http://git-wip-us.apache.org/repos/asf/hbase-site/blob/476c54ed/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html index da61859..4fa61f8 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.html @@ -366,598 +366,607 @@ <span class="sourceLineNo">358</span> return this.oldsources;<a name="line.358"></a> <span class="sourceLineNo">359</span> }<a name="line.359"></a> <span class="sourceLineNo">360</span><a name="line.360"></a> -<span class="sourceLineNo">361</span> @VisibleForTesting<a name="line.361"></a> -<span class="sourceLineNo">362</span> List<String> getAllQueues() {<a name="line.362"></a> -<span class="sourceLineNo">363</span> return replicationQueues.getAllQueues();<a name="line.363"></a> -<span class="sourceLineNo">364</span> }<a name="line.364"></a> -<span class="sourceLineNo">365</span><a name="line.365"></a> -<span class="sourceLineNo">366</span> void preLogRoll(Path newLog) throws IOException {<a name="line.366"></a> -<span class="sourceLineNo">367</span> recordLog(newLog);<a name="line.367"></a> -<span class="sourceLineNo">368</span> String logName = newLog.getName();<a name="line.368"></a> -<span class="sourceLineNo">369</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.369"></a> -<span class="sourceLineNo">370</span> synchronized (latestPaths) {<a name="line.370"></a> -<span class="sourceLineNo">371</span> Iterator<Path> iterator = latestPaths.iterator();<a name="line.371"></a> -<span class="sourceLineNo">372</span> while (iterator.hasNext()) {<a name="line.372"></a> -<span class="sourceLineNo">373</span> Path path = iterator.next();<a name="line.373"></a> -<span class="sourceLineNo">374</span> if (path.getName().contains(logPrefix)) {<a name="line.374"></a> -<span class="sourceLineNo">375</span> iterator.remove();<a name="line.375"></a> -<span class="sourceLineNo">376</span> break;<a name="line.376"></a> -<span class="sourceLineNo">377</span> }<a name="line.377"></a> -<span class="sourceLineNo">378</span> }<a name="line.378"></a> -<span class="sourceLineNo">379</span> this.latestPaths.add(newLog);<a name="line.379"></a> -<span class="sourceLineNo">380</span> }<a name="line.380"></a> -<span class="sourceLineNo">381</span> }<a name="line.381"></a> -<span class="sourceLineNo">382</span><a name="line.382"></a> -<span class="sourceLineNo">383</span> /**<a name="line.383"></a> -<span class="sourceLineNo">384</span> * Check and enqueue the given log to the correct source. If there's still no source for the<a name="line.384"></a> -<span class="sourceLineNo">385</span> * group to which the given log belongs, create one<a name="line.385"></a> -<span class="sourceLineNo">386</span> * @param logPath the log path to check and enqueue<a name="line.386"></a> -<span class="sourceLineNo">387</span> * @throws IOException<a name="line.387"></a> -<span class="sourceLineNo">388</span> */<a name="line.388"></a> -<span class="sourceLineNo">389</span> private void recordLog(Path logPath) throws IOException {<a name="line.389"></a> -<span class="sourceLineNo">390</span> String logName = logPath.getName();<a name="line.390"></a> -<span class="sourceLineNo">391</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.391"></a> -<span class="sourceLineNo">392</span> // update replication queues on ZK<a name="line.392"></a> -<span class="sourceLineNo">393</span> // synchronize on replicationPeers to avoid adding source for the to-be-removed peer<a name="line.393"></a> -<span class="sourceLineNo">394</span> synchronized (replicationPeers) {<a name="line.394"></a> -<span class="sourceLineNo">395</span> for (String id : replicationPeers.getConnectedPeerIds()) {<a name="line.395"></a> -<span class="sourceLineNo">396</span> try {<a name="line.396"></a> -<span class="sourceLineNo">397</span> this.replicationQueues.addLog(id, logName);<a name="line.397"></a> -<span class="sourceLineNo">398</span> } catch (ReplicationException e) {<a name="line.398"></a> -<span class="sourceLineNo">399</span> throw new IOException("Cannot add log to replication queue"<a name="line.399"></a> -<span class="sourceLineNo">400</span> + " when creating a new source, queueId=" + id + ", filename=" + logName, e);<a name="line.400"></a> -<span class="sourceLineNo">401</span> }<a name="line.401"></a> -<span class="sourceLineNo">402</span> }<a name="line.402"></a> -<span class="sourceLineNo">403</span> }<a name="line.403"></a> -<span class="sourceLineNo">404</span> // update walsById map<a name="line.404"></a> -<span class="sourceLineNo">405</span> synchronized (walsById) {<a name="line.405"></a> -<span class="sourceLineNo">406</span> for (Map.Entry<String, Map<String, SortedSet<String>>> entry : this.walsById.entrySet()) {<a name="line.406"></a> -<span class="sourceLineNo">407</span> String peerId = entry.getKey();<a name="line.407"></a> -<span class="sourceLineNo">408</span> Map<String, SortedSet<String>> walsByPrefix = entry.getValue();<a name="line.408"></a> -<span class="sourceLineNo">409</span> boolean existingPrefix = false;<a name="line.409"></a> -<span class="sourceLineNo">410</span> for (Map.Entry<String, SortedSet<String>> walsEntry : walsByPrefix.entrySet()) {<a name="line.410"></a> -<span class="sourceLineNo">411</span> SortedSet<String> wals = walsEntry.getValue();<a name="line.411"></a> -<span class="sourceLineNo">412</span> if (this.sources.isEmpty()) {<a name="line.412"></a> -<span class="sourceLineNo">413</span> // If there's no slaves, don't need to keep the old wals since<a name="line.413"></a> -<span class="sourceLineNo">414</span> // we only consider the last one when a new slave comes in<a name="line.414"></a> -<span class="sourceLineNo">415</span> wals.clear();<a name="line.415"></a> -<span class="sourceLineNo">416</span> }<a name="line.416"></a> -<span class="sourceLineNo">417</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.417"></a> -<span class="sourceLineNo">418</span> wals.add(logName);<a name="line.418"></a> -<span class="sourceLineNo">419</span> existingPrefix = true;<a name="line.419"></a> -<span class="sourceLineNo">420</span> }<a name="line.420"></a> -<span class="sourceLineNo">421</span> }<a name="line.421"></a> -<span class="sourceLineNo">422</span> if (!existingPrefix) {<a name="line.422"></a> -<span class="sourceLineNo">423</span> // The new log belongs to a new group, add it into this peer<a name="line.423"></a> -<span class="sourceLineNo">424</span> LOG.debug("Start tracking logs for wal group " + logPrefix + " for peer " + peerId);<a name="line.424"></a> -<span class="sourceLineNo">425</span> SortedSet<String> wals = new TreeSet<>();<a name="line.425"></a> -<span class="sourceLineNo">426</span> wals.add(logName);<a name="line.426"></a> -<span class="sourceLineNo">427</span> walsByPrefix.put(logPrefix, wals);<a name="line.427"></a> -<span class="sourceLineNo">428</span> }<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> void postLogRoll(Path newLog) throws IOException {<a name="line.433"></a> -<span class="sourceLineNo">434</span> // This only updates the sources we own, not the recovered ones<a name="line.434"></a> -<span class="sourceLineNo">435</span> for (ReplicationSourceInterface source : this.sources) {<a name="line.435"></a> -<span class="sourceLineNo">436</span> source.enqueueLog(newLog);<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> @VisibleForTesting<a name="line.440"></a> -<span class="sourceLineNo">441</span> public AtomicLong getTotalBufferUsed() {<a name="line.441"></a> -<span class="sourceLineNo">442</span> return totalBufferUsed;<a name="line.442"></a> -<span class="sourceLineNo">443</span> }<a name="line.443"></a> -<span class="sourceLineNo">444</span><a name="line.444"></a> -<span class="sourceLineNo">445</span> /**<a name="line.445"></a> -<span class="sourceLineNo">446</span> * Factory method to create a replication source<a name="line.446"></a> -<span class="sourceLineNo">447</span> * @param conf the configuration to use<a name="line.447"></a> -<span class="sourceLineNo">448</span> * @param fs the file system to use<a name="line.448"></a> -<span class="sourceLineNo">449</span> * @param manager the manager to use<a name="line.449"></a> -<span class="sourceLineNo">450</span> * @param server the server object for this region server<a name="line.450"></a> -<span class="sourceLineNo">451</span> * @param peerId the id of the peer cluster<a name="line.451"></a> -<span class="sourceLineNo">452</span> * @return the created source<a name="line.452"></a> -<span class="sourceLineNo">453</span> * @throws IOException<a name="line.453"></a> -<span class="sourceLineNo">454</span> */<a name="line.454"></a> -<span class="sourceLineNo">455</span> protected ReplicationSourceInterface getReplicationSource(final Configuration conf,<a name="line.455"></a> -<span class="sourceLineNo">456</span> final FileSystem fs, final ReplicationSourceManager manager,<a name="line.456"></a> -<span class="sourceLineNo">457</span> final ReplicationQueues replicationQueues, final ReplicationPeers replicationPeers,<a name="line.457"></a> -<span class="sourceLineNo">458</span> final Server server, final String peerId, final UUID clusterId,<a name="line.458"></a> -<span class="sourceLineNo">459</span> final ReplicationPeerConfig peerConfig, final ReplicationPeer replicationPeer)<a name="line.459"></a> -<span class="sourceLineNo">460</span> throws IOException {<a name="line.460"></a> -<span class="sourceLineNo">461</span> RegionServerCoprocessorHost rsServerHost = null;<a name="line.461"></a> -<span class="sourceLineNo">462</span> TableDescriptors tableDescriptors = null;<a name="line.462"></a> -<span class="sourceLineNo">463</span> if (server instanceof HRegionServer) {<a name="line.463"></a> -<span class="sourceLineNo">464</span> rsServerHost = ((HRegionServer) server).getRegionServerCoprocessorHost();<a name="line.464"></a> -<span class="sourceLineNo">465</span> tableDescriptors = ((HRegionServer) server).getTableDescriptors();<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> ReplicationSourceInterface src = ReplicationSourceFactory.create(conf, peerId);<a name="line.468"></a> -<span class="sourceLineNo">469</span><a name="line.469"></a> -<span class="sourceLineNo">470</span> ReplicationEndpoint replicationEndpoint = null;<a name="line.470"></a> -<span class="sourceLineNo">471</span> try {<a name="line.471"></a> -<span class="sourceLineNo">472</span> String replicationEndpointImpl = peerConfig.getReplicationEndpointImpl();<a name="line.472"></a> -<span class="sourceLineNo">473</span> if (replicationEndpointImpl == null) {<a name="line.473"></a> -<span class="sourceLineNo">474</span> // Default to HBase inter-cluster replication endpoint<a name="line.474"></a> -<span class="sourceLineNo">475</span> replicationEndpointImpl = HBaseInterClusterReplicationEndpoint.class.getName();<a name="line.475"></a> -<span class="sourceLineNo">476</span> }<a name="line.476"></a> -<span class="sourceLineNo">477</span> @SuppressWarnings("rawtypes")<a name="line.477"></a> -<span class="sourceLineNo">478</span> Class c = Class.forName(replicationEndpointImpl);<a name="line.478"></a> -<span class="sourceLineNo">479</span> replicationEndpoint = (ReplicationEndpoint) c.newInstance();<a name="line.479"></a> -<span class="sourceLineNo">480</span> if(rsServerHost != null) {<a name="line.480"></a> -<span class="sourceLineNo">481</span> ReplicationEndpoint newReplicationEndPoint = rsServerHost<a name="line.481"></a> -<span class="sourceLineNo">482</span> .postCreateReplicationEndPoint(replicationEndpoint);<a name="line.482"></a> -<span class="sourceLineNo">483</span> if(newReplicationEndPoint != null) {<a name="line.483"></a> -<span class="sourceLineNo">484</span> // Override the newly created endpoint from the hook with configured end point<a name="line.484"></a> -<span class="sourceLineNo">485</span> replicationEndpoint = newReplicationEndPoint;<a name="line.485"></a> -<span class="sourceLineNo">486</span> }<a name="line.486"></a> -<span class="sourceLineNo">487</span> }<a name="line.487"></a> -<span class="sourceLineNo">488</span> } catch (Exception e) {<a name="line.488"></a> -<span class="sourceLineNo">489</span> LOG.warn("Passed replication endpoint implementation throws errors"<a name="line.489"></a> -<span class="sourceLineNo">490</span> + " while initializing ReplicationSource for peer: " + peerId, e);<a name="line.490"></a> -<span class="sourceLineNo">491</span> throw new IOException(e);<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> MetricsSource metrics = new MetricsSource(peerId);<a name="line.494"></a> -<span class="sourceLineNo">495</span> // init replication source<a name="line.495"></a> -<span class="sourceLineNo">496</span> src.init(conf, fs, manager, replicationQueues, replicationPeers, server, peerId,<a name="line.496"></a> -<span class="sourceLineNo">497</span> clusterId, replicationEndpoint, metrics);<a name="line.497"></a> -<span class="sourceLineNo">498</span><a name="line.498"></a> -<span class="sourceLineNo">499</span> // init replication endpoint<a name="line.499"></a> -<span class="sourceLineNo">500</span> replicationEndpoint.init(new ReplicationEndpoint.Context(replicationPeer.getConfiguration(),<a name="line.500"></a> -<span class="sourceLineNo">501</span> fs, peerId, clusterId, replicationPeer, metrics, tableDescriptors, server));<a name="line.501"></a> -<span class="sourceLineNo">502</span><a name="line.502"></a> -<span class="sourceLineNo">503</span> return src;<a name="line.503"></a> -<span class="sourceLineNo">504</span> }<a name="line.504"></a> -<span class="sourceLineNo">505</span><a name="line.505"></a> -<span class="sourceLineNo">506</span> /**<a name="line.506"></a> -<span class="sourceLineNo">507</span> * Transfer all the queues of the specified to this region server.<a name="line.507"></a> -<span class="sourceLineNo">508</span> * First it tries to grab a lock and if it works it will move the<a name="line.508"></a> -<span class="sourceLineNo">509</span> * znodes and finally will delete the old znodes.<a name="line.509"></a> -<span class="sourceLineNo">510</span> *<a name="line.510"></a> -<span class="sourceLineNo">511</span> * It creates one old source for any type of source of the old rs.<a name="line.511"></a> -<span class="sourceLineNo">512</span> * @param rsZnode<a name="line.512"></a> -<span class="sourceLineNo">513</span> */<a name="line.513"></a> -<span class="sourceLineNo">514</span> private void transferQueues(String rsZnode) {<a name="line.514"></a> -<span class="sourceLineNo">515</span> NodeFailoverWorker transfer =<a name="line.515"></a> -<span class="sourceLineNo">516</span> new NodeFailoverWorker(rsZnode, this.replicationQueues, this.replicationPeers,<a name="line.516"></a> -<span class="sourceLineNo">517</span> this.clusterId);<a name="line.517"></a> -<span class="sourceLineNo">518</span> try {<a name="line.518"></a> -<span class="sourceLineNo">519</span> this.executor.execute(transfer);<a name="line.519"></a> -<span class="sourceLineNo">520</span> } catch (RejectedExecutionException ex) {<a name="line.520"></a> -<span class="sourceLineNo">521</span> LOG.info("Cancelling the transfer of " + rsZnode + " because of " + ex.getMessage());<a name="line.521"></a> -<span class="sourceLineNo">522</span> }<a name="line.522"></a> -<span class="sourceLineNo">523</span> }<a name="line.523"></a> -<span class="sourceLineNo">524</span><a name="line.524"></a> -<span class="sourceLineNo">525</span> /**<a name="line.525"></a> -<span class="sourceLineNo">526</span> * Clear the references to the specified old source<a name="line.526"></a> -<span class="sourceLineNo">527</span> * @param src source to clear<a name="line.527"></a> -<span class="sourceLineNo">528</span> */<a name="line.528"></a> -<span class="sourceLineNo">529</span> public void closeRecoveredQueue(ReplicationSourceInterface src) {<a name="line.529"></a> -<span class="sourceLineNo">530</span> LOG.info("Done with the recovered queue " + src.getPeerClusterZnode());<a name="line.530"></a> -<span class="sourceLineNo">531</span> if (src instanceof ReplicationSource) {<a name="line.531"></a> -<span class="sourceLineNo">532</span> ((ReplicationSource) src).getSourceMetrics().clear();<a name="line.532"></a> -<span class="sourceLineNo">533</span> }<a name="line.533"></a> -<span class="sourceLineNo">534</span> this.oldsources.remove(src);<a name="line.534"></a> -<span class="sourceLineNo">535</span> deleteSource(src.getPeerClusterZnode(), false);<a name="line.535"></a> -<span class="sourceLineNo">536</span> this.walsByIdRecoveredQueues.remove(src.getPeerClusterZnode());<a name="line.536"></a> +<span class="sourceLineNo">361</span> /**<a name="line.361"></a> +<span class="sourceLineNo">362</span> * Get the normal source for a given peer<a name="line.362"></a> +<span class="sourceLineNo">363</span> * @param peerId<a name="line.363"></a> +<span class="sourceLineNo">364</span> * @return the normal source for the give peer if it exists, otherwise null.<a name="line.364"></a> +<span class="sourceLineNo">365</span> */<a name="line.365"></a> +<span class="sourceLineNo">366</span> public ReplicationSourceInterface getSource(String peerId) {<a name="line.366"></a> +<span class="sourceLineNo">367</span> for (ReplicationSourceInterface source: getSources()) {<a name="line.367"></a> +<span class="sourceLineNo">368</span> if (source.getPeerId().equals(peerId)) {<a name="line.368"></a> +<span class="sourceLineNo">369</span> return source;<a name="line.369"></a> +<span class="sourceLineNo">370</span> }<a name="line.370"></a> +<span class="sourceLineNo">371</span> }<a name="line.371"></a> +<span class="sourceLineNo">372</span> return null;<a name="line.372"></a> +<span class="sourceLineNo">373</span> }<a name="line.373"></a> +<span class="sourceLineNo">374</span><a name="line.374"></a> +<span class="sourceLineNo">375</span> @VisibleForTesting<a name="line.375"></a> +<span class="sourceLineNo">376</span> List<String> getAllQueues() {<a name="line.376"></a> +<span class="sourceLineNo">377</span> return replicationQueues.getAllQueues();<a name="line.377"></a> +<span class="sourceLineNo">378</span> }<a name="line.378"></a> +<span class="sourceLineNo">379</span><a name="line.379"></a> +<span class="sourceLineNo">380</span> void preLogRoll(Path newLog) throws IOException {<a name="line.380"></a> +<span class="sourceLineNo">381</span> recordLog(newLog);<a name="line.381"></a> +<span class="sourceLineNo">382</span> String logName = newLog.getName();<a name="line.382"></a> +<span class="sourceLineNo">383</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.383"></a> +<span class="sourceLineNo">384</span> synchronized (latestPaths) {<a name="line.384"></a> +<span class="sourceLineNo">385</span> Iterator<Path> iterator = latestPaths.iterator();<a name="line.385"></a> +<span class="sourceLineNo">386</span> while (iterator.hasNext()) {<a name="line.386"></a> +<span class="sourceLineNo">387</span> Path path = iterator.next();<a name="line.387"></a> +<span class="sourceLineNo">388</span> if (path.getName().contains(logPrefix)) {<a name="line.388"></a> +<span class="sourceLineNo">389</span> iterator.remove();<a name="line.389"></a> +<span class="sourceLineNo">390</span> break;<a name="line.390"></a> +<span class="sourceLineNo">391</span> }<a name="line.391"></a> +<span class="sourceLineNo">392</span> }<a name="line.392"></a> +<span class="sourceLineNo">393</span> this.latestPaths.add(newLog);<a name="line.393"></a> +<span class="sourceLineNo">394</span> }<a name="line.394"></a> +<span class="sourceLineNo">395</span> }<a name="line.395"></a> +<span class="sourceLineNo">396</span><a name="line.396"></a> +<span class="sourceLineNo">397</span> /**<a name="line.397"></a> +<span class="sourceLineNo">398</span> * Check and enqueue the given log to the correct source. If there's still no source for the<a name="line.398"></a> +<span class="sourceLineNo">399</span> * group to which the given log belongs, create one<a name="line.399"></a> +<span class="sourceLineNo">400</span> * @param logPath the log path to check and enqueue<a name="line.400"></a> +<span class="sourceLineNo">401</span> * @throws IOException<a name="line.401"></a> +<span class="sourceLineNo">402</span> */<a name="line.402"></a> +<span class="sourceLineNo">403</span> private void recordLog(Path logPath) throws IOException {<a name="line.403"></a> +<span class="sourceLineNo">404</span> String logName = logPath.getName();<a name="line.404"></a> +<span class="sourceLineNo">405</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.405"></a> +<span class="sourceLineNo">406</span> // update replication queues on ZK<a name="line.406"></a> +<span class="sourceLineNo">407</span> // synchronize on replicationPeers to avoid adding source for the to-be-removed peer<a name="line.407"></a> +<span class="sourceLineNo">408</span> synchronized (replicationPeers) {<a name="line.408"></a> +<span class="sourceLineNo">409</span> for (String id : replicationPeers.getConnectedPeerIds()) {<a name="line.409"></a> +<span class="sourceLineNo">410</span> try {<a name="line.410"></a> +<span class="sourceLineNo">411</span> this.replicationQueues.addLog(id, logName);<a name="line.411"></a> +<span class="sourceLineNo">412</span> } catch (ReplicationException e) {<a name="line.412"></a> +<span class="sourceLineNo">413</span> throw new IOException("Cannot add log to replication queue"<a name="line.413"></a> +<span class="sourceLineNo">414</span> + " when creating a new source, queueId=" + id + ", filename=" + logName, e);<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> }<a name="line.417"></a> +<span class="sourceLineNo">418</span> // update walsById map<a name="line.418"></a> +<span class="sourceLineNo">419</span> synchronized (walsById) {<a name="line.419"></a> +<span class="sourceLineNo">420</span> for (Map.Entry<String, Map<String, SortedSet<String>>> entry : this.walsById.entrySet()) {<a name="line.420"></a> +<span class="sourceLineNo">421</span> String peerId = entry.getKey();<a name="line.421"></a> +<span class="sourceLineNo">422</span> Map<String, SortedSet<String>> walsByPrefix = entry.getValue();<a name="line.422"></a> +<span class="sourceLineNo">423</span> boolean existingPrefix = false;<a name="line.423"></a> +<span class="sourceLineNo">424</span> for (Map.Entry<String, SortedSet<String>> walsEntry : walsByPrefix.entrySet()) {<a name="line.424"></a> +<span class="sourceLineNo">425</span> SortedSet<String> wals = walsEntry.getValue();<a name="line.425"></a> +<span class="sourceLineNo">426</span> if (this.sources.isEmpty()) {<a name="line.426"></a> +<span class="sourceLineNo">427</span> // If there's no slaves, don't need to keep the old wals since<a name="line.427"></a> +<span class="sourceLineNo">428</span> // we only consider the last one when a new slave comes in<a name="line.428"></a> +<span class="sourceLineNo">429</span> wals.clear();<a name="line.429"></a> +<span class="sourceLineNo">430</span> }<a name="line.430"></a> +<span class="sourceLineNo">431</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.431"></a> +<span class="sourceLineNo">432</span> wals.add(logName);<a name="line.432"></a> +<span class="sourceLineNo">433</span> existingPrefix = true;<a name="line.433"></a> +<span class="sourceLineNo">434</span> }<a name="line.434"></a> +<span class="sourceLineNo">435</span> }<a name="line.435"></a> +<span class="sourceLineNo">436</span> if (!existingPrefix) {<a name="line.436"></a> +<span class="sourceLineNo">437</span> // The new log belongs to a new group, add it into this peer<a name="line.437"></a> +<span class="sourceLineNo">438</span> LOG.debug("Start tracking logs for wal group " + logPrefix + " for peer " + peerId);<a name="line.438"></a> +<span class="sourceLineNo">439</span> SortedSet<String> wals = new TreeSet<>();<a name="line.439"></a> +<span class="sourceLineNo">440</span> wals.add(logName);<a name="line.440"></a> +<span class="sourceLineNo">441</span> walsByPrefix.put(logPrefix, wals);<a name="line.441"></a> +<span class="sourceLineNo">442</span> }<a name="line.442"></a> +<span class="sourceLineNo">443</span> }<a name="line.443"></a> +<span class="sourceLineNo">444</span> }<a name="line.444"></a> +<span class="sourceLineNo">445</span> }<a name="line.445"></a> +<span class="sourceLineNo">446</span><a name="line.446"></a> +<span class="sourceLineNo">447</span> void postLogRoll(Path newLog) throws IOException {<a name="line.447"></a> +<span class="sourceLineNo">448</span> // This only updates the sources we own, not the recovered ones<a name="line.448"></a> +<span class="sourceLineNo">449</span> for (ReplicationSourceInterface source : this.sources) {<a name="line.449"></a> +<span class="sourceLineNo">450</span> source.enqueueLog(newLog);<a name="line.450"></a> +<span class="sourceLineNo">451</span> }<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> @VisibleForTesting<a name="line.454"></a> +<span class="sourceLineNo">455</span> public AtomicLong getTotalBufferUsed() {<a name="line.455"></a> +<span class="sourceLineNo">456</span> return totalBufferUsed;<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> * Factory method to create a replication source<a name="line.460"></a> +<span class="sourceLineNo">461</span> * @param conf the configuration to use<a name="line.461"></a> +<span class="sourceLineNo">462</span> * @param fs the file system to use<a name="line.462"></a> +<span class="sourceLineNo">463</span> * @param manager the manager to use<a name="line.463"></a> +<span class="sourceLineNo">464</span> * @param server the server object for this region server<a name="line.464"></a> +<span class="sourceLineNo">465</span> * @param peerId the id of the peer cluster<a name="line.465"></a> +<span class="sourceLineNo">466</span> * @return the created source<a name="line.466"></a> +<span class="sourceLineNo">467</span> * @throws IOException<a name="line.467"></a> +<span class="sourceLineNo">468</span> */<a name="line.468"></a> +<span class="sourceLineNo">469</span> protected ReplicationSourceInterface getReplicationSource(final Configuration conf,<a name="line.469"></a> +<span class="sourceLineNo">470</span> final FileSystem fs, final ReplicationSourceManager manager,<a name="line.470"></a> +<span class="sourceLineNo">471</span> final ReplicationQueues replicationQueues, final ReplicationPeers replicationPeers,<a name="line.471"></a> +<span class="sourceLineNo">472</span> final Server server, final String peerId, final UUID clusterId,<a name="line.472"></a> +<span class="sourceLineNo">473</span> final ReplicationPeerConfig peerConfig, final ReplicationPeer replicationPeer)<a name="line.473"></a> +<span class="sourceLineNo">474</span> throws IOException {<a name="line.474"></a> +<span class="sourceLineNo">475</span> RegionServerCoprocessorHost rsServerHost = null;<a name="line.475"></a> +<span class="sourceLineNo">476</span> TableDescriptors tableDescriptors = null;<a name="line.476"></a> +<span class="sourceLineNo">477</span> if (server instanceof HRegionServer) {<a name="line.477"></a> +<span class="sourceLineNo">478</span> rsServerHost = ((HRegionServer) server).getRegionServerCoprocessorHost();<a name="line.478"></a> +<span class="sourceLineNo">479</span> tableDescriptors = ((HRegionServer) server).getTableDescriptors();<a name="line.479"></a> +<span class="sourceLineNo">480</span> }<a name="line.480"></a> +<span class="sourceLineNo">481</span><a name="line.481"></a> +<span class="sourceLineNo">482</span> ReplicationSourceInterface src = ReplicationSourceFactory.create(conf, peerId);<a name="line.482"></a> +<span class="sourceLineNo">483</span><a name="line.483"></a> +<span class="sourceLineNo">484</span> ReplicationEndpoint replicationEndpoint = null;<a name="line.484"></a> +<span class="sourceLineNo">485</span> try {<a name="line.485"></a> +<span class="sourceLineNo">486</span> String replicationEndpointImpl = peerConfig.getReplicationEndpointImpl();<a name="line.486"></a> +<span class="sourceLineNo">487</span> if (replicationEndpointImpl == null) {<a name="line.487"></a> +<span class="sourceLineNo">488</span> // Default to HBase inter-cluster replication endpoint<a name="line.488"></a> +<span class="sourceLineNo">489</span> replicationEndpointImpl = HBaseInterClusterReplicationEndpoint.class.getName();<a name="line.489"></a> +<span class="sourceLineNo">490</span> }<a name="line.490"></a> +<span class="sourceLineNo">491</span> @SuppressWarnings("rawtypes")<a name="line.491"></a> +<span class="sourceLineNo">492</span> Class c = Class.forName(replicationEndpointImpl);<a name="line.492"></a> +<span class="sourceLineNo">493</span> replicationEndpoint = (ReplicationEndpoint) c.newInstance();<a name="line.493"></a> +<span class="sourceLineNo">494</span> if(rsServerHost != null) {<a name="line.494"></a> +<span class="sourceLineNo">495</span> ReplicationEndpoint newReplicationEndPoint = rsServerHost<a name="line.495"></a> +<span class="sourceLineNo">496</span> .postCreateReplicationEndPoint(replicationEndpoint);<a name="line.496"></a> +<span class="sourceLineNo">497</span> if(newReplicationEndPoint != null) {<a name="line.497"></a> +<span class="sourceLineNo">498</span> // Override the newly created endpoint from the hook with configured end point<a name="line.498"></a> +<span class="sourceLineNo">499</span> replicationEndpoint = newReplicationEndPoint;<a name="line.499"></a> +<span class="sourceLineNo">500</span> }<a name="line.500"></a> +<span class="sourceLineNo">501</span> }<a name="line.501"></a> +<span class="sourceLineNo">502</span> } catch (Exception e) {<a name="line.502"></a> +<span class="sourceLineNo">503</span> LOG.warn("Passed replication endpoint implementation throws errors"<a name="line.503"></a> +<span class="sourceLineNo">504</span> + " while initializing ReplicationSource for peer: " + peerId, e);<a name="line.504"></a> +<span class="sourceLineNo">505</span> throw new IOException(e);<a name="line.505"></a> +<span class="sourceLineNo">506</span> }<a name="line.506"></a> +<span class="sourceLineNo">507</span><a name="line.507"></a> +<span class="sourceLineNo">508</span> MetricsSource metrics = new MetricsSource(peerId);<a name="line.508"></a> +<span class="sourceLineNo">509</span> // init replication source<a name="line.509"></a> +<span class="sourceLineNo">510</span> src.init(conf, fs, manager, replicationQueues, replicationPeers, server, peerId,<a name="line.510"></a> +<span class="sourceLineNo">511</span> clusterId, replicationEndpoint, metrics);<a name="line.511"></a> +<span class="sourceLineNo">512</span><a name="line.512"></a> +<span class="sourceLineNo">513</span> // init replication endpoint<a name="line.513"></a> +<span class="sourceLineNo">514</span> replicationEndpoint.init(new ReplicationEndpoint.Context(replicationPeer.getConfiguration(),<a name="line.514"></a> +<span class="sourceLineNo">515</span> fs, peerId, clusterId, replicationPeer, metrics, tableDescriptors, server));<a name="line.515"></a> +<span class="sourceLineNo">516</span><a name="line.516"></a> +<span class="sourceLineNo">517</span> return src;<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> * Transfer all the queues of the specified to this region server.<a name="line.521"></a> +<span class="sourceLineNo">522</span> * First it tries to grab a lock and if it works it will move the<a name="line.522"></a> +<span class="sourceLineNo">523</span> * znodes and finally will delete the old znodes.<a name="line.523"></a> +<span class="sourceLineNo">524</span> *<a name="line.524"></a> +<span class="sourceLineNo">525</span> * It creates one old source for any type of source of the old rs.<a name="line.525"></a> +<span class="sourceLineNo">526</span> * @param rsZnode<a name="line.526"></a> +<span class="sourceLineNo">527</span> */<a name="line.527"></a> +<span class="sourceLineNo">528</span> private void transferQueues(String rsZnode) {<a name="line.528"></a> +<span class="sourceLineNo">529</span> NodeFailoverWorker transfer =<a name="line.529"></a> +<span class="sourceLineNo">530</span> new NodeFailoverWorker(rsZnode, this.replicationQueues, this.replicationPeers,<a name="line.530"></a> +<span class="sourceLineNo">531</span> this.clusterId);<a name="line.531"></a> +<span class="sourceLineNo">532</span> try {<a name="line.532"></a> +<span class="sourceLineNo">533</span> this.executor.execute(transfer);<a name="line.533"></a> +<span class="sourceLineNo">534</span> } catch (RejectedExecutionException ex) {<a name="line.534"></a> +<span class="sourceLineNo">535</span> LOG.info("Cancelling the transfer of " + rsZnode + " because of " + ex.getMessage());<a name="line.535"></a> +<span class="sourceLineNo">536</span> }<a name="line.536"></a> <span class="sourceLineNo">537</span> }<a name="line.537"></a> <span class="sourceLineNo">538</span><a name="line.538"></a> <span class="sourceLineNo">539</span> /**<a name="line.539"></a> <span class="sourceLineNo">540</span> * Clear the references to the specified old source<a name="line.540"></a> <span class="sourceLineNo">541</span> * @param src source to clear<a name="line.541"></a> <span class="sourceLineNo">542</span> */<a name="line.542"></a> -<span class="sourceLineNo">543</span> public void closeQueue(ReplicationSourceInterface src) {<a name="line.543"></a> -<span class="sourceLineNo">544</span> LOG.info("Done with the queue " + src.getPeerClusterZnode());<a name="line.544"></a> +<span class="sourceLineNo">543</span> public void closeRecoveredQueue(ReplicationSourceInterface src) {<a name="line.543"></a> +<span class="sourceLineNo">544</span> LOG.info("Done with the recovered queue " + src.getPeerClusterZnode());<a name="line.544"></a> <span class="sourceLineNo">545</span> if (src instanceof ReplicationSource) {<a name="line.545"></a> <span class="sourceLineNo">546</span> ((ReplicationSource) src).getSourceMetrics().clear();<a name="line.546"></a> <span class="sourceLineNo">547</span> }<a name="line.547"></a> -<span class="sourceLineNo">548</span> this.sources.remove(src);<a name="line.548"></a> -<span class="sourceLineNo">549</span> deleteSource(src.getPeerClusterZnode(), true);<a name="line.549"></a> -<span class="sourceLineNo">550</span> this.walsById.remove(src.getPeerClusterZnode());<a name="line.550"></a> +<span class="sourceLineNo">548</span> this.oldsources.remove(src);<a name="line.548"></a> +<span class="sourceLineNo">549</span> deleteSource(src.getPeerClusterZnode(), false);<a name="line.549"></a> +<span class="sourceLineNo">550</span> this.walsByIdRecoveredQueues.remove(src.getPeerClusterZnode());<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> /**<a name="line.553"></a> -<span class="sourceLineNo">554</span> * Thie method first deletes all the recovered sources for the specified<a name="line.554"></a> -<span class="sourceLineNo">555</span> * id, then deletes the normal source (deleting all related data in ZK).<a name="line.555"></a> -<span class="sourceLineNo">556</span> * @param id The id of the peer cluster<a name="line.556"></a> -<span class="sourceLineNo">557</span> */<a name="line.557"></a> -<span class="sourceLineNo">558</span> public void removePeer(String id) {<a name="line.558"></a> -<span class="sourceLineNo">559</span> LOG.info("Closing the following queue " + id + ", currently have "<a name="line.559"></a> -<span class="sourceLineNo">560</span> + sources.size() + " and another "<a name="line.560"></a> -<span class="sourceLineNo">561</span> + oldsources.size() + " that were recovered");<a name="line.561"></a> -<span class="sourceLineNo">562</span> String terminateMessage = "Replication stream was removed by a user";<a name="line.562"></a> -<span class="sourceLineNo">563</span> List<ReplicationSourceInterface> oldSourcesToDelete = new ArrayList<>();<a name="line.563"></a> -<span class="sourceLineNo">564</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.564"></a> -<span class="sourceLineNo">565</span> // see NodeFailoverWorker.run<a name="line.565"></a> -<span class="sourceLineNo">566</span> synchronized (oldsources) {<a name="line.566"></a> -<span class="sourceLineNo">567</span> // First close all the recovered sources for this peer<a name="line.567"></a> -<span class="sourceLineNo">568</span> for (ReplicationSourceInterface src : oldsources) {<a name="line.568"></a> -<span class="sourceLineNo">569</span> if (id.equals(src.getPeerId())) {<a name="line.569"></a> -<span class="sourceLineNo">570</span> oldSourcesToDelete.add(src);<a name="line.570"></a> -<span class="sourceLineNo">571</span> }<a name="line.571"></a> -<span class="sourceLineNo">572</span> }<a name="line.572"></a> -<span class="sourceLineNo">573</span> for (ReplicationSourceInterface src : oldSourcesToDelete) {<a name="line.573"></a> -<span class="sourceLineNo">574</span> src.terminate(terminateMessage);<a name="line.574"></a> -<span class="sourceLineNo">575</span> closeRecoveredQueue(src);<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> LOG.info("Number of deleted recovered sources for " + id + ": "<a name="line.578"></a> -<span class="sourceLineNo">579</span> + oldSourcesToDelete.size());<a name="line.579"></a> -<span class="sourceLineNo">580</span> // Now look for the one on this cluster<a name="line.580"></a> -<span class="sourceLineNo">581</span> List<ReplicationSourceInterface> srcToRemove = new ArrayList<>();<a name="line.581"></a> -<span class="sourceLineNo">582</span> // synchronize on replicationPeers to avoid adding source for the to-be-removed peer<a name="line.582"></a> -<span class="sourceLineNo">583</span> synchronized (this.replicationPeers) {<a name="line.583"></a> -<span class="sourceLineNo">584</span> for (ReplicationSourceInterface src : this.sources) {<a name="line.584"></a> -<span class="sourceLineNo">585</span> if (id.equals(src.getPeerId())) {<a name="line.585"></a> -<span class="sourceLineNo">586</span> srcToRemove.add(src);<a name="line.586"></a> -<span class="sourceLineNo">587</span> }<a name="line.587"></a> +<span class="sourceLineNo">554</span> * Clear the references to the specified old source<a name="line.554"></a> +<span class="sourceLineNo">555</span> * @param src source to clear<a name="line.555"></a> +<span class="sourceLineNo">556</span> */<a name="line.556"></a> +<span class="sourceLineNo">557</span> public void closeQueue(ReplicationSourceInterface src) {<a name="line.557"></a> +<span class="sourceLineNo">558</span> LOG.info("Done with the queue " + src.getPeerClusterZnode());<a name="line.558"></a> +<span class="sourceLineNo">559</span> src.getSourceMetrics().clear();<a name="line.559"></a> +<span class="sourceLineNo">560</span> this.sources.remove(src);<a name="line.560"></a> +<span class="sourceLineNo">561</span> deleteSource(src.getPeerClusterZnode(), true);<a name="line.561"></a> +<span class="sourceLineNo">562</span> this.walsById.remove(src.getPeerClusterZnode());<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> /**<a name="line.565"></a> +<span class="sourceLineNo">566</span> * Thie method first deletes all the recovered sources for the specified<a name="line.566"></a> +<span class="sourceLineNo">567</span> * id, then deletes the normal source (deleting all related data in ZK).<a name="line.567"></a> +<span class="sourceLineNo">568</span> * @param id The id of the peer cluster<a name="line.568"></a> +<span class="sourceLineNo">569</span> */<a name="line.569"></a> +<span class="sourceLineNo">570</span> public void removePeer(String id) {<a name="line.570"></a> +<span class="sourceLineNo">571</span> LOG.info("Closing the following queue " + id + ", currently have "<a name="line.571"></a> +<span class="sourceLineNo">572</span> + sources.size() + " and another "<a name="line.572"></a> +<span class="sourceLineNo">573</span> + oldsources.size() + " that were recovered");<a name="line.573"></a> +<span class="sourceLineNo">574</span> String terminateMessage = "Replication stream was removed by a user";<a name="line.574"></a> +<span class="sourceLineNo">575</span> List<ReplicationSourceInterface> oldSourcesToDelete = new ArrayList<>();<a name="line.575"></a> +<span class="sourceLineNo">576</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.576"></a> +<span class="sourceLineNo">577</span> // see NodeFailoverWorker.run<a name="line.577"></a> +<span class="sourceLineNo">578</span> synchronized (oldsources) {<a name="line.578"></a> +<span class="sourceLineNo">579</span> // First close all the recovered sources for this peer<a name="line.579"></a> +<span class="sourceLineNo">580</span> for (ReplicationSourceInterface src : oldsources) {<a name="line.580"></a> +<span class="sourceLineNo">581</span> if (id.equals(src.getPeerId())) {<a name="line.581"></a> +<span class="sourceLineNo">582</span> oldSourcesToDelete.add(src);<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> for (ReplicationSourceInterface src : oldSourcesToDelete) {<a name="line.585"></a> +<span class="sourceLineNo">586</span> src.terminate(terminateMessage);<a name="line.586"></a> +<span class="sourceLineNo">587</span> closeRecoveredQueue(src);<a name="line.587"></a> <span class="sourceLineNo">588</span> }<a name="line.588"></a> -<span class="sourceLineNo">589</span> if (srcToRemove.isEmpty()) {<a name="line.589"></a> -<span class="sourceLineNo">590</span> LOG.error("The peer we wanted to remove is missing a ReplicationSourceInterface. " +<a name="line.590"></a> -<span class="sourceLineNo">591</span> "This could mean that ReplicationSourceInterface initialization failed for this peer " +<a name="line.591"></a> -<span class="sourceLineNo">592</span> "and that replication on this peer may not be caught up. peerId=" + id);<a name="line.592"></a> -<span class="sourceLineNo">593</span> }<a name="line.593"></a> -<span class="sourceLineNo">594</span> for (ReplicationSourceInterface toRemove : srcToRemove) {<a name="line.594"></a> -<span class="sourceLineNo">595</span> toRemove.terminate(terminateMessage);<a name="line.595"></a> -<span class="sourceLineNo">596</span> if (toRemove instanceof ReplicationSource) {<a name="line.596"></a> -<span class="sourceLineNo">597</span> ((ReplicationSource) toRemove).getSourceMetrics().clear();<a name="line.597"></a> -<span class="sourceLineNo">598</span> }<a name="line.598"></a> -<span class="sourceLineNo">599</span> this.sources.remove(toRemove);<a name="line.599"></a> +<span class="sourceLineNo">589</span> }<a name="line.589"></a> +<span class="sourceLineNo">590</span> LOG.info("Number of deleted recovered sources for " + id + ": "<a name="line.590"></a> +<span class="sourceLineNo">591</span> + oldSourcesToDelete.size());<a name="line.591"></a> +<span class="sourceLineNo">592</span> // Now look for the one on this cluster<a name="line.592"></a> +<span class="sourceLineNo">593</span> List<ReplicationSourceInterface> srcToRemove = new ArrayList<>();<a name="line.593"></a> +<span class="sourceLineNo">594</span> // synchronize on replicationPeers to avoid adding source for the to-be-removed peer<a name="line.594"></a> +<span class="sourceLineNo">595</span> synchronized (this.replicationPeers) {<a name="line.595"></a> +<span class="sourceLineNo">596</span> for (ReplicationSourceInterface src : this.sources) {<a name="line.596"></a> +<span class="sourceLineNo">597</span> if (id.equals(src.getPeerId())) {<a name="line.597"></a> +<span class="sourceLineNo">598</span> srcToRemove.add(src);<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> deleteSource(id, true);<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><a name="line.604"></a> -<span class="sourceLineNo">605</span> @Override<a name="line.605"></a> -<span class="sourceLineNo">606</span> public void regionServerRemoved(String regionserver) {<a name="line.606"></a> -<span class="sourceLineNo">607</span> transferQueues(regionserver);<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> @Override<a name="line.610"></a> -<span class="sourceLineNo">611</span> public void peerRemoved(String peerId) {<a name="line.611"></a> -<span class="sourceLineNo">612</span> removePeer(peerId);<a name="line.612"></a> -<span class="sourceLineNo">613</span> this.replicationQueues.removePeerFromHFileRefs(peerId);<a name="line.613"></a> -<span class="sourceLineNo">614</span> }<a name="line.614"></a> -<span class="sourceLineNo">615</span><a name="line.615"></a> -<span class="sourceLineNo">616</span> @Override<a name="line.616"></a> -<span class="sourceLineNo">617</span> public void peerListChanged(List<String> peerIds) {<a name="line.617"></a> -<span class="sourceLineNo">618</span> for (String id : peerIds) {<a name="line.618"></a> -<span class="sourceLineNo">619</span> try {<a name="line.619"></a> -<span class="sourceLineNo">620</span> boolean added = this.replicationPeers.peerConnected(id);<a name="line.620"></a> -<span class="sourceLineNo">621</span> if (added) {<a name="line.621"></a> -<span class="sourceLineNo">622</span> addSource(id);<a name="line.622"></a> -<span class="sourceLineNo">623</span> if (replicationForBulkLoadDataEnabled) {<a name="line.623"></a> -<span class="sourceLineNo">624</span> this.replicationQueues.addPeerToHFileRefs(id);<a name="line.624"></a> -<span class="sourceLineNo">625</span> }<a name="line.625"></a> -<span class="sourceLineNo">626</span> }<a name="line.626"></a> -<span class="sourceLineNo">627</span> } catch (Exception e) {<a name="line.627"></a> -<span class="sourceLineNo">628</span> LOG.error("Error while adding a new peer", e);<a name="line.628"></a> -<span class="sourceLineNo">629</span> }<a name="line.629"></a> -<span class="sourceLineNo">630</span> }<a name="line.630"></a> -<span class="sourceLineNo">631</span> }<a name="line.631"></a> -<span class="sourceLineNo">632</span><a name="line.632"></a> -<span class="sourceLineNo">633</span> /**<a name="line.633"></a> -<span class="sourceLineNo">634</span> * Class responsible to setup new ReplicationSources to take care of the<a name="line.634"></a> -<span class="sourceLineNo">635</span> * queues from dead region servers.<a name="line.635"></a> -<span class="sourceLineNo">636</span> */<a name="line.636"></a> -<span class="sourceLineNo">637</span> class NodeFailoverWorker extends Thread {<a name="line.637"></a> -<span class="sourceLineNo">638</span><a name="line.638"></a> -<span class="sourceLineNo">639</span> private String rsZnode;<a name="line.639"></a> -<span class="sourceLineNo">640</span> private final ReplicationQueues rq;<a name="line.640"></a> -<span class="sourceLineNo">641</span> private final ReplicationPeers rp;<a name="line.641"></a> -<span class="sourceLineNo">642</span> private final UUID clusterId;<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> * @param rsZnode<a name="line.645"></a> -<span class="sourceLineNo">646</span> */<a name="line.646"></a> -<span class="sourceLineNo">647</span> public NodeFailoverWorker(String rsZnode) {<a name="line.647"></a> -<span class="sourceLineNo">648</span> this(rsZnode, replicationQueues, replicationPeers, ReplicationSourceManager.this.clusterId);<a name="line.648"></a> -<span class="sourceLineNo">649</span> }<a name="line.649"></a> -<span class="sourceLineNo">650</span><a name="line.650"></a> -<span class="sourceLineNo">651</span> public NodeFailoverWorker(String rsZnode, final ReplicationQueues replicationQueues,<a name="line.651"></a> -<span class="sourceLineNo">652</span> final ReplicationPeers replicationPeers, final UUID clusterId) {<a name="line.652"></a> -<span class="sourceLineNo">653</span> super("Failover-for-"+rsZnode);<a name="line.653"></a> -<span class="sourceLineNo">654</span> this.rsZnode = rsZnode;<a name="line.654"></a> -<span class="sourceLineNo">655</span> this.rq = replicationQueues;<a name="line.655"></a> -<span class="sourceLineNo">656</span> this.rp = replicationPeers;<a name="line.656"></a> -<span class="sourceLineNo">657</span> this.clusterId = clusterId;<a name="line.657"></a> +<span class="sourceLineNo">601</span> if (srcToRemove.isEmpty()) {<a name="line.601"></a> +<span class="sourceLineNo">602</span> LOG.error("The peer we wanted to remove is missing a ReplicationSourceInterface. " +<a name="line.602"></a> +<span class="sourceLineNo">603</span> "This could mean that ReplicationSourceInterface initialization failed for this peer " +<a name="line.603"></a> +<span class="sourceLineNo">604</span> "and that replication on this peer may not be caught up. peerId=" + id);<a name="line.604"></a> +<span class="sourceLineNo">605</span> }<a name="line.605"></a> +<span class="sourceLineNo">606</span> for (ReplicationSourceInterface toRemove : srcToRemove) {<a name="line.606"></a> +<span class="sourceLineNo">607</span> toRemove.terminate(terminateMessage);<a name="line.607"></a> +<span class="sourceLineNo">608</span> closeQueue(toRemove);<a name="line.608"></a> +<span class="sourceLineNo">609</span> }<a name="line.609"></a> +<span class="sourceLineNo">610</span> deleteSource(id, true);<a name="line.610"></a> +<span class="sourceLineNo">611</span> }<a name="line.611"></a> +<span class="sourceLineNo">612</span> }<a name="line.612"></a> +<span class="sourceLineNo">613</span><a name="line.613"></a> +<span class="sourceLineNo">614</span> @Override<a name="line.614"></a> +<span class="sourceLineNo">615</span> public void regionServerRemoved(String regionserver) {<a name="line.615"></a> +<span class="sourceLineNo">616</span> transferQueues(regionserver);<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> @Override<a name="line.619"></a> +<span class="sourceLineNo">620</span> public void peerRemoved(String peerId) {<a name="line.620"></a> +<span class="sourceLineNo">621</span> removePeer(peerId);<a name="line.621"></a> +<span class="sourceLineNo">622</span> this.replicationQueues.removePeerFromHFileRefs(peerId);<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> @Override<a name="line.625"></a> +<span class="sourceLineNo">626</span> public void peerListChanged(List<String> peerIds) {<a name="line.626"></a> +<span class="sourceLineNo">627</span> for (String id : peerIds) {<a name="line.627"></a> +<span class="sourceLineNo">628</span> try {<a name="line.628"></a> +<span class="sourceLineNo">629</span> boolean added = this.replicationPeers.peerConnected(id);<a name="line.629"></a> +<span class="sourceLineNo">630</span> if (added) {<a name="line.630"></a> +<span class="sourceLineNo">631</span> addSource(id);<a name="line.631"></a> +<span class="sourceLineNo">632</span> if (replicationForBulkLoadDataEnabled) {<a name="line.632"></a> +<span class="sourceLineNo">633</span> this.replicationQueues.addPeerToHFileRefs(id);<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> } catch (Exception e) {<a name="line.636"></a> +<span class="sourceLineNo">637</span> LOG.error("Error while adding a new peer", e);<a name="line.637"></a> +<span class="sourceLineNo">638</span> }<a name="line.638"></a> +<span class="sourceLineNo">639</span> }<a name="line.639"></a> +<span class="sourceLineNo">640</span> }<a name="line.640"></a> +<span class="sourceLineNo">641</span><a name="line.641"></a> +<span class="sourceLineNo">642</span> /**<a name="line.642"></a> +<span class="sourceLineNo">643</span> * Class responsible to setup new ReplicationSources to take care of the<a name="line.643"></a> +<span class="sourceLineNo">644</span> * queues from dead region servers.<a name="line.644"></a> +<span class="sourceLineNo">645</span> */<a name="line.645"></a> +<span class="sourceLineNo">646</span> class NodeFailoverWorker extends Thread {<a name="line.646"></a> +<span class="sourceLineNo">647</span><a name="line.647"></a> +<span class="sourceLineNo">648</span> private String rsZnode;<a name="line.648"></a> +<span class="sourceLineNo">649</span> private final ReplicationQueues rq;<a name="line.649"></a> +<span class="sourceLineNo">650</span> private final ReplicationPeers rp;<a name="line.650"></a> +<span class="sourceLineNo">651</span> private final UUID clusterId;<a name="line.651"></a> +<span class="sourceLineNo">652</span><a name="line.652"></a> +<span class="sourceLineNo">653</span> /**<a name="line.653"></a> +<span class="sourceLineNo">654</span> * @param rsZnode<a name="line.654"></a> +<span class="sourceLineNo">655</span> */<a name="line.655"></a> +<span class="sourceLineNo">656</span> public NodeFailoverWorker(String rsZnode) {<a name="line.656"></a> +<span class="sourceLineNo">657</span> this(rsZnode, replicationQueues, replicationPeers, ReplicationSourceManager.this.clusterId);<a name="line.657"></a> <span class="sourceLineNo">658</span> }<a name="line.658"></a> <span class="sourceLineNo">659</span><a name="line.659"></a> -<span class="sourceLineNo">660</span> @Override<a name="line.660"></a> -<span class="sourceLineNo">661</span> public void run() {<a name="line.661"></a> -<span class="sourceLineNo">662</span> if (this.rq.isThisOurRegionServer(rsZnode)) {<a name="line.662"></a> -<span class="sourceLineNo">663</span> return;<a name="line.663"></a> -<span class="sourceLineNo">664</span> }<a name="line.664"></a> -<span class="sourceLineNo">665</span> // Wait a bit before transferring the queues, we may be shutting down.<a name="line.665"></a> -<span class="sourceLineNo">666</span> // This sleep may not be enough in some cases.<a name="line.666"></a> -<span class="sourceLineNo">667</span> try {<a name="line.667"></a> -<span class="sourceLineNo">668</span> Thread.sleep(sleepBeforeFailover + (long) (rand.nextFloat() * sleepBeforeFailover));<a name="line.668"></a> -<span class="sourceLineNo">669</span> } catch (InterruptedException e) {<a name="line.669"></a> -<span class="sourceLineNo">670</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.670"></a> -<span class="sourceLineNo">671</span> Thread.currentThread().interrupt();<a name="line.671"></a> -<span class="sourceLineNo">672</span> }<a name="line.672"></a> -<span class="sourceLineNo">673</span> // We try to lock that rs' queue directory<a name="line.673"></a> -<span class="sourceLineNo">674</span> if (server.isStopped()) {<a name="line.674"></a> -<span class="sourceLineNo">675</span> LOG.info("Not transferring queue since we are shutting down");<a name="line.675"></a> -<span class="sourceLineNo">676</span> return;<a name="line.676"></a> -<span class="sourceLineNo">677</span> }<a name="line.677"></a> -<span class="sourceLineNo">678</span> Map<String, Set<String>> newQueues = new HashMap<>();<a name="line.678"></a> -<span class="sourceLineNo">679</span> List<String> peers = rq.getUnClaimedQueueIds(rsZnode);<a name="line.679"></a> -<span class="sourceLineNo">680</span> while (peers != null && !peers.isEmpty()) {<a name="line.680"></a> -<span class="sourceLineNo">681</span> Pair<String, SortedSet<String>> peer = this.rq.claimQueue(rsZnode,<a name="line.681"></a> -<span class="sourceLineNo">682</span> peers.get(rand.nextInt(peers.size())));<a name="line.682"></a> -<span class="sourceLineNo">683</span> long sleep = sleepBeforeFailover/2;<a name="line.683"></a> -<span class="sourceLineNo">684</span> if (peer != null) {<a name="line.684"></a> -<span class="sourceLineNo">685</span> newQueues.put(peer.getFirst(), peer.getSecond());<a name="line.685"></a> -<span class="sourceLineNo">686</span> sleep = sleepBeforeFailover;<a name="line.686"></a> -<span class="sourceLineNo">687</span> }<a name="line.687"></a> -<span class="sourceLineNo">688</span> try {<a name="line.688"></a> -<span class="sourceLineNo">689</span> Thread.sleep(sleep);<a name="line.689"></a> -<span class="sourceLineNo">690</span> } catch (InterruptedException e) {<a name="line.690"></a> -<span class="sourceLineNo">691</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.691"></a> -<span class="sourceLineNo">692</span> Thread.currentThread().interrupt();<a name="line.692"></a> -<span class="sourceLineNo">693</span> }<a name="line.693"></a> -<span class="sourceLineNo">694</span> peers = rq.getUnClaimedQueueIds(rsZnode);<a name="line.694"></a> -<span class="sourceLineNo">695</span> }<a name="line.695"></a> -<span class="sourceLineNo">696</span> if (peers != null) {<a name="line.696"></a> -<span class="sourceLineNo">697</span> rq.removeReplicatorIfQueueIsEmpty(rsZnode);<a name="line.697"></a> -<span class="sourceLineNo">698</span> }<a name="line.698"></a> -<span class="sourceLineNo">699</span> // Copying over the failed queue is completed.<a name="line.699"></a> -<span class="sourceLineNo">700</span> if (newQueues.isEmpty()) {<a name="line.700"></a> -<span class="sourceLineNo">701</span> // We either didn't get the lock or the failed region server didn't have any outstanding<a name="line.701"></a> -<span class="sourceLineNo">702</span> // WALs to replicate, so we are done.<a name="line.702"></a> -<span class="sourceLineNo">703</span> return;<a name="line.703"></a> +<span class="sourceLineNo">660</span> public NodeFailoverWorker(String rsZnode, final ReplicationQueues replicationQueues,<a name="line.660"></a> +<span class="sourceLineNo">661</span> final ReplicationPeers replicationPeers, final UUID clusterId) {<a name="line.661"></a> +<span class="sourceLineNo">662</span> super("Failover-for-"+rsZnode);<a name="line.662"></a> +<span class="sourceLineNo">663</span> this.rsZnode = rsZnode;<a name="line.663"></a> +<span class="sourceLineNo">664</span> this.rq = replicationQueues;<a name="line.664"></a> +<span class="sourceLineNo">665</span> this.rp = replicationPeers;<a name="line.665"></a> +<span class="sourceLineNo">666</span> this.clusterId = clusterId;<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> @Override<a name="line.669"></a> +<span class="sourceLineNo">670</span> public void run() {<a name="line.670"></a> +<span class="sourceLineNo">671</span> if (this.rq.isThisOurRegionServer(rsZnode)) {<a name="line.671"></a> +<span class="sourceLineNo">672</span> return;<a name="line.672"></a> +<span class="sourceLineNo">673</span> }<a name="line.673"></a> +<span class="sourceLineNo">674</span> // Wait a bit before transferring the queues, we may be shutting down.<a name="line.674"></a> +<span class="sourceLineNo">675</span> // This sleep may not be enough in some cases.<a name="line.675"></a> +<span class="sourceLineNo">676</span> try {<a name="line.676"></a> +<span class="sourceLineNo">677</span> Thread.sleep(sleepBeforeFailover + (long) (rand.nextFloat() * sleepBeforeFailover));<a name="line.677"></a> +<span class="sourceLineNo">678</span> } catch (InterruptedException e) {<a name="line.678"></a> +<span class="sourceLineNo">679</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.679"></a> +<span class="sourceLineNo">680</span> Thread.currentThread().interrupt();<a name="line.680"></a> +<span class="sourceLineNo">681</span> }<a name="line.681"></a> +<span class="sourceLineNo">682</span> // We try to lock that rs' queue directory<a name="line.682"></a> +<span class="sourceLineNo">683</span> if (server.isStopped()) {<a name="line.683"></a> +<span class="sourceLineNo">684</span> LOG.info("Not transferring queue since we are shutting down");<a name="line.684"></a> +<span class="sourceLineNo">685</span> return;<a name="line.685"></a> +<span class="sourceLineNo">686</span> }<a name="line.686"></a> +<span class="sourceLineNo">687</span> Map<String, Set<String>> newQueues = new HashMap<>();<a name="line.687"></a> +<span class="sourceLineNo">688</span> List<String> peers = rq.getUnClaimedQueueIds(rsZnode);<a name="line.688"></a> +<span class="sourceLineNo">689</span> while (peers != null && !peers.isEmpty()) {<a name="line.689"></a> +<span class="sourceLineNo">690</span> Pair<String, SortedSet<String>> peer = this.rq.claimQueue(rsZnode,<a name="line.690"></a> +<span class="sourceLineNo">691</span> peers.get(rand.nextInt(peers.size())));<a name="line.691"></a> +<span class="sourceLineNo">692</span> long sleep = sleepBeforeFailover/2;<a name="line.692"></a> +<span class="sourceLineNo">693</span> if (peer != null) {<a name="line.693"></a> +<span class="sourceLineNo">694</span> newQueues.put(peer.getFirst(), peer.getSecond());<a name="line.694"></a> +<span class="sourceLineNo">695</span> sleep = sleepBeforeFailover;<a name="line.695"></a> +<span class="sourceLineNo">696</span> }<a name="line.696"></a> +<span class="sourceLineNo">697</span> try {<a name="line.697"></a> +<span class="sourceLineNo">698</span> Thread.sleep(sleep);<a name="line.698"></a> +<span class="sourceLineNo">699</span> } catch (InterruptedException e) {<a name="line.699"></a> +<span class="sourceLineNo">700</span> LOG.warn("Interrupted while waiting before transferring a queue.");<a name="line.700"></a> +<span class="sourceLineNo">701</span> Thread.currentThread().interrupt();<a name="line.701"></a> +<span class="sourceLineNo">702</span> }<a name="line.702"></a> +<span class="sourceLineNo">703</span> peers = rq.getUnClaimedQueueIds(rsZnode);<a name="line.703"></a> <span class="sourceLineNo">704</span> }<a name="line.704"></a> -<span class="sourceLineNo">705</span><a name="line.705"></a> -<span class="sourceLineNo">706</span> for (Map.Entry<String, Set<String>> entry : newQueues.entrySet()) {<a name="line.706"></a> -<span class="sourceLineNo">707</span> String peerId = entry.getKey();<a name="line.707"></a> -<span class="sourceLineNo">708</span> Set<String> walsSet = entry.getValue();<a name="line.708"></a> -<span class="sourceLineNo">709</span> try {<a name="line.709"></a> -<span class="sourceLineNo">710</span> // there is not an actual peer defined corresponding to peerId for the failover.<a name="line.710"></a> -<span class="sourceLineNo">711</span> ReplicationQueueInfo replicationQueueInfo = new ReplicationQueueInfo(peerId);<a name="line.711"></a> -<span class="sourceLineNo">712</span> String actualPeerId = replicationQueueInfo.getPeerId();<a name="line.712"></a> -<span class="sourceLineNo">713</span> ReplicationPeer peer = replicationPeers.getConnectedPeer(actualPeerId);<a name="line.713"></a> -<span class="sourceLineNo">714</span> ReplicationPeerConfig peerConfig = null;<a name="line.714"></a> -<span class="sourceLineNo">715</span> try {<a name="line.715"></a> -<span class="sourceLineNo">716</span> peerConfig = replicationPeers.getReplicationPeerConfig(actualPeerId);<a name="line.716"></a> -<span class="sourceLineNo">717</span> } catch (ReplicationException ex) {<a name="line.717"></a> -<span class="sourceLineNo">718</span> LOG.warn("Received exception while getting replication peer config, skipping replay"<a name="line.718"></a> -<span class="sourceLineNo">719</span> + ex);<a name="line.719"></a> -<span class="sourceLineNo">720</span> }<a name="line.720"></a> -<span class="sourceLineNo">721</span> if (peer == null || peerConfig == null) {<a name="line.721"></a> -<span class="sourceLineNo">722</span> LOG.warn("Skipping failover for peer:" + actualPeerId + " of node" + rsZnode);<a name="line.722"></a> -<span class="sourceLineNo">723</span> replicationQueues.removeQueue(peerId);<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> // track sources in walsByIdRecoveredQueues<a name="line.726"></a> -<span class="sourceLineNo">727</span> Map<String, SortedSet<String>> walsByGroup = new HashMap<>();<a name="line.727"></a> -<span class="sourceLineNo">728</span> walsByIdRecoveredQueues.put(peerId, walsByGroup);<a name="line.728"></a> -<span class="sourceLineNo">729</span> for (String wal : walsSet) {<a name="line.729"></a> -<span class="sourceLineNo">730</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(wal);<a name="line.730"></a> -<span class="sourceLineNo">731</span> SortedSet<String> wals = walsByGroup.get(walPrefix);<a name="line.731"></a> -<span class="sourceLineNo">732</span> if (wals == null) {<a name="line.732"></a> -<span class="sourceLineNo">733</span> wals = new TreeSet<>();<a name="line.733"></a> -<span class="sourceLineNo">734</span> walsByGroup.put(walPrefix, wals);<a name="line.734"></a> -<span class="sourceLineNo">735</span> }<a name="line.735"></a> -<span class="sourceLineNo">736</span> wals.add(wal);<a name="line.736"></a> -<span class="sourceLineNo">737</span> }<a name="line.737"></a> -<span class="sourceLineNo">738</span><a name="line.738"></a> -<span class="sourceLineNo">739</span> // enqueue sources<a name="line.739"></a> -<span class="sourceLineNo">740</span> ReplicationSourceInterface src =<a name="line.740"></a> -<span class="sourceLineNo">741</span> getReplicationSource(conf, fs, ReplicationSourceManager.this, this.rq, this.rp,<a name="line.741"></a> -<span class="sourceLineNo">742</span> server, peerId, this.clusterId, peerConfig, peer);<a name="line.742"></a> -<span class="sourceLineNo">743</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.743"></a> -<span class="sourceLineNo">744</span> // see removePeer<a name="line.744"></a> -<span class="sourceLineNo">745</span> synchronized (oldsources) {<a name="line.745"></a> -<span class="sourceLineNo">746</span> if (!this.rp.getConnectedPeerIds().contains(src.getPeerId())) {<a name="line.746"></a> -<span class="sourceLineNo">747</span> src.terminate("Recovered queue doesn't belong to any current peer");<a name="line.747"></a> -<span class="sourceLineNo">748</span> closeRecoveredQueue(src);<a name="line.748"></a> -<span class="sourceLineNo">749</span> continue;<a name="line.749"></a> -<span class="sourceLineNo">750</span> }<a name="line.750"></a> -<span class="sourceLineNo">751</span> oldsources.add(src);<a name="line.751"></a> -<span class="sourceLineNo">752</span> for (String wal : walsSet) {<a name="line.752"></a> -<span class="sourceLineNo">753</span> src.enqueueLog(new Path(oldLogDir, wal));<a name="line.753"></a> -<span class="sourceLineNo">754</span> }<a name="line.754"></a> -<span class="sourceLineNo">755</span> src.startup();<a name="line.755"></a> -<span class="sourceLineNo">756</span> }<a name="line.756"></a> -<span class="sourceLineNo">757</span> } catch (IOException e) {<a name="line.757"></a> -<span class="sourceLineNo">758</span> // TODO manage it<a name="line.758"></a> -<span class="sourceLineNo">759</span> LOG.error("Failed creating a source", e);<a name="line.759"></a> -<span class="sourceLineNo">760</span> }<a name="line.760"></a> -<span class="sourceLineNo">761</span> }<a name="line.761"></a> -<span class="sourceLineNo">762</span> }<a name="line.762"></a> -<span class="sourceLineNo">763</span> }<a name="line.763"></a> -<span class="sourceLineNo">764</span><a name="line.764"></a> -<span class="sourceLineNo">765</span> class AdoptAbandonedQueuesWorker extends Thread{<a name="line.765"></a> -<span class="sourceLineNo">766</span><a name="line.766"></a> -<span class="sourceLineNo">767</span> public AdoptAbandonedQueuesWorker() {}<a name="line.767"></a> -<span class="sourceLineNo">768</span><a name="line.768"></a> -<span class="sourceLineNo">769</span> @Override<a name="line.769"></a> -<span class="sourceLineNo">770</span> public void run() {<a name="line.770"></a> -<span class="sourceLineNo">771</span> List<String> currentReplicators = replicationQueues.getListOfReplicators();<a name="line.771"></a> -<span class="sourceLineNo">772</span> if (currentReplicators == null || currentReplicators.isEmpty()) {<a name="line.772"></a> -<span class="sourceLineNo">773</span> return;<a name="line.773"></a> -<span class="sourceLineNo">774</span> }<a name="line.774"></a> -<span class="sourceLineNo">775</span> List<String> otherRegionServers = replicationTracker.getListOfRegionServers();<a name="line.775"></a> -<span class="sourceLineNo">776</span> LOG.info("Current list of replicators: " + currentReplicators + " other RSs: "<a name="line.776"></a> -<span class="sourceLineNo">777</span> + otherRegionServers);<a name="line.777"></a> -<span class="sourceLineNo">778</span><a name="line.778"></a> -<span class="sourceLineNo">779</span> // Look if there's anything to process after a restart<a name="line.779"></a> -<span class="sourceLineNo">780</span> for (String rs : currentReplicators) {<a name="line.780"></a> -<span class="sourceLineNo">781</span> if (!otherRegionServers.contains(rs)) {<a name="line.781"></a> -<span class="sourceLineNo">782</span> transferQueues(rs);<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> }<a name="line.786"></a> +<span class="sourceLineNo">705</span> if (peers != null) {<a name="line.705"></a> +<span class="sourceLineNo">706</span> rq.removeReplicatorIfQueueIsEmpty(rsZnode);<a name="line.706"></a> +<span class="sourceLineNo">707</span> }<a name="line.707"></a> +<span class="sourceLineNo">708</span> // Copying over the failed queue is completed.<a name="line.708"></a> +<span class="sourceLineNo">709</span> if (newQueues.isEmpty()) {<a name="line.709"></a> +<span class="sourceLineNo">710</span> // We either didn't get the lock or the failed region server didn't have any outstanding<a name="line.710"></a> +<span class="sourceLineNo">711</span> // WALs to replicate, so we are done.<a name="line.711"></a> +<span class="sourceLineNo">712</span> return;<a name="line.712"></a> +<span class="sourceLineNo">713</span> }<a name="line.713"></a> +<span class="sourceLineNo">714</span><a name="line.714"></a> +<span class="sourceLineNo">715</span> for (Map.Entry<String, Set<String>> entry : newQueues.entrySet()) {<a name="line.715"></a> +<span class="sourceLineNo">716</span> String peerId = entry.getKey();<a name="line.716"></a> +<span class="sourceLineNo">717</span> Set<String> walsSet = entry.getValue();<a name="line.717"></a> +<span class="sourceLineNo">718</span> try {<a name="line.718"></a> +<span class="sourceLineNo">719</span> // there is not an actual peer defined corresponding to peerId for the failover.<a name="line.719"></a> +<span class="sourceLineNo">720</span> ReplicationQueueInfo replicationQueueInfo = new ReplicationQueueInfo(peerId);<a name="line.720"></a> +<span class="sourceLineNo">721</span> String actualPeerId = replicationQueueInfo.getPeerId();<a name="line.721"></a> +<span class="sourceLineNo">722</span> ReplicationPeer peer = replicationPeers.getConnectedPeer(actualPeerId);<a name="line.722"></a> +<span class="sourceLineNo">723</span> ReplicationPeerConfig peerConfig = null;<a name="line.723"></a> +<span class="sourceLineNo">724</span> try {<a name="line.724"></a> +<span class="sourceLineNo">725</span> peerConfig = replicationPeers.getReplicationPeerConfig(actualPeerId);<a name="line.725"></a> +<span class="sourceLineNo">726</span> } catch (ReplicationException ex) {<a name="line.726"></a> +<span class="sourceLineNo">727</span> LOG.warn("Received exception while getting replication peer config, skipping replay"<a name="line.727"></a> +<span class="sourceLineNo">728</span> + ex);<a name="line.728"></a> +<span class="sourceLineNo">729</span> }<a name="line.729"></a> +<span class="sourceLineNo">730</span> if (peer == null || peerConfig == null) {<a name="line.730"></a> +<span class="sourceLineNo">731</span> LOG.warn("Skipping failover for peer:" + actualPeerId + " of node" + rsZnode);<a name="line.731"></a> +<span class="sourceLineNo">732</span> replicationQueues.removeQueue(peerId);<a name="line.732"></a> +<span class="sourceLineNo">733</span> continue;<a name="line.733"></a> +<span class="sourceLineNo">734</span> }<a name="line.734"></a> +<span class="sourceLineNo">735</span> // track sources in walsByIdRecoveredQueues<a name="line.735"></a> +<span class="sourceLineNo">736</span> Map<String, SortedSet<String>> walsByGroup = new HashMap<>();<a name="line.736"></a> +<span class="sourceLineNo">737</span> walsByIdRecoveredQueues.put(peerId, walsByGroup);<a name="line.737"></a> +<span class="sourceLineNo">738</span> for (String wal : walsSet) {<a name="line.738"></a> +<span class="sourceLineNo">739</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(wal);<a name="line.739"></a> +<span class="sourceLineNo">740</span> SortedSet<String> wals = walsByGroup.get(walPrefix);<a name="line.740"></a> +<span class="sourceLineNo">741</span> if (wals == null) {<a name="line.741"></a> +<span class="sourceLineNo">742</span> wals = new TreeSet<>();<a name="line.742"></a> +<span class="sourceLineNo">743</span> walsByGroup.put(walPrefix, wals);<a name="line.743"></a> +<span class="sourceLineNo">744</span> }<a name="line.744"></a> +<span class="sourceLineNo">745</span> wals.add(wal);<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> // enqueue sources<a name="line.748"></a> +<span class="sourceLineNo">749</span> ReplicationSourceInterface src =<a name="line.749"></a> +<span class="sourceLineNo">750</span> getReplicationSource(conf, fs, ReplicationSourceManager.this, this.rq, this.rp,<a name="line.750"></a> +<span class="sourceLineNo">751</span> server, peerId, this.clusterId, peerConfig, peer);<a name="line.751"></a> +<span class="sourceLineNo">752</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.752"></a> +<span class="sourceLineNo">753</span> // see removePeer<a name="line.753"></a> +<span class="sourceLineNo">754</span> synchronized (oldsources) {<a name="line.754"></a> +<span class="sourceLineNo">755</span> if (!this.rp.getConnectedPeerIds().contains(src.getPeerId())) {<a name="line.755"></a> +<span class="sourceLineNo">756</span> src.terminate("Recovered queue doesn't belong to any current peer");<a name="line.756"></a> +<span class="sourceLineNo">757</span> closeRecoveredQueue(src);<a name="line.757"></a> +<span class="sourceLineNo">758</span> continue;<a name="line.758"></a> +<span class="sourceLineNo">759</span> }<a name="line.759"></a> +<span class="sourceLineNo">760</span> oldsources.add(src);<a name="line.760"></a> +<span class="sourceLineNo">761</span> for (String wal : walsSet) {<a name="line.761"></a> +<span class="sourceLineNo">762</span> src.enqueueLog(new Path(oldLogDir, wal));<a name="line.762"></a> +<span class="sourceLineNo">763</span> }<a name="line.763"></a> +<span class="sourceLineNo">764</span> src.startup();<a name="line.764"></a> +<span class="sourceLineNo">765</span> }<a name="line.765"></a> +<span class="sourceLineNo">766</span> } catch (IOException e) {<a name="line.766"></a> +<span class="sourceLineNo">767</span> // TODO manage it<a name="line.767"></a> +<span class="sourceLineNo">768</span> LOG.error("Failed creating a source", e);<a name="line.768"></a> +<span class="sourceLineNo">769</span> }<a name="line.769"></a> +<span class="sourceLineNo">770</span> }<a name="line.770"></a> +<span class="sourceLineNo">771</span> }<a name="line.771"></a> +<span class="sourceLineNo">772</span> }<a name="line.772"></a> +<span class="sourceLineNo">773</span><a name="line.773"></a> +<span class="sourceLineNo">774</span> class AdoptAbandonedQueuesWorker extends Thread{<a name="line.774"></a> +<span class="sourceLineNo">775</span><a name="line.775"></a> +<span class="sourceLineNo">776</span> public AdoptAbandonedQueuesWorker() {}<a name="line.776"></a> +<span class="sourceLineNo">777</span><a name="line.777"></a> +<span class="sourceLineNo">778</span> @Override<a name="line.778"></a> +<span class="sourceLineNo">779</span> public void run() {<a name="line.779"></a> +<span class="sourceLineNo">780</span> List<String> currentReplicators = replicationQueues.getListOfReplicators();<a name="line.780"></a> +<span class="sourceLineNo">781</span> if (currentReplicators == null || currentReplicators.isEmpty()) {<a name="line.781"></a> +<span class="sourceLineNo">782</span> return;<a name="line.782"></a> +<span class="sourceLineNo">783</span> }<a name="line.783"></a> +<span class="sourceLineNo">784</span> List<String> otherRegionServers = replicationTracker.getListOfRegionServers();<a name="line.784"></a> +<span class="sourceLineNo">785</span> LOG.info("Current list of replicators: " + currentReplicators + " other RSs: "<a name="line.785"></a> +<span class="sourceLineNo">786</span> + otherRegionServers);<a name="line.786"></a> <span class="sourceLineNo">787</span><a name="line.787"></a> -<span class="sourceLineNo">788</span> /**<a name="line.788"></a> -<span class="sourceLineNo">789</span> * Get the directory where wals are archived<a name="line.789"></a> -<span class="sourceLineNo">790</span> * @return the directory where wals are archived<a name="line.790"></a> -<span class="sourceLineNo">791</span> */<a name="line.791"></a> -<span class="sourceLineNo">792</span> public Path getOldLogDir() {<a name="line.792"></a> -<span class="sourceLineNo">793</span> return this.oldLogDir;<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> /**<a name="line.796"></a> -<span class="sourceLineNo">797</span> * Get the directory where wals are stored by their RSs<a name="line.797"></a> -<span class="sourceLineNo">798</span> * @return the directory where wals are stored by their RSs<a name="line.798"></a> -<span class="sourceLineNo">799</span> */<a name="line.799"></a> -<span class="sourceLineNo">800</span> public Path getLogDir() {<a name="line.800"></a> -<span class="sourceLineNo">801</span> return this.logDir;<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> /**<a name="line.804"></a> -<span class="sourceLineNo">805</span> * Get the handle on the local file system<a name="line.805"></a> -<span class="sourceLineNo">806</span> * @return Handle on the local file system<a name="line.806"></a> -<span class="sourceLineNo">807</span> */<a name="line.807"></a> -<span class="sourceLineNo">808</span> public FileSystem getFs() {<a name="line.808"></a> -<span class="sourceLineNo">809</span> return this.fs;<a name="line.809"></a> -<span class="sourceLineNo">810</span> }<a name="line.810"></a> -<span class="sourceLineNo">811</span><a name="line.811"></a> -<span class="sourceLineNo">812</span> public Connection getConnection() {<a name="line.812"></a> -<span class="sourceLineNo">813</span> return this.connection;<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> * Get the ReplicationPeers used by this ReplicationSourceManager<a name="line.817"></a> -<span class="sourceLineNo">818</span> * @return the ReplicationPeers used by this ReplicationSourceManager<a name="line.818"></a> -<span class="sourceLineNo">819</span> */<a name="line.819"></a> -<span class="sourceLineNo">820</span> public ReplicationPeers getReplicationPeers() {return this.replicationPeers;}<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> * Get a string representation of all the sources' metrics<a name="line.823"></a> -<span class="sourceLineNo">824</span> */<a name="line.824"></a> -<span class="sourceLineNo">825</span> public String getStats() {<a name="line.825"></a> -<span class="sourceLineNo">826</span> StringBuffer stats = new StringBuffer();<a name="line.826"></a> -<span class="sourceLineNo">827</span> for (ReplicationSourceInterface source : sources) {<a name="line.827"></a> -<span class="sourceLineNo">828</span> stats.append("Normal source for cluster " + source.getPeerId() + ": ");<a name="line.828"></a> -<span class="sourceLineNo">829</span> stats.append(source.getStats() + "\n");<a name="line.829"></a> -<span class="sourceLineNo">830</span> }<a name="line.830"></a> -<span class="sourceLineNo">831</span> for (ReplicationSourceInterface oldSource : oldsources) {<a name="line.831"></a> -<span class="sourceLineNo">832</span> stats.append("Recovered source for cluster/machine(s) " + oldSource.getPeerId()+": ");<a name="line.832"></a> -<span class="sourceLineNo">833</span> stats.append(oldSource.getStats()+ "\n");<a name="line.833"></a> -<span class="sourceLineNo">834</span> }<a name="line.834"></a> -<span class="sourceLineNo">835</span> return stats.toString();<a name="line.835"></a> -<span class="sourceLineNo">836</span> }<a name="line.836"></a> -<span class="sourceLineNo">837</span><a name="line.837"></a> -<span class="sourceLineNo">838</span> public void addHFileRefs(TableName tableName, byte[] family, List<Pair<Path, Path>> pairs)<a name="line.838"></a> -<span class="sourceLineNo">839</span> throws ReplicationException {<a name="line.839"></a> -<span class="sourceLineNo">840</span> for (ReplicationSourceInterface source : thi
<TRUNCATED>
