http://git-wip-us.apache.org/repos/asf/hbase/blob/ef7355e1/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.html index fc37c91..9c149ef 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/SplitLogManager.html @@ -407,441 +407,440 @@ <span class="sourceLineNo">399</span> * It removes recovering regions under /hbase/recovering-regions/[encoded region name] so that the<a name="line.399"></a> <span class="sourceLineNo">400</span> * region server hosting the region can allow reads to the recovered region<a name="line.400"></a> <span class="sourceLineNo">401</span> * @param serverNames servers which are just recovered<a name="line.401"></a> -<span class="sourceLineNo">402</span> * @param isMetaRecovery whether current recovery is for the meta region on<a name="line.402"></a> -<span class="sourceLineNo">403</span> * <code>serverNames<code><a name="line.403"></a> -<span class="sourceLineNo">404</span> */<a name="line.404"></a> -<span class="sourceLineNo">405</span> private void removeRecoveringRegions(final Set<ServerName> serverNames, Boolean isMetaRecovery) {<a name="line.405"></a> -<span class="sourceLineNo">406</span> if (!isLogReplaying()) {<a name="line.406"></a> -<span class="sourceLineNo">407</span> // the function is only used in WALEdit direct replay mode<a name="line.407"></a> -<span class="sourceLineNo">408</span> return;<a name="line.408"></a> -<span class="sourceLineNo">409</span> }<a name="line.409"></a> -<span class="sourceLineNo">410</span> if (serverNames == null || serverNames.isEmpty()) return;<a name="line.410"></a> -<span class="sourceLineNo">411</span><a name="line.411"></a> -<span class="sourceLineNo">412</span> Set<String> recoveredServerNameSet = new HashSet<String>();<a name="line.412"></a> -<span class="sourceLineNo">413</span> for (ServerName tmpServerName : serverNames) {<a name="line.413"></a> -<span class="sourceLineNo">414</span> recoveredServerNameSet.add(tmpServerName.getServerName());<a name="line.414"></a> -<span class="sourceLineNo">415</span> }<a name="line.415"></a> -<span class="sourceLineNo">416</span> <a name="line.416"></a> -<span class="sourceLineNo">417</span> this.recoveringRegionLock.lock();<a name="line.417"></a> -<span class="sourceLineNo">418</span> try {<a name="line.418"></a> -<span class="sourceLineNo">419</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.419"></a> -<span class="sourceLineNo">420</span> .getSplitLogManagerCoordination().removeRecoveringRegions(recoveredServerNameSet,<a name="line.420"></a> -<span class="sourceLineNo">421</span> isMetaRecovery);<a name="line.421"></a> -<span class="sourceLineNo">422</span> } catch (IOException e) {<a name="line.422"></a> -<span class="sourceLineNo">423</span> LOG.warn("removeRecoveringRegions got exception. Will retry", e);<a name="line.423"></a> -<span class="sourceLineNo">424</span> if (serverNames != null && !serverNames.isEmpty()) {<a name="line.424"></a> -<span class="sourceLineNo">425</span> this.failedRecoveringRegionDeletions.add(new Pair<Set<ServerName>, Boolean>(serverNames,<a name="line.425"></a> -<span class="sourceLineNo">426</span> isMetaRecovery));<a name="line.426"></a> -<span class="sourceLineNo">427</span> }<a name="line.427"></a> -<span class="sourceLineNo">428</span> } finally {<a name="line.428"></a> -<span class="sourceLineNo">429</span> this.recoveringRegionLock.unlock();<a name="line.429"></a> -<span class="sourceLineNo">430</span> }<a name="line.430"></a> -<span class="sourceLineNo">431</span> }<a name="line.431"></a> -<span class="sourceLineNo">432</span><a name="line.432"></a> -<span class="sourceLineNo">433</span> /**<a name="line.433"></a> -<span class="sourceLineNo">434</span> * It removes stale recovering regions under /hbase/recovering-regions/[encoded region name]<a name="line.434"></a> -<span class="sourceLineNo">435</span> * during master initialization phase.<a name="line.435"></a> -<span class="sourceLineNo">436</span> * @param failedServers A set of known failed servers<a name="line.436"></a> -<span class="sourceLineNo">437</span> * @throws IOException<a name="line.437"></a> -<span class="sourceLineNo">438</span> */<a name="line.438"></a> -<span class="sourceLineNo">439</span> void removeStaleRecoveringRegions(final Set<ServerName> failedServers) throws IOException,<a name="line.439"></a> -<span class="sourceLineNo">440</span> InterruptedIOException {<a name="line.440"></a> -<span class="sourceLineNo">441</span> Set<String> knownFailedServers = new HashSet<String>();<a name="line.441"></a> -<span class="sourceLineNo">442</span> if (failedServers != null) {<a name="line.442"></a> -<span class="sourceLineNo">443</span> for (ServerName tmpServerName : failedServers) {<a name="line.443"></a> -<span class="sourceLineNo">444</span> knownFailedServers.add(tmpServerName.getServerName());<a name="line.444"></a> -<span class="sourceLineNo">445</span> }<a name="line.445"></a> -<span class="sourceLineNo">446</span> }<a name="line.446"></a> -<span class="sourceLineNo">447</span><a name="line.447"></a> -<span class="sourceLineNo">448</span> this.recoveringRegionLock.lock();<a name="line.448"></a> -<span class="sourceLineNo">449</span> try {<a name="line.449"></a> -<span class="sourceLineNo">450</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.450"></a> -<span class="sourceLineNo">451</span> .getSplitLogManagerCoordination().removeStaleRecoveringRegions(knownFailedServers);<a name="line.451"></a> -<span class="sourceLineNo">452</span> } finally {<a name="line.452"></a> -<span class="sourceLineNo">453</span> this.recoveringRegionLock.unlock();<a name="line.453"></a> -<span class="sourceLineNo">454</span> }<a name="line.454"></a> -<span class="sourceLineNo">455</span> }<a name="line.455"></a> -<span class="sourceLineNo">456</span><a name="line.456"></a> -<span class="sourceLineNo">457</span> /**<a name="line.457"></a> -<span class="sourceLineNo">458</span> * @param path<a name="line.458"></a> -<span class="sourceLineNo">459</span> * @param batch<a name="line.459"></a> -<span class="sourceLineNo">460</span> * @return null on success, existing task on error<a name="line.460"></a> -<span class="sourceLineNo">461</span> */<a name="line.461"></a> -<span class="sourceLineNo">462</span> private Task createTaskIfAbsent(String path, TaskBatch batch) {<a name="line.462"></a> -<span class="sourceLineNo">463</span> Task oldtask;<a name="line.463"></a> -<span class="sourceLineNo">464</span> // batch.installed is only changed via this function and<a name="line.464"></a> -<span class="sourceLineNo">465</span> // a single thread touches batch.installed.<a name="line.465"></a> -<span class="sourceLineNo">466</span> Task newtask = new Task();<a name="line.466"></a> -<span class="sourceLineNo">467</span> newtask.batch = batch;<a name="line.467"></a> -<span class="sourceLineNo">468</span> oldtask = tasks.putIfAbsent(path, newtask);<a name="line.468"></a> -<span class="sourceLineNo">469</span> if (oldtask == null) {<a name="line.469"></a> -<span class="sourceLineNo">470</span> batch.installed++;<a name="line.470"></a> -<span class="sourceLineNo">471</span> return null;<a name="line.471"></a> -<span class="sourceLineNo">472</span> }<a name="line.472"></a> -<span class="sourceLineNo">473</span> // new task was not used.<a name="line.473"></a> -<span class="sourceLineNo">474</span> synchronized (oldtask) {<a name="line.474"></a> -<span class="sourceLineNo">475</span> if (oldtask.isOrphan()) {<a name="line.475"></a> -<span class="sourceLineNo">476</span> if (oldtask.status == SUCCESS) {<a name="line.476"></a> -<span class="sourceLineNo">477</span> // The task is already done. Do not install the batch for this<a name="line.477"></a> -<span class="sourceLineNo">478</span> // task because it might be too late for setDone() to update<a name="line.478"></a> -<span class="sourceLineNo">479</span> // batch.done. There is no need for the batch creator to wait for<a name="line.479"></a> -<span class="sourceLineNo">480</span> // this task to complete.<a name="line.480"></a> -<span class="sourceLineNo">481</span> return (null);<a name="line.481"></a> -<span class="sourceLineNo">482</span> }<a name="line.482"></a> -<span class="sourceLineNo">483</span> if (oldtask.status == IN_PROGRESS) {<a name="line.483"></a> -<span class="sourceLineNo">484</span> oldtask.batch = batch;<a name="line.484"></a> -<span class="sourceLineNo">485</span> batch.installed++;<a name="line.485"></a> -<span class="sourceLineNo">486</span> LOG.debug("Previously orphan task " + path + " is now being waited upon");<a name="line.486"></a> -<span class="sourceLineNo">487</span> return null;<a name="line.487"></a> -<span class="sourceLineNo">488</span> }<a name="line.488"></a> -<span class="sourceLineNo">489</span> while (oldtask.status == FAILURE) {<a name="line.489"></a> -<span class="sourceLineNo">490</span> LOG.debug("wait for status of task " + path + " to change to DELETED");<a name="line.490"></a> -<span class="sourceLineNo">491</span> SplitLogCounters.tot_mgr_wait_for_zk_delete.incrementAndGet();<a name="line.491"></a> -<span class="sourceLineNo">492</span> try {<a name="line.492"></a> -<span class="sourceLineNo">493</span> oldtask.wait();<a name="line.493"></a> -<span class="sourceLineNo">494</span> } catch (InterruptedException e) {<a name="line.494"></a> -<span class="sourceLineNo">495</span> Thread.currentThread().interrupt();<a name="line.495"></a> -<span class="sourceLineNo">496</span> LOG.warn("Interrupted when waiting for znode delete callback");<a name="line.496"></a> -<span class="sourceLineNo">497</span> // fall through to return failure<a name="line.497"></a> -<span class="sourceLineNo">498</span> break;<a name="line.498"></a> -<span class="sourceLineNo">499</span> }<a name="line.499"></a> -<span class="sourceLineNo">500</span> }<a name="line.500"></a> -<span class="sourceLineNo">501</span> if (oldtask.status != DELETED) {<a name="line.501"></a> -<span class="sourceLineNo">502</span> LOG.warn("Failure because previously failed task"<a name="line.502"></a> -<span class="sourceLineNo">503</span> + " state still present. Waiting for znode delete callback" + " path=" + path);<a name="line.503"></a> -<span class="sourceLineNo">504</span> return oldtask;<a name="line.504"></a> -<span class="sourceLineNo">505</span> }<a name="line.505"></a> -<span class="sourceLineNo">506</span> // reinsert the newTask and it must succeed this time<a name="line.506"></a> -<span class="sourceLineNo">507</span> Task t = tasks.putIfAbsent(path, newtask);<a name="line.507"></a> -<span class="sourceLineNo">508</span> if (t == null) {<a name="line.508"></a> -<span class="sourceLineNo">509</span> batch.installed++;<a name="line.509"></a> -<span class="sourceLineNo">510</span> return null;<a name="line.510"></a> -<span class="sourceLineNo">511</span> }<a name="line.511"></a> -<span class="sourceLineNo">512</span> LOG.fatal("Logic error. Deleted task still present in tasks map");<a name="line.512"></a> -<span class="sourceLineNo">513</span> assert false : "Deleted task still present in tasks map";<a name="line.513"></a> -<span class="sourceLineNo">514</span> return t;<a name="line.514"></a> -<span class="sourceLineNo">515</span> }<a name="line.515"></a> -<span class="sourceLineNo">516</span> LOG.warn("Failure because two threads can't wait for the same task; path=" + path);<a name="line.516"></a> -<span class="sourceLineNo">517</span> return oldtask;<a name="line.517"></a> -<span class="sourceLineNo">518</span> }<a name="line.518"></a> -<span class="sourceLineNo">519</span> }<a name="line.519"></a> -<span class="sourceLineNo">520</span><a name="line.520"></a> -<span class="sourceLineNo">521</span> Task findOrCreateOrphanTask(String path) {<a name="line.521"></a> -<span class="sourceLineNo">522</span> Task orphanTask = new Task();<a name="line.522"></a> -<span class="sourceLineNo">523</span> Task task;<a name="line.523"></a> -<span class="sourceLineNo">524</span> task = tasks.putIfAbsent(path, orphanTask);<a name="line.524"></a> -<span class="sourceLineNo">525</span> if (task == null) {<a name="line.525"></a> -<span class="sourceLineNo">526</span> LOG.info("creating orphan task " + path);<a name="line.526"></a> -<span class="sourceLineNo">527</span> SplitLogCounters.tot_mgr_orphan_task_acquired.incrementAndGet();<a name="line.527"></a> -<span class="sourceLineNo">528</span> task = orphanTask;<a name="line.528"></a> -<span class="sourceLineNo">529</span> }<a name="line.529"></a> -<span class="sourceLineNo">530</span> return task;<a name="line.530"></a> -<span class="sourceLineNo">531</span> }<a name="line.531"></a> -<span class="sourceLineNo">532</span><a name="line.532"></a> -<span class="sourceLineNo">533</span> public void stop() {<a name="line.533"></a> -<span class="sourceLineNo">534</span> if (choreService != null) {<a name="line.534"></a> -<span class="sourceLineNo">535</span> choreService.shutdown();<a name="line.535"></a> -<span class="sourceLineNo">536</span> }<a name="line.536"></a> -<span class="sourceLineNo">537</span> if (timeoutMonitor != null) {<a name="line.537"></a> -<span class="sourceLineNo">538</span> timeoutMonitor.cancel(true);<a name="line.538"></a> -<span class="sourceLineNo">539</span> }<a name="line.539"></a> -<span class="sourceLineNo">540</span> }<a name="line.540"></a> -<span class="sourceLineNo">541</span><a name="line.541"></a> -<span class="sourceLineNo">542</span> void handleDeadWorker(ServerName workerName) {<a name="line.542"></a> -<span class="sourceLineNo">543</span> // resubmit the tasks on the TimeoutMonitor thread. Makes it easier<a name="line.543"></a> -<span class="sourceLineNo">544</span> // to reason about concurrency. Makes it easier to retry.<a name="line.544"></a> -<span class="sourceLineNo">545</span> synchronized (deadWorkersLock) {<a name="line.545"></a> -<span class="sourceLineNo">546</span> if (deadWorkers == null) {<a name="line.546"></a> -<span class="sourceLineNo">547</span> deadWorkers = new HashSet<ServerName>(100);<a name="line.547"></a> -<span class="sourceLineNo">548</span> }<a name="line.548"></a> -<span class="sourceLineNo">549</span> deadWorkers.add(workerName);<a name="line.549"></a> -<span class="sourceLineNo">550</span> }<a name="line.550"></a> -<span class="sourceLineNo">551</span> LOG.info("dead splitlog worker " + workerName);<a name="line.551"></a> -<span class="sourceLineNo">552</span> }<a name="line.552"></a> -<span class="sourceLineNo">553</span><a name="line.553"></a> -<span class="sourceLineNo">554</span> void handleDeadWorkers(Set<ServerName> serverNames) {<a name="line.554"></a> -<span class="sourceLineNo">555</span> synchronized (deadWorkersLock) {<a name="line.555"></a> -<span class="sourceLineNo">556</span> if (deadWorkers == null) {<a name="line.556"></a> -<span class="sourceLineNo">557</span> deadWorkers = new HashSet<ServerName>(100);<a name="line.557"></a> -<span class="sourceLineNo">558</span> }<a name="line.558"></a> -<span class="sourceLineNo">559</span> deadWorkers.addAll(serverNames);<a name="line.559"></a> -<span class="sourceLineNo">560</span> }<a name="line.560"></a> -<span class="sourceLineNo">561</span> LOG.info("dead splitlog workers " + serverNames);<a name="line.561"></a> -<span class="sourceLineNo">562</span> }<a name="line.562"></a> -<span class="sourceLineNo">563</span><a name="line.563"></a> -<span class="sourceLineNo">564</span> /**<a name="line.564"></a> -<span class="sourceLineNo">565</span> * This function is to set recovery mode from outstanding split log tasks from before or current<a name="line.565"></a> -<span class="sourceLineNo">566</span> * configuration setting<a name="line.566"></a> -<span class="sourceLineNo">567</span> * @param isForInitialization<a name="line.567"></a> -<span class="sourceLineNo">568</span> * @throws IOException throws if it's impossible to set recovery mode<a name="line.568"></a> -<span class="sourceLineNo">569</span> */<a name="line.569"></a> -<span class="sourceLineNo">570</span> public void setRecoveryMode(boolean isForInitialization) throws IOException {<a name="line.570"></a> -<span class="sourceLineNo">571</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.571"></a> -<span class="sourceLineNo">572</span> .getSplitLogManagerCoordination().setRecoveryMode(isForInitialization);<a name="line.572"></a> -<span class="sourceLineNo">573</span><a name="line.573"></a> -<span class="sourceLineNo">574</span> }<a name="line.574"></a> -<span class="sourceLineNo">575</span><a name="line.575"></a> -<span class="sourceLineNo">576</span> public void markRegionsRecovering(ServerName server, Set<HRegionInfo> userRegions)<a name="line.576"></a> -<span class="sourceLineNo">577</span> throws InterruptedIOException, IOException {<a name="line.577"></a> -<span class="sourceLineNo">578</span> if (userRegions == null || (!isLogReplaying())) {<a name="line.578"></a> -<span class="sourceLineNo">579</span> return;<a name="line.579"></a> -<span class="sourceLineNo">580</span> }<a name="line.580"></a> -<span class="sourceLineNo">581</span> try {<a name="line.581"></a> -<span class="sourceLineNo">582</span> this.recoveringRegionLock.lock();<a name="line.582"></a> -<span class="sourceLineNo">583</span> // mark that we're creating recovering regions<a name="line.583"></a> -<span class="sourceLineNo">584</span> ((BaseCoordinatedStateManager) this.server.getCoordinatedStateManager())<a name="line.584"></a> -<span class="sourceLineNo">585</span> .getSplitLogManagerCoordination().markRegionsRecovering(server, userRegions);<a name="line.585"></a> -<span class="sourceLineNo">586</span> } finally {<a name="line.586"></a> -<span class="sourceLineNo">587</span> this.recoveringRegionLock.unlock();<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> }<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> * @return whether log is replaying<a name="line.593"></a> -<span class="sourceLineNo">594</span> */<a name="line.594"></a> -<span class="sourceLineNo">595</span> public boolean isLogReplaying() {<a name="line.595"></a> -<span class="sourceLineNo">596</span> CoordinatedStateManager m = server.getCoordinatedStateManager();<a name="line.596"></a> -<span class="sourceLineNo">597</span> if (m == null) return false;<a name="line.597"></a> -<span class="sourceLineNo">598</span> return ((BaseCoordinatedStateManager)m).getSplitLogManagerCoordination().isReplaying();<a name="line.598"></a> -<span class="sourceLineNo">599</span> }<a name="line.599"></a> -<span class="sourceLineNo">600</span><a name="line.600"></a> -<span class="sourceLineNo">601</span> /**<a name="line.601"></a> -<span class="sourceLineNo">602</span> * @return whether log is splitting<a name="line.602"></a> -<span class="sourceLineNo">603</span> */<a name="line.603"></a> -<span class="sourceLineNo">604</span> public boolean isLogSplitting() {<a name="line.604"></a> -<span class="sourceLineNo">605</span> if (server.getCoordinatedStateManager() == null) return false;<a name="line.605"></a> -<span class="sourceLineNo">606</span> return ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.606"></a> -<span class="sourceLineNo">607</span> .getSplitLogManagerCoordination().isSplitting();<a name="line.607"></a> -<span class="sourceLineNo">608</span> }<a name="line.608"></a> -<span class="sourceLineNo">609</span><a name="line.609"></a> -<span class="sourceLineNo">610</span> /**<a name="line.610"></a> -<span class="sourceLineNo">611</span> * @return the current log recovery mode<a name="line.611"></a> -<span class="sourceLineNo">612</span> */<a name="line.612"></a> -<span class="sourceLineNo">613</span> public RecoveryMode getRecoveryMode() {<a name="line.613"></a> -<span class="sourceLineNo">614</span> return ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.614"></a> -<span class="sourceLineNo">615</span> .getSplitLogManagerCoordination().getRecoveryMode();<a name="line.615"></a> -<span class="sourceLineNo">616</span> }<a name="line.616"></a> -<span class="sourceLineNo">617</span><a name="line.617"></a> -<span class="sourceLineNo">618</span> /**<a name="line.618"></a> -<span class="sourceLineNo">619</span> * Keeps track of the batch of tasks submitted together by a caller in splitLogDistributed().<a name="line.619"></a> -<span class="sourceLineNo">620</span> * Clients threads use this object to wait for all their tasks to be done.<a name="line.620"></a> -<span class="sourceLineNo">621</span> * <p><a name="line.621"></a> -<span class="sourceLineNo">622</span> * All access is synchronized.<a name="line.622"></a> -<span class="sourceLineNo">623</span> */<a name="line.623"></a> -<span class="sourceLineNo">624</span> @InterfaceAudience.Private<a name="line.624"></a> -<span class="sourceLineNo">625</span> public static class TaskBatch {<a name="line.625"></a> -<span class="sourceLineNo">626</span> public int installed = 0;<a name="line.626"></a> -<span class="sourceLineNo">627</span> public int done = 0;<a name="line.627"></a> -<span class="sourceLineNo">628</span> public int error = 0;<a name="line.628"></a> -<span class="sourceLineNo">629</span> public volatile boolean isDead = false;<a name="line.629"></a> -<span class="sourceLineNo">630</span><a name="line.630"></a> -<span class="sourceLineNo">631</span> @Override<a name="line.631"></a> -<span class="sourceLineNo">632</span> public String toString() {<a name="line.632"></a> -<span class="sourceLineNo">633</span> return ("installed = " + installed + " done = " + done + " error = " + error);<a name="line.633"></a> -<span class="sourceLineNo">634</span> }<a name="line.634"></a> -<span class="sourceLineNo">635</span> }<a name="line.635"></a> -<span class="sourceLineNo">636</span><a name="line.636"></a> -<span class="sourceLineNo">637</span> /**<a name="line.637"></a> -<span class="sourceLineNo">638</span> * in memory state of an active task.<a name="line.638"></a> -<span class="sourceLineNo">639</span> */<a name="line.639"></a> -<span class="sourceLineNo">640</span> @InterfaceAudience.Private<a name="line.640"></a> -<span class="sourceLineNo">641</span> public static class Task {<a name="line.641"></a> -<span class="sourceLineNo">642</span> public volatile long last_update;<a name="line.642"></a> -<span class="sourceLineNo">643</span> public volatile int last_version;<a name="line.643"></a> -<span class="sourceLineNo">644</span> public volatile ServerName cur_worker_name;<a name="line.644"></a> -<span class="sourceLineNo">645</span> public volatile TaskBatch batch;<a name="line.645"></a> -<span class="sourceLineNo">646</span> public volatile TerminationStatus status;<a name="line.646"></a> -<span class="sourceLineNo">647</span> public volatile int incarnation;<a name="line.647"></a> -<span class="sourceLineNo">648</span> public final AtomicInteger unforcedResubmits = new AtomicInteger();<a name="line.648"></a> -<span class="sourceLineNo">649</span> public volatile boolean resubmitThresholdReached;<a name="line.649"></a> -<span class="sourceLineNo">650</span><a name="line.650"></a> -<span class="sourceLineNo">651</span> @Override<a name="line.651"></a> -<span class="sourceLineNo">652</span> public String toString() {<a name="line.652"></a> -<span class="sourceLineNo">653</span> return ("last_update = " + last_update + " last_version = " + last_version<a name="line.653"></a> -<span class="sourceLineNo">654</span> + " cur_worker_name = " + cur_worker_name + " status = " + status + " incarnation = "<a name="line.654"></a> -<span class="sourceLineNo">655</span> + incarnation + " resubmits = " + unforcedResubmits.get() + " batch = " + batch);<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> public Task() {<a name="line.658"></a> -<span class="sourceLineNo">659</span> incarnation = 0;<a name="line.659"></a> -<span class="sourceLineNo">660</span> last_version = -1;<a name="line.660"></a> -<span class="sourceLineNo">661</span> status = IN_PROGRESS;<a name="line.661"></a> -<span class="sourceLineNo">662</span> setUnassigned();<a name="line.662"></a> -<span class="sourceLineNo">663</span> }<a name="line.663"></a> -<span class="sourceLineNo">664</span><a name="line.664"></a> -<span class="sourceLineNo">665</span> public boolean isOrphan() {<a name="line.665"></a> -<span class="sourceLineNo">666</span> return (batch == null || batch.isDead);<a name="line.666"></a> -<span class="sourceLineNo">667</span> }<a name="line.667"></a> -<span class="sourceLineNo">668</span><a name="line.668"></a> -<span class="sourceLineNo">669</span> public boolean isUnassigned() {<a name="line.669"></a> -<span class="sourceLineNo">670</span> return (cur_worker_name == null);<a name="line.670"></a> -<span class="sourceLineNo">671</span> }<a name="line.671"></a> -<span class="sourceLineNo">672</span><a name="line.672"></a> -<span class="sourceLineNo">673</span> public void heartbeatNoDetails(long time) {<a name="line.673"></a> -<span class="sourceLineNo">674</span> last_update = time;<a name="line.674"></a> -<span class="sourceLineNo">675</span> }<a name="line.675"></a> -<span class="sourceLineNo">676</span><a name="line.676"></a> -<span class="sourceLineNo">677</span> public void heartbeat(long time, int version, ServerName worker) {<a name="line.677"></a> -<span class="sourceLineNo">678</span> last_version = version;<a name="line.678"></a> -<span class="sourceLineNo">679</span> last_update = time;<a name="line.679"></a> -<span class="sourceLineNo">680</span> cur_worker_name = worker;<a name="line.680"></a> -<span class="sourceLineNo">681</span> }<a name="line.681"></a> -<span class="sourceLineNo">682</span><a name="line.682"></a> -<span class="sourceLineNo">683</span> public void setUnassigned() {<a name="line.683"></a> -<span class="sourceLineNo">684</span> cur_worker_name = null;<a name="line.684"></a> -<span class="sourceLineNo">685</span> last_update = -1;<a name="line.685"></a> -<span class="sourceLineNo">686</span> }<a name="line.686"></a> -<span class="sourceLineNo">687</span> }<a name="line.687"></a> -<span class="sourceLineNo">688</span><a name="line.688"></a> -<span class="sourceLineNo">689</span> /**<a name="line.689"></a> -<span class="sourceLineNo">690</span> * Periodically checks all active tasks and resubmits the ones that have timed out<a name="line.690"></a> -<span class="sourceLineNo">691</span> */<a name="line.691"></a> -<span class="sourceLineNo">692</span> private class TimeoutMonitor extends ScheduledChore {<a name="line.692"></a> -<span class="sourceLineNo">693</span> private long lastLog = 0;<a name="line.693"></a> -<span class="sourceLineNo">694</span><a name="line.694"></a> -<span class="sourceLineNo">695</span> public TimeoutMonitor(final int period, Stoppable stopper) {<a name="line.695"></a> -<span class="sourceLineNo">696</span> super("SplitLogManager Timeout Monitor", stopper, period);<a name="line.696"></a> -<span class="sourceLineNo">697</span> }<a name="line.697"></a> -<span class="sourceLineNo">698</span><a name="line.698"></a> -<span class="sourceLineNo">699</span> @Override<a name="line.699"></a> -<span class="sourceLineNo">700</span> protected void chore() {<a name="line.700"></a> -<span class="sourceLineNo">701</span> int resubmitted = 0;<a name="line.701"></a> -<span class="sourceLineNo">702</span> int unassigned = 0;<a name="line.702"></a> -<span class="sourceLineNo">703</span> int tot = 0;<a name="line.703"></a> -<span class="sourceLineNo">704</span> boolean found_assigned_task = false;<a name="line.704"></a> -<span class="sourceLineNo">705</span> Set<ServerName> localDeadWorkers;<a name="line.705"></a> -<span class="sourceLineNo">706</span><a name="line.706"></a> -<span class="sourceLineNo">707</span> synchronized (deadWorkersLock) {<a name="line.707"></a> -<span class="sourceLineNo">708</span> localDeadWorkers = deadWorkers;<a name="line.708"></a> -<span class="sourceLineNo">709</span> deadWorkers = null;<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> for (Map.Entry<String, Task> e : tasks.entrySet()) {<a name="line.712"></a> -<span class="sourceLineNo">713</span> String path = e.getKey();<a name="line.713"></a> -<span class="sourceLineNo">714</span> Task task = e.getValue();<a name="line.714"></a> -<span class="sourceLineNo">715</span> ServerName cur_worker = task.cur_worker_name;<a name="line.715"></a> -<span class="sourceLineNo">716</span> tot++;<a name="line.716"></a> -<span class="sourceLineNo">717</span> // don't easily resubmit a task which hasn't been picked up yet. It<a name="line.717"></a> -<span class="sourceLineNo">718</span> // might be a long while before a SplitLogWorker is free to pick up a<a name="line.718"></a> -<span class="sourceLineNo">719</span> // task. This is because a SplitLogWorker picks up a task one at a<a name="line.719"></a> -<span class="sourceLineNo">720</span> // time. If we want progress when there are no region servers then we<a name="line.720"></a> -<span class="sourceLineNo">721</span> // will have to run a SplitLogWorker thread in the Master.<a name="line.721"></a> -<span class="sourceLineNo">722</span> if (task.isUnassigned()) {<a name="line.722"></a> -<span class="sourceLineNo">723</span> unassigned++;<a name="line.723"></a> -<span class="sourceLineNo">724</span> continue;<a name="line.724"></a> -<span class="sourceLineNo">725</span> }<a name="line.725"></a> -<span class="sourceLineNo">726</span> found_assigned_task = true;<a name="line.726"></a> -<span class="sourceLineNo">727</span> if (localDeadWorkers != null && localDeadWorkers.contains(cur_worker)) {<a name="line.727"></a> -<span class="sourceLineNo">728</span> SplitLogCounters.tot_mgr_resubmit_dead_server_task.incrementAndGet();<a name="line.728"></a> -<span class="sourceLineNo">729</span> if (((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.729"></a> -<span class="sourceLineNo">730</span> .getSplitLogManagerCoordination().resubmitTask(path, task, FORCE)) {<a name="line.730"></a> -<span class="sourceLineNo">731</span> resubmitted++;<a name="line.731"></a> -<span class="sourceLineNo">732</span> } else {<a name="line.732"></a> -<span class="sourceLineNo">733</span> handleDeadWorker(cur_worker);<a name="line.733"></a> -<span class="sourceLineNo">734</span> LOG.warn("Failed to resubmit task " + path + " owned by dead " + cur_worker<a name="line.734"></a> -<span class="sourceLineNo">735</span> + ", will retry.");<a name="line.735"></a> -<span class="sourceLineNo">736</span> }<a name="line.736"></a> -<span class="sourceLineNo">737</span> } else if (((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.737"></a> -<span class="sourceLineNo">738</span> .getSplitLogManagerCoordination().resubmitTask(path, task, CHECK)) {<a name="line.738"></a> -<span class="sourceLineNo">739</span> resubmitted++;<a name="line.739"></a> -<span class="sourceLineNo">740</span> }<a name="line.740"></a> -<span class="sourceLineNo">741</span> }<a name="line.741"></a> -<span class="sourceLineNo">742</span> if (tot > 0) {<a name="line.742"></a> -<span class="sourceLineNo">743</span> long now = EnvironmentEdgeManager.currentTime();<a name="line.743"></a> -<span class="sourceLineNo">744</span> if (now > lastLog + 5000) {<a name="line.744"></a> -<span class="sourceLineNo">745</span> lastLog = now;<a name="line.745"></a> -<span class="sourceLineNo">746</span> LOG.info("total tasks = " + tot + " unassigned = " + unassigned + " tasks=" + tasks);<a name="line.746"></a> -<span class="sourceLineNo">747</span> }<a name="line.747"></a> -<span class="sourceLineNo">748</span> }<a name="line.748"></a> -<span class="sourceLineNo">749</span> if (resubmitted > 0) {<a name="line.749"></a> -<span class="sourceLineNo">750</span> LOG.info("resubmitted " + resubmitted + " out of " + tot + " tasks");<a name="line.750"></a> -<span class="sourceLineNo">751</span> }<a name="line.751"></a> -<span class="sourceLineNo">752</span> // If there are pending tasks and all of them have been unassigned for<a name="line.752"></a> -<span class="sourceLineNo">753</span> // some time then put up a RESCAN node to ping the workers.<a name="line.753"></a> -<span class="sourceLineNo">754</span> // ZKSplitlog.DEFAULT_UNASSIGNED_TIMEOUT is of the order of minutes<a name="line.754"></a> -<span class="sourceLineNo">755</span> // because a. it is very unlikely that every worker had a<a name="line.755"></a> -<span class="sourceLineNo">756</span> // transient error when trying to grab the task b. if there are no<a name="line.756"></a> -<span class="sourceLineNo">757</span> // workers then all tasks wills stay unassigned indefinitely and the<a name="line.757"></a> -<span class="sourceLineNo">758</span> // manager will be indefinitely creating RESCAN nodes. TODO may be the<a name="line.758"></a> -<span class="sourceLineNo">759</span> // master should spawn both a manager and a worker thread to guarantee<a name="line.759"></a> -<span class="sourceLineNo">760</span> // that there is always one worker in the system<a name="line.760"></a> -<span class="sourceLineNo">761</span> if (tot > 0<a name="line.761"></a> -<span class="sourceLineNo">762</span> && !found_assigned_task<a name="line.762"></a> -<span class="sourceLineNo">763</span> && ((EnvironmentEdgeManager.currentTime() - lastTaskCreateTime) > unassignedTimeout)) {<a name="line.763"></a> -<span class="sourceLineNo">764</span> for (Map.Entry<String, Task> e : tasks.entrySet()) {<a name="line.764"></a> -<span class="sourceLineNo">765</span> String key = e.getKey();<a name="line.765"></a> -<span class="sourceLineNo">766</span> Task task = e.getValue();<a name="line.766"></a> -<span class="sourceLineNo">767</span> // we have to do task.isUnassigned() check again because tasks might<a name="line.767"></a> -<span class="sourceLineNo">768</span> // have been asynchronously assigned. There is no locking required<a name="line.768"></a> -<span class="sourceLineNo">769</span> // for these checks ... it is OK even if tryGetDataSetWatch() is<a name="line.769"></a> -<span class="sourceLineNo">770</span> // called unnecessarily for a taskpath<a name="line.770"></a> -<span class="sourceLineNo">771</span> if (task.isUnassigned() && (task.status != FAILURE)) {<a name="line.771"></a> -<span class="sourceLineNo">772</span> // We just touch the znode to make sure its still there<a name="line.772"></a> -<span class="sourceLineNo">773</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.773"></a> -<span class="sourceLineNo">774</span> .getSplitLogManagerCoordination().checkTaskStillAvailable(key);<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> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.777"></a> -<span class="sourceLineNo">778</span> .getSplitLogManagerCoordination().checkTasks();<a name="line.778"></a> -<span class="sourceLineNo">779</span> SplitLogCounters.tot_mgr_resubmit_unassigned.incrementAndGet();<a name="line.779"></a> -<span class="sourceLineNo">780</span> LOG.debug("resubmitting unassigned task(s) after timeout");<a name="line.780"></a> -<span class="sourceLineNo">781</span> }<a name="line.781"></a> -<span class="sourceLineNo">782</span> Set<String> failedDeletions =<a name="line.782"></a> -<span class="sourceLineNo">783</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.783"></a> -<span class="sourceLineNo">784</span> .getSplitLogManagerCoordination().getDetails().getFailedDeletions();<a name="line.784"></a> -<span class="sourceLineNo">785</span> // Retry previously failed deletes<a name="line.785"></a> -<span class="sourceLineNo">786</span> if (failedDeletions.size() > 0) {<a name="line.786"></a> -<span class="sourceLineNo">787</span> List<String> tmpPaths = new ArrayList<String>(failedDeletions);<a name="line.787"></a> -<span class="sourceLineNo">788</span> for (String tmpPath : tmpPaths) {<a name="line.788"></a> -<span class="sourceLineNo">789</span> // deleteNode is an async call<a name="line.789"></a> -<span class="sourceLineNo">790</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.790"></a> -<span class="sourceLineNo">791</span> .getSplitLogManagerCoordination().deleteTask(tmpPath);<a name="line.791"></a> -<span class="sourceLineNo">792</span> }<a name="line.792"></a> -<span class="sourceLineNo">793</span> failedDeletions.removeAll(tmpPaths);<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> // Garbage collect left-over<a name="line.796"></a> -<span class="sourceLineNo">797</span> long timeInterval =<a name="line.797"></a> -<span class="sourceLineNo">798</span> EnvironmentEdgeManager.currentTime()<a name="line.798"></a> -<span class="sourceLineNo">799</span> - ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.799"></a> -<span class="sourceLineNo">800</span> .getSplitLogManagerCoordination().getLastRecoveryTime();<a name="line.800"></a> -<span class="sourceLineNo">801</span> if (!failedRecoveringRegionDeletions.isEmpty()<a name="line.801"></a> -<span class="sourceLineNo">802</span> || (tot == 0 && tasks.size() == 0 && (timeInterval > checkRecoveringTimeThreshold))) {<a name="line.802"></a> -<span class="sourceLineNo">803</span> // inside the function there have more checks before GC anything<a name="line.803"></a> -<span class="sourceLineNo">804</span> if (!failedRecoveringRegionDeletions.isEmpty()) {<a name="line.804"></a> -<span class="sourceLineNo">805</span> List<Pair<Set<ServerName>, Boolean>> previouslyFailedDeletions =<a name="line.805"></a> -<span class="sourceLineNo">806</span> new ArrayList<Pair<Set<ServerName>, Boolean>>(failedRecoveringRegionDeletions);<a name="line.806"></a> -<span class="sourceLineNo">807</span> failedRecoveringRegionDeletions.removeAll(previouslyFailedDeletions);<a name="line.807"></a> -<span class="sourceLineNo">808</span> for (Pair<Set<ServerName>, Boolean> failedDeletion : previouslyFailedDeletions) {<a name="line.808"></a> -<span class="sourceLineNo">809</span> removeRecoveringRegions(failedDeletion.getFirst(), failedDeletion.getSecond());<a name="line.809"></a> -<span class="sourceLineNo">810</span> }<a name="line.810"></a> -<span class="sourceLineNo">811</span> } else {<a name="line.811"></a> -<span class="sourceLineNo">812</span> removeRecoveringRegions(null, null);<a name="line.812"></a> -<span class="sourceLineNo">813</span> }<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> }<a name="line.816"></a> -<span class="sourceLineNo">817</span><a name="line.817"></a> -<span class="sourceLineNo">818</span> public enum ResubmitDirective {<a name="line.818"></a> -<span class="sourceLineNo">819</span> CHECK(), FORCE();<a name="line.819"></a> -<span class="sourceLineNo">820</span> }<a name="line.820"></a> -<span class="sourceLineNo">821</span><a name="line.821"></a> -<span class="sourceLineNo">822</span> public enum TerminationStatus {<a name="line.822"></a> -<span class="sourceLineNo">823</span> IN_PROGRESS("in_progress"), SUCCESS("success"), FAILURE("failure"), DELETED("deleted");<a name="line.823"></a> -<span class="sourceLineNo">824</span><a name="line.824"></a> -<span class="sourceLineNo">825</span> String statusMsg;<a name="line.825"></a> -<span class="sourceLineNo">826</span><a name="line.826"></a> -<span class="sourceLineNo">827</span> TerminationStatus(String msg) {<a name="line.827"></a> -<span class="sourceLineNo">828</span> statusMsg = msg;<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> @Override<a name="line.831"></a> -<span class="sourceLineNo">832</span> public String toString() {<a name="line.832"></a> -<span class="sourceLineNo">833</span> return statusMsg;<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>}<a name="line.836"></a> +<span class="sourceLineNo">402</span> * @param isMetaRecovery whether current recovery is for the meta region on {@code serverNames}<a name="line.402"></a> +<span class="sourceLineNo">403</span> */<a name="line.403"></a> +<span class="sourceLineNo">404</span> private void removeRecoveringRegions(final Set<ServerName> serverNames, Boolean isMetaRecovery) {<a name="line.404"></a> +<span class="sourceLineNo">405</span> if (!isLogReplaying()) {<a name="line.405"></a> +<span class="sourceLineNo">406</span> // the function is only used in WALEdit direct replay mode<a name="line.406"></a> +<span class="sourceLineNo">407</span> return;<a name="line.407"></a> +<span class="sourceLineNo">408</span> }<a name="line.408"></a> +<span class="sourceLineNo">409</span> if (serverNames == null || serverNames.isEmpty()) return;<a name="line.409"></a> +<span class="sourceLineNo">410</span><a name="line.410"></a> +<span class="sourceLineNo">411</span> Set<String> recoveredServerNameSet = new HashSet<String>();<a name="line.411"></a> +<span class="sourceLineNo">412</span> for (ServerName tmpServerName : serverNames) {<a name="line.412"></a> +<span class="sourceLineNo">413</span> recoveredServerNameSet.add(tmpServerName.getServerName());<a name="line.413"></a> +<span class="sourceLineNo">414</span> }<a name="line.414"></a> +<span class="sourceLineNo">415</span> <a name="line.415"></a> +<span class="sourceLineNo">416</span> this.recoveringRegionLock.lock();<a name="line.416"></a> +<span class="sourceLineNo">417</span> try {<a name="line.417"></a> +<span class="sourceLineNo">418</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.418"></a> +<span class="sourceLineNo">419</span> .getSplitLogManagerCoordination().removeRecoveringRegions(recoveredServerNameSet,<a name="line.419"></a> +<span class="sourceLineNo">420</span> isMetaRecovery);<a name="line.420"></a> +<span class="sourceLineNo">421</span> } catch (IOException e) {<a name="line.421"></a> +<span class="sourceLineNo">422</span> LOG.warn("removeRecoveringRegions got exception. Will retry", e);<a name="line.422"></a> +<span class="sourceLineNo">423</span> if (serverNames != null && !serverNames.isEmpty()) {<a name="line.423"></a> +<span class="sourceLineNo">424</span> this.failedRecoveringRegionDeletions.add(new Pair<Set<ServerName>, Boolean>(serverNames,<a name="line.424"></a> +<span class="sourceLineNo">425</span> isMetaRecovery));<a name="line.425"></a> +<span class="sourceLineNo">426</span> }<a name="line.426"></a> +<span class="sourceLineNo">427</span> } finally {<a name="line.427"></a> +<span class="sourceLineNo">428</span> this.recoveringRegionLock.unlock();<a name="line.428"></a> +<span class="sourceLineNo">429</span> }<a name="line.429"></a> +<span class="sourceLineNo">430</span> }<a name="line.430"></a> +<span class="sourceLineNo">431</span><a name="line.431"></a> +<span class="sourceLineNo">432</span> /**<a name="line.432"></a> +<span class="sourceLineNo">433</span> * It removes stale recovering regions under /hbase/recovering-regions/[encoded region name]<a name="line.433"></a> +<span class="sourceLineNo">434</span> * during master initialization phase.<a name="line.434"></a> +<span class="sourceLineNo">435</span> * @param failedServers A set of known failed servers<a name="line.435"></a> +<span class="sourceLineNo">436</span> * @throws IOException<a name="line.436"></a> +<span class="sourceLineNo">437</span> */<a name="line.437"></a> +<span class="sourceLineNo">438</span> void removeStaleRecoveringRegions(final Set<ServerName> failedServers) throws IOException,<a name="line.438"></a> +<span class="sourceLineNo">439</span> InterruptedIOException {<a name="line.439"></a> +<span class="sourceLineNo">440</span> Set<String> knownFailedServers = new HashSet<String>();<a name="line.440"></a> +<span class="sourceLineNo">441</span> if (failedServers != null) {<a name="line.441"></a> +<span class="sourceLineNo">442</span> for (ServerName tmpServerName : failedServers) {<a name="line.442"></a> +<span class="sourceLineNo">443</span> knownFailedServers.add(tmpServerName.getServerName());<a name="line.443"></a> +<span class="sourceLineNo">444</span> }<a name="line.444"></a> +<span class="sourceLineNo">445</span> }<a name="line.445"></a> +<span class="sourceLineNo">446</span><a name="line.446"></a> +<span class="sourceLineNo">447</span> this.recoveringRegionLock.lock();<a name="line.447"></a> +<span class="sourceLineNo">448</span> try {<a name="line.448"></a> +<span class="sourceLineNo">449</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.449"></a> +<span class="sourceLineNo">450</span> .getSplitLogManagerCoordination().removeStaleRecoveringRegions(knownFailedServers);<a name="line.450"></a> +<span class="sourceLineNo">451</span> } finally {<a name="line.451"></a> +<span class="sourceLineNo">452</span> this.recoveringRegionLock.unlock();<a name="line.452"></a> +<span class="sourceLineNo">453</span> }<a name="line.453"></a> +<span class="sourceLineNo">454</span> }<a name="line.454"></a> +<span class="sourceLineNo">455</span><a name="line.455"></a> +<span class="sourceLineNo">456</span> /**<a name="line.456"></a> +<span class="sourceLineNo">457</span> * @param path<a name="line.457"></a> +<span class="sourceLineNo">458</span> * @param batch<a name="line.458"></a> +<span class="sourceLineNo">459</span> * @return null on success, existing task on error<a name="line.459"></a> +<span class="sourceLineNo">460</span> */<a name="line.460"></a> +<span class="sourceLineNo">461</span> private Task createTaskIfAbsent(String path, TaskBatch batch) {<a name="line.461"></a> +<span class="sourceLineNo">462</span> Task oldtask;<a name="line.462"></a> +<span class="sourceLineNo">463</span> // batch.installed is only changed via this function and<a name="line.463"></a> +<span class="sourceLineNo">464</span> // a single thread touches batch.installed.<a name="line.464"></a> +<span class="sourceLineNo">465</span> Task newtask = new Task();<a name="line.465"></a> +<span class="sourceLineNo">466</span> newtask.batch = batch;<a name="line.466"></a> +<span class="sourceLineNo">467</span> oldtask = tasks.putIfAbsent(path, newtask);<a name="line.467"></a> +<span class="sourceLineNo">468</span> if (oldtask == null) {<a name="line.468"></a> +<span class="sourceLineNo">469</span> batch.installed++;<a name="line.469"></a> +<span class="sourceLineNo">470</span> return null;<a name="line.470"></a> +<span class="sourceLineNo">471</span> }<a name="line.471"></a> +<span class="sourceLineNo">472</span> // new task was not used.<a name="line.472"></a> +<span class="sourceLineNo">473</span> synchronized (oldtask) {<a name="line.473"></a> +<span class="sourceLineNo">474</span> if (oldtask.isOrphan()) {<a name="line.474"></a> +<span class="sourceLineNo">475</span> if (oldtask.status == SUCCESS) {<a name="line.475"></a> +<span class="sourceLineNo">476</span> // The task is already done. Do not install the batch for this<a name="line.476"></a> +<span class="sourceLineNo">477</span> // task because it might be too late for setDone() to update<a name="line.477"></a> +<span class="sourceLineNo">478</span> // batch.done. There is no need for the batch creator to wait for<a name="line.478"></a> +<span class="sourceLineNo">479</span> // this task to complete.<a name="line.479"></a> +<span class="sourceLineNo">480</span> return (null);<a name="line.480"></a> +<span class="sourceLineNo">481</span> }<a name="line.481"></a> +<span class="sourceLineNo">482</span> if (oldtask.status == IN_PROGRESS) {<a name="line.482"></a> +<span class="sourceLineNo">483</span> oldtask.batch = batch;<a name="line.483"></a> +<span class="sourceLineNo">484</span> batch.installed++;<a name="line.484"></a> +<span class="sourceLineNo">485</span> LOG.debug("Previously orphan task " + path + " is now being waited upon");<a name="line.485"></a> +<span class="sourceLineNo">486</span> return null;<a name="line.486"></a> +<span class="sourceLineNo">487</span> }<a name="line.487"></a> +<span class="sourceLineNo">488</span> while (oldtask.status == FAILURE) {<a name="line.488"></a> +<span class="sourceLineNo">489</span> LOG.debug("wait for status of task " + path + " to change to DELETED");<a name="line.489"></a> +<span class="sourceLineNo">490</span> SplitLogCounters.tot_mgr_wait_for_zk_delete.incrementAndGet();<a name="line.490"></a> +<span class="sourceLineNo">491</span> try {<a name="line.491"></a> +<span class="sourceLineNo">492</span> oldtask.wait();<a name="line.492"></a> +<span class="sourceLineNo">493</span> } catch (InterruptedException e) {<a name="line.493"></a> +<span class="sourceLineNo">494</span> Thread.currentThread().interrupt();<a name="line.494"></a> +<span class="sourceLineNo">495</span> LOG.warn("Interrupted when waiting for znode delete callback");<a name="line.495"></a> +<span class="sourceLineNo">496</span> // fall through to return failure<a name="line.496"></a> +<span class="sourceLineNo">497</span> break;<a name="line.497"></a> +<span class="sourceLineNo">498</span> }<a name="line.498"></a> +<span class="sourceLineNo">499</span> }<a name="line.499"></a> +<span class="sourceLineNo">500</span> if (oldtask.status != DELETED) {<a name="line.500"></a> +<span class="sourceLineNo">501</span> LOG.warn("Failure because previously failed task"<a name="line.501"></a> +<span class="sourceLineNo">502</span> + " state still present. Waiting for znode delete callback" + " path=" + path);<a name="line.502"></a> +<span class="sourceLineNo">503</span> return oldtask;<a name="line.503"></a> +<span class="sourceLineNo">504</span> }<a name="line.504"></a> +<span class="sourceLineNo">505</span> // reinsert the newTask and it must succeed this time<a name="line.505"></a> +<span class="sourceLineNo">506</span> Task t = tasks.putIfAbsent(path, newtask);<a name="line.506"></a> +<span class="sourceLineNo">507</span> if (t == null) {<a name="line.507"></a> +<span class="sourceLineNo">508</span> batch.installed++;<a name="line.508"></a> +<span class="sourceLineNo">509</span> return null;<a name="line.509"></a> +<span class="sourceLineNo">510</span> }<a name="line.510"></a> +<span class="sourceLineNo">511</span> LOG.fatal("Logic error. Deleted task still present in tasks map");<a name="line.511"></a> +<span class="sourceLineNo">512</span> assert false : "Deleted task still present in tasks map";<a name="line.512"></a> +<span class="sourceLineNo">513</span> return t;<a name="line.513"></a> +<span class="sourceLineNo">514</span> }<a name="line.514"></a> +<span class="sourceLineNo">515</span> LOG.warn("Failure because two threads can't wait for the same task; path=" + path);<a name="line.515"></a> +<span class="sourceLineNo">516</span> return oldtask;<a name="line.516"></a> +<span class="sourceLineNo">517</span> }<a name="line.517"></a> +<span class="sourceLineNo">518</span> }<a name="line.518"></a> +<span class="sourceLineNo">519</span><a name="line.519"></a> +<span class="sourceLineNo">520</span> Task findOrCreateOrphanTask(String path) {<a name="line.520"></a> +<span class="sourceLineNo">521</span> Task orphanTask = new Task();<a name="line.521"></a> +<span class="sourceLineNo">522</span> Task task;<a name="line.522"></a> +<span class="sourceLineNo">523</span> task = tasks.putIfAbsent(path, orphanTask);<a name="line.523"></a> +<span class="sourceLineNo">524</span> if (task == null) {<a name="line.524"></a> +<span class="sourceLineNo">525</span> LOG.info("creating orphan task " + path);<a name="line.525"></a> +<span class="sourceLineNo">526</span> SplitLogCounters.tot_mgr_orphan_task_acquired.incrementAndGet();<a name="line.526"></a> +<span class="sourceLineNo">527</span> task = orphanTask;<a name="line.527"></a> +<span class="sourceLineNo">528</span> }<a name="line.528"></a> +<span class="sourceLineNo">529</span> return task;<a name="line.529"></a> +<span class="sourceLineNo">530</span> }<a name="line.530"></a> +<span class="sourceLineNo">531</span><a name="line.531"></a> +<span class="sourceLineNo">532</span> public void stop() {<a name="line.532"></a> +<span class="sourceLineNo">533</span> if (choreService != null) {<a name="line.533"></a> +<span class="sourceLineNo">534</span> choreService.shutdown();<a name="line.534"></a> +<span class="sourceLineNo">535</span> }<a name="line.535"></a> +<span class="sourceLineNo">536</span> if (timeoutMonitor != null) {<a name="line.536"></a> +<span class="sourceLineNo">537</span> timeoutMonitor.cancel(true);<a name="line.537"></a> +<span class="sourceLineNo">538</span> }<a name="line.538"></a> +<span class="sourceLineNo">539</span> }<a name="line.539"></a> +<span class="sourceLineNo">540</span><a name="line.540"></a> +<span class="sourceLineNo">541</span> void handleDeadWorker(ServerName workerName) {<a name="line.541"></a> +<span class="sourceLineNo">542</span> // resubmit the tasks on the TimeoutMonitor thread. Makes it easier<a name="line.542"></a> +<span class="sourceLineNo">543</span> // to reason about concurrency. Makes it easier to retry.<a name="line.543"></a> +<span class="sourceLineNo">544</span> synchronized (deadWorkersLock) {<a name="line.544"></a> +<span class="sourceLineNo">545</span> if (deadWorkers == null) {<a name="line.545"></a> +<span class="sourceLineNo">546</span> deadWorkers = new HashSet<ServerName>(100);<a name="line.546"></a> +<span class="sourceLineNo">547</span> }<a name="line.547"></a> +<span class="sourceLineNo">548</span> deadWorkers.add(workerName);<a name="line.548"></a> +<span class="sourceLineNo">549</span> }<a name="line.549"></a> +<span class="sourceLineNo">550</span> LOG.info("dead splitlog worker " + workerName);<a name="line.550"></a> +<span class="sourceLineNo">551</span> }<a name="line.551"></a> +<span class="sourceLineNo">552</span><a name="line.552"></a> +<span class="sourceLineNo">553</span> void handleDeadWorkers(Set<ServerName> serverNames) {<a name="line.553"></a> +<span class="sourceLineNo">554</span> synchronized (deadWorkersLock) {<a name="line.554"></a> +<span class="sourceLineNo">555</span> if (deadWorkers == null) {<a name="line.555"></a> +<span class="sourceLineNo">556</span> deadWorkers = new HashSet<ServerName>(100);<a name="line.556"></a> +<span class="sourceLineNo">557</span> }<a name="line.557"></a> +<span class="sourceLineNo">558</span> deadWorkers.addAll(serverNames);<a name="line.558"></a> +<span class="sourceLineNo">559</span> }<a name="line.559"></a> +<span class="sourceLineNo">560</span> LOG.info("dead splitlog workers " + serverNames);<a name="line.560"></a> +<span class="sourceLineNo">561</span> }<a name="line.561"></a> +<span class="sourceLineNo">562</span><a name="line.562"></a> +<span class="sourceLineNo">563</span> /**<a name="line.563"></a> +<span class="sourceLineNo">564</span> * This function is to set recovery mode from outstanding split log tasks from before or current<a name="line.564"></a> +<span class="sourceLineNo">565</span> * configuration setting<a name="line.565"></a> +<span class="sourceLineNo">566</span> * @param isForInitialization<a name="line.566"></a> +<span class="sourceLineNo">567</span> * @throws IOException throws if it's impossible to set recovery mode<a name="line.567"></a> +<span class="sourceLineNo">568</span> */<a name="line.568"></a> +<span class="sourceLineNo">569</span> public void setRecoveryMode(boolean isForInitialization) throws IOException {<a name="line.569"></a> +<span class="sourceLineNo">570</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.570"></a> +<span class="sourceLineNo">571</span> .getSplitLogManagerCoordination().setRecoveryMode(isForInitialization);<a name="line.571"></a> +<span class="sourceLineNo">572</span><a name="line.572"></a> +<span class="sourceLineNo">573</span> }<a name="line.573"></a> +<span class="sourceLineNo">574</span><a name="line.574"></a> +<span class="sourceLineNo">575</span> public void markRegionsRecovering(ServerName server, Set<HRegionInfo> userRegions)<a name="line.575"></a> +<span class="sourceLineNo">576</span> throws InterruptedIOException, IOException {<a name="line.576"></a> +<span class="sourceLineNo">577</span> if (userRegions == null || (!isLogReplaying())) {<a name="line.577"></a> +<span class="sourceLineNo">578</span> return;<a name="line.578"></a> +<span class="sourceLineNo">579</span> }<a name="line.579"></a> +<span class="sourceLineNo">580</span> try {<a name="line.580"></a> +<span class="sourceLineNo">581</span> this.recoveringRegionLock.lock();<a name="line.581"></a> +<span class="sourceLineNo">582</span> // mark that we're creating recovering regions<a name="line.582"></a> +<span class="sourceLineNo">583</span> ((BaseCoordinatedStateManager) this.server.getCoordinatedStateManager())<a name="line.583"></a> +<span class="sourceLineNo">584</span> .getSplitLogManagerCoordination().markRegionsRecovering(server, userRegions);<a name="line.584"></a> +<span class="sourceLineNo">585</span> } finally {<a name="line.585"></a> +<span class="sourceLineNo">586</span> this.recoveringRegionLock.unlock();<a name="line.586"></a> +<span class="sourceLineNo">587</span> }<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><a name="line.590"></a> +<span class="sourceLineNo">591</span> /**<a name="line.591"></a> +<span class="sourceLineNo">592</span> * @return whether log is replaying<a name="line.592"></a> +<span class="sourceLineNo">593</span> */<a name="line.593"></a> +<span class="sourceLineNo">594</span> public boolean isLogReplaying() {<a name="line.594"></a> +<span class="sourceLineNo">595</span> CoordinatedStateManager m = server.getCoordinatedStateManager();<a name="line.595"></a> +<span class="sourceLineNo">596</span> if (m == null) return false;<a name="line.596"></a> +<span class="sourceLineNo">597</span> return ((BaseCoordinatedStateManager)m).getSplitLogManagerCoordination().isReplaying();<a name="line.597"></a> +<span class="sourceLineNo">598</span> }<a name="line.598"></a> +<span class="sourceLineNo">599</span><a name="line.599"></a> +<span class="sourceLineNo">600</span> /**<a name="line.600"></a> +<span class="sourceLineNo">601</span> * @return whether log is splitting<a name="line.601"></a> +<span class="sourceLineNo">602</span> */<a name="line.602"></a> +<span class="sourceLineNo">603</span> public boolean isLogSplitting() {<a name="line.603"></a> +<span class="sourceLineNo">604</span> if (server.getCoordinatedStateManager() == null) return false;<a name="line.604"></a> +<span class="sourceLineNo">605</span> return ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.605"></a> +<span class="sourceLineNo">606</span> .getSplitLogManagerCoordination().isSplitting();<a name="line.606"></a> +<span class="sourceLineNo">607</span> }<a name="line.607"></a> +<span class="sourceLineNo">608</span><a name="line.608"></a> +<span class="sourceLineNo">609</span> /**<a name="line.609"></a> +<span class="sourceLineNo">610</span> * @return the current log recovery mode<a name="line.610"></a> +<span class="sourceLineNo">611</span> */<a name="line.611"></a> +<span class="sourceLineNo">612</span> public RecoveryMode getRecoveryMode() {<a name="line.612"></a> +<span class="sourceLineNo">613</span> return ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.613"></a> +<span class="sourceLineNo">614</span> .getSplitLogManagerCoordination().getRecoveryMode();<a name="line.614"></a> +<span class="sourceLineNo">615</span> }<a name="line.615"></a> +<span class="sourceLineNo">616</span><a name="line.616"></a> +<span class="sourceLineNo">617</span> /**<a name="line.617"></a> +<span class="sourceLineNo">618</span> * Keeps track of the batch of tasks submitted together by a caller in splitLogDistributed().<a name="line.618"></a> +<span class="sourceLineNo">619</span> * Clients threads use this object to wait for all their tasks to be done.<a name="line.619"></a> +<span class="sourceLineNo">620</span> * <p><a name="line.620"></a> +<span class="sourceLineNo">621</span> * All access is synchronized.<a name="line.621"></a> +<span class="sourceLineNo">622</span> */<a name="line.622"></a> +<span class="sourceLineNo">623</span> @InterfaceAudience.Private<a name="line.623"></a> +<span class="sourceLineNo">624</span> public static class TaskBatch {<a name="line.624"></a> +<span class="sourceLineNo">625</span> public int installed = 0;<a name="line.625"></a> +<span class="sourceLineNo">626</span> public int done = 0;<a name="line.626"></a> +<span class="sourceLineNo">627</span> public int error = 0;<a name="line.627"></a> +<span class="sourceLineNo">628</span> public volatile boolean isDead = false;<a name="line.628"></a> +<span class="sourceLineNo">629</span><a name="line.629"></a> +<span class="sourceLineNo">630</span> @Override<a name="line.630"></a> +<span class="sourceLineNo">631</span> public String toString() {<a name="line.631"></a> +<span class="sourceLineNo">632</span> return ("installed = " + installed + " done = " + done + " error = " + error);<a name="line.632"></a> +<span class="sourceLineNo">633</span> }<a name="line.633"></a> +<span class="sourceLineNo">634</span> }<a name="line.634"></a> +<span class="sourceLineNo">635</span><a name="line.635"></a> +<span class="sourceLineNo">636</span> /**<a name="line.636"></a> +<span class="sourceLineNo">637</span> * in memory state of an active task.<a name="line.637"></a> +<span class="sourceLineNo">638</span> */<a name="line.638"></a> +<span class="sourceLineNo">639</span> @InterfaceAudience.Private<a name="line.639"></a> +<span class="sourceLineNo">640</span> public static class Task {<a name="line.640"></a> +<span class="sourceLineNo">641</span> public volatile long last_update;<a name="line.641"></a> +<span class="sourceLineNo">642</span> public volatile int last_version;<a name="line.642"></a> +<span class="sourceLineNo">643</span> public volatile ServerName cur_worker_name;<a name="line.643"></a> +<span class="sourceLineNo">644</span> public volatile TaskBatch batch;<a name="line.644"></a> +<span class="sourceLineNo">645</span> public volatile TerminationStatus status;<a name="line.645"></a> +<span class="sourceLineNo">646</span> public volatile int incarnation;<a name="line.646"></a> +<span class="sourceLineNo">647</span> public final AtomicInteger unforcedResubmits = new AtomicInteger();<a name="line.647"></a> +<span class="sourceLineNo">648</span> public volatile boolean resubmitThresholdReached;<a name="line.648"></a> +<span class="sourceLineNo">649</span><a name="line.649"></a> +<span class="sourceLineNo">650</span> @Override<a name="line.650"></a> +<span class="sourceLineNo">651</span> public String toString() {<a name="line.651"></a> +<span class="sourceLineNo">652</span> return ("last_update = " + last_update + " last_version = " + last_version<a name="line.652"></a> +<span class="sourceLineNo">653</span> + " cur_worker_name = " + cur_worker_name + " status = " + status + " incarnation = "<a name="line.653"></a> +<span class="sourceLineNo">654</span> + incarnation + " resubmits = " + unforcedResubmits.get() + " batch = " + batch);<a name="line.654"></a> +<span class="sourceLineNo">655</span> }<a name="line.655"></a> +<span class="sourceLineNo">656</span><a name="line.656"></a> +<span class="sourceLineNo">657</span> public Task() {<a name="line.657"></a> +<span class="sourceLineNo">658</span> incarnation = 0;<a name="line.658"></a> +<span class="sourceLineNo">659</span> last_version = -1;<a name="line.659"></a> +<span class="sourceLineNo">660</span> status = IN_PROGRESS;<a name="line.660"></a> +<span class="sourceLineNo">661</span> setUnassigned();<a name="line.661"></a> +<span class="sourceLineNo">662</span> }<a name="line.662"></a> +<span class="sourceLineNo">663</span><a name="line.663"></a> +<span class="sourceLineNo">664</span> public boolean isOrphan() {<a name="line.664"></a> +<span class="sourceLineNo">665</span> return (batch == null || batch.isDead);<a name="line.665"></a> +<span class="sourceLineNo">666</span> }<a name="line.666"></a> +<span class="sourceLineNo">667</span><a name="line.667"></a> +<span class="sourceLineNo">668</span> public boolean isUnassigned() {<a name="line.668"></a> +<span class="sourceLineNo">669</span> return (cur_worker_name == null);<a name="line.669"></a> +<span class="sourceLineNo">670</span> }<a name="line.670"></a> +<span class="sourceLineNo">671</span><a name="line.671"></a> +<span class="sourceLineNo">672</span> public void heartbeatNoDetails(long time) {<a name="line.672"></a> +<span class="sourceLineNo">673</span> last_update = time;<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> public void heartbeat(long time, int version, ServerName worker) {<a name="line.676"></a> +<span class="sourceLineNo">677</span> last_version = version;<a name="line.677"></a> +<span class="sourceLineNo">678</span> last_update = time;<a name="line.678"></a> +<span class="sourceLineNo">679</span> cur_worker_name = worker;<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> public void setUnassigned() {<a name="line.682"></a> +<span class="sourceLineNo">683</span> cur_worker_name = null;<a name="line.683"></a> +<span class="sourceLineNo">684</span> last_update = -1;<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><a name="line.687"></a> +<span class="sourceLineNo">688</span> /**<a name="line.688"></a> +<span class="sourceLineNo">689</span> * Periodically checks all active tasks and resubmits the ones that have timed out<a name="line.689"></a> +<span class="sourceLineNo">690</span> */<a name="line.690"></a> +<span class="sourceLineNo">691</span> private class TimeoutMonitor extends ScheduledChore {<a name="line.691"></a> +<span class="sourceLineNo">692</span> private long lastLog = 0;<a name="line.692"></a> +<span class="sourceLineNo">693</span><a name="line.693"></a> +<span class="sourceLineNo">694</span> public TimeoutMonitor(final int period, Stoppable stopper) {<a name="line.694"></a> +<span class="sourceLineNo">695</span> super("SplitLogManager Timeout Monitor", stopper, period);<a name="line.695"></a> +<span class="sourceLineNo">696</span> }<a name="line.696"></a> +<span class="sourceLineNo">697</span><a name="line.697"></a> +<span class="sourceLineNo">698</span> @Override<a name="line.698"></a> +<span class="sourceLineNo">699</span> protected void chore() {<a name="line.699"></a> +<span class="sourceLineNo">700</span> int resubmitted = 0;<a name="line.700"></a> +<span class="sourceLineNo">701</span> int unassigned = 0;<a name="line.701"></a> +<span class="sourceLineNo">702</span> int tot = 0;<a name="line.702"></a> +<span class="sourceLineNo">703</span> boolean found_assigned_task = false;<a name="line.703"></a> +<span class="sourceLineNo">704</span> Set<ServerName> localDeadWorkers;<a name="line.704"></a> +<span class="sourceLineNo">705</span><a name="line.705"></a> +<span class="sourceLineNo">706</span> synchronized (deadWorkersLock) {<a name="line.706"></a> +<span class="sourceLineNo">707</span> localDeadWorkers = deadWorkers;<a name="line.707"></a> +<span class="sourceLineNo">708</span> deadWorkers = null;<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> for (Map.Entry<String, Task> e : tasks.entrySet()) {<a name="line.711"></a> +<span class="sourceLineNo">712</span> String path = e.getKey();<a name="line.712"></a> +<span class="sourceLineNo">713</span> Task task = e.getValue();<a name="line.713"></a> +<span class="sourceLineNo">714</span> ServerName cur_worker = task.cur_worker_name;<a name="line.714"></a> +<span class="sourceLineNo">715</span> tot++;<a name="line.715"></a> +<span class="sourceLineNo">716</span> // don't easily resubmit a task which hasn't been picked up yet. It<a name="line.716"></a> +<span class="sourceLineNo">717</span> // might be a long while before a SplitLogWorker is free to pick up a<a name="line.717"></a> +<span class="sourceLineNo">718</span> // task. This is because a SplitLogWorker picks up a task one at a<a name="line.718"></a> +<span class="sourceLineNo">719</span> // time. If we want progress when there are no region servers then we<a name="line.719"></a> +<span class="sourceLineNo">720</span> // will have to run a SplitLogWorker thread in the Master.<a name="line.720"></a> +<span class="sourceLineNo">721</span> if (task.isUnassigned()) {<a name="line.721"></a> +<span class="sourceLineNo">722</span> unassigned++;<a name="line.722"></a> +<span class="sourceLineNo">723</span> continue;<a name="line.723"></a> +<span class="sourceLineNo">724</span> }<a name="line.724"></a> +<span class="sourceLineNo">725</span> found_assigned_task = true;<a name="line.725"></a> +<span class="sourceLineNo">726</span> if (localDeadWorkers != null && localDeadWorkers.contains(cur_worker)) {<a name="line.726"></a> +<span class="sourceLineNo">727</span> SplitLogCounters.tot_mgr_resubmit_dead_server_task.incrementAndGet();<a name="line.727"></a> +<span class="sourceLineNo">728</span> if (((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.728"></a> +<span class="sourceLineNo">729</span> .getSplitLogManagerCoordination().resubmitTask(path, task, FORCE)) {<a name="line.729"></a> +<span class="sourceLineNo">730</span> resubmitted++;<a name="line.730"></a> +<span class="sourceLineNo">731</span> } else {<a name="line.731"></a> +<span class="sourceLineNo">732</span> handleDeadWorker(cur_worker);<a name="line.732"></a> +<span class="sourceLineNo">733</span> LOG.warn("Failed to resubmit task " + path + " owned by dead " + cur_worker<a name="line.733"></a> +<span class="sourceLineNo">734</span> + ", will retry.");<a name="line.734"></a> +<span class="sourceLineNo">735</span> }<a name="line.735"></a> +<span class="sourceLineNo">736</span> } else if (((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.736"></a> +<span class="sourceLineNo">737</span> .getSplitLogManagerCoordination().resubmitTask(path, task, CHECK)) {<a name="line.737"></a> +<span class="sourceLineNo">738</span> resubmitted++;<a name="line.738"></a> +<span class="sourceLineNo">739</span> }<a name="line.739"></a> +<span class="sourceLineNo">740</span> }<a name="line.740"></a> +<span class="sourceLineNo">741</span> if (tot > 0) {<a name="line.741"></a> +<span class="sourceLineNo">742</span> long now = EnvironmentEdgeManager.currentTime();<a name="line.742"></a> +<span class="sourceLineNo">743</span> if (now > lastLog + 5000) {<a name="line.743"></a> +<span class="sourceLineNo">744</span> lastLog = now;<a name="line.744"></a> +<span class="sourceLineNo">745</span> LOG.info("total tasks = " + tot + " unassigned = " + unassigned + " tasks=" + tasks);<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> if (resubmitted > 0) {<a name="line.748"></a> +<span class="sourceLineNo">749</span> LOG.info("resubmitted " + resubmitted + " out of " + tot + " tasks");<a name="line.749"></a> +<span class="sourceLineNo">750</span> }<a name="line.750"></a> +<span class="sourceLineNo">751</span> // If there are pending tasks and all of them have been unassigned for<a name="line.751"></a> +<span class="sourceLineNo">752</span> // some time then put up a RESCAN node to ping the workers.<a name="line.752"></a> +<span class="sourceLineNo">753</span> // ZKSplitlog.DEFAULT_UNASSIGNED_TIMEOUT is of the order of minutes<a name="line.753"></a> +<span class="sourceLineNo">754</span> // because a. it is very unlikely that every worker had a<a name="line.754"></a> +<span class="sourceLineNo">755</span> // transient error when trying to grab the task b. if there are no<a name="line.755"></a> +<span class="sourceLineNo">756</span> // workers then all tasks wills stay unassigned indefinitely and the<a name="line.756"></a> +<span class="sourceLineNo">757</span> // manager will be indefinitely creating RESCAN nodes. TODO may be the<a name="line.757"></a> +<span class="sourceLineNo">758</span> // master should spawn both a manager and a worker thread to guarantee<a name="line.758"></a> +<span class="sourceLineNo">759</span> // that there is always one worker in the system<a name="line.759"></a> +<span class="sourceLineNo">760</span> if (tot > 0<a name="line.760"></a> +<span class="sourceLineNo">761</span> && !found_assigned_task<a name="line.761"></a> +<span class="sourceLineNo">762</span> && ((EnvironmentEdgeManager.currentTime() - lastTaskCreateTime) > unassignedTimeout)) {<a name="line.762"></a> +<span class="sourceLineNo">763</span> for (Map.Entry<String, Task> e : tasks.entrySet()) {<a name="line.763"></a> +<span class="sourceLineNo">764</span> String key = e.getKey();<a name="line.764"></a> +<span class="sourceLineNo">765</span> Task task = e.getValue();<a name="line.765"></a> +<span class="sourceLineNo">766</span> // we have to do task.isUnassigned() check again because tasks might<a name="line.766"></a> +<span class="sourceLineNo">767</span> // have been asynchronously assigned. There is no locking required<a name="line.767"></a> +<span class="sourceLineNo">768</span> // for these checks ... it is OK even if tryGetDataSetWatch() is<a name="line.768"></a> +<span class="sourceLineNo">769</span> // called unnecessarily for a taskpath<a name="line.769"></a> +<span class="sourceLineNo">770</span> if (task.isUnassigned() && (task.status != FAILURE)) {<a name="line.770"></a> +<span class="sourceLineNo">771</span> // We just touch the znode to make sure its still there<a name="line.771"></a> +<span class="sourceLineNo">772</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.772"></a> +<span class="sourceLineNo">773</span> .getSplitLogManagerCoordination().checkTaskStillAvailable(key);<a name="line.773"></a> +<span class="sourceLineNo">774</span> }<a name="line.774"></a> +<span class="sourceLineNo">775</span> }<a name="line.775"></a> +<span class="sourceLineNo">776</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.776"></a> +<span class="sourceLineNo">777</span> .getSplitLogManagerCoordination().checkTasks();<a name="line.777"></a> +<span class="sourceLineNo">778</span> SplitLogCounters.tot_mgr_resubmit_unassigned.incrementAndGet();<a name="line.778"></a> +<span class="sourceLineNo">779</span> LOG.debug("resubmitting unassigned task(s) after timeout");<a name="line.779"></a> +<span class="sourceLineNo">780</span> }<a name="line.780"></a> +<span class="sourceLineNo">781</span> Set<String> failedDeletions =<a name="line.781"></a> +<span class="sourceLineNo">782</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.782"></a> +<span class="sourceLineNo">783</span> .getSplitLogManagerCoordination().getDetails().getFailedDeletions();<a name="line.783"></a> +<span class="sourceLineNo">784</span> // Retry previously failed deletes<a name="line.784"></a> +<span class="sourceLineNo">785</span> if (failedDeletions.size() > 0) {<a name="line.785"></a> +<span class="sourceLineNo">786</span> List<String> tmpPaths = new ArrayList<String>(failedDeletions);<a name="line.786"></a> +<span class="sourceLineNo">787</span> for (String tmpPath : tmpPaths) {<a name="line.787"></a> +<span class="sourceLineNo">788</span> // deleteNode is an async call<a name="line.788"></a> +<span class="sourceLineNo">789</span> ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.789"></a> +<span class="sourceLineNo">790</span> .getSplitLogManagerCoordination().deleteTask(tmpPath);<a name="line.790"></a> +<span class="sourceLineNo">791</span> }<a name="line.791"></a> +<span class="sourceLineNo">792</span> failedDeletions.removeAll(tmpPaths);<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> // Garbage collect left-over<a name="line.795"></a> +<span class="sourceLineNo">796</span> long timeInterval =<a name="line.796"></a> +<span class="sourceLineNo">797</span> EnvironmentEdgeManager.currentTime()<a name="line.797"></a> +<span class="sourceLineNo">798</span> - ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())<a name="line.798"></a> +<span class="sourceLineNo">799</span> .getSplitLogManagerCoordination().getLastRecoveryTime();<a name="line.799"></a> +<span class="sourceLineNo">800</span> if (!failedRecoveringRegionDeletions.isEmpty()<a name="line.800"></a> +<span class="sourceLineNo">801</span> || (tot == 0 && tasks.size() == 0 && (timeInterval > checkRecoveringTimeThreshold))) {<a name="line.801"></a> +<span class="sourceLineNo">802</span> // inside the function there have more checks before GC anything<a name="line.802"></a> +<span class="sourceLineNo">803</span> if (!failedRecoveringRegionDeletions.isEmpty()) {<a name="line.803"></a> +<span class="sourceLineNo">804</span> List<Pair<Set<ServerName>, Boolean>> previouslyFailedDeletions =<a name="line.804"></a> +<span class="sourceLineNo">805</span> new ArrayList<Pair<Set<ServerName>, Boolean>>(failedRecoveringRegionDeletions);<a name="line.805"></a> +<span class="sourceLineNo">806</span> failedRecoveringRegionDeletions.removeAll(previouslyFailedDeletions);<a name="line.806"></a> +<span class="sourceLineNo">807</span> for (Pair<Set<ServerName>, Boolean> failedDeletion : previouslyFailedDeletions) {<a name="line.807"></a> +<span class="sourceLineNo">808</span> removeRecoveringRegions(failedDeletion.getFirst(), failedDeletion.getSecond());<a name="line.808"></a> +<span class="sourceLineNo">809</span> }<a name="line.809"></a> +<span class="sourceLineNo">810</span> } else {<a name="line.810"></a> +<span class="sourceLineNo">811</span> removeRecoveringRegions(null, null);<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> }<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> public enum ResubmitDirective {<a name="line.817"></a> +<span class="sourceLineNo">818</span> CHECK(), FORCE();<a name="line.818"></a> +<span class="sourceLineNo">819</span> }<a name="line.819"></a> +<span class="sourceLineNo">820</span><a name="line.820"></a> +<span class="sourceLineNo">821</span> public enum TerminationStatus {<a name="line.821"></a> +<span class="sourceLineNo">822</span> IN_PROGRESS("in_progress"), SUCCESS("success"), FAILURE("failure"), DELETED("deleted");<a name="line.822"></a> +<span class="sourceLineNo">823</span><a name="line.823"></a> +<span class="sourceLineNo">824</span> String statusMsg;<a name="line.824"></a> +<span class="sourceLineNo">825</span><a name="line.825"></a> +<span class="sourceLineNo">826</span> TerminationStatus(String msg) {<a name="line.826"></a> +<span class="sourceLineNo">827</span> statusMsg = msg;<a name="line.827"></a> +<span class="sourceLineNo">828</span> }<a name="line.828"></a> +<span class="sourceLineNo">829</span><a name="line.829"></a> +<span class="sourceLineNo">830</span> @Override<a name="line.830"></a> +<span class="sourceLineNo">831</span> public String toString() {<a name="line.831"></a> +<span class="sourceLineNo">832</span> return statusMsg;<a name="line.832"></a> +<span class="sourceLineNo">833</span> }<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>
http://git-wip-us.apache.org/repos/asf/hbase/blob/ef7355e1/devapidocs/src-html/org/apache/hadoop/hbase/mob/mapreduce/SweepMapper.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/mob/mapreduce/SweepMapper.html b/devapidocs/src-html/org/apache/hadoop/hbase/mob/mapreduce/SweepMapper.html index c1f2e61..5dce057 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/mob/mapreduce/SweepMapper.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/mob/mapreduce/SweepMapper.html @@ -43,8 +43,8 @@ <span class="sourceLineNo">035</span><a name="line.35"></a> <span class="sourceLineNo">036</span>/**<a name="line.36"></a> <span class="sourceLineNo">037</span> * The mapper of a sweep job.<a name="line.37"></a> -<span class="sourceLineNo">038</span> * Takes the rows from the table and their results and map to <filename:Text, mobValue:KeyValue><a name="line.38"></a> -<span class="sourceLineNo">039</span> * where mobValue is the actual cell in HBase.<a name="line.39"></a> +<span class="sourceLineNo">038</span> * Takes the rows from the table and their results and map to {@literal <filename:Text,<a name="line.38"></a> +<span class="sourceLineNo">039</span> * mobValue:KeyValue>} where mobValue is the actual cell in HBase.<a name="line.39"></a> <span class="sourceLineNo">040</span> */<a name="line.40"></a> <span class="sourceLineNo">041</span>@InterfaceAudience.Private<a name="line.41"></a> <span class="sourceLineNo">042</span>public class SweepMapper extends TableMapper<Text, KeyValue> {<a name="line.42"></a>
