourceLineNo">915</span>      if (traceEnabled) {<a name="line.915"></a>
+<span class="sourceLineNo">916</span>        LOG.trace("Waiting for pid=" + oldProcId.longValue() + " to be submitted");<a name="line.916"></a>
+<span class="sourceLineNo">917</span>      }<a name="line.917"></a>
+<span class="sourceLineNo">918</span>      Threads.sleep(100);<a name="line.918"></a>
+<span class="sourceLineNo">919</span>    }<a name="line.919"></a>
+<span class="sourceLineNo">920</span>    return oldProcId.longValue();<a name="line.920"></a>
+<span class="sourceLineNo">921</span>  }<a name="line.921"></a>
 <span class="sourceLineNo">922</span><a name="line.922"></a>
-<span class="sourceLineNo">923</span>    public FailedProcedure(long procId, String procName, User owner,<a name="line.923"></a>
-<span class="sourceLineNo">924</span>        NonceKey nonceKey, IOException exception) {<a name="line.924"></a>
-<span class="sourceLineNo">925</span>      this.procName = procName;<a name="line.925"></a>
-<span class="sourceLineNo">926</span>      setProcId(procId);<a name="line.926"></a>
-<span class="sourceLineNo">927</span>      setState(ProcedureState.ROLLEDBACK);<a name="line.927"></a>
-<span class="sourceLineNo">928</span>      setOwner(owner);<a name="line.928"></a>
-<span class="sourceLineNo">929</span>      setNonceKey(nonceKey);<a name="line.929"></a>
-<span class="sourceLineNo">930</span>      long currentTime = EnvironmentEdgeManager.currentTime();<a name="line.930"></a>
-<span class="sourceLineNo">931</span>      setSubmittedTime(currentTime);<a name="line.931"></a>
-<span class="sourceLineNo">932</span>      setLastUpdate(currentTime);<a name="line.932"></a>
-<span class="sourceLineNo">933</span>      setFailure(Objects.toString(exception.getMessage(), ""), exception);<a name="line.933"></a>
-<span class="sourceLineNo">934</span>    }<a name="line.934"></a>
-<span class="sourceLineNo">935</span><a name="line.935"></a>
-<span class="sourceLineNo">936</span>    @Override<a name="line.936"></a>
-<span class="sourceLineNo">937</span>    public String getProcName() {<a name="line.937"></a>
-<span class="sourceLineNo">938</span>      return procName;<a name="line.938"></a>
-<span class="sourceLineNo">939</span>    }<a name="line.939"></a>
-<span class="sourceLineNo">940</span><a name="line.940"></a>
-<span class="sourceLineNo">941</span>    @Override<a name="line.941"></a>
-<span class="sourceLineNo">942</span>    protected Procedure&lt;TEnvironment&gt;[] execute(TEnvironment env)<a name="line.942"></a>
-<span class="sourceLineNo">943</span>        throws ProcedureYieldException, ProcedureSuspendedException,<a name="line.943"></a>
-<span class="sourceLineNo">944</span>        InterruptedException {<a name="line.944"></a>
-<span class="sourceLineNo">945</span>      throw new UnsupportedOperationException();<a name="line.945"></a>
-<span class="sourceLineNo">946</span>    }<a name="line.946"></a>
-<span class="sourceLineNo">947</span><a name="line.947"></a>
-<span class="sourceLineNo">948</span>    @Override<a name="line.948"></a>
-<span class="sourceLineNo">949</span>    protected void rollback(TEnvironment env)<a name="line.949"></a>
-<span class="sourceLineNo">950</span>        throws IOException, InterruptedException {<a name="line.950"></a>
-<span class="sourceLineNo">951</span>      throw new UnsupportedOperationException();<a name="line.951"></a>
-<span class="sourceLineNo">952</span>    }<a name="line.952"></a>
-<span class="sourceLineNo">953</span><a name="line.953"></a>
-<span class="sourceLineNo">954</span>    @Override<a name="line.954"></a>
-<span class="sourceLineNo">955</span>    protected boolean abort(TEnvironment env) {<a name="line.955"></a>
-<span class="sourceLineNo">956</span>      throw new UnsupportedOperationException();<a name="line.956"></a>
-<span class="sourceLineNo">957</span>    }<a name="line.957"></a>
-<span class="sourceLineNo">958</span><a name="line.958"></a>
-<span class="sourceLineNo">959</span>    @Override<a name="line.959"></a>
-<span class="sourceLineNo">960</span>    protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.960"></a>
-<span class="sourceLineNo">961</span>        throws IOException {<a name="line.961"></a>
-<span class="sourceLineNo">962</span>    }<a name="line.962"></a>
-<span class="sourceLineNo">963</span><a name="line.963"></a>
-<span class="sourceLineNo">964</span>    @Override<a name="line.964"></a>
-<span class="sourceLineNo">965</span>    protected void deserializeStateData(ProcedureStateSerializer serializer)<a name="line.965"></a>
-<span class="sourceLineNo">966</span>        throws IOException {<a name="line.966"></a>
-<span class="sourceLineNo">967</span>    }<a name="line.967"></a>
-<span class="sourceLineNo">968</span>  }<a name="line.968"></a>
+<span class="sourceLineNo">923</span>  /**<a name="line.923"></a>
+<span class="sourceLineNo">924</span>   * Remove the NonceKey if the procedure was not submitted to the executor.<a name="line.924"></a>
+<span class="sourceLineNo">925</span>   * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.925"></a>
+<span class="sourceLineNo">926</span>   */<a name="line.926"></a>
+<span class="sourceLineNo">927</span>  public void unregisterNonceIfProcedureWasNotSubmitted(final NonceKey nonceKey) {<a name="line.927"></a>
+<span class="sourceLineNo">928</span>    if (nonceKey == null) return;<a name="line.928"></a>
+<span class="sourceLineNo">929</span><a name="line.929"></a>
+<span class="sourceLineNo">930</span>    final Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.930"></a>
+<span class="sourceLineNo">931</span>    if (procId == null) return;<a name="line.931"></a>
+<span class="sourceLineNo">932</span><a name="line.932"></a>
+<span class="sourceLineNo">933</span>    // if the procedure was not submitted, remove the nonce<a name="line.933"></a>
+<span class="sourceLineNo">934</span>    if (!(procedures.containsKey(procId) || completed.containsKey(procId))) {<a name="line.934"></a>
+<span class="sourceLineNo">935</span>      nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.935"></a>
+<span class="sourceLineNo">936</span>    }<a name="line.936"></a>
+<span class="sourceLineNo">937</span>  }<a name="line.937"></a>
+<span class="sourceLineNo">938</span><a name="line.938"></a>
+<span class="sourceLineNo">939</span>  public static class FailedProcedure&lt;TEnvironment&gt; extends Procedure&lt;TEnvironment&gt; {<a name="line.939"></a>
+<span class="sourceLineNo">940</span>    private String procName;<a name="line.940"></a>
+<span class="sourceLineNo">941</span><a name="line.941"></a>
+<span class="sourceLineNo">942</span>    public FailedProcedure() {<a name="line.942"></a>
+<span class="sourceLineNo">943</span>    }<a name="line.943"></a>
+<span class="sourceLineNo">944</span><a name="line.944"></a>
+<span class="sourceLineNo">945</span>    public FailedProcedure(long procId, String procName, User owner,<a name="line.945"></a>
+<span class="sourceLineNo">946</span>        NonceKey nonceKey, IOException exception) {<a name="line.946"></a>
+<span class="sourceLineNo">947</span>      this.procName = procName;<a name="line.947"></a>
+<span class="sourceLineNo">948</span>      setProcId(procId);<a name="line.948"></a>
+<span class="sourceLineNo">949</span>      setState(ProcedureState.ROLLEDBACK);<a name="line.949"></a>
+<span class="sourceLineNo">950</span>      setOwner(owner);<a name="line.950"></a>
+<span class="sourceLineNo">951</span>      setNonceKey(nonceKey);<a name="line.951"></a>
+<span class="sourceLineNo">952</span>      long currentTime = EnvironmentEdgeManager.currentTime();<a name="line.952"></a>
+<span class="sourceLineNo">953</span>      setSubmittedTime(currentTime);<a name="line.953"></a>
+<span class="sourceLineNo">954</span>      setLastUpdate(currentTime);<a name="line.954"></a>
+<span class="sourceLineNo">955</span>      setFailure(Objects.toString(exception.getMessage(), ""), exception);<a name="line.955"></a>
+<span class="sourceLineNo">956</span>    }<a name="line.956"></a>
+<span class="sourceLineNo">957</span><a name="line.957"></a>
+<span class="sourceLineNo">958</span>    @Override<a name="line.958"></a>
+<span class="sourceLineNo">959</span>    public String getProcName() {<a name="line.959"></a>
+<span class="sourceLineNo">960</span>      return procName;<a name="line.960"></a>
+<span class="sourceLineNo">961</span>    }<a name="line.961"></a>
+<span class="sourceLineNo">962</span><a name="line.962"></a>
+<span class="sourceLineNo">963</span>    @Override<a name="line.963"></a>
+<span class="sourceLineNo">964</span>    protected Procedure&lt;TEnvironment&gt;[] execute(TEnvironment env)<a name="line.964"></a>
+<span class="sourceLineNo">965</span>        throws ProcedureYieldException, ProcedureSuspendedException,<a name="line.965"></a>
+<span class="sourceLineNo">966</span>        InterruptedException {<a name="line.966"></a>
+<span class="sourceLineNo">967</span>      throw new UnsupportedOperationException();<a name="line.967"></a>
+<span class="sourceLineNo">968</span>    }<a name="line.968"></a>
 <span class="sourceLineNo">969</span><a name="line.969"></a>
-<span class="sourceLineNo">970</span>  /**<a name="line.970"></a>
-<span class="sourceLineNo">971</span>   * If the failure failed before submitting it, we may want to give back the<a name="line.971"></a>
-<span class="sourceLineNo">972</span>   * same error to the requests with the same nonceKey.<a name="line.972"></a>
-<span class="sourceLineNo">973</span>   *<a name="line.973"></a>
-<span class="sourceLineNo">974</span>   * @param nonceKey A unique identifier for this operation from the client or process<a name="line.974"></a>
-<span class="sourceLineNo">975</span>   * @param procName name of the procedure, used to inform the user<a name="line.975"></a>
-<span class="sourceLineNo">976</span>   * @param procOwner name of the owner of the procedure, used to inform the user<a name="line.976"></a>
-<span class="sourceLineNo">977</span>   * @param exception the failure to report to the user<a name="line.977"></a>
-<span class="sourceLineNo">978</span>   */<a name="line.978"></a>
-<span class="sourceLineNo">979</span>  public void setFailureResultForNonce(NonceKey nonceKey, String procName, User procOwner,<a name="line.979"></a>
-<span class="sourceLineNo">980</span>      IOException exception) {<a name="line.980"></a>
-<span class="sourceLineNo">981</span>    if (nonceKey == null) {<a name="line.981"></a>
-<span class="sourceLineNo">982</span>      return;<a name="line.982"></a>
-<span class="sourceLineNo">983</span>    }<a name="line.983"></a>
-<span class="sourceLineNo">984</span><a name="line.984"></a>
-<span class="sourceLineNo">985</span>    Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.985"></a>
-<span class="sourceLineNo">986</span>    if (procId == null || completed.containsKey(procId)) {<a name="line.986"></a>
-<span class="sourceLineNo">987</span>      return;<a name="line.987"></a>
-<span class="sourceLineNo">988</span>    }<a name="line.988"></a>
-<span class="sourceLineNo">989</span><a name="line.989"></a>
-<span class="sourceLineNo">990</span>    Procedure&lt;TEnvironment&gt; proc =<a name="line.990"></a>
-<span class="sourceLineNo">991</span>      new FailedProcedure&lt;&gt;(procId.longValue(), procName, procOwner, nonceKey, exception);<a name="line.991"></a>
-<span class="sourceLineNo">992</span><a name="line.992"></a>
-<span class="sourceLineNo">993</span>    completed.putIfAbsent(procId, new CompletedProcedureRetainer&lt;&gt;(proc));<a name="line.993"></a>
-<span class="sourceLineNo">994</span>  }<a name="line.994"></a>
-<span class="sourceLineNo">995</span><a name="line.995"></a>
-<span class="sourceLineNo">996</span>  // ==========================================================================<a name="line.996"></a>
-<span class="sourceLineNo">997</span>  //  Submit/Abort Procedure<a name="line.997"></a>
-<span class="sourceLineNo">998</span>  // ==========================================================================<a name="line.998"></a>
-<span class="sourceLineNo">999</span>  /**<a name="line.999"></a>
-<span class="sourceLineNo">1000</span>   * Add a new root-procedure to the executor.<a name="line.1000"></a>
-<span class="sourceLineNo">1001</span>   * @param proc the new procedure to execute.<a name="line.1001"></a>
-<span class="sourceLineNo">1002</span>   * @return the procedure id, that can be used to monitor the operation<a name="line.1002"></a>
-<span class="sourceLineNo">1003</span>   */<a name="line.1003"></a>
-<span class="sourceLineNo">1004</span>  public long submitProcedure(Procedure&lt;TEnvironment&gt; proc) {<a name="line.1004"></a>
-<span class="sourceLineNo">1005</span>    return submitProcedure(proc, null);<a name="line.1005"></a>
-<span class="sourceLineNo">1006</span>  }<a name="line.1006"></a>
-<span class="sourceLineNo">1007</span><a name="line.1007"></a>
-<span class="sourceLineNo">1008</span>  /**<a name="line.1008"></a>
-<span class="sourceLineNo">1009</span>   * Bypass a procedure. If the procedure is set to bypass, all the logic in<a name="line.1009"></a>
-<span class="sourceLineNo">1010</span>   * execute/rollback will be ignored and it will return success, whatever.<a name="line.1010"></a>
-<span class="sourceLineNo">1011</span>   * It is used to recover buggy stuck procedures, releasing the lock resources<a name="line.1011"></a>
-<span class="sourceLineNo">1012</span>   * and letting other procedures run. Bypassing one procedure (and its ancestors will<a name="line.1012"></a>
-<span class="sourceLineNo">1013</span>   * be bypassed automatically) may leave the cluster in a middle state, e.g. region<a name="line.1013"></a>
-<span class="sourceLineNo">1014</span>   * not assigned, or some hdfs files left behind. After getting rid of those stuck procedures,<a name="line.1014"></a>
-<span class="sourceLineNo">1015</span>   * the operators may have to do some clean up on hdfs or schedule some assign procedures<a name="line.1015"></a>
-<span class="sourceLineNo">1016</span>   * to let region online. DO AT YOUR OWN RISK.<a name="line.1016"></a>
-<span class="sourceLineNo">1017</span>   * &lt;p&gt;<a name="line.1017"></a>
-<span class="sourceLineNo">1018</span>   * A procedure can be bypassed only if<a name="line.1018"></a>
-<span class="sourceLineNo">1019</span>   * 1. The procedure is in state of RUNNABLE, WAITING, WAITING_TIMEOUT<a name="line.1019"></a>
-<span class="sourceLineNo">1020</span>   * or it is a root procedure without any child.<a name="line.1020"></a>
-<span class="sourceLineNo">1021</span>   * 2. No other worker thread is executing it<a name="line.1021"></a>
-<span class="sourceLineNo">1022</span>   * 3. No child procedure has been submitted<a name="line.1022"></a>
-<span class="sourceLineNo">1023</span>   *<a name="line.1023"></a>
-<span class="sourceLineNo">1024</span>   * &lt;p&gt;<a name="line.1024"></a>
-<span class="sourceLineNo">1025</span>   * If all the requirements are meet, the procedure and its ancestors will be<a name="line.1025"></a>
-<span class="sourceLineNo">1026</span>   * bypassed and persisted to WAL.<a name="line.1026"></a>
-<span class="sourceLineNo">1027</span>   *<a name="line.1027"></a>
-<span class="sourceLineNo">1028</span>   * &lt;p&gt;<a name="line.1028"></a>
-<span class="sourceLineNo">1029</span>   * If the procedure is in WAITING state, will set it to RUNNABLE add it to run queue.<a name="line.1029"></a>
-<span class="sourceLineNo">1030</span>   * TODO: What about WAITING_TIMEOUT?<a name="line.1030"></a>
-<span class="sourceLineNo">1031</span>   * @param pids the procedure id<a name="line.1031"></a>
-<span class="sourceLineNo">1032</span>   * @param lockWait time to wait lock<a name="line.1032"></a>
-<span class="sourceLineNo">1033</span>   * @param force if force set to true, we will bypass the procedure even if it is executing.<a name="line.1033"></a>
-<span class="sourceLineNo">1034</span>   *              This is for procedures which can't break out during executing(due to bug, mostly)<a name="line.1034"></a>
-<span class="sourceLineNo">1035</span>   *              In this case, bypassing the procedure is not enough, since it is already stuck<a name="line.1035"></a>
-<span class="sourceLineNo">1036</span>   *              there. We need to restart the master after bypassing, and letting the problematic<a name="line.1036"></a>
-<span class="sourceLineNo">1037</span>   *              procedure to execute wth bypass=true, so in that condition, the procedure can be<a name="line.1037"></a>
-<span class="sourceLineNo">1038</span>   *              successfully bypassed.<a name="line.1038"></a>
-<span class="sourceLineNo">1039</span>   * @param recursive We will do an expensive search for children of each pid. EXPENSIVE!<a name="line.1039"></a>
-<span class="sourceLineNo">1040</span>   * @return true if bypass success<a name="line.1040"></a>
-<span class="sourceLineNo">1041</span>   * @throws IOException IOException<a name="line.1041"></a>
-<span class="sourceLineNo">1042</span>   */<a name="line.1042"></a>
-<span class="sourceLineNo">1043</span>  public List&lt;Boolean&gt; bypassProcedure(List&lt;Long&gt; pids, long lockWait, boolean force,<a name="line.1043"></a>
-<span class="sourceLineNo">1044</span>      boolean recursive)<a name="line.1044"></a>
-<span class="sourceLineNo">1045</span>      throws IOException {<a name="line.1045"></a>
-<span class="sourceLineNo">1046</span>    List&lt;Boolean&gt; result = new ArrayList&lt;Boolean&gt;(pids.size());<a name="line.1046"></a>
-<span class="sourceLineNo">1047</span>    for(long pid: pids) {<a name="line.1047"></a>
-<span class="sourceLineNo">1048</span>      result.add(bypassProcedure(pid, lockWait, force, recursive));<a name="line.1048"></a>
-<span class="sourceLineNo">1049</span>    }<a name="line.1049"></a>
-<span class="sourceLineNo">1050</span>    return result;<a name="line.1050"></a>
-<span class="sourceLineNo">1051</span>  }<a name="line.1051"></a>
-<span class="sourceLineNo">1052</span><a name="line.1052"></a>
-<span class="sourceLineNo">1053</span>  boolean bypassProcedure(long pid, long lockWait, boolean override, boolean recursive)<a name="line.1053"></a>
-<span class="sourceLineNo">1054</span>      throws IOException {<a name="line.1054"></a>
-<span class="sourceLineNo">1055</span>    Preconditions.checkArgument(lockWait &gt; 0, "lockWait should be positive");<a name="line.1055"></a>
-<span class="sourceLineNo">1056</span>    final Procedure&lt;TEnvironment&gt; procedure = getProcedure(pid);<a name="line.1056"></a>
-<span class="sourceLineNo">1057</span>    if (procedure == null) {<a name="line.1057"></a>
-<span class="sourceLineNo">1058</span>      LOG.debug("Procedure pid={} does not exist, skipping bypass", pid);<a name="line.1058"></a>
-<span class="sourceLineNo">1059</span>      return false;<a name="line.1059"></a>
-<span class="sourceLineNo">1060</span>    }<a name="line.1060"></a>
-<span class="sourceLineNo">1061</span><a name="line.1061"></a>
-<span class="sourceLineNo">1062</span>    LOG.debug("Begin bypass {} with lockWait={}, override={}, recursive={}",<a name="line.1062"></a>
-<span class="sourceLineNo">1063</span>        procedure, lockWait, override, recursive);<a name="line.1063"></a>
-<span class="sourceLineNo">1064</span>    IdLock.Entry lockEntry = procExecutionLock.tryLockEntry(procedure.getProcId(), lockWait);<a name="line.1064"></a>
-<span class="sourceLineNo">1065</span>    if (lockEntry == null &amp;&amp; !override) {<a name="line.1065"></a>
-<span class="sourceLineNo">1066</span>      LOG.debug("Waited {} ms, but {} is still running, skipping bypass with force={}",<a name="line.1066"></a>
-<span class="sourceLineNo">1067</span>          lockWait, procedure, override);<a name="line.1067"></a>
-<span class="sourceLineNo">1068</span>      return false;<a name="line.1068"></a>
-<span class="sourceLineNo">1069</span>    } else if (lockEntry == null) {<a name="line.1069"></a>
-<span class="sourceLineNo">1070</span>      LOG.debug("Waited {} ms, but {} is still running, begin bypass with force={}",<a name="line.1070"></a>
-<span class="sourceLineNo">1071</span>          lockWait, procedure, override);<a name="line.1071"></a>
-<span class="sourceLineNo">1072</span>    }<a name="line.1072"></a>
-<span class="sourceLineNo">1073</span>    try {<a name="line.1073"></a>
-<span class="sourceLineNo">1074</span>      // check whether the procedure is already finished<a name="line.1074"></a>
-<span class="sourceLineNo">1075</span>      if (procedure.isFinished()) {<a name="line.1075"></a>
-<span class="sourceLineNo">1076</span>        LOG.debug("{} is already finished, skipping bypass", procedure);<a name="line.1076"></a>
-<span class="sourceLineNo">1077</span>        return false;<a name="line.1077"></a>
-<span class="sourceLineNo">1078</span>      }<a name="line.1078"></a>
-<span class="sourceLineNo">1079</span><a name="line.1079"></a>
-<span class="sourceLineNo">1080</span>      if (procedure.hasChildren()) {<a name="line.1080"></a>
-<span class="sourceLineNo">1081</span>        if (recursive) {<a name="line.1081"></a>
-<span class="sourceLineNo">1082</span>          // EXPENSIVE. Checks each live procedure of which there could be many!!!<a name="line.1082"></a>
-<span class="sourceLineNo">1083</span>          // Is there another way to get children of a procedure?<a name="line.1083"></a>
-<span class="sourceLineNo">1084</span>          LOG.info("Recursive bypass on children of pid={}", procedure.getProcId());<a name="line.1084"></a>
-<span class="sourceLineNo">1085</span>          this.procedures.forEachValue(1 /*Single-threaded*/,<a name="line.1085"></a>
-<span class="sourceLineNo">1086</span>            // Transformer<a name="line.1086"></a>
-<span class="sourceLineNo">1087</span>            v -&gt; v.getParentProcId() == procedure.getProcId()? v: null,<a name="line.1087"></a>
-<span class="sourceLineNo">1088</span>            // Consumer<a name="line.1088"></a>
-<span class="sourceLineNo">1089</span>            v -&gt; {<a name="line.1089"></a>
-<span class="sourceLineNo">1090</span>              try {<a name="line.1090"></a>
-<span class="sourceLineNo">1091</span>                bypassProcedure(v.getProcId(), lockWait, override, recursive);<a name="line.1091"></a>
-<span class="sourceLineNo">1092</span>              } catch (IOException e) {<a name="line.1092"></a>
-<span class="sourceLineNo">1093</span>                LOG.warn("Recursive bypass of pid={}", v.getProcId(), e);<a name="line.1093"></a>
-<span class="sourceLineNo">1094</span>              }<a name="line.1094"></a>
-<span class="sourceLineNo">1095</span>            });<a name="line.1095"></a>
-<span class="sourceLineNo">1096</span>        } else {<a name="line.1096"></a>
-<span class="sourceLineNo">1097</span>          LOG.debug("{} has children, skipping bypass", procedure);<a name="line.1097"></a>
-<span class="sourceLineNo">1098</span>          return false;<a name="line.1098"></a>
-<span class="sourceLineNo">1099</span>        }<a name="line.1099"></a>
+<span class="sourceLineNo">970</span>    @Override<a name="line.970"></a>
+<span class="sourceLineNo">971</span>    protected void rollback(TEnvironment env)<a name="line.971"></a>
+<span class="sourceLineNo">972</span>        throws IOException, InterruptedException {<a name="line.972"></a>
+<span class="sourceLineNo">973</span>      throw new UnsupportedOperationException();<a name="line.973"></a>
+<span class="sourceLineNo">974</span>    }<a name="line.974"></a>
+<span class="sourceLineNo">975</span><a name="line.975"></a>
+<span class="sourceLineNo">976</span>    @Override<a name="line.976"></a>
+<span class="sourceLineNo">977</span>    protected boolean abort(TEnvironment env) {<a name="line.977"></a>
+<span class="sourceLineNo">978</span>      throw new UnsupportedOperationException();<a name="line.978"></a>
+<span class="sourceLineNo">979</span>    }<a name="line.979"></a>
+<span class="sourceLineNo">980</span><a name="line.980"></a>
+<span class="sourceLineNo">981</span>    @Override<a name="line.981"></a>
+<span class="sourceLineNo">982</span>    protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.982"></a>
+<span class="sourceLineNo">983</span>        throws IOException {<a name="line.983"></a>
+<span class="sourceLineNo">984</span>    }<a name="line.984"></a>
+<span class="sourceLineNo">985</span><a name="line.985"></a>
+<span class="sourceLineNo">986</span>    @Override<a name="line.986"></a>
+<span class="sourceLineNo">987</span>    protected void deserializeStateData(ProcedureStateSerializer serializer)<a name="line.987"></a>
+<span class="sourceLineNo">988</span>        throws IOException {<a name="line.988"></a>
+<span class="sourceLineNo">989</span>    }<a name="line.989"></a>
+<span class="sourceLineNo">990</span>  }<a name="line.990"></a>
+<span class="sourceLineNo">991</span><a name="line.991"></a>
+<span class="sourceLineNo">992</span>  /**<a name="line.992"></a>
+<span class="sourceLineNo">993</span>   * If the failure failed before submitting it, we may want to give back the<a name="line.993"></a>
+<span class="sourceLineNo">994</span>   * same error to the requests with the same nonceKey.<a name="line.994"></a>
+<span class="sourceLineNo">995</span>   *<a name="line.995"></a>
+<span class="sourceLineNo">996</span>   * @param nonceKey A unique identifier for this operation from the client or process<a name="line.996"></a>
+<span class="sourceLineNo">997</span>   * @param procName name of the procedure, used to inform the user<a name="line.997"></a>
+<span class="sourceLineNo">998</span>   * @param procOwner name of the owner of the procedure, used to inform the user<a name="line.998"></a>
+<span class="sourceLineNo">999</span>   * @param exception the failure to report to the user<a name="line.999"></a>
+<span class="sourceLineNo">1000</span>   */<a name="line.1000"></a>
+<span class="sourceLineNo">1001</span>  public void setFailureResultForNonce(NonceKey nonceKey, String procName, User procOwner,<a name="line.1001"></a>
+<span class="sourceLineNo">1002</span>      IOException exception) {<a name="line.1002"></a>
+<span class="sourceLineNo">1003</span>    if (nonceKey == null) {<a name="line.1003"></a>
+<span class="sourceLineNo">1004</span>      return;<a name="line.1004"></a>
+<span class="sourceLineNo">1005</span>    }<a name="line.1005"></a>
+<span class="sourceLineNo">1006</span><a name="line.1006"></a>
+<span class="sourceLineNo">1007</span>    Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.1007"></a>
+<span class="sourceLineNo">1008</span>    if (procId == null || completed.containsKey(procId)) {<a name="line.1008"></a>
+<span class="sourceLineNo">1009</span>      return;<a name="line.1009"></a>
+<span class="sourceLineNo">1010</span>    }<a name="line.1010"></a>
+<span class="sourceLineNo">1011</span><a name="line.1011"></a>
+<span class="sourceLineNo">1012</span>    Procedure&lt;TEnvironment&gt; proc =<a name="line.1012"></a>
+<span class="sourceLineNo">1013</span>      new FailedProcedure&lt;&gt;(procId.longValue(), procName, procOwner, nonceKey, exception);<a name="line.1013"></a>
+<span class="sourceLineNo">1014</span><a name="line.1014"></a>
+<span class="sourceLineNo">1015</span>    completed.putIfAbsent(procId, new CompletedProcedureRetainer&lt;&gt;(proc));<a name="line.1015"></a>
+<span class="sourceLineNo">1016</span>  }<a name="line.1016"></a>
+<span class="sourceLineNo">1017</span><a name="line.1017"></a>
+<span class="sourceLineNo">1018</span>  // ==========================================================================<a name="line.1018"></a>
+<span class="sourceLineNo">1019</span>  //  Submit/Abort Procedure<a name="line.1019"></a>
+<span class="sourceLineNo">1020</span>  // ==========================================================================<a name="line.1020"></a>
+<span class="sourceLineNo">1021</span>  /**<a name="line.1021"></a>
+<span class="sourceLineNo">1022</span>   * Add a new root-procedure to the executor.<a name="line.1022"></a>
+<span class="sourceLineNo">1023</span>   * @param proc the new procedure to execute.<a name="line.1023"></a>
+<span class="sourceLineNo">1024</span>   * @return the procedure id, that can be used to monitor the operation<a name="line.1024"></a>
+<span class="sourceLineNo">1025</span>   */<a name="line.1025"></a>
+<span class="sourceLineNo">1026</span>  public long submitProcedure(Procedure&lt;TEnvironment&gt; proc) {<a name="line.1026"></a>
+<span class="sourceLineNo">1027</span>    return submitProcedure(proc, null);<a name="line.1027"></a>
+<span class="sourceLineNo">1028</span>  }<a name="line.1028"></a>
+<span class="sourceLineNo">1029</span><a name="line.1029"></a>
+<span class="sourceLineNo">1030</span>  /**<a name="line.1030"></a>
+<span class="sourceLineNo">1031</span>   * Bypass a procedure. If the procedure is set to bypass, all the logic in<a name="line.1031"></a>
+<span class="sourceLineNo">1032</span>   * execute/rollback will be ignored and it will return success, whatever.<a name="line.1032"></a>
+<span class="sourceLineNo">1033</span>   * It is used to recover buggy stuck procedures, releasing the lock resources<a name="line.1033"></a>
+<span class="sourceLineNo">1034</span>   * and letting other procedures run. Bypassing one procedure (and its ancestors will<a name="line.1034"></a>
+<span class="sourceLineNo">1035</span>   * be bypassed automatically) may leave the cluster in a middle state, e.g. region<a name="line.1035"></a>
+<span class="sourceLineNo">1036</span>   * not assigned, or some hdfs files left behind. After getting rid of those stuck procedures,<a name="line.1036"></a>
+<span class="sourceLineNo">1037</span>   * the operators may have to do some clean up on hdfs or schedule some assign procedures<a name="line.1037"></a>
+<span class="sourceLineNo">1038</span>   * to let region online. DO AT YOUR OWN RISK.<a name="line.1038"></a>
+<span class="sourceLineNo">1039</span>   * &lt;p&gt;<a name="line.1039"></a>
+<span class="sourceLineNo">1040</span>   * A procedure can be bypassed only if<a name="line.1040"></a>
+<span class="sourceLineNo">1041</span>   * 1. The procedure is in state of RUNNABLE, WAITING, WAITING_TIMEOUT<a name="line.1041"></a>
+<span class="sourceLineNo">1042</span>   * or it is a root procedure without any child.<a name="line.1042"></a>
+<span class="sourceLineNo">1043</span>   * 2. No other worker thread is executing it<a name="line.1043"></a>
+<span class="sourceLineNo">1044</span>   * 3. No child procedure has been submitted<a name="line.1044"></a>
+<span class="sourceLineNo">1045</span>   *<a name="line.1045"></a>
+<span class="sourceLineNo">1046</span>   * &lt;p&gt;<a name="line.1046"></a>
+<span class="sourceLineNo">1047</span>   * If all the requirements are meet, the procedure and its ancestors will be<a name="line.1047"></a>
+<span class="sourceLineNo">1048</span>   * bypassed and persisted to WAL.<a name="line.1048"></a>
+<span class="sourceLineNo">1049</span>   *<a name="line.1049"></a>
+<span class="sourceLineNo">1050</span>   * &lt;p&gt;<a name="line.1050"></a>
+<span class="sourceLineNo">1051</span>   * If the procedure is in WAITING state, will set it to RUNNABLE add it to run queue.<a name="line.1051"></a>
+<span class="sourceLineNo">1052</span>   * TODO: What about WAITING_TIMEOUT?<a name="line.1052"></a>
+<span class="sourceLineNo">1053</span>   * @param pids the procedure id<a name="line.1053"></a>
+<span class="sourceLineNo">1054</span>   * @param lockWait time to wait lock<a name="line.1054"></a>
+<span class="sourceLineNo">1055</span>   * @param force if force set to true, we will bypass the procedure even if it is executing.<a name="line.1055"></a>
+<span class="sourceLineNo">1056</span>   *              This is for procedures which can't break out during executing(due to bug, mostly)<a name="line.1056"></a>
+<span class="sourceLineNo">1057</span>   *              In this case, bypassing the procedure is not enough, since it is already stuck<a name="line.1057"></a>
+<span class="sourceLineNo">1058</span>   *              there. We need to restart the master after bypassing, and letting the problematic<a name="line.1058"></a>
+<span class="sourceLineNo">1059</span>   *              procedure to execute wth bypass=true, so in that condition, the procedure can be<a name="line.1059"></a>
+<span class="sourceLineNo">1060</span>   *              successfully bypassed.<a name="line.1060"></a>
+<span class="sourceLineNo">1061</span>   * @param recursive We will do an expensive search for children of each pid. EXPENSIVE!<a name="line.1061"></a>
+<span class="sourceLineNo">1062</span>   * @return true if bypass success<a name="line.1062"></a>
+<span class="sourceLineNo">1063</span>   * @throws IOException IOException<a name="line.1063"></a>
+<span class="sourceLineNo">1064</span>   */<a name="line.1064"></a>
+<span class="sourceLineNo">1065</span>  public List&lt;Boolean&gt; bypassProcedure(List&lt;Long&gt; pids, long lockWait, boolean force,<a name="line.1065"></a>
+<span class="sourceLineNo">1066</span>      boolean recursive)<a name="line.1066"></a>
+<span class="sourceLineNo">1067</span>      throws IOException {<a name="line.1067"></a>
+<span class="sourceLineNo">1068</span>    List&lt;Boolean&gt; result = new ArrayList&lt;Boolean&gt;(pids.size());<a name="line.1068"></a>
+<span class="sourceLineNo">1069</span>    for(long pid: pids) {<a name="line.1069"></a>
+<span class="sourceLineNo">1070</span>      result.add(bypassProcedure(pid, lockWait, force, recursive));<a name="line.1070"></a>
+<span class="sourceLineNo">1071</span>    }<a name="line.1071"></a>
+<span class="sourceLineNo">1072</span>    return result;<a name="line.1072"></a>
+<span class="sourceLineNo">1073</span>  }<a name="line.1073"></a>
+<span class="sourceLineNo">1074</span><a name="line.1074"></a>
+<span class="sourceLineNo">1075</span>  boolean bypassProcedure(long pid, long lockWait, boolean override, boolean recursive)<a name="line.1075"></a>
+<span class="sourceLineNo">1076</span>      throws IOException {<a name="line.1076"></a>
+<span class="sourceLineNo">1077</span>    Preconditions.checkArgument(lockWait &gt; 0, "lockWait should be positive");<a name="line.1077"></a>
+<span class="sourceLineNo">1078</span>    final Procedure&lt;TEnvironment&gt; procedure = getProcedure(pid);<a name="line.1078"></a>
+<span class="sourceLineNo">1079</span>    if (procedure == null) {<a name="line.1079"></a>
+<span class="sourceLineNo">1080</span>      LOG.debug("Procedure pid={} does not exist, skipping bypass", pid);<a name="line.1080"></a>
+<span class="sourceLineNo">1081</span>      return false;<a name="line.1081"></a>
+<span class="sourceLineNo">1082</span>    }<a name="line.1082"></a>
+<span class="sourceLineNo">1083</span><a name="line.1083"></a>
+<span class="sourceLineNo">1084</span>    LOG.debug("Begin bypass {} with lockWait={}, override={}, recursive={}",<a name="line.1084"></a>
+<span class="sourceLineNo">1085</span>        procedure, lockWait, override, recursive);<a name="line.1085"></a>
+<span class="sourceLineNo">1086</span>    IdLock.Entry lockEntry = procExecutionLock.tryLockEntry(procedure.getProcId(), lockWait);<a name="line.1086"></a>
+<span class="sourceLineNo">1087</span>    if (lockEntry == null &amp;&amp; !override) {<a name="line.1087"></a>
+<span class="sourceLineNo">1088</span>      LOG.debug("Waited {} ms, but {} is still running, skipping bypass with force={}",<a name="line.1088"></a>
+<span class="sourceLineNo">1089</span>          lockWait, procedure, override);<a name="line.1089"></a>
+<span class="sourceLineNo">1090</span>      return false;<a name="line.1090"></a>
+<span class="sourceLineNo">1091</span>    } else if (lockEntry == null) {<a name="line.1091"></a>
+<span class="sourceLineNo">1092</span>      LOG.debug("Waited {} ms, but {} is still running, begin bypass with force={}",<a name="line.1092"></a>
+<span class="sourceLineNo">1093</span>          lockWait, procedure, override);<a name="line.1093"></a>
+<span class="sourceLineNo">1094</span>    }<a name="line.1094"></a>
+<span class="sourceLineNo">1095</span>    try {<a name="line.1095"></a>
+<span class="sourceLineNo">1096</span>      // check whether the procedure is already finished<a name="line.1096"></a>
+<span class="sourceLineNo">1097</span>      if (procedure.isFinished()) {<a name="line.1097"></a>
+<span class="sourceLineNo">1098</span>        LOG.debug("{} is already finished, skipping bypass", procedure);<a name="line.1098"></a>
+<span class="sourceLineNo">1099</span>        return false;<a name="line.1099"></a>
 <span class="sourceLineNo">1100</span>      }<a name="line.1100"></a>
 <span class="sourceLineNo">1101</span><a name="line.1101"></a>
-<span class="sourceLineNo">1102</span>      // If the procedure has no parent or no child, we are safe to bypass it in whatever state<a name="line.1102"></a>
-<span class="sourceLineNo">1103</span>      if (procedure.hasParent() &amp;&amp; procedure.getState() != ProcedureState.RUNNABLE<a name="line.1103"></a>
-<span class="sourceLineNo">1104</span>          &amp;&amp; procedure.getState() != ProcedureState.WAITING<a name="line.1104"></a>
-<span class="sourceLineNo">1105</span>          &amp;&amp; procedure.getState() != ProcedureState.WAITING_TIMEOUT) {<a name="line.1105"></a>
-<span class="sourceLineNo">1106</span>        LOG.debug("Bypassing procedures in RUNNABLE, WAITING and WAITING_TIMEOUT states "<a name="line.1106"></a>
-<span class="sourceLineNo">1107</span>                + "(with no parent), {}",<a name="line.1107"></a>
-<span class="sourceLineNo">1108</span>            procedure);<a name="line.1108"></a>
-<span class="sourceLineNo">1109</span>        // Question: how is the bypass done here?<a name="line.1109"></a>
-<span class="sourceLineNo">1110</span>        return false;<a name="line.1110"></a>
-<span class="sourceLineNo">1111</span>      }<a name="line.1111"></a>
-<span class="sourceLineNo">1112</span><a name="line.1112"></a>
-<span class="sourceLineNo">1113</span>      // Now, the procedure is not finished, and no one can execute it since we take the lock now<a name="line.1113"></a>
-<span class="sourceLineNo">1114</span>      // And we can be sure that its ancestor is not running too, since th

<TRUNCATED>

Mime
View raw message