http://git-wip-us.apache.org/repos/asf/hbase-site/blob/de18d468/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html index c282308..094b592 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html @@ -426,483 +426,481 @@ <span class="sourceLineNo">418</span> */<a name="line.418"></a> <span class="sourceLineNo">419</span> void removeRecoveredSource(ReplicationSourceInterface src) {<a name="line.419"></a> <span class="sourceLineNo">420</span> LOG.info("Done with the recovered queue " + src.getQueueId());<a name="line.420"></a> -<span class="sourceLineNo">421</span> src.getSourceMetrics().clear();<a name="line.421"></a> -<span class="sourceLineNo">422</span> this.oldsources.remove(src);<a name="line.422"></a> -<span class="sourceLineNo">423</span> // Delete queue from storage and memory<a name="line.423"></a> -<span class="sourceLineNo">424</span> deleteQueue(src.getQueueId());<a name="line.424"></a> -<span class="sourceLineNo">425</span> this.walsByIdRecoveredQueues.remove(src.getQueueId());<a name="line.425"></a> -<span class="sourceLineNo">426</span> }<a name="line.426"></a> -<span class="sourceLineNo">427</span><a name="line.427"></a> -<span class="sourceLineNo">428</span> /**<a name="line.428"></a> -<span class="sourceLineNo">429</span> * Clear the metrics and related replication queue of the specified old source<a name="line.429"></a> -<span class="sourceLineNo">430</span> * @param src source to clear<a name="line.430"></a> -<span class="sourceLineNo">431</span> */<a name="line.431"></a> -<span class="sourceLineNo">432</span> void removeSource(ReplicationSourceInterface src) {<a name="line.432"></a> -<span class="sourceLineNo">433</span> LOG.info("Done with the queue " + src.getQueueId());<a name="line.433"></a> -<span class="sourceLineNo">434</span> src.getSourceMetrics().clear();<a name="line.434"></a> -<span class="sourceLineNo">435</span> this.sources.remove(src.getPeerId());<a name="line.435"></a> -<span class="sourceLineNo">436</span> // Delete queue from storage and memory<a name="line.436"></a> -<span class="sourceLineNo">437</span> deleteQueue(src.getQueueId());<a name="line.437"></a> -<span class="sourceLineNo">438</span> this.walsById.remove(src.getQueueId());<a name="line.438"></a> -<span class="sourceLineNo">439</span> }<a name="line.439"></a> -<span class="sourceLineNo">440</span><a name="line.440"></a> -<span class="sourceLineNo">441</span> /**<a name="line.441"></a> -<span class="sourceLineNo">442</span> * Delete a complete queue of wals associated with a replication source<a name="line.442"></a> -<span class="sourceLineNo">443</span> * @param queueId the id of replication queue to delete<a name="line.443"></a> -<span class="sourceLineNo">444</span> */<a name="line.444"></a> -<span class="sourceLineNo">445</span> private void deleteQueue(String queueId) {<a name="line.445"></a> -<span class="sourceLineNo">446</span> abortWhenFail(() -> this.queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.446"></a> -<span class="sourceLineNo">447</span> }<a name="line.447"></a> -<span class="sourceLineNo">448</span><a name="line.448"></a> -<span class="sourceLineNo">449</span> @FunctionalInterface<a name="line.449"></a> -<span class="sourceLineNo">450</span> private interface ReplicationQueueOperation {<a name="line.450"></a> -<span class="sourceLineNo">451</span> void exec() throws ReplicationException;<a name="line.451"></a> -<span class="sourceLineNo">452</span> }<a name="line.452"></a> -<span class="sourceLineNo">453</span><a name="line.453"></a> -<span class="sourceLineNo">454</span> private void abortWhenFail(ReplicationQueueOperation op) {<a name="line.454"></a> -<span class="sourceLineNo">455</span> try {<a name="line.455"></a> -<span class="sourceLineNo">456</span> op.exec();<a name="line.456"></a> -<span class="sourceLineNo">457</span> } catch (ReplicationException e) {<a name="line.457"></a> -<span class="sourceLineNo">458</span> server.abort("Failed to operate on replication queue", e);<a name="line.458"></a> -<span class="sourceLineNo">459</span> }<a name="line.459"></a> -<span class="sourceLineNo">460</span> }<a name="line.460"></a> -<span class="sourceLineNo">461</span><a name="line.461"></a> -<span class="sourceLineNo">462</span> private void throwIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.462"></a> -<span class="sourceLineNo">463</span> try {<a name="line.463"></a> -<span class="sourceLineNo">464</span> op.exec();<a name="line.464"></a> -<span class="sourceLineNo">465</span> } catch (ReplicationException e) {<a name="line.465"></a> -<span class="sourceLineNo">466</span> throw new IOException(e);<a name="line.466"></a> -<span class="sourceLineNo">467</span> }<a name="line.467"></a> -<span class="sourceLineNo">468</span> }<a name="line.468"></a> -<span class="sourceLineNo">469</span><a name="line.469"></a> -<span class="sourceLineNo">470</span> private void abortAndThrowIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.470"></a> -<span class="sourceLineNo">471</span> try {<a name="line.471"></a> -<span class="sourceLineNo">472</span> op.exec();<a name="line.472"></a> -<span class="sourceLineNo">473</span> } catch (ReplicationException e) {<a name="line.473"></a> -<span class="sourceLineNo">474</span> server.abort("Failed to operate on replication queue", e);<a name="line.474"></a> -<span class="sourceLineNo">475</span> throw new IOException(e);<a name="line.475"></a> -<span class="sourceLineNo">476</span> }<a name="line.476"></a> -<span class="sourceLineNo">477</span> }<a name="line.477"></a> -<span class="sourceLineNo">478</span><a name="line.478"></a> -<span class="sourceLineNo">479</span> /**<a name="line.479"></a> -<span class="sourceLineNo">480</span> * This method will log the current position to storage. And also clean old logs from the<a name="line.480"></a> -<span class="sourceLineNo">481</span> * replication queue.<a name="line.481"></a> -<span class="sourceLineNo">482</span> * @param queueId id of the replication queue<a name="line.482"></a> -<span class="sourceLineNo">483</span> * @param queueRecovered indicates if this queue comes from another region server<a name="line.483"></a> -<span class="sourceLineNo">484</span> * @param entryBatch the wal entry batch we just shipped<a name="line.484"></a> -<span class="sourceLineNo">485</span> */<a name="line.485"></a> -<span class="sourceLineNo">486</span> public void logPositionAndCleanOldLogs(String queueId, boolean queueRecovered,<a name="line.486"></a> -<span class="sourceLineNo">487</span> WALEntryBatch entryBatch) {<a name="line.487"></a> -<span class="sourceLineNo">488</span> String fileName = entryBatch.getLastWalPath().getName();<a name="line.488"></a> -<span class="sourceLineNo">489</span> abortWhenFail(() -> this.queueStorage.setWALPosition(server.getServerName(), queueId, fileName,<a name="line.489"></a> -<span class="sourceLineNo">490</span> entryBatch.getLastWalPosition(), entryBatch.getLastSeqIds()));<a name="line.490"></a> -<span class="sourceLineNo">491</span> cleanOldLogs(fileName, entryBatch.isEndOfFile(), queueId, queueRecovered);<a name="line.491"></a> -<span class="sourceLineNo">492</span> }<a name="line.492"></a> -<span class="sourceLineNo">493</span><a name="line.493"></a> -<span class="sourceLineNo">494</span> /**<a name="line.494"></a> -<span class="sourceLineNo">495</span> * Cleans a log file and all older logs from replication queue. Called when we are sure that a log<a name="line.495"></a> -<span class="sourceLineNo">496</span> * file is closed and has no more entries.<a name="line.496"></a> -<span class="sourceLineNo">497</span> * @param log Path to the log<a name="line.497"></a> -<span class="sourceLineNo">498</span> * @param inclusive whether we should also remove the given log file<a name="line.498"></a> -<span class="sourceLineNo">499</span> * @param queueId id of the replication queue<a name="line.499"></a> -<span class="sourceLineNo">500</span> * @param queueRecovered Whether this is a recovered queue<a name="line.500"></a> -<span class="sourceLineNo">501</span> */<a name="line.501"></a> -<span class="sourceLineNo">502</span> @VisibleForTesting<a name="line.502"></a> -<span class="sourceLineNo">503</span> void cleanOldLogs(String log, boolean inclusive, String queueId, boolean queueRecovered) {<a name="line.503"></a> -<span class="sourceLineNo">504</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log);<a name="line.504"></a> -<span class="sourceLineNo">505</span> if (queueRecovered) {<a name="line.505"></a> -<span class="sourceLineNo">506</span> NavigableSet<String> wals = walsByIdRecoveredQueues.get(queueId).get(logPrefix);<a name="line.506"></a> -<span class="sourceLineNo">507</span> if (wals != null) {<a name="line.507"></a> -<span class="sourceLineNo">508</span> cleanOldLogs(wals, log, inclusive, queueId);<a name="line.508"></a> -<span class="sourceLineNo">509</span> }<a name="line.509"></a> -<span class="sourceLineNo">510</span> } else {<a name="line.510"></a> -<span class="sourceLineNo">511</span> // synchronized on walsById to avoid race with preLogRoll<a name="line.511"></a> -<span class="sourceLineNo">512</span> synchronized (this.walsById) {<a name="line.512"></a> -<span class="sourceLineNo">513</span> NavigableSet<String> wals = walsById.get(queueId).get(logPrefix);<a name="line.513"></a> -<span class="sourceLineNo">514</span> if (wals != null) {<a name="line.514"></a> -<span class="sourceLineNo">515</span> cleanOldLogs(wals, log, inclusive, queueId);<a name="line.515"></a> -<span class="sourceLineNo">516</span> }<a name="line.516"></a> -<span class="sourceLineNo">517</span> }<a name="line.517"></a> -<span class="sourceLineNo">518</span> }<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> private void cleanOldLogs(NavigableSet<String> wals, String key, boolean inclusive, String id) {<a name="line.521"></a> -<span class="sourceLineNo">522</span> NavigableSet<String> walSet = wals.headSet(key, inclusive);<a name="line.522"></a> -<span class="sourceLineNo">523</span> if (walSet.isEmpty()) {<a name="line.523"></a> -<span class="sourceLineNo">524</span> return;<a name="line.524"></a> -<span class="sourceLineNo">525</span> }<a name="line.525"></a> -<span class="sourceLineNo">526</span> LOG.debug("Removing {} logs in the list: {}", walSet.size(), walSet);<a name="line.526"></a> -<span class="sourceLineNo">527</span> for (String wal : walSet) {<a name="line.527"></a> -<span class="sourceLineNo">528</span> abortWhenFail(() -> this.queueStorage.removeWAL(server.getServerName(), id, wal));<a name="line.528"></a> -<span class="sourceLineNo">529</span> }<a name="line.529"></a> -<span class="sourceLineNo">530</span> walSet.clear();<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 because of we call it in TestReplicationEmptyWALRecovery<a name="line.533"></a> -<span class="sourceLineNo">534</span> @VisibleForTesting<a name="line.534"></a> -<span class="sourceLineNo">535</span> public void preLogRoll(Path newLog) throws IOException {<a name="line.535"></a> -<span class="sourceLineNo">536</span> String logName = newLog.getName();<a name="line.536"></a> -<span class="sourceLineNo">537</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.537"></a> -<span class="sourceLineNo">538</span> // synchronized on latestPaths to avoid the new open source miss the new log<a name="line.538"></a> -<span class="sourceLineNo">539</span> synchronized (this.latestPaths) {<a name="line.539"></a> -<span class="sourceLineNo">540</span> // Add log to queue storage<a name="line.540"></a> -<span class="sourceLineNo">541</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.541"></a> -<span class="sourceLineNo">542</span> // If record log to queue storage failed, abort RS and throw exception to make log roll<a name="line.542"></a> -<span class="sourceLineNo">543</span> // failed<a name="line.543"></a> -<span class="sourceLineNo">544</span> abortAndThrowIOExceptionWhenFail(<a name="line.544"></a> -<span class="sourceLineNo">545</span> () -> this.queueStorage.addWAL(server.getServerName(), source.getQueueId(), logName));<a name="line.545"></a> -<span class="sourceLineNo">546</span> }<a name="line.546"></a> -<span class="sourceLineNo">547</span><a name="line.547"></a> -<span class="sourceLineNo">548</span> // synchronized on walsById to avoid race with cleanOldLogs<a name="line.548"></a> -<span class="sourceLineNo">549</span> synchronized (this.walsById) {<a name="line.549"></a> -<span class="sourceLineNo">550</span> // Update walsById map<a name="line.550"></a> -<span class="sourceLineNo">551</span> for (Map.Entry<String, Map<String, NavigableSet<String>>> entry : this.walsById<a name="line.551"></a> -<span class="sourceLineNo">552</span> .entrySet()) {<a name="line.552"></a> -<span class="sourceLineNo">553</span> String peerId = entry.getKey();<a name="line.553"></a> -<span class="sourceLineNo">554</span> Map<String, NavigableSet<String>> walsByPrefix = entry.getValue();<a name="line.554"></a> -<span class="sourceLineNo">555</span> boolean existingPrefix = false;<a name="line.555"></a> -<span class="sourceLineNo">556</span> for (Map.Entry<String, NavigableSet<String>> walsEntry : walsByPrefix.entrySet()) {<a name="line.556"></a> -<span class="sourceLineNo">557</span> SortedSet<String> wals = walsEntry.getValue();<a name="line.557"></a> -<span class="sourceLineNo">558</span> if (this.sources.isEmpty()) {<a name="line.558"></a> -<span class="sourceLineNo">559</span> // If there's no slaves, don't need to keep the old wals since<a name="line.559"></a> -<span class="sourceLineNo">560</span> // we only consider the last one when a new slave comes in<a name="line.560"></a> -<span class="sourceLineNo">561</span> wals.clear();<a name="line.561"></a> -<span class="sourceLineNo">562</span> }<a name="line.562"></a> -<span class="sourceLineNo">563</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.563"></a> -<span class="sourceLineNo">564</span> wals.add(logName);<a name="line.564"></a> -<span class="sourceLineNo">565</span> existingPrefix = true;<a name="line.565"></a> -<span class="sourceLineNo">566</span> }<a name="line.566"></a> -<span class="sourceLineNo">567</span> }<a name="line.567"></a> -<span class="sourceLineNo">568</span> if (!existingPrefix) {<a name="line.568"></a> -<span class="sourceLineNo">569</span> // The new log belongs to a new group, add it into this peer<a name="line.569"></a> -<span class="sourceLineNo">570</span> LOG.debug("Start tracking logs for wal group {} for peer {}", logPrefix, peerId);<a name="line.570"></a> -<span class="sourceLineNo">571</span> NavigableSet<String> wals = new TreeSet<>();<a name="line.571"></a> -<span class="sourceLineNo">572</span> wals.add(logName);<a name="line.572"></a> -<span class="sourceLineNo">573</span> walsByPrefix.put(logPrefix, wals);<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> }<a name="line.576"></a> -<span class="sourceLineNo">577</span><a name="line.577"></a> -<span class="sourceLineNo">578</span> // Add to latestPaths<a name="line.578"></a> -<span class="sourceLineNo">579</span> Iterator<Path> iterator = latestPaths.iterator();<a name="line.579"></a> -<span class="sourceLineNo">580</span> while (iterator.hasNext()) {<a name="line.580"></a> -<span class="sourceLineNo">581</span> Path path = iterator.next();<a name="line.581"></a> -<span class="sourceLineNo">582</span> if (path.getName().contains(logPrefix)) {<a name="line.582"></a> -<span class="sourceLineNo">583</span> iterator.remove();<a name="line.583"></a> -<span class="sourceLineNo">584</span> break;<a name="line.584"></a> -<span class="sourceLineNo">585</span> }<a name="line.585"></a> -<span class="sourceLineNo">586</span> }<a name="line.586"></a> -<span class="sourceLineNo">587</span> this.latestPaths.add(newLog);<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> // public because of we call it in TestReplicationEmptyWALRecovery<a name="line.591"></a> -<span class="sourceLineNo">592</span> @VisibleForTesting<a name="line.592"></a> -<span class="sourceLineNo">593</span> public void postLogRoll(Path newLog) throws IOException {<a name="line.593"></a> -<span class="sourceLineNo">594</span> // This only updates the sources we own, not the recovered ones<a name="line.594"></a> -<span class="sourceLineNo">595</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.595"></a> -<span class="sourceLineNo">596</span> source.enqueueLog(newLog);<a name="line.596"></a> -<span class="sourceLineNo">597</span> }<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> @Override<a name="line.600"></a> -<span class="sourceLineNo">601</span> public void regionServerRemoved(String regionserver) {<a name="line.601"></a> -<span class="sourceLineNo">602</span> transferQueues(ServerName.valueOf(regionserver));<a name="line.602"></a> -<span class="sourceLineNo">603</span> }<a name="line.603"></a> -<span class="sourceLineNo">604</span><a name="line.604"></a> -<span class="sourceLineNo">605</span> /**<a name="line.605"></a> -<span class="sourceLineNo">606</span> * Transfer all the queues of the specified to this region server. First it tries to grab a lock<a name="line.606"></a> -<span class="sourceLineNo">607</span> * and if it works it will move the old queues and finally will delete the old queues.<a name="line.607"></a> -<span class="sourceLineNo">608</span> * <p><a name="line.608"></a> -<span class="sourceLineNo">609</span> * It creates one old source for any type of source of the old rs.<a name="line.609"></a> -<span class="sourceLineNo">610</span> */<a name="line.610"></a> -<span class="sourceLineNo">611</span> private void transferQueues(ServerName deadRS) {<a name="line.611"></a> -<span class="sourceLineNo">612</span> if (server.getServerName().equals(deadRS)) {<a name="line.612"></a> -<span class="sourceLineNo">613</span> // it's just us, give up<a name="line.613"></a> -<span class="sourceLineNo">614</span> return;<a name="line.614"></a> -<span class="sourceLineNo">615</span> }<a name="line.615"></a> -<span class="sourceLineNo">616</span> NodeFailoverWorker transfer = new NodeFailoverWorker(deadRS);<a name="line.616"></a> -<span class="sourceLineNo">617</span> try {<a name="line.617"></a> -<span class="sourceLineNo">618</span> this.executor.execute(transfer);<a name="line.618"></a> -<span class="sourceLineNo">619</span> } catch (RejectedExecutionException ex) {<a name="line.619"></a> -<span class="sourceLineNo">620</span> LOG.info("Cancelling the transfer of " + deadRS + " because of " + ex.getMessage());<a name="line.620"></a> -<span class="sourceLineNo">621</span> }<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> * Class responsible to setup new ReplicationSources to take care of the queues from dead region<a name="line.625"></a> -<span class="sourceLineNo">626</span> * servers.<a name="line.626"></a> -<span class="sourceLineNo">627</span> */<a name="line.627"></a> -<span class="sourceLineNo">628</span> class NodeFailoverWorker extends Thread {<a name="line.628"></a> -<span class="sourceLineNo">629</span><a name="line.629"></a> -<span class="sourceLineNo">630</span> private final ServerName deadRS;<a name="line.630"></a> -<span class="sourceLineNo">631</span> // After claim the queues from dead region server, the NodeFailoverWorker will skip to start<a name="line.631"></a> -<span class="sourceLineNo">632</span> // the RecoveredReplicationSource if the peer has been removed. but there's possible that<a name="line.632"></a> -<span class="sourceLineNo">633</span> // remove a peer with peerId = 2 and add a peer with peerId = 2 again during the<a name="line.633"></a> -<span class="sourceLineNo">634</span> // NodeFailoverWorker. So we need a deep copied <peerId, peer> map to decide whether we<a name="line.634"></a> -<span class="sourceLineNo">635</span> // should start the RecoveredReplicationSource. If the latest peer is not the old peer when<a name="line.635"></a> -<span class="sourceLineNo">636</span> // NodeFailoverWorker begin, we should skip to start the RecoveredReplicationSource, Otherwise<a name="line.636"></a> -<span class="sourceLineNo">637</span> // the rs will abort (See HBASE-20475).<a name="line.637"></a> -<span class="sourceLineNo">638</span> private final Map<String, ReplicationPeerImpl> peersSnapshot;<a name="line.638"></a> -<span class="sourceLineNo">639</span><a name="line.639"></a> -<span class="sourceLineNo">640</span> @VisibleForTesting<a name="line.640"></a> -<span class="sourceLineNo">641</span> public NodeFailoverWorker(ServerName deadRS) {<a name="line.641"></a> -<span class="sourceLineNo">642</span> super("Failover-for-" + deadRS);<a name="line.642"></a> -<span class="sourceLineNo">643</span> this.deadRS = deadRS;<a name="line.643"></a> -<span class="sourceLineNo">644</span> peersSnapshot = new HashMap<>(replicationPeers.getPeerCache());<a name="line.644"></a> -<span class="sourceLineNo">645</span> }<a name="line.645"></a> -<span class="sourceLineNo">646</span><a name="line.646"></a> -<span class="sourceLineNo">647</span> private boolean isOldPeer(String peerId, ReplicationPeerImpl newPeerRef) {<a name="line.647"></a> -<span class="sourceLineNo">648</span> ReplicationPeerImpl oldPeerRef = peersSnapshot.get(peerId);<a name="line.648"></a> -<span class="sourceLineNo">649</span> return oldPeerRef != null && oldPeerRef == newPeerRef;<a name="line.649"></a> -<span class="sourceLineNo">650</span> }<a name="line.650"></a> -<span class="sourceLineNo">651</span><a name="line.651"></a> -<span class="sourceLineNo">652</span> @Override<a name="line.652"></a> -<span class="sourceLineNo">653</span> public void run() {<a name="line.653"></a> -<span class="sourceLineNo">654</span> // Wait a bit before transferring the queues, we may be shutting down.<a name="line.654"></a> -<span class="sourceLineNo">655</span> // This sleep may not be enough in some cases.<a name="line.655"></a> -<span class="sourceLineNo">656</span> try {<a name="line.656"></a> -<span class="sourceLineNo">657</span> Thread.sleep(sleepBeforeFailover +<a name="line.657"></a> -<span class="sourceLineNo">658</span> (long) (ThreadLocalRandom.current().nextFloat() * sleepBeforeFailover));<a name="line.658"></a> -<span class="sourceLineNo">659</span> } catch (InterruptedException e) {<a name="line.659"></a> -<span class="sourceLineNo">660</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.660"></a> -<span class="sourceLineNo">661</span> Thread.currentThread().interrupt();<a name="line.661"></a> -<span class="sourceLineNo">662</span> }<a name="line.662"></a> -<span class="sourceLineNo">663</span> // We try to lock that rs' queue directory<a name="line.663"></a> -<span class="sourceLineNo">664</span> if (server.isStopped()) {<a name="line.664"></a> -<span class="sourceLineNo">665</span> LOG.info("Not transferring queue since we are shutting down");<a name="line.665"></a> -<span class="sourceLineNo">666</span> return;<a name="line.666"></a> -<span class="sourceLineNo">667</span> }<a name="line.667"></a> -<span class="sourceLineNo">668</span> Map<String, Set<String>> newQueues = new HashMap<>();<a name="line.668"></a> -<span class="sourceLineNo">669</span> try {<a name="line.669"></a> -<span class="sourceLineNo">670</span> List<String> queues = queueStorage.getAllQueues(deadRS);<a name="line.670"></a> -<span class="sourceLineNo">671</span> while (!queues.isEmpty()) {<a name="line.671"></a> -<span class="sourceLineNo">672</span> Pair<String, SortedSet<String>> peer = queueStorage.claimQueue(deadRS,<a name="line.672"></a> -<span class="sourceLineNo">673</span> queues.get(ThreadLocalRandom.current().nextInt(queues.size())), server.getServerName());<a name="line.673"></a> -<span class="sourceLineNo">674</span> long sleep = sleepBeforeFailover / 2;<a name="line.674"></a> -<span class="sourceLineNo">675</span> if (!peer.getSecond().isEmpty()) {<a name="line.675"></a> -<span class="sourceLineNo">676</span> newQueues.put(peer.getFirst(), peer.getSecond());<a name="line.676"></a> -<span class="sourceLineNo">677</span> sleep = sleepBeforeFailover;<a name="line.677"></a> -<span class="sourceLineNo">678</span> }<a name="line.678"></a> -<span class="sourceLineNo">679</span> try {<a name="line.679"></a> -<span class="sourceLineNo">680</span> Thread.sleep(sleep);<a name="line.680"></a> -<span class="sourceLineNo">681</span> } catch (InterruptedException e) {<a name="line.681"></a> -<span class="sourceLineNo">682</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.682"></a> -<span class="sourceLineNo">683</span> Thread.currentThread().interrupt();<a name="line.683"></a> -<span class="sourceLineNo">684</span> }<a name="line.684"></a> -<span class="sourceLineNo">685</span> queues = queueStorage.getAllQueues(deadRS);<a name="line.685"></a> -<span class="sourceLineNo">686</span> }<a name="line.686"></a> -<span class="sourceLineNo">687</span> if (queues.isEmpty()) {<a name="line.687"></a> -<span class="sourceLineNo">688</span> queueStorage.removeReplicatorIfQueueIsEmpty(deadRS);<a name="line.688"></a> -<span class="sourceLineNo">689</span> }<a name="line.689"></a> -<span class="sourceLineNo">690</span> } catch (ReplicationException e) {<a name="line.690"></a> -<span class="sourceLineNo">691</span> server.abort("Failed to claim queue from dead regionserver", e);<a name="line.691"></a> -<span class="sourceLineNo">692</span> return;<a name="line.692"></a> -<span class="sourceLineNo">693</span> }<a name="line.693"></a> -<span class="sourceLineNo">694</span> // Copying over the failed queue is completed.<a name="line.694"></a> -<span class="sourceLineNo">695</span> if (newQueues.isEmpty()) {<a name="line.695"></a> -<span class="sourceLineNo">696</span> // We either didn't get the lock or the failed region server didn't have any outstanding<a name="line.696"></a> -<span class="sourceLineNo">697</span> // WALs to replicate, so we are done.<a name="line.697"></a> -<span class="sourceLineNo">698</span> return;<a name="line.698"></a> -<span class="sourceLineNo">699</span> }<a name="line.699"></a> -<span class="sourceLineNo">700</span><a name="line.700"></a> -<span class="sourceLineNo">701</span> for (Map.Entry<String, Set<String>> entry : newQueues.entrySet()) {<a name="line.701"></a> -<span class="sourceLineNo">702</span> String queueId = entry.getKey();<a name="line.702"></a> -<span class="sourceLineNo">703</span> Set<String> walsSet = entry.getValue();<a name="line.703"></a> -<span class="sourceLineNo">704</span> try {<a name="line.704"></a> -<span class="sourceLineNo">705</span> // there is not an actual peer defined corresponding to peerId for the failover.<a name="line.705"></a> -<span class="sourceLineNo">706</span> ReplicationQueueInfo replicationQueueInfo = new ReplicationQueueInfo(queueId);<a name="line.706"></a> -<span class="sourceLineNo">707</span> String actualPeerId = replicationQueueInfo.getPeerId();<a name="line.707"></a> -<span class="sourceLineNo">708</span><a name="line.708"></a> -<span class="sourceLineNo">709</span> ReplicationPeerImpl peer = replicationPeers.getPeer(actualPeerId);<a name="line.709"></a> -<span class="sourceLineNo">710</span> if (peer == null || !isOldPeer(actualPeerId, peer)) {<a name="line.710"></a> -<span class="sourceLineNo">711</span> LOG.warn("Skipping failover for peer {} of node {}, peer is null", actualPeerId,<a name="line.711"></a> -<span class="sourceLineNo">712</span> deadRS);<a name="line.712"></a> -<span class="sourceLineNo">713</span> abortWhenFail(() -> queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.713"></a> -<span class="sourceLineNo">714</span> continue;<a name="line.714"></a> -<span class="sourceLineNo">715</span> }<a name="line.715"></a> -<span class="sourceLineNo">716</span> if (server instanceof ReplicationSyncUp.DummyServer<a name="line.716"></a> -<span class="sourceLineNo">717</span> && peer.getPeerState().equals(PeerState.DISABLED)) {<a name="line.717"></a> -<span class="sourceLineNo">718</span> LOG.warn("Peer {} is disabled. ReplicationSyncUp tool will skip "<a name="line.718"></a> -<span class="sourceLineNo">719</span> + "replicating data to this peer.",<a name="line.719"></a> -<span class="sourceLineNo">720</span> actualPeerId);<a name="line.720"></a> -<span class="sourceLineNo">721</span> continue;<a name="line.721"></a> -<span class="sourceLineNo">722</span> }<a name="line.722"></a> -<span class="sourceLineNo">723</span> // track sources in walsByIdRecoveredQueues<a name="line.723"></a> -<span class="sourceLineNo">724</span> Map<String, NavigableSet<String>> walsByGroup = new HashMap<>();<a name="line.724"></a> -<span class="sourceLineNo">725</span> walsByIdRecoveredQueues.put(queueId, walsByGroup);<a name="line.725"></a> -<span class="sourceLineNo">726</span> for (String wal : walsSet) {<a name="line.726"></a> -<span class="sourceLineNo">727</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(wal);<a name="line.727"></a> -<span class="sourceLineNo">728</span> NavigableSet<String> wals = walsByGroup.get(walPrefix);<a name="line.728"></a> -<span class="sourceLineNo">729</span> if (wals == null) {<a name="line.729"></a> -<span class="sourceLineNo">730</span> wals = new TreeSet<>();<a name="line.730"></a> -<span class="sourceLineNo">731</span> walsByGroup.put(walPrefix, wals);<a name="line.731"></a> -<span class="sourceLineNo">732</span> }<a name="line.732"></a> -<span class="sourceLineNo">733</span> wals.add(wal);<a name="line.733"></a> -<span class="sourceLineNo">734</span> }<a name="line.734"></a> -<span class="sourceLineNo">735</span><a name="line.735"></a> -<span class="sourceLineNo">736</span> ReplicationSourceInterface src = createSource(queueId, peer);<a name="line.736"></a> -<span class="sourceLineNo">737</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.737"></a> -<span class="sourceLineNo">738</span> synchronized (oldsources) {<a name="line.738"></a> -<span class="sourceLineNo">739</span> peer = replicationPeers.getPeer(src.getPeerId());<a name="line.739"></a> -<span class="sourceLineNo">740</span> if (peer == null || !isOldPeer(src.getPeerId(), peer)) {<a name="line.740"></a> -<span class="sourceLineNo">741</span> src.terminate("Recovered queue doesn't belong to any current peer");<a name="line.741"></a> -<span class="sourceLineNo">742</span> removeRecoveredSource(src);<a name="line.742"></a> -<span class="sourceLineNo">743</span> continue;<a name="line.743"></a> -<span class="sourceLineNo">744</span> }<a name="line.744"></a> -<span class="sourceLineNo">745</span> oldsources.add(src);<a name="line.745"></a> -<span class="sourceLineNo">746</span> for (String wal : walsSet) {<a name="line.746"></a> -<span class="sourceLineNo">747</span> src.enqueueLog(new Path(oldLogDir, wal));<a name="line.747"></a> -<span class="sourceLineNo">748</span> }<a name="line.748"></a> -<span class="sourceLineNo">749</span> src.startup();<a name="line.749"></a> -<span class="sourceLineNo">750</span> }<a name="line.750"></a> -<span class="sourceLineNo">751</span> } catch (IOException e) {<a name="line.751"></a> -<span class="sourceLineNo">752</span> // TODO manage it<a name="line.752"></a> -<span class="sourceLineNo">753</span> LOG.error("Failed creating a source", e);<a name="line.753"></a> -<span class="sourceLineNo">754</span> }<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> }<a name="line.757"></a> -<span class="sourceLineNo">758</span><a name="line.758"></a> -<span class="sourceLineNo">759</span> /**<a name="line.759"></a> -<span class="sourceLineNo">760</span> * Terminate the replication on this region server<a name="line.760"></a> -<span class="sourceLineNo">761</span> */<a name="line.761"></a> -<span class="sourceLineNo">762</span> public void join() {<a name="line.762"></a> -<span class="sourceLineNo">763</span> this.executor.shutdown();<a name="line.763"></a> -<span class="sourceLineNo">764</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.764"></a> -<span class="sourceLineNo">765</span> source.terminate("Region server is closing");<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><a name="line.768"></a> -<span class="sourceLineNo">769</span> /**<a name="line.769"></a> -<span class="sourceLineNo">770</span> * Get a copy of the wals of the normal sources on this rs<a name="line.770"></a> -<span class="sourceLineNo">771</span> * @return a sorted set of wal names<a name="line.771"></a> -<span class="sourceLineNo">772</span> */<a name="line.772"></a> -<span class="sourceLineNo">773</span> @VisibleForTesting<a name="line.773"></a> -<span class="sourceLineNo">774</span> public Map<String, Map<String, NavigableSet<String>>> getWALs() {<a name="line.774"></a> -<span class="sourceLineNo">775</span> return Collections.unmodifiableMap(walsById);<a name="line.775"></a> -<span class="sourceLineNo">776</span> }<a name="line.776"></a> -<span class="sourceLineNo">777</span><a name="line.777"></a> -<span class="sourceLineNo">778</span> /**<a name="line.778"></a> -<span class="sourceLineNo">779</span> * Get a copy of the wals of the recovered sources on this rs<a name="line.779"></a> -<span class="sourceLineNo">780</span> * @return a sorted set of wal names<a name="line.780"></a> -<span class="sourceLineNo">781</span> */<a name="line.781"></a> -<span class="sourceLineNo">782</span> @VisibleForTesting<a name="line.782"></a> -<span class="sourceLineNo">783</span> Map<String, Map<String, NavigableSet<String>>> getWalsByIdRecoveredQueues() {<a name="line.783"></a> -<span class="sourceLineNo">784</span> return Collections.unmodifiableMap(walsByIdRecoveredQueues);<a name="line.784"></a> -<span class="sourceLineNo">785</span> }<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> * Get a list of all the normal sources of this rs<a name="line.788"></a> -<span class="sourceLineNo">789</span> * @return list of all normal sources<a name="line.789"></a> -<span class="sourceLineNo">790</span> */<a name="line.790"></a> -<span class="sourceLineNo">791</span> public List<ReplicationSourceInterface> getSources() {<a name="line.791"></a> -<span class="sourceLineNo">792</span> return new ArrayList<>(this.sources.values());<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> * Get a list of all the recovered sources of this rs<a name="line.796"></a> -<span class="sourceLineNo">797</span> * @return list of all recovered sources<a name="line.797"></a> -<span class="sourceLineNo">798</span> */<a name="line.798"></a> -<span class="sourceLineNo">799</span> public List<ReplicationSourceInterface> getOldSources() {<a name="line.799"></a> -<span class="sourceLineNo">800</span> return this.oldsources;<a name="line.800"></a> -<span class="sourceLineNo">801</span> }<a name="line.801"></a> -<span class="sourceLineNo">802</span><a name="line.802"></a> -<span class="sourceLineNo">803</span> /**<a name="line.803"></a> -<span class="sourceLineNo">804</span> * Get the normal source for a given peer<a name="line.804"></a> -<span class="sourceLineNo">805</span> * @return the normal source for the give peer if it exists, otherwise null.<a name="line.805"></a> -<span class="sourceLineNo">806</span> */<a name="line.806"></a> -<span class="sourceLineNo">807</span> @VisibleForTesting<a name="line.807"></a> -<span class="sourceLineNo">808</span> public ReplicationSourceInterface getSource(String peerId) {<a name="line.808"></a> -<span class="sourceLineNo">809</span> return this.sources.get(peerId);<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> @VisibleForTesting<a name="line.812"></a> -<span class="sourceLineNo">813</span> List<String> getAllQueues() throws IOException {<a name="line.813"></a> -<span class="sourceLineNo">814</span> List<String> allQueues = Collections.emptyList();<a name="line.814"></a> -<span class="sourceLineNo">815</span> try {<a name="line.815"></a> -<span class="sourceLineNo">816</span> allQueues = queueStorage.getAllQueues(server.getServerName());<a name="line.816"></a> -<span class="sourceLineNo">817</span> } catch (ReplicationException e) {<a name="line.817"></a> -<span class="sourceLineNo">818</span> throw new IOException(e);<a name="line.818"></a> -<span class="sourceLineNo">819</span> }<a name="line.819"></a> -<span class="sourceLineNo">820</span> return allQueues;<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> @VisibleForTesting<a name="line.823"></a> -<span class="sourceLineNo">824</span> int getSizeOfLatestPath() {<a name="line.824"></a> -<span class="sourceLineNo">825</span> synchronized (latestPaths) {<a name="line.825"></a> -<span class="sourceLineNo">826</span> return latestPaths.size();<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">829</span><a name="line.829"></a> -<span class="sourceLineNo">830</span> @VisibleForTesting<a name="line.830"></a> -<span class="sourceLineNo">831</span> public AtomicLong getTotalBufferUsed() {<a name="line.831"></a> -<span class="sourceLineNo">832</span> return totalBufferUsed;<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> -<span class="sourceLineNo">836</span> * Get the directory where wals are archived<a name="line.836"></a> -<span class="sourceLineNo">837</span> * @return the directory where wals are archived<a name="line.837"></a> -<span class="sourceLineNo">838</span> */<a name="line.838"></a> -<span class="sourceLineNo">839</span> public Path getOldLogDir() {<a name="line.839"></a> -<span class="sourceLineNo">840</span> return this.oldLogDir;<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> /**<a name="line.843"></a> -<span class="sourceLineNo">844</span> * Get the directory where wals are stored by their RSs<a name="line.844"></a> -<span class="sourceLineNo">845</span> * @return the directory where wals are stored by their RSs<a name="line.845"></a> -<span class="sourceLineNo">846</span> */<a name="line.846"></a> -<span class="sourceLineNo">847</span> public Path getLogDir() {<a name="line.847"></a> -<span class="sourceLineNo">848</span> return this.logDir;<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> /**<a name="line.851"></a> -<span class="sourceLineNo">852</span> * Get the handle on the local file system<a name="line.852"></a> -<span class="sourceLineNo">853</span> * @return Handle on the local file system<a name="line.853"></a> -<span class="sourceLineNo">854</span> */<a name="line.854"></a> -<span class="sourceLineNo">855</span> public FileSystem getFs() {<a name="line.855"></a> -<span class="sourceLineNo">856</span> return this.fs;<a name="line.856"></a> -<span class="sourceLineNo">857</span> }<a name="line.857"></a> -<span class="sourceLineNo">858</span><a name="line.858"></a> -<span class="sourceLineNo">859</span> /**<a name="line.859"></a> -<span class="sourceLineNo">860</span> * Get the ReplicationPeers used by this ReplicationSourceManager<a name="line.860"></a> -<span class="sourceLineNo">861</span> * @return the ReplicationPeers used by this ReplicationSourceManager<a name="line.861"></a> -<span class="sourceLineNo">862</span> */<a name="line.862"></a> -<span class="sourceLineNo">863</span> public ReplicationPeers getReplicationPeers() {<a name="line.863"></a> -<span class="sourceLineNo">864</span> return this.replicationPeers;<a name="line.864"></a> -<span class="sourceLineNo">865</span> }<a name="line.865"></a> -<span class="sourceLineNo">866</span><a name="line.866"></a> -<span class="sourceLineNo">867</span> /**<a name="line.867"></a> -<span class="sourceLineNo">868</span> * Get a string representation of all the sources' metrics<a name="line.868"></a> -<span class="sourceLineNo">869</span> */<a name="line.869"></a> -<span class="sourceLineNo">870</span> public String getStats() {<a name="line.870"></a> -<span class="sourceLineNo">871</span> StringBuilder stats = new StringBuilder();<a name="line.871"></a> -<span class="sourceLineNo">872</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.872"></a> -<span class="sourceLineNo">873</span> stats.append("Normal source for cluster " + source.getPeerId() + ": ");<a name="line.873"></a> -<span class="sourceLineNo">874</span> stats.append(source.getStats() + "\n");<a name="line.874"></a> -<span class="sourceLineNo">875</span> }<a name="line.875"></a> -<span class="sourceLineNo">876</span> for (ReplicationSourceInterface oldSource : oldsources) {<a name="line.876"></a> -<span class="sourceLineNo">877</span> stats.append("Recovered source for cluster/machine(s) " + oldSource.getPeerId() + ": ");<a name="line.877"></a> -<span class="sourceLineNo">878</span> stats.append(oldSource.getStats() + "\n");<a name="line.878"></a> -<span class="sourceLineNo">879</span> }<a name="line.879"></a> -<span class="sourceLineNo">880</span> return stats.toString();<a name="line.880"></a> -<span class="sourceLineNo">881</span> }<a name="line.881"></a> -<span class="sourceLineNo">882</span><a name="line.882"></a> -<span class="sourceLineNo">883</span> public void addHFileRefs(TableName tableName, byte[] family, List<Pair<Path, Path>> pairs)<a name="line.883"></a> -<span class="sourceLineNo">884</span> throws IOException {<a name="line.884"></a> -<span class="sourceLineNo">885</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.885"></a> -<span class="sourceLineNo">886</span> throwIOExceptionWhenFail(() -> source.addHFileRefs(tableName, family, pairs));<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><a name="line.889"></a> -<span class="sourceLineNo">890</span> public void cleanUpHFileRefs(String peerId, List<String> files) {<a name="line.890"></a> -<span class="sourceLineNo">891</span> abortWhenFail(() -> this.queueStorage.removeHFileRefs(peerId, files));<a name="line.891"></a> -<span class="sourceLineNo">892</span> }<a name="line.892"></a> -<span class="sourceLineNo">893</span><a name="line.893"></a> -<span class="sourceLineNo">894</span> int activeFailoverTaskCount() {<a name="line.894"></a> -<span class="sourceLineNo">895</span> return executor.getActiveCount();<a name="line.895"></a> -<span class="sourceLineNo">896</span> }<a name="line.896"></a> -<span class="sourceLineNo">897</span>}<a name="line.897"></a> +<span class="sourceLineNo">421</span> this.oldsources.remove(src);<a name="line.421"></a> +<span class="sourceLineNo">422</span> // Delete queue from storage and memory<a name="line.422"></a> +<span class="sourceLineNo">423</span> deleteQueue(src.getQueueId());<a name="line.423"></a> +<span class="sourceLineNo">424</span> this.walsByIdRecoveredQueues.remove(src.getQueueId());<a name="line.424"></a> +<span class="sourceLineNo">425</span> }<a name="line.425"></a> +<span class="sourceLineNo">426</span><a name="line.426"></a> +<span class="sourceLineNo">427</span> /**<a name="line.427"></a> +<span class="sourceLineNo">428</span> * Clear the metrics and related replication queue of the specified old source<a name="line.428"></a> +<span class="sourceLineNo">429</span> * @param src source to clear<a name="line.429"></a> +<span class="sourceLineNo">430</span> */<a name="line.430"></a> +<span class="sourceLineNo">431</span> void removeSource(ReplicationSourceInterface src) {<a name="line.431"></a> +<span class="sourceLineNo">432</span> LOG.info("Done with the queue " + src.getQueueId());<a name="line.432"></a> +<span class="sourceLineNo">433</span> this.sources.remove(src.getPeerId());<a name="line.433"></a> +<span class="sourceLineNo">434</span> // Delete queue from storage and memory<a name="line.434"></a> +<span class="sourceLineNo">435</span> deleteQueue(src.getQueueId());<a name="line.435"></a> +<span class="sourceLineNo">436</span> this.walsById.remove(src.getQueueId());<a name="line.436"></a> +<span class="sourceLineNo">437</span> }<a name="line.437"></a> +<span class="sourceLineNo">438</span><a name="line.438"></a> +<span class="sourceLineNo">439</span> /**<a name="line.439"></a> +<span class="sourceLineNo">440</span> * Delete a complete queue of wals associated with a replication source<a name="line.440"></a> +<span class="sourceLineNo">441</span> * @param queueId the id of replication queue to delete<a name="line.441"></a> +<span class="sourceLineNo">442</span> */<a name="line.442"></a> +<span class="sourceLineNo">443</span> private void deleteQueue(String queueId) {<a name="line.443"></a> +<span class="sourceLineNo">444</span> abortWhenFail(() -> this.queueStorage.removeQueue(server.getServerName(), queueId));<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> @FunctionalInterface<a name="line.447"></a> +<span class="sourceLineNo">448</span> private interface ReplicationQueueOperation {<a name="line.448"></a> +<span class="sourceLineNo">449</span> void exec() throws ReplicationException;<a name="line.449"></a> +<span class="sourceLineNo">450</span> }<a name="line.450"></a> +<span class="sourceLineNo">451</span><a name="line.451"></a> +<span class="sourceLineNo">452</span> private void abortWhenFail(ReplicationQueueOperation op) {<a name="line.452"></a> +<span class="sourceLineNo">453</span> try {<a name="line.453"></a> +<span class="sourceLineNo">454</span> op.exec();<a name="line.454"></a> +<span class="sourceLineNo">455</span> } catch (ReplicationException e) {<a name="line.455"></a> +<span class="sourceLineNo">456</span> server.abort("Failed to operate on replication queue", e);<a name="line.456"></a> +<span class="sourceLineNo">457</span> }<a name="line.457"></a> +<span class="sourceLineNo">458</span> }<a name="line.458"></a> +<span class="sourceLineNo">459</span><a name="line.459"></a> +<span class="sourceLineNo">460</span> private void throwIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.460"></a> +<span class="sourceLineNo">461</span> try {<a name="line.461"></a> +<span class="sourceLineNo">462</span> op.exec();<a name="line.462"></a> +<span class="sourceLineNo">463</span> } catch (ReplicationException e) {<a name="line.463"></a> +<span class="sourceLineNo">464</span> throw new IOException(e);<a name="line.464"></a> +<span class="sourceLineNo">465</span> }<a name="line.465"></a> +<span class="sourceLineNo">466</span> }<a name="line.466"></a> +<span class="sourceLineNo">467</span><a name="line.467"></a> +<span class="sourceLineNo">468</span> private void abortAndThrowIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.468"></a> +<span class="sourceLineNo">469</span> try {<a name="line.469"></a> +<span class="sourceLineNo">470</span> op.exec();<a name="line.470"></a> +<span class="sourceLineNo">471</span> } catch (ReplicationException e) {<a name="line.471"></a> +<span class="sourceLineNo">472</span> server.abort("Failed to operate on replication queue", e);<a name="line.472"></a> +<span class="sourceLineNo">473</span> throw new IOException(e);<a name="line.473"></a> +<span class="sourceLineNo">474</span> }<a name="line.474"></a> +<span class="sourceLineNo">475</span> }<a name="line.475"></a> +<span class="sourceLineNo">476</span><a name="line.476"></a> +<span class="sourceLineNo">477</span> /**<a name="line.477"></a> +<span class="sourceLineNo">478</span> * This method will log the current position to storage. And also clean old logs from the<a name="line.478"></a> +<span class="sourceLineNo">479</span> * replication queue.<a name="line.479"></a> +<span class="sourceLineNo">480</span> * @param queueId id of the replication queue<a name="line.480"></a> +<span class="sourceLineNo">481</span> * @param queueRecovered indicates if this queue comes from another region server<a name="line.481"></a> +<span class="sourceLineNo">482</span> * @param entryBatch the wal entry batch we just shipped<a name="line.482"></a> +<span class="sourceLineNo">483</span> */<a name="line.483"></a> +<span class="sourceLineNo">484</span> public void logPositionAndCleanOldLogs(String queueId, boolean queueRecovered,<a name="line.484"></a> +<span class="sourceLineNo">485</span> WALEntryBatch entryBatch) {<a name="line.485"></a> +<span class="sourceLineNo">486</span> String fileName = entryBatch.getLastWalPath().getName();<a name="line.486"></a> +<span class="sourceLineNo">487</span> abortWhenFail(() -> this.queueStorage.setWALPosition(server.getServerName(), queueId, fileName,<a name="line.487"></a> +<span class="sourceLineNo">488</span> entryBatch.getLastWalPosition(), entryBatch.getLastSeqIds()));<a name="line.488"></a> +<span class="sourceLineNo">489</span> cleanOldLogs(fileName, entryBatch.isEndOfFile(), queueId, queueRecovered);<a name="line.489"></a> +<span class="sourceLineNo">490</span> }<a name="line.490"></a> +<span class="sourceLineNo">491</span><a name="line.491"></a> +<span class="sourceLineNo">492</span> /**<a name="line.492"></a> +<span class="sourceLineNo">493</span> * Cleans a log file and all older logs from replication queue. Called when we are sure that a log<a name="line.493"></a> +<span class="sourceLineNo">494</span> * file is closed and has no more entries.<a name="line.494"></a> +<span class="sourceLineNo">495</span> * @param log Path to the log<a name="line.495"></a> +<span class="sourceLineNo">496</span> * @param inclusive whether we should also remove the given log file<a name="line.496"></a> +<span class="sourceLineNo">497</span> * @param queueId id of the replication queue<a name="line.497"></a> +<span class="sourceLineNo">498</span> * @param queueRecovered Whether this is a recovered queue<a name="line.498"></a> +<span class="sourceLineNo">499</span> */<a name="line.499"></a> +<span class="sourceLineNo">500</span> @VisibleForTesting<a name="line.500"></a> +<span class="sourceLineNo">501</span> void cleanOldLogs(String log, boolean inclusive, String queueId, boolean queueRecovered) {<a name="line.501"></a> +<span class="sourceLineNo">502</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log);<a name="line.502"></a> +<span class="sourceLineNo">503</span> if (queueRecovered) {<a name="line.503"></a> +<span class="sourceLineNo">504</span> NavigableSet<String> wals = walsByIdRecoveredQueues.get(queueId).get(logPrefix);<a name="line.504"></a> +<span class="sourceLineNo">505</span> if (wals != null) {<a name="line.505"></a> +<span class="sourceLineNo">506</span> cleanOldLogs(wals, log, inclusive, queueId);<a name="line.506"></a> +<span class="sourceLineNo">507</span> }<a name="line.507"></a> +<span class="sourceLineNo">508</span> } else {<a name="line.508"></a> +<span class="sourceLineNo">509</span> // synchronized on walsById to avoid race with preLogRoll<a name="line.509"></a> +<span class="sourceLineNo">510</span> synchronized (this.walsById) {<a name="line.510"></a> +<span class="sourceLineNo">511</span> NavigableSet<String> wals = walsById.get(queueId).get(logPrefix);<a name="line.511"></a> +<span class="sourceLineNo">512</span> if (wals != null) {<a name="line.512"></a> +<span class="sourceLineNo">513</span> cleanOldLogs(wals, log, inclusive, queueId);<a name="line.513"></a> +<span class="sourceLineNo">514</span> }<a name="line.514"></a> +<span class="sourceLineNo">515</span> }<a name="line.515"></a> +<span class="sourceLineNo">516</span> }<a name="line.516"></a> +<span class="sourceLineNo">517</span> }<a name="line.517"></a> +<span class="sourceLineNo">518</span><a name="line.518"></a> +<span class="sourceLineNo">519</span> private void cleanOldLogs(NavigableSet<String> wals, String key, boolean inclusive, String id) {<a name="line.519"></a> +<span class="sourceLineNo">520</span> NavigableSet<String> walSet = wals.headSet(key, inclusive);<a name="line.520"></a> +<span class="sourceLineNo">521</span> if (walSet.isEmpty()) {<a name="line.521"></a> +<span class="sourceLineNo">522</span> return;<a name="line.522"></a> +<span class="sourceLineNo">523</span> }<a name="line.523"></a> +<span class="sourceLineNo">524</span> LOG.debug("Removing {} logs in the list: {}", walSet.size(), walSet);<a name="line.524"></a> +<span class="sourceLineNo">525</span> for (String wal : walSet) {<a name="line.525"></a> +<span class="sourceLineNo">526</span> abortWhenFail(() -> this.queueStorage.removeWAL(server.getServerName(), id, wal));<a name="line.526"></a> +<span class="sourceLineNo">527</span> }<a name="line.527"></a> +<span class="sourceLineNo">528</span> walSet.clear();<a name="line.528"></a> +<span class="sourceLineNo">529</span> }<a name="line.529"></a> +<span class="sourceLineNo">530</span><a name="line.530"></a> +<span class="sourceLineNo">531</span> // public because of we call it in TestReplicationEmptyWALRecovery<a name="line.531"></a> +<span class="sourceLineNo">532</span> @VisibleForTesting<a name="line.532"></a> +<span class="sourceLineNo">533</span> public void preLogRoll(Path newLog) throws IOException {<a name="line.533"></a> +<span class="sourceLineNo">534</span> String logName = newLog.getName();<a name="line.534"></a> +<span class="sourceLineNo">535</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.535"></a> +<span class="sourceLineNo">536</span> // synchronized on latestPaths to avoid the new open source miss the new log<a name="line.536"></a> +<span class="sourceLineNo">537</span> synchronized (this.latestPaths) {<a name="line.537"></a> +<span class="sourceLineNo">538</span> // Add log to queue storage<a name="line.538"></a> +<span class="sourceLineNo">539</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.539"></a> +<span class="sourceLineNo">540</span> // If record log to queue storage failed, abort RS and throw exception to make log roll<a name="line.540"></a> +<span class="sourceLineNo">541</span> // failed<a name="line.541"></a> +<span class="sourceLineNo">542</span> abortAndThrowIOExceptionWhenFail(<a name="line.542"></a> +<span class="sourceLineNo">543</span> () -> this.queueStorage.addWAL(server.getServerName(), source.getQueueId(), logName));<a name="line.543"></a> +<span class="sourceLineNo">544</span> }<a name="line.544"></a> +<span class="sourceLineNo">545</span><a name="line.545"></a> +<span class="sourceLineNo">546</span> // synchronized on walsById to avoid race with cleanOldLogs<a name="line.546"></a> +<span class="sourceLineNo">547</span> synchronized (this.walsById) {<a name="line.547"></a> +<span class="sourceLineNo">548</span> // Update walsById map<a name="line.548"></a> +<span class="sourceLineNo">549</span> for (Map.Entry<String, Map<String, NavigableSet<String>>> entry : this.walsById<a name="line.549"></a> +<span class="sourceLineNo">550</span> .entrySet()) {<a name="line.550"></a> +<span class="sourceLineNo">551</span> String peerId = entry.getKey();<a name="line.551"></a> +<span class="sourceLineNo">552</span> Map<String, NavigableSet<String>> walsByPrefix = entry.getValue();<a name="line.552"></a> +<span class="sourceLineNo">553</span> boolean existingPrefix = false;<a name="line.553"></a> +<span class="sourceLineNo">554</span> for (Map.Entry<String, NavigableSet<String>> walsEntry : walsByPrefix.entrySet()) {<a name="line.554"></a> +<span class="sourceLineNo">555</span> SortedSet<String> wals = walsEntry.getValue();<a name="line.555"></a> +<span class="sourceLineNo">556</span> if (this.sources.isEmpty()) {<a name="line.556"></a> +<span class="sourceLineNo">557</span> // If there's no slaves, don't need to keep the old wals since<a name="line.557"></a> +<span class="sourceLineNo">558</span> // we only consider the last one when a new slave comes in<a name="line.558"></a> +<span class="sourceLineNo">559</span> wals.clear();<a name="line.559"></a> +<span class="sourceLineNo">560</span> }<a name="line.560"></a> +<span class="sourceLineNo">561</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.561"></a> +<span class="sourceLineNo">562</span> wals.add(logName);<a name="line.562"></a> +<span class="sourceLineNo">563</span> existingPrefix = true;<a name="line.563"></a> +<span class="sourceLineNo">564</span> }<a name="line.564"></a> +<span class="sourceLineNo">565</span> }<a name="line.565"></a> +<span class="sourceLineNo">566</span> if (!existingPrefix) {<a name="line.566"></a> +<span class="sourceLineNo">567</span> // The new log belongs to a new group, add it into this peer<a name="line.567"></a> +<span class="sourceLineNo">568</span> LOG.debug("Start tracking logs for wal group {} for peer {}", logPrefix, peerId);<a name="line.568"></a> +<span class="sourceLineNo">569</span> NavigableSet<String> wals = new TreeSet<>();<a name="line.569"></a> +<span class="sourceLineNo">570</span> wals.add(logName);<a name="line.570"></a> +<span class="sourceLineNo">571</span> walsByPrefix.put(logPrefix, wals);<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><a name="line.575"></a> +<span class="sourceLineNo">576</span> // Add to latestPaths<a name="line.576"></a> +<span class="sourceLineNo">577</span> Iterator<Path> iterator = latestPaths.iterator();<a name="line.577"></a> +<span class="sourceLineNo">578</span> while (iterator.hasNext()) {<a name="line.578"></a> +<span class="sourceLineNo">579</span> Path path = iterator.next();<a name="line.579"></a> +<span class="sourceLineNo">580</span> if (path.getName().contains(logPrefix)) {<a name="line.580"></a> +<span class="sourceLineNo">581</span> iterator.remove();<a name="line.581"></a> +<span class="sourceLineNo">582</span> break;<a name="line.582"></a> +<span class="sourceLineNo">583</span> }<a name="line.583"></a> +<span class="sourceLineNo">584</span> }<a name="line.584"></a> +<span class="sourceLineNo">585</span> this.latestPaths.add(newLog);<a name="line.585"></a> +<span class="sourceLineNo">586</span> }<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> // public because of we call it in TestReplicationEmptyWALRecovery<a name="line.589"></a> +<span class="sourceLineNo">590</span> @VisibleForTesting<a name="line.590"></a> +<span class="sourceLineNo">591</span> public void postLogRoll(Path newLog) throws IOException {<a name="line.591"></a> +<span class="sourceLineNo">592</span> // This only updates the sources we own, not the recovered ones<a name="line.592"></a> +<span class="sourceLineNo">593</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.593"></a> +<span class="sourceLineNo">594</span> source.enqueueLog(newLog);<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> @Override<a name="line.598"></a> +<span class="sourceLineNo">599</span> public void regionServerRemoved(String regionserver) {<a name="line.599"></a> +<span class="sourceLineNo">600</span> transferQueues(ServerName.valueOf(regionserver));<a name="line.600"></a> +<span class="sourceLineNo">601</span> }<a name="line.601"></a> +<span class="sourceLineNo">602</span><a name="line.602"></a> +<span class="sourceLineNo">603</span> /**<a name="line.603"></a> +<span class="sourceLineNo">604</span> * Transfer all the queues of the specified to this region server. First it tries to grab a lock<a name="line.604"></a> +<span class="sourceLineNo">605</span> * and if it works it will move the old queues and finally will delete the old queues.<a name="line.605"></a> +<span class="sourceLineNo">606</span> * <p><a name="line.606"></a> +<span class="sourceLineNo">607</span> * It creates one old source for any type of source of the old rs.<a name="line.607"></a> +<span class="sourceLineNo">608</span> */<a name="line.608"></a> +<span class="sourceLineNo">609</span> private void transferQueues(ServerName deadRS) {<a name="line.609"></a> +<span class="sourceLineNo">610</span> if (server.getServerName().equals(deadRS)) {<a name="line.610"></a> +<span class="sourceLineNo">611</span> // it's just us, give up<a name="line.611"></a> +<span class="sourceLineNo">612</span> return;<a name="line.612"></a> +<span class="sourceLineNo">613</span> }<a name="line.613"></a> +<span class="sourceLineNo">614</span> NodeFailoverWorker transfer = new NodeFailoverWorker(deadRS);<a name="line.614"></a> +<span class="sourceLineNo">615</span> try {<a name="line.615"></a> +<span class="sourceLineNo">616</span> this.executor.execute(transfer);<a name="line.616"></a> +<span class="sourceLineNo">617</span> } catch (RejectedExecutionException ex) {<a name="line.617"></a> +<span class="sourceLineNo">618</span> LOG.info("Cancelling the transfer of " + deadRS + " because of " + ex.getMessage());<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><a name="line.621"></a> +<span class="sourceLineNo">622</span> /**<a name="line.622"></a> +<span class="sourceLineNo">623</span> * Class responsible to setup new ReplicationSources to take care of the queues from dead region<a name="line.623"></a> +<span class="sourceLineNo">624</span> * servers.<a name="line.624"></a> +<span class="sourceLineNo">625</span> */<a name="line.625"></a> +<span class="sourceLineNo">626</span> class NodeFailoverWorker extends Thread {<a name="line.626"></a> +<span class="sourceLineNo">627</span><a name="line.627"></a> +<span class="sourceLineNo">628</span> private final ServerName deadRS;<a name="line.628"></a> +<span class="sourceLineNo">629</span> // After claim the queues from dead region server, the NodeFailoverWorker will skip to start<a name="line.629"></a> +<span class="sourceLineNo">630</span> // the RecoveredReplicationSource if the peer has been removed. but there's possible that<a name="line.630"></a> +<span class="sourceLineNo">631</span> // remove a peer with peerId = 2 and add a peer with peerId = 2 again during the<a name="line.631"></a> +<span class="sourceLineNo">632</span> // NodeFailoverWorker. So we need a deep copied <peerId, peer> map to decide whether we<a name="line.632"></a> +<span class="sourceLineNo">633</span> // should start the RecoveredReplicationSource. If the latest peer is not the old peer when<a name="line.633"></a> +<span class="sourceLineNo">634</span> // NodeFailoverWorker begin, we should skip to start the RecoveredReplicationSource, Otherwise<a name="line.634"></a> +<span class="sourceLineNo">635</span> // the rs will abort (See HBASE-20475).<a name="line.635"></a> +<span class="sourceLineNo">636</span> private final Map<String, ReplicationPeerImpl> peersSnapshot;<a name="line.636"></a> +<span class="sourceLineNo">637</span><a name="line.637"></a> +<span class="sourceLineNo">638</span> @VisibleForTesting<a name="line.638"></a> +<span class="sourceLineNo">639</span> public NodeFailoverWorker(ServerName deadRS) {<a name="line.639"></a> +<span class="sourceLineNo">640</span> super("Failover-for-" + deadRS);<a name="line.640"></a> +<span class="sourceLineNo">641</span> this.deadRS = deadRS;<a name="line.641"></a> +<span class="sourceLineNo">642</span> peersSnapshot = new HashMap<>(replicationPeers.getPeerCache());<a name="line.642"></a> +<span class="sourceLineNo">643</span> }<a name="line.643"></a> +<span class="sourceLineNo">644</span><a name="line.644"></a> +<span class="sourceLineNo">645</span> private boolean isOldPeer(String peerId, ReplicationPeerImpl newPeerRef) {<a name="line.645"></a> +<span class="sourceLineNo">646</span> ReplicationPeerImpl oldPeerRef = peersSnapshot.get(peerId);<a name="line.646"></a> +<span class="sourceLineNo">647</span> return oldPeerRef != null && oldPeerRef == newPeerRef;<a name="line.647"></a> +<span class="sourceLineNo">648</span> }<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 void run() {<a name="line.651"></a> +<span class="sourceLineNo">652</span> // Wait a bit before transferring the queues, we may be shutting down.<a name="line.652"></a> +<span class="sourceLineNo">653</span> // This sleep may not be enough in some cases.<a name="line.653"></a> +<span class="sourceLineNo">654</span> try {<a name="line.654"></a> +<span class="sourceLineNo">655</span> Thread.sleep(sleepBeforeFailover +<a name="line.655"></a> +<span class="sourceLineNo">656</span> (long) (ThreadLocalRandom.current().nextFloat() * sleepBeforeFailover));<a name="line.656"></a> +<span class="sourceLineNo">657</span> } catch (InterruptedException e) {<a name="line.657"></a> +<span class="sourceLineNo">658</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.658"></a> +<span class="sourceLineNo">659</span> Thread.currentThread().interrupt();<a name="line.659"></a> +<span class="sourceLineNo">660</span> }<a name="line.660"></a> +<span class="sourceLineNo">661</span> // We try to lock that rs' queue directory<a name="line.661"></a> +<span class="sourceLineNo">662</span> if (server.isStopped()) {<a name="line.662"></a> +<span class="sourceLineNo">663</span> LOG.info("Not transferring queue since we are shutting down");<a name="line.663"></a> +<span class="sourceLineNo">664</span> return;<a name="line.664"></a> +<span class="sourceLineNo">665</span> }<a name="line.665"></a> +<span class="sourceLineNo">666</span> Map<String, Set<String>> newQueues = new HashMap<>();<a name="line.666"></a> +<span class="sourceLineNo">667</span> try {<a name="line.667"></a> +<span class="sourceLineNo">668</span> List<String> queues = queueStorage.getAllQueues(deadRS);<a name="line.668"></a> +<span class="sourceLineNo">669</span> while (!queues.isEmpty()) {<a name="line.669"></a> +<span class="sourceLineNo">670</span> Pair<String, SortedSet<String>> peer = queueStorage.claimQueue(deadRS,<a name="line.670"></a> +<span class="sourceLineNo">671</span> queues.get(ThreadLocalRandom.current().nextInt(queues.size())), server.getServerName());<a name="line.671"></a> +<span class="sourceLineNo">672</span> long sleep = sleepBeforeFailover / 2;<a name="line.672"></a> +<span class="sourceLineNo">673</span> if (!peer.getSecond().isEmpty()) {<a name="line.673"></a> +<span class="sourceLineNo">674</span> newQueues.put(peer.getFirst(), peer.getSecond());<a name="line.674"></a> +<span class="sourceLineNo">675</span> sleep = sleepBeforeFailover;<a name="line.675"></a> +<span class="sourceLineNo">676</span> }<a name="line.676"></a> +<span class="sourceLineNo">677</span> try {<a name="line.677"></a> +<span class="sourceLineNo">678</span> Thread.sleep(sleep);<a name="line.678"></a> +<span class="sourceLineNo">679</span> } catch (InterruptedException e) {<a name="line.679"></a> +<span class="sourceLineNo">680</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.680"></a> +<span class="sourceLineNo">681</span> Thread.currentThread().interrupt();<a name="line.681"></a> +<span class="sourceLineNo">682</span> }<a name="line.682"></a> +<span class="sourceLineNo">683</span> queues = queueStorage.getAllQueues(deadRS);<a name="line.683"></a> +<span class="sourceLineNo">684</span> }<a name="line.684"></a> +<span class="sourceLineNo">685</span> if (queues.isEmpty()) {<a name="line.685"></a> +<span class="sourceLineNo">686</span> queueStorage.removeReplicatorIfQueueIsEmpty(deadRS);<a name="line.686"></a> +<span class="sourceLineNo">687</span> }<a name="line.687"></a> +<span class="sourceLineNo">688</span> } catch (ReplicationException e) {<a name="line.688"></a> +<span class="sourceLineNo">689</span> server.abort("Failed to claim queue from dead regionserver", e);<a name="line.689"></a> +<span class="sourceLineNo">690</span> return;<a name="line.690"></a> +<span class="sourceLineNo">691</span> }<a name="line.691"></a> +<span class="sourceLineNo">692</span> // Copying over the failed queue is completed.<a name="line.692"></a> +<span class="sourceLineNo">693</span> if (newQueues.isEmpty()) {<a name="line.693"></a> +<span class="sourceLineNo">694</span> // We either didn't get the lock or the failed region server didn't have any outstanding<a name="line.694"></a> +<span class="sourceLineNo">695</span> // WALs to replicate, so we are done.<a name="line.695"></a> +<span class="sourceLineNo">696</span> return;<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> for (Map.Entry<String, Set<String>> entry : newQueues.entrySet()) {<a name="line.699"></a> +<span class="sourceLineNo">700</span> String queueId = entry.getKey();<a name="line.700"></a> +<span class="sourceLineNo">701</span> Set<String> walsSet = entry.getValue();<a name="line.701"></a> +<span class="sourceLineNo">702</span> try {<a name="line.702"></a> +<span class="sourceLineNo">703</span> // there is not an actual peer defined corresponding to peerId for the failover.<a name="line.703"></a> +<span class="sourceLineNo">704</span> ReplicationQueueInfo replicationQueueInfo = new ReplicationQueueInfo(queueId);<a name="line.704"></a> +<span class="sourceLineNo">705</span> String actualPeerId = replicationQueueInfo.getPeerId();<a name="line.705"></a> +<span class="sourceLineNo">706</span><a name="line.706"></a> +<span class="sourceLineNo">707</span> ReplicationPeerImpl peer = replicationPeers.getPeer(actualPeerId);<a name="line.707"></a> +<span class="sourceLineNo">708</span> if (peer == null || !isOldPeer(actualPeerId, peer)) {<a name="line.708"></a> +<span class="sourceLineNo">709</span> LOG.warn("Skipping failover for peer {} of node {}, peer is null", actualPeerId,<a name="line.709"></a> +<span class="sourceLineNo">710</span> deadRS);<a name="line.710"></a> +<span class="sourceLineNo">711</span> abortWhenFail(() -> queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.711"></a> +<span class="sourceLineNo">712</span> continue;<a name="line.712"></a> +<span class="sourceLineNo">713</span> }<a name="line.713"></a> +<span class="sourceLineNo">714</span> if (server instanceof ReplicationSyncUp.DummyServer<a name="line.714"></a> +<span class="sourceLineNo">715</span> && peer.getPeerState().equals(PeerState.DISABLED)) {<a name="line.715"></a> +<span class="sourceLineNo">716</span> LOG.warn("Peer {} is disabled. ReplicationSyncUp tool will skip "<a name="line.716"></a> +<span class="sourceLineNo">717</span> + "replicating data to this peer.",<a name="line.717"></a> +<span class="sourceLineNo">718</span> actualPeerId);<a name="line.718"></a> +<span class="sourceLineNo">719</span> continue;<a name="line.719"></a> +<span class="sourceLineNo">720</span> }<a name="line.720"></a> +<span class="sourceLineNo">721</span> // track sources in walsByIdRecoveredQueues<a name="line.721"></a> +<span class="sourceLineNo">722</span> Map<String, NavigableSet<String>> walsByGroup = new HashMap<>();<a name="line.722"></a> +<span class="sourceLineNo">723</span> walsByIdRecoveredQueues.put(queueId, walsByGroup);<a name="line.723"></a> +<span class="sourceLineNo">724</span> for (String wal : walsSet) {<a name="line.724"></a> +<span class="sourceLineNo">725</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(wal);<a name="line.725"></a> +<span class="sourceLineNo">726</span> NavigableSet<String> wals = walsByGroup.get(walPrefix);<a name="line.726"></a> +<span class="sourceLineNo">727</span> if (wals == null) {<a name="line.727"></a> +<span class="sourceLineNo">728</span> wals = new TreeSet<>();<a name="line.728"></a> +<span class="sourceLineNo">729</span> walsByGroup.put(walPrefix, wals);<a name="line.729"></a> +<span class="sourceLineNo">730</span> }<a name="line.730"></a> +<span class="sourceLineNo">731</span> wals.add(wal);<a name="line.731"></a> +<span class="sourceLineNo">732</span> }<a name="line.732"></a> +<span class="sourceLineNo">733</span><a name="line.733"></a> +<span class="sourceLineNo">734</span> ReplicationSourceInterface src = createSource(queueId, peer);<a name="line.734"></a> +<span class="sourceLineNo">735</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.735"></a> +<span class="sourceLineNo">736</span> synchronized (oldsources) {<a name="line.736"></a> +<span class="sourceLineNo">737</span> peer = replicationPeers.getPeer(src.getPeerId());<a name="line.737"></a> +<span class="sourceLineNo">738</span> if (peer == null || !isOldPeer(src.getPeerId(), peer)) {<a name="line.738"></a> +<span class="sourceLineNo">739</span> src.terminate("Recovered queue doesn't belong to any current peer");<a name="line.739"></a> +<span class="sourceLineNo">740</span> removeRecoveredSource(src);<a name="line.740"></a> +<span class="sourceLineNo">741</span> continue;<a name="line.741"></a> +<span class="sourceLineNo">742</span> }<a name="line.742"></a> +<span class="sourceLineNo">743</span> oldsources.add(src);<a name="line.743"></a> +<span class="sourceLineNo">744</span> for (String wal : walsSet) {<a name="line.744"></a> +<span class="sourceLineNo">745</span> src.enqueueLog(new Path(oldLogDir, wal));<a name="line.745"></a> +<span class="sourceLineNo">746</span> }<a name="line.746"></a> +<span class="sourceLineNo">747</span> src.startup();<a name="line.747"></a> +<span class="sourceLineNo">748</span> }<a name="line.748"></a> +<span class="sourceLineNo">749</span> } catch (IOException e) {<a name="line.749"></a> +<span class="sourceLineNo">750</span> // TODO manage it<a name="line.750"></a> +<span class="sourceLineNo">751</span> LOG.error("Failed creating a source", e);<a name="line.751"></a> +<span class="sourceLineNo">752</span> }<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> }<a name="line.755"></a> +<span class="sourceLineNo">756</span><a name="line.756"></a> +<span class="sourceLineNo">757</span> /**<a name="line.757"></a> +<span class="sourceLineNo">758</span> * Terminate the replication on this region server<a name="line.758"></a> +<span class="sourceLineNo">759</span> */<a name="line.759"></a> +<span class="sourceLineNo">760</span> public void join() {<a name="line.760"></a> +<span class="sourceLineNo">761</span> this.executor.shutdown();<a name="line.761"></a> +<span class="sourceLineNo">762</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.762"></a> +<span class="sourceLineNo">763</span> source.terminate("Region server is closing");<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><a name="line.766"></a> +<span class="sourceLineNo">767</span> /**<a name="line.767"></a> +<span class="sourceLineNo">768</span> * Get a copy of the wals of the normal sources on this rs<a name="line.768"></a> +<span class="sourceLineNo">769</span> * @return a sorted set of wal names<a name="line.769"></a> +<span class="sourceLineNo">770</span> */<a name="line.770"></a> +<span class="sourceLineNo">771</span> @VisibleForTesting<a name="line.771"></a> +<span class="sourceLineNo">772</span> public Map<String, Map<String, NavigableSet<String>>> getWALs() {<a name="line.772"></a> +<span class="sourceLineNo">773</span> return Collections.unmodifiableMap(walsById);<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> /**<a name="line.776"></a> +<span class="sourceLineNo">777</span> * Get a copy of the wals of the recovered sources on this rs<a name="line.777"></a> +<span class="sourceLineNo">778</span> * @return a sorted set of wal names<a name="line.778"></a> +<span class="sourceLineNo">779</span> */<a name="line.779"></a> +<span class="sourceLineNo">780</span> @VisibleForTesting<a name="line.780"></a> +<span class="sourceLineNo">781</span> Map<String, Map<String, NavigableSet<String>>> getWalsByIdRecoveredQueues() {<a name="line.781"></a> +<span class="sourceLineNo">782</span> return Collections.unmodifiableMap(walsByIdRecoveredQueues);<a name="line.782"></a> +<span class="sourceLineNo">783</span> }<a name="line.783"></a> +<span class="sourceLineNo">784</span><a name="line.784"></a> +<span class="sourceLineNo">785</span> /**<a name="line.785"></a> +<span class="sourceLineNo">786</span> * Get a list of all the normal sources of this rs<a name="line.786"></a> +<span class="sourceLineNo">787</span> * @return list of all normal sources<a name="line.787"></a> +<span class="sourceLineNo">788</span> */<a name="line.788"></a> +<span class="sourceLineNo">789</span> public List<ReplicationSourceInterface> getSources() {<a name="line.789"></a> +<span class="sourceLineNo">790</span> return new ArrayList<>(this.sources.values());<a name="line.790"></a> +<span class="sourceLineNo">791</span> }<a name="line.791"></a> +<span class="sourceLineNo">792</span><a name="line.792"></a> +<span class="sourceLineNo">793</span> /**<a name="line.793"></a> +<span class="sourceLineNo">794</span> * Get a list of all the recovered sources of this rs<a name="line.794"></a> +<span class="sourceLineNo">795</span> * @return list of all recovered sources<a name="line.795"></a> +<span class="sourceLineNo">796</span> */<a name="line.796"></a> +<span class="sourceLineNo">797</span> public List<ReplicationSourceInterface> getOldSources() {<a name="line.797"></a> +<span class="sourceLineNo">798</span> return this.oldsources;<a name="line.798"></a> +<span class="sourceLineNo">799</span> }<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> * Get the normal source for a given peer<a name="line.802"></a> +<span class="sourceLineNo">803</span> * @return the normal source for the give peer if it exists, otherwise null.<a name="line.803"></a> +<span class="sourceLineNo">804</span> */<a name="line.804"></a> +<span class="sourceLineNo">805</span> @VisibleForTesting<a name="line.805"></a> +<span class="sourceLineNo">806</span> public ReplicationSourceInterface getSource(String peerId) {<a name="line.806"></a> +<span class="sourceLineNo">807</span> return this.sources.get(peerId);<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> @VisibleForTesting<a name="line.810"></a> +<span class="sourceLineNo">811</span> List<String> getAllQueues() throws IOException {<a name="line.811"></a> +<span class="sourceLineNo">812</span> List<String> allQueues = Collections.emptyList();<a name="line.812"></a> +<span class="sourceLineNo">813</span> try {<a name="line.813"></a> +<span class="sourceLineNo">814</span> allQueues = queueStorage.getAllQueues(server.getServerName());<a name="line.814"></a> +<span class="sourceLineNo">815</span> } catch (ReplicationException e) {<a name="line.815"></a> +<span class="sourceLineNo">816</span> throw new IOException(e);<a name="line.816"></a> +<span class="sourceLineNo">817</span> }<a name="line.817"></a> +<span class="sourceLineNo">818</span> return allQueues;<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> @VisibleForTesting<a name="line.821"></a> +<span class="sourceLineNo">822</span> int getSizeOfLatestPath() {<a name="line.822"></a> +<span class="sourceLineNo">823</span> synchronized (latestPaths) {<a name="line.823"></a> +<span class="sourceLineNo">824</span> return latestPaths.size();<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> @VisibleForTesting<a name="line.828"></a> +<span class="sourceLineNo">829</span> public AtomicLong getTotalBufferUsed() {<a name="line.829"></a> +<span class="sourceLineNo">830</span> return totalBufferUsed;<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> * Get the directory where wals are archived<a name="line.834"></a> +<span class="sourceLineNo">835</span> * @return the directory where wals are archived<a name="line.835"></a> +<span class="sourceLineNo">836</span> */<a name="line.836"></a> +<span class="sourceLineNo">837</span> public Path getOldLogDir() {<a name="line.837"></a> +<span class="sourceLineNo">838</span> return this.oldLogDir;<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> /**<a name="line.841"></a> +<span class="sourceLineNo">842</span> * Get the directory where wals are stored by their RSs<a name="line.842"></a> +<span class="sourceLineNo">843</span> * @return the directory where wals are stored by their RSs<a name="line.843"></a> +<span class="sourceLineNo">844</span> */
<TRUNCATED>