http://git-wip-us.apache.org/repos/asf/hbase-site/blob/6b8b907f/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.KeepAliveWorkerThread.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.KeepAliveWorkerThread.html b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.KeepAliveWorkerThread.html index 688bd4e..37bf508 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.KeepAliveWorkerThread.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.KeepAliveWorkerThread.html @@ -589,1651 +589,1649 @@ <span class="sourceLineNo">581</span> LOG.debug("Loading {}", proc);<a name="line.581"></a> <span class="sourceLineNo">582</span><a name="line.582"></a> <span class="sourceLineNo">583</span> Long rootProcId = getRootProcedureId(proc);<a name="line.583"></a> -<span class="sourceLineNo">584</span> if (rootProcId == null) {<a name="line.584"></a> -<span class="sourceLineNo">585</span> // The 'proc' was ready to run but the root procedure was rolledback?<a name="line.585"></a> -<span class="sourceLineNo">586</span> scheduler.addBack(proc);<a name="line.586"></a> -<span class="sourceLineNo">587</span> continue;<a name="line.587"></a> -<span class="sourceLineNo">588</span> }<a name="line.588"></a> -<span class="sourceLineNo">589</span><a name="line.589"></a> -<span class="sourceLineNo">590</span> if (proc.hasParent()) {<a name="line.590"></a> -<span class="sourceLineNo">591</span> Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());<a name="line.591"></a> -<span class="sourceLineNo">592</span> // corrupted procedures are handled later at step 3<a name="line.592"></a> -<span class="sourceLineNo">593</span> if (parent != null && !proc.isFinished()) {<a name="line.593"></a> -<span class="sourceLineNo">594</span> parent.incChildrenLatch();<a name="line.594"></a> -<span class="sourceLineNo">595</span> }<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> RootProcedureState<TEnvironment> procStack = rollbackStack.get(rootProcId);<a name="line.598"></a> -<span class="sourceLineNo">599</span> procStack.loadStack(proc);<a name="line.599"></a> -<span class="sourceLineNo">600</span><a name="line.600"></a> -<span class="sourceLineNo">601</span> proc.setRootProcId(rootProcId);<a name="line.601"></a> -<span class="sourceLineNo">602</span> switch (proc.getState()) {<a name="line.602"></a> -<span class="sourceLineNo">603</span> case RUNNABLE:<a name="line.603"></a> -<span class="sourceLineNo">604</span> runnableList.add(proc);<a name="line.604"></a> -<span class="sourceLineNo">605</span> break;<a name="line.605"></a> -<span class="sourceLineNo">606</span> case WAITING:<a name="line.606"></a> -<span class="sourceLineNo">607</span> waitingList.add(proc);<a name="line.607"></a> -<span class="sourceLineNo">608</span> break;<a name="line.608"></a> -<span class="sourceLineNo">609</span> case WAITING_TIMEOUT:<a name="line.609"></a> -<span class="sourceLineNo">610</span> waitingTimeoutList.add(proc);<a name="line.610"></a> -<span class="sourceLineNo">611</span> break;<a name="line.611"></a> -<span class="sourceLineNo">612</span> case FAILED:<a name="line.612"></a> -<span class="sourceLineNo">613</span> failedList.add(proc);<a name="line.613"></a> -<span class="sourceLineNo">614</span> break;<a name="line.614"></a> -<span class="sourceLineNo">615</span> case ROLLEDBACK:<a name="line.615"></a> -<span class="sourceLineNo">616</span> case INITIALIZING:<a name="line.616"></a> -<span class="sourceLineNo">617</span> String msg = "Unexpected " + proc.getState() + " state for " + proc;<a name="line.617"></a> -<span class="sourceLineNo">618</span> LOG.error(msg);<a name="line.618"></a> -<span class="sourceLineNo">619</span> throw new UnsupportedOperationException(msg);<a name="line.619"></a> -<span class="sourceLineNo">620</span> default:<a name="line.620"></a> -<span class="sourceLineNo">621</span> break;<a name="line.621"></a> -<span class="sourceLineNo">622</span> }<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> // 4. Check the waiting procedures to see if some of them can be added to runnable.<a name="line.625"></a> -<span class="sourceLineNo">626</span> waitingList.forEach(proc -> {<a name="line.626"></a> -<span class="sourceLineNo">627</span> if (!proc.hasChildren()) {<a name="line.627"></a> -<span class="sourceLineNo">628</span> // Normally, WAITING procedures should be waken by its children.<a name="line.628"></a> -<span class="sourceLineNo">629</span> // But, there is a case that, all the children are successful and before<a name="line.629"></a> -<span class="sourceLineNo">630</span> // they can wake up their parent procedure, the master was killed.<a name="line.630"></a> -<span class="sourceLineNo">631</span> // So, during recovering the procedures from ProcedureWal, its children<a name="line.631"></a> -<span class="sourceLineNo">632</span> // are not loaded because of their SUCCESS state.<a name="line.632"></a> -<span class="sourceLineNo">633</span> // So we need to continue to run this WAITING procedure. But before<a name="line.633"></a> -<span class="sourceLineNo">634</span> // executing, we need to set its state to RUNNABLE, otherwise, a exception<a name="line.634"></a> -<span class="sourceLineNo">635</span> // will throw:<a name="line.635"></a> -<span class="sourceLineNo">636</span> // Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,<a name="line.636"></a> -<span class="sourceLineNo">637</span> // "NOT RUNNABLE! " + procedure.toString());<a name="line.637"></a> -<span class="sourceLineNo">638</span> proc.setState(ProcedureState.RUNNABLE);<a name="line.638"></a> -<span class="sourceLineNo">639</span> runnableList.add(proc);<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> // 5. Push the procedures to the timeout executor<a name="line.643"></a> -<span class="sourceLineNo">644</span> waitingTimeoutList.forEach(proc -> {<a name="line.644"></a> -<span class="sourceLineNo">645</span> proc.afterReplay(getEnvironment());<a name="line.645"></a> -<span class="sourceLineNo">646</span> timeoutExecutor.add(proc);<a name="line.646"></a> -<span class="sourceLineNo">647</span> });<a name="line.647"></a> -<span class="sourceLineNo">648</span> // 6. restore locks<a name="line.648"></a> -<span class="sourceLineNo">649</span> restoreLocks();<a name="line.649"></a> -<span class="sourceLineNo">650</span> // 7. Push the procedure to the scheduler<a name="line.650"></a> -<span class="sourceLineNo">651</span> failedList.forEach(scheduler::addBack);<a name="line.651"></a> -<span class="sourceLineNo">652</span> runnableList.forEach(p -> {<a name="line.652"></a> -<span class="sourceLineNo">653</span> p.afterReplay(getEnvironment());<a name="line.653"></a> -<span class="sourceLineNo">654</span> if (!p.hasParent()) {<a name="line.654"></a> -<span class="sourceLineNo">655</span> sendProcedureLoadedNotification(p.getProcId());<a name="line.655"></a> -<span class="sourceLineNo">656</span> }<a name="line.656"></a> -<span class="sourceLineNo">657</span> scheduler.addBack(p);<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><a name="line.660"></a> -<span class="sourceLineNo">661</span> /**<a name="line.661"></a> -<span class="sourceLineNo">662</span> * Initialize the procedure executor, but do not start workers. We will start them later.<a name="line.662"></a> -<span class="sourceLineNo">663</span> * <p/><a name="line.663"></a> -<span class="sourceLineNo">664</span> * It calls ProcedureStore.recoverLease() and ProcedureStore.load() to recover the lease, and<a name="line.664"></a> -<span class="sourceLineNo">665</span> * ensure a single executor, and start the procedure replay to resume and recover the previous<a name="line.665"></a> -<span class="sourceLineNo">666</span> * pending and in-progress procedures.<a name="line.666"></a> -<span class="sourceLineNo">667</span> * @param numThreads number of threads available for procedure execution.<a name="line.667"></a> -<span class="sourceLineNo">668</span> * @param abortOnCorruption true if you want to abort your service in case a corrupted procedure<a name="line.668"></a> -<span class="sourceLineNo">669</span> * is found on replay. otherwise false.<a name="line.669"></a> -<span class="sourceLineNo">670</span> */<a name="line.670"></a> -<span class="sourceLineNo">671</span> public void init(int numThreads, boolean abortOnCorruption) throws IOException {<a name="line.671"></a> -<span class="sourceLineNo">672</span> // We have numThreads executor + one timer thread used for timing out<a name="line.672"></a> -<span class="sourceLineNo">673</span> // procedures and triggering periodic procedures.<a name="line.673"></a> -<span class="sourceLineNo">674</span> this.corePoolSize = numThreads;<a name="line.674"></a> -<span class="sourceLineNo">675</span> this.maxPoolSize = 10 * numThreads;<a name="line.675"></a> -<span class="sourceLineNo">676</span> LOG.info("Starting {} core workers (bigger of cpus/4 or 16) with max (burst) worker count={}",<a name="line.676"></a> -<span class="sourceLineNo">677</span> corePoolSize, maxPoolSize);<a name="line.677"></a> -<span class="sourceLineNo">678</span><a name="line.678"></a> -<span class="sourceLineNo">679</span> this.threadGroup = new ThreadGroup("PEWorkerGroup");<a name="line.679"></a> -<span class="sourceLineNo">680</span> this.timeoutExecutor = new TimeoutExecutorThread<>(this, threadGroup);<a name="line.680"></a> -<span class="sourceLineNo">681</span><a name="line.681"></a> -<span class="sourceLineNo">682</span> // Create the workers<a name="line.682"></a> -<span class="sourceLineNo">683</span> workerId.set(0);<a name="line.683"></a> -<span class="sourceLineNo">684</span> workerThreads = new CopyOnWriteArrayList<>();<a name="line.684"></a> -<span class="sourceLineNo">685</span> for (int i = 0; i < corePoolSize; ++i) {<a name="line.685"></a> -<span class="sourceLineNo">686</span> workerThreads.add(new WorkerThread(threadGroup));<a name="line.686"></a> -<span class="sourceLineNo">687</span> }<a name="line.687"></a> +<span class="sourceLineNo">584</span> // The orphan procedures will be passed to handleCorrupted, so add an assert here<a name="line.584"></a> +<span class="sourceLineNo">585</span> assert rootProcId != null;<a name="line.585"></a> +<span class="sourceLineNo">586</span><a name="line.586"></a> +<span class="sourceLineNo">587</span> if (proc.hasParent()) {<a name="line.587"></a> +<span class="sourceLineNo">588</span> Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());<a name="line.588"></a> +<span class="sourceLineNo">589</span> if (parent != null && !proc.isFinished()) {<a name="line.589"></a> +<span class="sourceLineNo">590</span> parent.incChildrenLatch();<a name="line.590"></a> +<span class="sourceLineNo">591</span> }<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> RootProcedureState<TEnvironment> procStack = rollbackStack.get(rootProcId);<a name="line.594"></a> +<span class="sourceLineNo">595</span> procStack.loadStack(proc);<a name="line.595"></a> +<span class="sourceLineNo">596</span><a name="line.596"></a> +<span class="sourceLineNo">597</span> proc.setRootProcId(rootProcId);<a name="line.597"></a> +<span class="sourceLineNo">598</span> switch (proc.getState()) {<a name="line.598"></a> +<span class="sourceLineNo">599</span> case RUNNABLE:<a name="line.599"></a> +<span class="sourceLineNo">600</span> runnableList.add(proc);<a name="line.600"></a> +<span class="sourceLineNo">601</span> break;<a name="line.601"></a> +<span class="sourceLineNo">602</span> case WAITING:<a name="line.602"></a> +<span class="sourceLineNo">603</span> waitingList.add(proc);<a name="line.603"></a> +<span class="sourceLineNo">604</span> break;<a name="line.604"></a> +<span class="sourceLineNo">605</span> case WAITING_TIMEOUT:<a name="line.605"></a> +<span class="sourceLineNo">606</span> waitingTimeoutList.add(proc);<a name="line.606"></a> +<span class="sourceLineNo">607</span> break;<a name="line.607"></a> +<span class="sourceLineNo">608</span> case FAILED:<a name="line.608"></a> +<span class="sourceLineNo">609</span> failedList.add(proc);<a name="line.609"></a> +<span class="sourceLineNo">610</span> break;<a name="line.610"></a> +<span class="sourceLineNo">611</span> case ROLLEDBACK:<a name="line.611"></a> +<span class="sourceLineNo">612</span> case INITIALIZING:<a name="line.612"></a> +<span class="sourceLineNo">613</span> String msg = "Unexpected " + proc.getState() + " state for " + proc;<a name="line.613"></a> +<span class="sourceLineNo">614</span> LOG.error(msg);<a name="line.614"></a> +<span class="sourceLineNo">615</span> throw new UnsupportedOperationException(msg);<a name="line.615"></a> +<span class="sourceLineNo">616</span> default:<a name="line.616"></a> +<span class="sourceLineNo">617</span> break;<a name="line.617"></a> +<span class="sourceLineNo">618</span> }<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> // 3. Check the waiting procedures to see if some of them can be added to runnable.<a name="line.621"></a> +<span class="sourceLineNo">622</span> waitingList.forEach(proc -> {<a name="line.622"></a> +<span class="sourceLineNo">623</span> if (!proc.hasChildren()) {<a name="line.623"></a> +<span class="sourceLineNo">624</span> // Normally, WAITING procedures should be waken by its children.<a name="line.624"></a> +<span class="sourceLineNo">625</span> // But, there is a case that, all the children are successful and before<a name="line.625"></a> +<span class="sourceLineNo">626</span> // they can wake up their parent procedure, the master was killed.<a name="line.626"></a> +<span class="sourceLineNo">627</span> // So, during recovering the procedures from ProcedureWal, its children<a name="line.627"></a> +<span class="sourceLineNo">628</span> // are not loaded because of their SUCCESS state.<a name="line.628"></a> +<span class="sourceLineNo">629</span> // So we need to continue to run this WAITING procedure. But before<a name="line.629"></a> +<span class="sourceLineNo">630</span> // executing, we need to set its state to RUNNABLE, otherwise, a exception<a name="line.630"></a> +<span class="sourceLineNo">631</span> // will throw:<a name="line.631"></a> +<span class="sourceLineNo">632</span> // Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,<a name="line.632"></a> +<span class="sourceLineNo">633</span> // "NOT RUNNABLE! " + procedure.toString());<a name="line.633"></a> +<span class="sourceLineNo">634</span> proc.setState(ProcedureState.RUNNABLE);<a name="line.634"></a> +<span class="sourceLineNo">635</span> runnableList.add(proc);<a name="line.635"></a> +<span class="sourceLineNo">636</span> } else {<a name="line.636"></a> +<span class="sourceLineNo">637</span> proc.afterReplay(getEnvironment());<a name="line.637"></a> +<span class="sourceLineNo">638</span> }<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> // 4. Push the procedures to the timeout executor<a name="line.641"></a> +<span class="sourceLineNo">642</span> waitingTimeoutList.forEach(proc -> {<a name="line.642"></a> +<span class="sourceLineNo">643</span> proc.afterReplay(getEnvironment());<a name="line.643"></a> +<span class="sourceLineNo">644</span> timeoutExecutor.add(proc);<a name="line.644"></a> +<span class="sourceLineNo">645</span> });<a name="line.645"></a> +<span class="sourceLineNo">646</span> // 5. restore locks<a name="line.646"></a> +<span class="sourceLineNo">647</span> restoreLocks();<a name="line.647"></a> +<span class="sourceLineNo">648</span> // 6. Push the procedure to the scheduler<a name="line.648"></a> +<span class="sourceLineNo">649</span> failedList.forEach(scheduler::addBack);<a name="line.649"></a> +<span class="sourceLineNo">650</span> runnableList.forEach(p -> {<a name="line.650"></a> +<span class="sourceLineNo">651</span> p.afterReplay(getEnvironment());<a name="line.651"></a> +<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> * <p/><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<>(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<>();<a name="line.682"></a> +<span class="sourceLineNo">683</span> for (int i = 0; i < 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> long st, et;<a name="line.689"></a> -<span class="sourceLineNo">690</span><a name="line.690"></a> -<span class="sourceLineNo">691</span> // Acquire the store lease.<a name="line.691"></a> -<span class="sourceLineNo">692</span> st = System.nanoTime();<a name="line.692"></a> -<span class="sourceLineNo">693</span> store.recoverLease();<a name="line.693"></a> -<span class="sourceLineNo">694</span> et = System.nanoTime();<a name="line.694"></a> -<span class="sourceLineNo">695</span> LOG.info("Recovered {} lease in {}", store.getClass().getSimpleName(),<a name="line.695"></a> -<span class="sourceLineNo">696</span> StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.696"></a> -<span class="sourceLineNo">697</span><a name="line.697"></a> -<span class="sourceLineNo">698</span> // start the procedure scheduler<a name="line.698"></a> -<span class="sourceLineNo">699</span> scheduler.start();<a name="line.699"></a> -<span class="sourceLineNo">700</span><a name="line.700"></a> -<span class="sourceLineNo">701</span> // TODO: Split in two steps.<a name="line.701"></a> -<span class="sourceLineNo">702</span> // TODO: Handle corrupted procedures (currently just a warn)<a name="line.702"></a> -<span class="sourceLineNo">703</span> // The first one will make sure that we have the latest id,<a name="line.703"></a> -<span class="sourceLineNo">704</span> // so we can start the threads and accept new procedures.<a name="line.704"></a> -<span class="sourceLineNo">705</span> // The second step will do the actual load of old procedures.<a name="line.705"></a> -<span class="sourceLineNo">706</span> st = System.nanoTime();<a name="line.706"></a> -<span class="sourceLineNo">707</span> load(abortOnCorruption);<a name="line.707"></a> -<span class="sourceLineNo">708</span> et = System.nanoTime();<a name="line.708"></a> -<span class="sourceLineNo">709</span> LOG.info("Loaded {} in {}", store.getClass().getSimpleName(),<a name="line.709"></a> -<span class="sourceLineNo">710</span> StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.710"></a> -<span class="sourceLineNo">711</span> }<a name="line.711"></a> -<span class="sourceLineNo">712</span><a name="line.712"></a> -<span class="sourceLineNo">713</span> /**<a name="line.713"></a> -<span class="sourceLineNo">714</span> * Start the workers.<a name="line.714"></a> -<span class="sourceLineNo">715</span> */<a name="line.715"></a> -<span class="sourceLineNo">716</span> public void startWorkers() throws IOException {<a name="line.716"></a> -<span class="sourceLineNo">717</span> if (!running.compareAndSet(false, true)) {<a name="line.717"></a> -<span class="sourceLineNo">718</span> LOG.warn("Already running");<a name="line.718"></a> -<span class="sourceLineNo">719</span> return;<a name="line.719"></a> -<span class="sourceLineNo">720</span> }<a name="line.720"></a> -<span class="sourceLineNo">721</span> // Start the executors. Here we must have the lastProcId set.<a name="line.721"></a> -<span class="sourceLineNo">722</span> LOG.trace("Start workers {}", workerThreads.size());<a name="line.722"></a> -<span class="sourceLineNo">723</span> timeoutExecutor.start();<a name="line.723"></a> -<span class="sourceLineNo">724</span> for (WorkerThread worker: workerThreads) {<a name="line.724"></a> -<span class="sourceLineNo">725</span> worker.start();<a name="line.725"></a> -<span class="sourceLineNo">726</span> }<a name="line.726"></a> -<span class="sourceLineNo">727</span><a name="line.727"></a> -<span class="sourceLineNo">728</span> // Internal chores<a name="line.728"></a> -<span class="sourceLineNo">729</span> timeoutExecutor.add(new WorkerMonitor());<a name="line.729"></a> -<span class="sourceLineNo">730</span><a name="line.730"></a> -<span class="sourceLineNo">731</span> // Add completed cleaner chore<a name="line.731"></a> -<span class="sourceLineNo">732</span> addChore(new CompletedProcedureCleaner<>(conf, store, completed, nonceKeysToProcIdsMap));<a name="line.732"></a> -<span class="sourceLineNo">733</span> }<a name="line.733"></a> -<span class="sourceLineNo">734</span><a name="line.734"></a> -<span class="sourceLineNo">735</span> public void stop() {<a name="line.735"></a> -<span class="sourceLineNo">736</span> if (!running.getAndSet(false)) {<a name="line.736"></a> -<span class="sourceLineNo">737</span> return;<a name="line.737"></a> -<span class="sourceLineNo">738</span> }<a name="line.738"></a> -<span class="sourceLineNo">739</span><a name="line.739"></a> -<span class="sourceLineNo">740</span> LOG.info("Stopping");<a name="line.740"></a> -<span class="sourceLineNo">741</span> scheduler.stop();<a name="line.741"></a> -<span class="sourceLineNo">742</span> timeoutExecutor.sendStopSignal();<a name="line.742"></a> -<span class="sourceLineNo">743</span> }<a name="line.743"></a> -<span class="sourceLineNo">744</span><a name="line.744"></a> -<span class="sourceLineNo">745</span> @VisibleForTesting<a name="line.745"></a> -<span class="sourceLineNo">746</span> public void join() {<a name="line.746"></a> -<span class="sourceLineNo">747</span> assert !isRunning() : "expected not running";<a name="line.747"></a> -<span class="sourceLineNo">748</span><a name="line.748"></a> -<span class="sourceLineNo">749</span> // stop the timeout executor<a name="line.749"></a> -<span class="sourceLineNo">750</span> timeoutExecutor.awaitTermination();<a name="line.750"></a> -<span class="sourceLineNo">751</span><a name="line.751"></a> -<span class="sourceLineNo">752</span> // stop the worker threads<a name="line.752"></a> -<span class="sourceLineNo">753</span> for (WorkerThread worker: workerThreads) {<a name="line.753"></a> -<span class="sourceLineNo">754</span> worker.awaitTermination();<a name="line.754"></a> -<span class="sourceLineNo">755</span> }<a name="line.755"></a> -<span class="sourceLineNo">756</span><a name="line.756"></a> -<span class="sourceLineNo">757</span> // Destroy the Thread Group for the executors<a name="line.757"></a> -<span class="sourceLineNo">758</span> // TODO: Fix. #join is not place to destroy resources.<a name="line.758"></a> -<span class="sourceLineNo">759</span> try {<a name="line.759"></a> -<span class="sourceLineNo">760</span> threadGroup.destroy();<a name="line.760"></a> -<span class="sourceLineNo">761</span> } catch (IllegalThreadStateException e) {<a name="line.761"></a> -<span class="sourceLineNo">762</span> LOG.error("ThreadGroup {} contains running threads; {}: See STDOUT",<a name="line.762"></a> -<span class="sourceLineNo">763</span> this.threadGroup, e.getMessage());<a name="line.763"></a> -<span class="sourceLineNo">764</span> // This dumps list of threads on STDOUT.<a name="line.764"></a> -<span class="sourceLineNo">765</span> this.threadGroup.list();<a name="line.765"></a> -<span class="sourceLineNo">766</span> }<a name="line.766"></a> -<span class="sourceLineNo">767</span><a name="line.767"></a> -<span class="sourceLineNo">768</span> // reset the in-memory state for testing<a name="line.768"></a> -<span class="sourceLineNo">769</span> completed.clear();<a name="line.769"></a> -<span class="sourceLineNo">770</span> rollbackStack.clear();<a name="line.770"></a> -<span class="sourceLineNo">771</span> procedures.clear();<a name="line.771"></a> -<span class="sourceLineNo">772</span> nonceKeysToProcIdsMap.clear();<a name="line.772"></a> -<span class="sourceLineNo">773</span> scheduler.clear();<a name="line.773"></a> -<span class="sourceLineNo">774</span> lastProcId.set(-1);<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> public void refreshConfiguration(final Configuration conf) {<a name="line.777"></a> -<span class="sourceLineNo">778</span> this.conf = conf;<a name="line.778"></a> -<span class="sourceLineNo">779</span> setKeepAliveTime(conf.getLong(WORKER_KEEP_ALIVE_TIME_CONF_KEY,<a name="line.779"></a> -<span class="sourceLineNo">780</span> DEFAULT_WORKER_KEEP_ALIVE_TIME), TimeUnit.MILLISECONDS);<a name="line.780"></a> -<span class="sourceLineNo">781</span> }<a name="line.781"></a> -<span class="sourceLineNo">782</span><a name="line.782"></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">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">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<>(conf, store, completed, nonceKeysToProcIdsMap));<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">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> // Accessors<a name="line.784"></a> -<span class="sourceLineNo">785</span> // ==========================================================================<a name="line.785"></a> -<span class="sourceLineNo">786</span> public boolean isRunning() {<a name="line.786"></a> -<span class="sourceLineNo">787</span> return running.get();<a name="line.787"></a> -<span class="sourceLineNo">788</span> }<a name="line.788"></a> -<span class="sourceLineNo">789</span><a name="line.789"></a> -<span class="sourceLineNo">790</span> /**<a name="line.790"></a> -<span class="sourceLineNo">791</span> * @return the current number of worker threads.<a name="line.791"></a> -<span class="sourceLineNo">792</span> */<a name="line.792"></a> -<span class="sourceLineNo">793</span> public int getWorkerThreadCount() {<a name="line.793"></a> -<span class="sourceLineNo">794</span> return workerThreads.size();<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> /**<a name="line.797"></a> -<span class="sourceLineNo">798</span> * @return the core pool size settings.<a name="line.798"></a> -<span class="sourceLineNo">799</span> */<a name="line.799"></a> -<span class="sourceLineNo">800</span> public int getCorePoolSize() {<a name="line.800"></a> -<span class="sourceLineNo">801</span> return corePoolSize;<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> public int getActiveExecutorCount() {<a name="line.804"></a> -<span class="sourceLineNo">805</span> return activeExecutorCount.get();<a name="line.805"></a> -<span class="sourceLineNo">806</span> }<a name="line.806"></a> -<span class="sourceLineNo">807</span><a name="line.807"></a> -<span class="sourceLineNo">808</span> public TEnvironment getEnvironment() {<a name="line.808"></a> -<span class="sourceLineNo">809</span> return this.environment;<a name="line.809"></a> -<span class="sourceLineNo">810</span> }<a name="line.810"></a> -<span class="sourceLineNo">811</span><a name="line.811"></a> -<span class="sourceLineNo">812</span> public ProcedureStore getStore() {<a name="line.812"></a> -<span class="sourceLineNo">813</span> return this.store;<a name="line.813"></a> -<span class="sourceLineNo">814</span> }<a name="line.814"></a> -<span class="sourceLineNo">815</span><a name="line.815"></a> -<span class="sourceLineNo">816</span> ProcedureScheduler getScheduler() {<a name="line.816"></a> -<span class="sourceLineNo">817</span> return scheduler;<a name="line.817"></a> -<span class="sourceLineNo">818</span> }<a name="line.818"></a> -<span class="sourceLineNo">819</span><a name="line.819"></a> -<span class="sourceLineNo">820</span> public void setKeepAliveTime(final long keepAliveTime, final TimeUnit timeUnit) {<a name="line.820"></a> -<span class="sourceLineNo">821</span> this.keepAliveTime = timeUnit.toMillis(keepAliveTime);<a name="line.821"></a> -<span class="sourceLineNo">822</span> this.scheduler.signalAll();<a name="line.822"></a> -<span class="sourceLineNo">823</span> }<a name="line.823"></a> -<span class="sourceLineNo">824</span><a name="line.824"></a> -<span class="sourceLineNo">825</span> public long getKeepAliveTime(final TimeUnit timeUnit) {<a name="line.825"></a> -<span class="sourceLineNo">826</span> return timeUnit.convert(keepAliveTime, TimeUnit.MILLISECONDS);<a name="line.826"></a> -<span class="sourceLineNo">827</span> }<a name="line.827"></a> -<span class="sourceLineNo">828</span><a name="line.828"></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">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">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> // Submit/Remove Chores<a name="line.830"></a> -<span class="sourceLineNo">831</span> // ==========================================================================<a name="line.831"></a> -<span class="sourceLineNo">832</span><a name="line.832"></a> -<span class="sourceLineNo">833</span> /**<a name="line.833"></a> -<span class="sourceLineNo">834</span> * Add a chore procedure to the executor<a name="line.834"></a> -<span class="sourceLineNo">835</span> * @param chore the chore to add<a name="line.835"></a> -<span class="sourceLineNo">836</span> */<a name="line.836"></a> -<span class="sourceLineNo">837</span> public void addChore(ProcedureInMemoryChore<TEnvironment> chore) {<a name="line.837"></a> -<span class="sourceLineNo">838</span> chore.setState(ProcedureState.WAITING_TIMEOUT);<a name="line.838"></a> -<span class="sourceLineNo">839</span> timeoutExecutor.add(chore);<a name="line.839"></a> -<span class="sourceLineNo">840</span> }<a name="line.840"></a> -<span class="sourceLineNo">841</span><a name="line.841"></a> -<span class="sourceLineNo">842</span> /**<a name="line.842"></a> -<span class="sourceLineNo">843</span> * Remove a chore procedure from the executor<a name="line.843"></a> -<span class="sourceLineNo">844</span> * @param chore the chore to remove<a name="line.844"></a> -<span class="sourceLineNo">845</span> * @return whether the chore is removed, or it will be removed later<a name="line.845"></a> -<span class="sourceLineNo">846</span> */<a name="line.846"></a> -<span class="sourceLineNo">847</span> public boolean removeChore(ProcedureInMemoryChore<TEnvironment> chore) {<a name="line.847"></a> -<span class="sourceLineNo">848</span> chore.setState(ProcedureState.SUCCESS);<a name="line.848"></a> -<span class="sourceLineNo">849</span> return timeoutExecutor.remove(chore);<a name="line.849"></a> -<span class="sourceLineNo">850</span> }<a name="line.850"></a> -<span class="sourceLineNo">851</span><a name="line.851"></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<TEnvironment> 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">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<TEnvironment> 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">853</span> // Nonce Procedure helpers<a name="line.853"></a> -<span class="sourceLineNo">854</span> // ==========================================================================<a name="line.854"></a> -<span class="sourceLineNo">855</span> /**<a name="line.855"></a> -<span class="sourceLineNo">856</span> * Create a NoneKey from the specified nonceGroup and nonce.<a name="line.856"></a> -<span class="sourceLineNo">857</span> * @param nonceGroup<a name="line.857"></a> -<span class="sourceLineNo">858</span> * @param nonce<a name="line.858"></a> -<span class="sourceLineNo">859</span> * @return the generated NonceKey<a name="line.859"></a> -<span class="sourceLineNo">860</span> */<a name="line.860"></a> -<span class="sourceLineNo">861</span> public NonceKey createNonceKey(final long nonceGroup, final long nonce) {<a name="line.861"></a> -<span class="sourceLineNo">862</span> return (nonce == HConstants.NO_NONCE) ? null : new NonceKey(nonceGroup, nonce);<a name="line.862"></a> -<span class="sourceLineNo">863</span> }<a name="line.863"></a> -<span class="sourceLineNo">864</span><a name="line.864"></a> -<span class="sourceLineNo">865</span> /**<a name="line.865"></a> -<span class="sourceLineNo">866</span> * Register a nonce for a procedure that is going to be submitted.<a name="line.866"></a> -<span class="sourceLineNo">867</span> * A procId will be reserved and on submitProcedure(),<a name="line.867"></a> -<span class="sourceLineNo">868</span> * the procedure with the specified nonce will take the reserved ProcId.<a name="line.868"></a> -<span class="sourceLineNo">869</span> * If someone already reserved the nonce, this method will return the procId reserved,<a name="line.869"></a> -<span class="sourceLineNo">870</span> * otherwise an invalid procId will be returned. and the caller should procede<a name="line.870"></a> -<span class="sourceLineNo">871</span> * and submit the procedure.<a name="line.871"></a> -<span class="sourceLineNo">872</span> *<a name="line.872"></a> -<span class="sourceLineNo">873</span> * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.873"></a> -<span class="sourceLineNo">874</span> * @return the procId associated with the nonce, if any otherwise an invalid procId.<a name="line.874"></a> -<span class="sourceLineNo">875</span> */<a name="line.875"></a> -<span class="sourceLineNo">876</span> public long registerNonce(final NonceKey nonceKey) {<a name="line.876"></a> -<span class="sourceLineNo">877</span> if (nonceKey == null) return -1;<a name="line.877"></a> -<span class="sourceLineNo">878</span><a name="line.878"></a> -<span class="sourceLineNo">879</span> // check if we have already a Reserved ID for the nonce<a name="line.879"></a> -<span class="sourceLineNo">880</span> Long oldProcId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.880"></a> -<span class="sourceLineNo">881</span> if (oldProcId == null) {<a name="line.881"></a> -<span class="sourceLineNo">882</span> // reserve a new Procedure ID, this will be associated with the nonce<a name="line.882"></a> -<span class="sourceLineNo">883</span> // and the procedure submitted with the specified nonce will use this ID.<a name="line.883"></a> -<span class="sourceLineNo">884</span> final long newProcId = nextProcId();<a name="line.884"></a> -<span class="sourceLineNo">885</span> oldProcId = nonceKeysToProcIdsMap.putIfAbsent(nonceKey, newProcId);<a name="line.885"></a> -<span class="sourceLineNo">886</span> if (oldProcId == null) return -1;<a name="line.886"></a> -<span class="sourceLineNo">887</span> }<a name="line.887"></a> -<span class="sourceLineNo">888</span><a name="line.888"></a> -<span class="sourceLineNo">889</span> // we found a registered nonce, but the procedure may not have been submitted yet.<a name="line.889"></a> -<span class="sourceLineNo">890</span> // since the client expect the procedure to be submitted, spin here until it is.<a name="line.890"></a> -<span class="sourceLineNo">891</span> final boolean traceEnabled = LOG.isTraceEnabled();<a name="line.891"></a> -<span class="sourceLineNo">892</span> while (isRunning() &&<a name="line.892"></a> -<span class="sourceLineNo">893</span> !(procedures.containsKey(oldProcId) || completed.containsKey(oldProcId)) &&<a name="line.893"></a> -<span class="sourceLineNo">894</span> nonceKeysToProcIdsMap.containsKey(nonceKey)) {<a name="line.894"></a> -<span class="sourceLineNo">895</span> if (traceEnabled) {<a name="line.895"></a> -<span class="sourceLineNo">896</span> LOG.trace("Waiting for pid=" + oldProcId.longValue() + " to be submitted");<a name="line.896"></a> -<span class="sourceLineNo">897</span> }<a name="line.897"></a> -<span class="sourceLineNo">898</span> Threads.sleep(100);<a name="line.898"></a> -<span class="sourceLineNo">899</span> }<a name="line.899"></a> -<span class="sourceLineNo">900</span> return oldProcId.longValue();<a name="line.900"></a> -<span class="sourceLineNo">901</span> }<a name="line.901"></a> -<span class="sourceLineNo">902</span><a name="line.902"></a> -<span class="sourceLineNo">903</span> /**<a name="line.903"></a> -<span class="sourceLineNo">904</span> * Remove the NonceKey if the procedure was not submitted to the executor.<a name="line.904"></a> -<span class="sourceLineNo">905</span> * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.905"></a> -<span class="sourceLineNo">906</span> */<a name="line.906"></a> -<span class="sourceLineNo">907</span> public void unregisterNonceIfProcedureWasNotSubmitted(final NonceKey nonceKey) {<a name="line.907"></a> -<span class="sourceLineNo">908</span> if (nonceKey == null) return;<a name="line.908"></a> -<span class="sourceLineNo">909</span><a name="line.909"></a> -<span class="sourceLineNo">910</span> final Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.910"></a> -<span class="sourceLineNo">911</span> if (procId == null) return;<a name="line.911"></a> -<span class="sourceLineNo">912</span><a name="line.912"></a> -<span class="sourceLineNo">913</span> // if the procedure was not submitted, remove the nonce<a name="line.913"></a> -<span class="sourceLineNo">914</span> if (!(procedures.containsKey(procId) || completed.containsKey(procId))) {<a name="line.914"></a> -<span class="sourceLineNo">915</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.915"></a> -<span class="sourceLineNo">916</span> }<a name="line.916"></a> -<span class="sourceLineNo">917</span> }<a name="line.917"></a> -<span class="sourceLineNo">918</span><a name="line.918"></a> -<span class="sourceLineNo">919</span> public static class FailedProcedure<TEnvironment> extends Procedure<TEnvironment> {<a name="line.919"></a> -<span class="sourceLineNo">920</span> private String procName;<a name="line.920"></a> -<span class="sourceLineNo">921</span><a name="line.921"></a> -<span class="sourceLineNo">922</span> public FailedProcedure() {<a name="line.922"></a> -<span class="sourceLineNo">923</span> }<a name="line.923"></a> -<span class="sourceLineNo">924</span><a name="line.924"></a> -<span class="sourceLineNo">925</span> public FailedProcedure(long procId, String procName, User owner,<a name="line.925"></a> -<span class="sourceLineNo">926</span> NonceKey nonceKey, IOException exception) {<a name="line.926"></a> -<span class="sourceLineNo">927</span> this.procName = procName;<a name="line.927"></a> -<span class="sourceLineNo">928</span> setProcId(procId);<a name="line.928"></a> -<span class="sourceLineNo">929</span> setState(ProcedureState.ROLLEDBACK);<a name="line.929"></a> -<span class="sourceLineNo">930</span> setOwner(owner);<a name="line.930"></a> -<span class="sourceLineNo">931</span> setNonceKey(nonceKey);<a name="line.931"></a> -<span class="sourceLineNo">932</span> long currentTime = EnvironmentEdgeManager.currentTime();<a name="line.932"></a> -<span class="sourceLineNo">933</span> setSubmittedTime(currentTime);<a name="line.933"></a> -<span class="sourceLineNo">934</span> setLastUpdate(currentTime);<a name="line.934"></a> -<span class="sourceLineNo">935</span> setFailure(Objects.toString(exception.getMessage(), ""), exception);<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> @Override<a name="line.938"></a> -<span class="sourceLineNo">939</span> public String getProcName() {<a name="line.939"></a> -<span class="sourceLineNo">940</span> return procName;<a name="line.940"></a> -<span class="sourceLineNo">941</span> }<a name="line.941"></a> -<span class="sourceLineNo">942</span><a name="line.942"></a> -<span class="sourceLineNo">943</span> @Override<a name="line.943"></a> -<span class="sourceLineNo">944</span> protected Procedure<TEnvironment>[] execute(TEnvironment env)<a name="line.944"></a> -<span class="sourceLineNo">945</span> throws ProcedureYieldException, ProcedureSuspendedException,<a name="line.945"></a> -<span class="sourceLineNo">946</span> InterruptedException {<a name="line.946"></a> -<span class="sourceLineNo">947</span> throw new UnsupportedOperationException();<a name="line.947"></a> -<span class="sourceLineNo">948</span> }<a name="line.948"></a> -<span class="sourceLineNo">949</span><a name="line.949"></a> -<span class="sourceLineNo">950</span> @Override<a name="line.950"></a> -<span class="sourceLineNo">951</span> protected void rollback(TEnvironment env)<a name="line.951"></a> -<span class="sourceLineNo">952</span> throws IOException, InterruptedException {<a name="line.952"></a> -<span class="sourceLineNo">953</span> throw new UnsupportedOperationException();<a name="line.953"></a> -<span class="sourceLineNo">954</span> }<a name="line.954"></a> -<span class="sourceLineNo">955</span><a name="line.955"></a> -<span class="sourceLineNo">956</span> @Override<a name="line.956"></a> -<span class="sourceLineNo">957</span> protected boolean abort(TEnvironment env) {<a name="line.957"></a> -<span class="sourceLineNo">958</span> throw new UnsupportedOperationException();<a name="line.958"></a> -<span class="sourceLineNo">959</span> }<a name="line.959"></a> -<span class="sourceLineNo">960</span><a name="line.960"></a> -<span class="sourceLineNo">961</span> @Override<a name="line.961"></a> -<span class="sourceLineNo">962</span> protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.962"></a> -<span class="sourceLineNo">963</span> throws IOException {<a name="line.963"></a> -<span class="sourceLineNo">964</span> }<a name="line.964"></a> -<span class="sourceLineNo">965</span><a name="line.965"></a> -<span class="sourceLineNo">966</span> @Override<a name="line.966"></a> -<span class="sourceLineNo">967</span> protected void deserializeStateData(ProcedureStateSerializer serializer)<a name="line.967"></a> -<span class="sourceLineNo">968</span> throws IOException {<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><a name="line.971"></a> -<span class="sourceLineNo">972</span> /**<a name="line.972"></a> -<span class="sourceLineNo">973</span> * If the failure failed before submitting it, we may want to give back the<a name="line.973"></a> -<span class="sourceLineNo">974</span> * same error to the requests with the same nonceKey.<a name="line.974"></a> -<span class="sourceLineNo">975</span> *<a name="line.975"></a> -<span class="sourceLineNo">976</span> * @param nonceKey A unique identifier for this operation from the client or process<a name="line.976"></a> -<span class="sourceLineNo">977</span> * @param procName name of the procedure, used to inform the user<a name="line.977"></a> -<span class="sourceLineNo">978</span> * @param procOwner name of the owner of the procedure, used to inform the user<a name="line.978"></a> -<span class="sourceLineNo">979</span> * @param exception the failure to report to the user<a name="line.979"></a> -<span class="sourceLineNo">980</span> */<a name="line.980"></a> -<span class="sourceLineNo">981</span> public void setFailureResultForNonce(NonceKey nonceKey, String procName, User procOwner,<a name="line.981"></a> -<span class="sourceLineNo">982</span> IOException exception) {<a name="line.982"></a> -<span class="sourceLineNo">983</span> if (nonceKey == null) {<a name="line.983"></a> -<span class="sourceLineNo">984</span> return;<a name="line.984"></a> -<span class="sourceLineNo">985</span> }<a name="line.985"></a> -<span class="sourceLineNo">986</span><a name="line.986"></a> -<span class="sourceLineNo">987</span> Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.987"></a> -<span class="sourceLineNo">988</span> if (procId == null || completed.containsKey(procId)) {<a name="line.988"></a> -<span class="sourceLineNo">989</span> return;<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> Procedure<TEnvironment> proc =<a name="line.992"></a> -<span class="sourceLineNo">993</span> new FailedProcedure<>(procId.longValue(), procName, procOwner, nonceKey, exception);<a name="line.993"></a> -<span class="sourceLineNo">994</span><a name="line.994"></a> -<span class="sourceLineNo">995</span> completed.putIfAbsent(procId, new CompletedProcedureRetainer<>(proc));<a name="line.995"></a> -<span class="sourceLineNo">996</span> }<a name="line.996"></a> -<span class="sourceLineNo">997</span><a name="line.997"></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() &&<a name="line.890"></a> +<span class="sourceLineNo">891</span> !(procedures.containsKey(oldProcId) || completed.containsKey(oldProcId)) &&<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<TEnvironment> extends Procedure<TEnvironment> {<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">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<TEnvironment>[] 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">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<TEnvironment> proc =<a name="line.990"></a> +<span class="sourceLineNo">991</span> new FailedProcedure<>(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<>(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> // Submit/Abort Procedure<a name="line.999"></a> -<span class="sourceLineNo">1000</span> // ==========================================================================<a name="line.1000"></a> -<span class="sourceLineNo">1001</span> /**<a name="line.1001"></a> -<span class="sourceLineNo">1002</span> * Add a new root-procedure to the executor.<a name="line.1002"></a> -<span class="sourceLineNo">1003</span> * @param proc the new procedure to execute.<a name="line.1003"></a> -<span class="sourceLineNo">1004</span> * @return the procedure id, that can be used to monitor the operation<a name="line.1004"></a> -<span class="sourceLineNo">1005</span> */<a name="line.1005"></a> -<span class="sourceLineNo">1006</span> public long submitProcedure(Procedure<TEnvironment> proc) {<a name="line.1006"></a> -<span class="sourceLineNo">1007</span> return submitProcedure(proc, null);<a name="line.1007"></a> -<span class="sourceLineNo">1008</span> }<a name="line.1008"></a> -<span class="sourceLineNo">1009</span><a name="line.1009"></a> -<span class="sourceLineNo">1010</span> /**<a name="line.1010"></a> -<span class="sourceLineNo">1011</span> * Bypass a procedure. If the procedure is set to bypass, all the logic in<a name="line.1011"></a> -<span class="sourceLineNo">1012</span> * execute/rollback will be ignored and it will return success, whatever.<a name="line.1012"></a> -<span class="sourceLineNo">1013</span> * It is used to recover buggy stuck procedures, releasing the lock resources<a name="line.1013"></a> -<span class="sourceLineNo">1014</span> * and letting other procedures run. Bypassing one procedure (and its ancestors will<a name="line.1014"></a> -<span class="sourceLineNo">1015</span> * be bypassed automatically) may leave the cluster in a middle state, e.g. region<a name="line.1015"></a> -<span class="sourceLineNo">1016</span> * not assigned, or some hdfs files left behind. After getting rid of those stuck procedures,<a name="line.1016"></a> -<span class="sourceLineNo">1017</span> * the operators may have to do some clean up on hdfs or schedule some assign procedures<a name="line.1017"></a> -<span class="sourceLineNo">1018</span> * to let region online. DO AT YOUR OWN RISK.<a name="line.1018"></a> -<span class="sourceLineNo">1019</span> * <p><a name="line.1019"></a> -<span class="sourceLineNo">1020</span> * A procedure can be bypassed only if<a name="line.1020"></a> -<span class="sourceLineNo">1021</span> * 1. The procedure is in state of RUNNABLE, WAITING, WAITING_TIMEOUT<a name="line.1021"></a> -<span class="sourceLineNo">1022</span> * or it is a root procedure without any child.<a name="line.1022"></a> -<span class="sourceLineNo">1023</span> * 2. No other worker thread is executing it<a name="line.1023"></a> -<span class="sourceLineNo">1024</span> * 3. No child procedure has been submitted<a name="line.1024"></a> -<span class="sourceLineNo">1025</span> *<a name="line.1025"></a> -<span class="sourceLineNo">1026</span> * <p><a name="line.1026"></a> -<span class="sourceLineNo">1027</span> * If all the requirements are meet, the procedure and its ancestors will be<a name="line.1027"></a> -<span class="sourceLineNo">1028</span> * bypassed and persisted to WAL.<a name="line.1028"></a> -<span class="sourceLineNo">1029</span> *<a name="line.1029"></a> -<span class="sourceLineNo">1030</span> * <p><a name="line.1030"></a> -<span class="sourceLineNo">1031</span> * If the procedure is in WAITING state, will set it to RUNNABLE add it to run queue.<a name="line.1031"></a> -<span class="sourceLineNo">1032</span> * TODO: What about WAITING_TIMEOUT?<a name="line.1032"></a> -<span class="sourceLineNo">1033</span> * @param pids the procedure id<a name="line.1033"></a> -<span class="sourceLineNo">1034</span> * @param lockWait time to wait lock<a name="line.1034"></a> -<span class="sourceLineNo">1035</span> * @param force if force set to true, we will bypass the procedure even if it is executing.<a name="line.1035"></a> -<span class="sourceLineNo">1036</span> * This is for procedures which can't break out during executing(due to bug, mostly)<a name="line.1036"></a> -<span class="sourceLineNo">1037</span> * In this case, bypassing the procedure is not enough, since it is already stuck<a name="line.1037"></a> -<span class="sourceLineNo">1038</span> * there. We need to restart the master after bypassing, and letting the problematic<a name="line.1038"></a> -<span class="sourceLineNo">1039</span> * procedure to execute wth bypass=true, so in that condition, the procedure can be<a name="line.1039"></a> -<span class="sourceLineNo">1040</span> * successfully bypassed.<a name="line.1040"></a> -<span class="sourceLineNo">1041</span> * @param recursive We will do an expensive search for children of each pid. EXPENSIVE!<a name="line.1041"></a> -<span class="sourceLineNo">1042</span> * @return true if bypass success<a name="line.1042"></a> -<span class="sourceLineNo">1043</span> * @throws IOException IOException<a name="line.1043"></a> -<span class="sourceLineNo">1044</span> */<a name="line.1044"></a> -<span class="sourceLineNo">1045</span> public List<Boolean> bypassProcedure(List<Long> pids, long lockWait, boolean force,<a name="line.1045"></a> -<span class="sourceLineNo">1046</span> boolean recursive)<a name="line.1046"></a> -<span class="sourceLineNo">1047</span> throws IOException {<a name="line.1047"></a> -<span class="sourceLineNo">1048</span> List<Boolean> result = new ArrayList<Boolean>(pids.size());<a name="line.1048"></a> -<span class="sourceLineNo">1049</span> for(long pid: pids) {<a name="line.1049"></a> -<span class="sourceLineNo">1050</span> result.add(bypassProcedure(pid, lockWait, force, recursive));<a name="line.1050"></a> -<span class="sourceLineNo">1051</span> }<a name="line.1051"></a> -<span class="sourceLineNo">1052</span> return result;<a name="line.1052"></a> -<span class="sourceLineNo">1053</span> }<a name="line.1053"></a> -<span class="sourceLineNo">1054</span><a name="line.1054"></a> -<span class="sourceLineNo">1055</span> boolean bypassProcedure(long pid, long lockWait, boolean override, boolean recursive)<a name="line.1055"></a> -<span class="sourceLineNo">1056</span> throws IOException {<a name="line.1056"></a> -<span class="sourceLineNo">1057</span> Preconditions.checkArgument(lockWait > 0, "lockWait should be positive");<a name="line.1057"></a> -<span class="sourceLineNo">1058</span> final Procedure<TEnvironment> procedure = getProcedure(pid);<a name="line.1058"></a> -<span class="sourceLineNo">1059</span> if (procedure == null) {<a name="line.1059"></a> -<span class="sourceLineNo">1060</span> LOG.debug("Procedure pid={} does not exist, skipping bypass", pid);<a name="line.1060"></a> -<span class="sourceLineNo">1061</span> return false;<a name="line.1061"></a> -<span class="sourceLineNo">1062</span> }<a name="line.1062"></a> -<span class="sourceLineNo">1063</span><a name="line.1063"></a> -<span class="sourceLineNo">1064</span> LOG.debug("Begin bypass {} with lockWait={}, override={}, recursive={}",<a name="line.1064"></a> -<span class="sourceLineNo">1065</span> procedure, lockWait, override, recursive);<a name="line.1065"></a> -<span class="sourceLineNo">1066</span> IdLock.Entry lockEntry = procExecutionLock.tryLockEntry(procedure.getProcId(), lockWait);<a name="line.1066"></a> -<span class="sourceLineNo">1067</span> if (lockEntry == null && !override) {<a name="line.1067"></a> -<span class="sourceLineNo">1068</span> LOG.debug("Waited {} ms, but {} is still running, skipping bypass with force={}",<a name="line.1068"></a> -<span class="sourceLineNo">1069</span> lockWait, procedure, override);<a name="line.1069"></a> -<span class="sourceLineNo">1070</span> return false;<a name="line.1070"></a> -<span class="sourceLineNo">1071</span> } else if (lockEntry == null) {<a name="line.1071"></a> -<span class="sourceLineNo">1072</span> LOG.debug("Waited {} ms, but {} is still running, begin bypass with force={}",<a name="line.1072"></a> -<span class="sourceLineNo">1073</span> lockWait, procedure, override);<a name="line.1073"></a> -<span class="sourceLineNo">1074</span> }<a name="line.1074"></a> -<span class="sourceLineNo">1075</span> try {<a name="line.1075"></a> -<span class="sourceLineNo">1076</span> // check whether the procedure is already finished<a name="line.1076"></a> -<span class="sourceLineNo">1077</span> if (procedure.isFinished()) {<a name="line.1077"></a> -<span class="sourceLineNo">1078</span> LOG.debug("{} is already finished, skipping bypass", procedure);<a name="line.1078"></a> -<span class="sourceLineNo">1079</span> return false;<a name="line.1079"></a> -<span class="sourceLineNo">1080</span> }<a name="line.1080"></a> -<span class="sourceLineNo">1081</span><a name="line.1081"></a> -<span class="sourceLineNo">1082</span> if (procedure.hasChildren()) {<a name="line.1082"></a> -<span class="sourceLineNo">1083</span> if (recursive) {<a name="line.1083"></a> -<span class="sourceLineNo">1084</span> // EXPENSIVE. Checks each live procedure of which there could be many!!!<a name="line.1084"></a> -<span class="sourceLineNo">1085</span> // Is there another way to get children of a procedure?<a name="line.1085"></a> -<span class="sourceLineNo">1086</span> LOG.info("Recursive bypass on children of pid={}", procedure.getProcId());<a name="line.1086"></a> -<span class="sourceLineNo">1087</span> this.procedures.forEachValue(1 /*Single-threaded*/,<a name="line.1087"></a> -<span class="sourceLineNo">1088</span> // Transformer<a name="line.1088"></a> -<span class="sourceLineNo">1089</span> v -> v.getParentProcId() == procedure.getProcId()? v: null,<a name="line.1089"></a> -<span class="sourceLineNo">1090</span> // Consumer<a name="line.1090"></a> -<span class=
<TRUNCATED>