http://git-wip-us.apache.org/repos/asf/hbase-site/blob/3469cbc0/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html index a88ff57..6458b43 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html @@ -565,502 +565,508 @@ <span class="sourceLineNo">557</span> /*<a name="line.557"></a> <span class="sourceLineNo">558</span> * Expire the passed server. Add it to list of dead servers and queue a<a name="line.558"></a> <span class="sourceLineNo">559</span> * shutdown processing.<a name="line.559"></a> -<span class="sourceLineNo">560</span> */<a name="line.560"></a> -<span class="sourceLineNo">561</span> public synchronized void expireServer(final ServerName serverName) {<a name="line.561"></a> -<span class="sourceLineNo">562</span> if (serverName.equals(master.getServerName())) {<a name="line.562"></a> -<span class="sourceLineNo">563</span> if (!(master.isAborted() || master.isStopped())) {<a name="line.563"></a> -<span class="sourceLineNo">564</span> master.stop("We lost our znode?");<a name="line.564"></a> -<span class="sourceLineNo">565</span> }<a name="line.565"></a> -<span class="sourceLineNo">566</span> return;<a name="line.566"></a> -<span class="sourceLineNo">567</span> }<a name="line.567"></a> -<span class="sourceLineNo">568</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.568"></a> -<span class="sourceLineNo">569</span> LOG.info("Master doesn't enable ServerShutdownHandler during initialization, "<a name="line.569"></a> -<span class="sourceLineNo">570</span> + "delay expiring server " + serverName);<a name="line.570"></a> -<span class="sourceLineNo">571</span> // Even we delay expire this server, we still need to handle Meta's RIT<a name="line.571"></a> -<span class="sourceLineNo">572</span> // that are against the crashed server; since when we do RecoverMetaProcedure,<a name="line.572"></a> -<span class="sourceLineNo">573</span> // the SCP is not enable yet and Meta's RIT may be suspend forever. See HBase-19287<a name="line.573"></a> -<span class="sourceLineNo">574</span> master.getAssignmentManager().handleMetaRITOnCrashedServer(serverName);<a name="line.574"></a> -<span class="sourceLineNo">575</span> this.queuedDeadServers.add(serverName);<a name="line.575"></a> -<span class="sourceLineNo">576</span> return;<a name="line.576"></a> -<span class="sourceLineNo">577</span> }<a name="line.577"></a> -<span class="sourceLineNo">578</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.578"></a> -<span class="sourceLineNo">579</span> // TODO: Can this happen? It shouldn't be online in this case?<a name="line.579"></a> -<span class="sourceLineNo">580</span> LOG.warn("Expiration of " + serverName +<a name="line.580"></a> -<span class="sourceLineNo">581</span> " but server shutdown already in progress");<a name="line.581"></a> -<span class="sourceLineNo">582</span> return;<a name="line.582"></a> -<span class="sourceLineNo">583</span> }<a name="line.583"></a> -<span class="sourceLineNo">584</span> moveFromOnlineToDeadServers(serverName);<a name="line.584"></a> -<span class="sourceLineNo">585</span><a name="line.585"></a> -<span class="sourceLineNo">586</span> // If cluster is going down, yes, servers are going to be expiring; don't<a name="line.586"></a> -<span class="sourceLineNo">587</span> // process as a dead server<a name="line.587"></a> -<span class="sourceLineNo">588</span> if (isClusterShutdown()) {<a name="line.588"></a> -<span class="sourceLineNo">589</span> LOG.info("Cluster shutdown set; " + serverName +<a name="line.589"></a> -<span class="sourceLineNo">590</span> " expired; onlineServers=" + this.onlineServers.size());<a name="line.590"></a> -<span class="sourceLineNo">591</span> if (this.onlineServers.isEmpty()) {<a name="line.591"></a> -<span class="sourceLineNo">592</span> master.stop("Cluster shutdown set; onlineServer=0");<a name="line.592"></a> -<span class="sourceLineNo">593</span> }<a name="line.593"></a> -<span class="sourceLineNo">594</span> return;<a name="line.594"></a> -<span class="sourceLineNo">595</span> }<a name="line.595"></a> -<span class="sourceLineNo">596</span> LOG.info("Processing expiration of " + serverName + " on " + this.master.getServerName());<a name="line.596"></a> -<span class="sourceLineNo">597</span> master.getAssignmentManager().submitServerCrash(serverName, true);<a name="line.597"></a> -<span class="sourceLineNo">598</span><a name="line.598"></a> -<span class="sourceLineNo">599</span> // Tell our listeners that a server was removed<a name="line.599"></a> -<span class="sourceLineNo">600</span> if (!this.listeners.isEmpty()) {<a name="line.600"></a> -<span class="sourceLineNo">601</span> for (ServerListener listener : this.listeners) {<a name="line.601"></a> -<span class="sourceLineNo">602</span> listener.serverRemoved(serverName);<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><a name="line.606"></a> -<span class="sourceLineNo">607</span> @VisibleForTesting<a name="line.607"></a> -<span class="sourceLineNo">608</span> public void moveFromOnlineToDeadServers(final ServerName sn) {<a name="line.608"></a> -<span class="sourceLineNo">609</span> synchronized (onlineServers) {<a name="line.609"></a> -<span class="sourceLineNo">610</span> if (!this.onlineServers.containsKey(sn)) {<a name="line.610"></a> -<span class="sourceLineNo">611</span> LOG.warn("Expiration of " + sn + " but server not online");<a name="line.611"></a> -<span class="sourceLineNo">612</span> }<a name="line.612"></a> -<span class="sourceLineNo">613</span> // Remove the server from the known servers lists and update load info BUT<a name="line.613"></a> -<span class="sourceLineNo">614</span> // add to deadservers first; do this so it'll show in dead servers list if<a name="line.614"></a> -<span class="sourceLineNo">615</span> // not in online servers list.<a name="line.615"></a> -<span class="sourceLineNo">616</span> this.deadservers.add(sn);<a name="line.616"></a> -<span class="sourceLineNo">617</span> this.onlineServers.remove(sn);<a name="line.617"></a> -<span class="sourceLineNo">618</span> onlineServers.notifyAll();<a name="line.618"></a> -<span class="sourceLineNo">619</span> }<a name="line.619"></a> -<span class="sourceLineNo">620</span> this.rsAdmins.remove(sn);<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> public synchronized void processDeadServer(final ServerName serverName, boolean shouldSplitWal) {<a name="line.623"></a> -<span class="sourceLineNo">624</span> // When assignment manager is cleaning up the zookeeper nodes and rebuilding the<a name="line.624"></a> -<span class="sourceLineNo">625</span> // in-memory region states, region servers could be down. Meta table can and<a name="line.625"></a> -<span class="sourceLineNo">626</span> // should be re-assigned, log splitting can be done too. However, it is better to<a name="line.626"></a> -<span class="sourceLineNo">627</span> // wait till the cleanup is done before re-assigning user regions.<a name="line.627"></a> -<span class="sourceLineNo">628</span> //<a name="line.628"></a> -<span class="sourceLineNo">629</span> // We should not wait in the server shutdown handler thread since it can clog<a name="line.629"></a> -<span class="sourceLineNo">630</span> // the handler threads and meta table could not be re-assigned in case<a name="line.630"></a> -<span class="sourceLineNo">631</span> // the corresponding server is down. So we queue them up here instead.<a name="line.631"></a> -<span class="sourceLineNo">632</span> if (!master.getAssignmentManager().isFailoverCleanupDone()) {<a name="line.632"></a> -<span class="sourceLineNo">633</span> requeuedDeadServers.put(serverName, shouldSplitWal);<a name="line.633"></a> -<span class="sourceLineNo">634</span> return;<a name="line.634"></a> -<span class="sourceLineNo">635</span> }<a name="line.635"></a> -<span class="sourceLineNo">636</span><a name="line.636"></a> -<span class="sourceLineNo">637</span> this.deadservers.add(serverName);<a name="line.637"></a> -<span class="sourceLineNo">638</span> master.getAssignmentManager().submitServerCrash(serverName, shouldSplitWal);<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> * Process the servers which died during master's initialization. It will be<a name="line.642"></a> -<span class="sourceLineNo">643</span> * called after HMaster#assignMeta and AssignmentManager#joinCluster.<a name="line.643"></a> -<span class="sourceLineNo">644</span> * */<a name="line.644"></a> -<span class="sourceLineNo">645</span> synchronized void processQueuedDeadServers() {<a name="line.645"></a> -<span class="sourceLineNo">646</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.646"></a> -<span class="sourceLineNo">647</span> LOG.info("Master hasn't enabled ServerShutdownHandler");<a name="line.647"></a> -<span class="sourceLineNo">648</span> }<a name="line.648"></a> -<span class="sourceLineNo">649</span> Iterator<ServerName> serverIterator = queuedDeadServers.iterator();<a name="line.649"></a> -<span class="sourceLineNo">650</span> while (serverIterator.hasNext()) {<a name="line.650"></a> -<span class="sourceLineNo">651</span> ServerName tmpServerName = serverIterator.next();<a name="line.651"></a> -<span class="sourceLineNo">652</span> expireServer(tmpServerName);<a name="line.652"></a> -<span class="sourceLineNo">653</span> serverIterator.remove();<a name="line.653"></a> -<span class="sourceLineNo">654</span> requeuedDeadServers.remove(tmpServerName);<a name="line.654"></a> -<span class="sourceLineNo">655</span> }<a name="line.655"></a> -<span class="sourceLineNo">656</span><a name="line.656"></a> -<span class="sourceLineNo">657</span> if (!master.getAssignmentManager().isFailoverCleanupDone()) {<a name="line.657"></a> -<span class="sourceLineNo">658</span> if (LOG.isTraceEnabled()) {<a name="line.658"></a> -<span class="sourceLineNo">659</span> LOG.trace("AssignmentManager failover cleanup not done.");<a name="line.659"></a> -<span class="sourceLineNo">660</span> }<a name="line.660"></a> +<span class="sourceLineNo">560</span> * @return True if we queued a ServerCrashProcedure else false if we did not (could happen<a name="line.560"></a> +<span class="sourceLineNo">561</span> * for many reasons including the fact that its this server that is going down or we already<a name="line.561"></a> +<span class="sourceLineNo">562</span> * have queued an SCP for this server or SCP processing is currently disabled because we are<a name="line.562"></a> +<span class="sourceLineNo">563</span> * in startup phase).<a name="line.563"></a> +<span class="sourceLineNo">564</span> */<a name="line.564"></a> +<span class="sourceLineNo">565</span> public synchronized boolean expireServer(final ServerName serverName) {<a name="line.565"></a> +<span class="sourceLineNo">566</span> // THIS server is going down... can't handle our own expiration.<a name="line.566"></a> +<span class="sourceLineNo">567</span> if (serverName.equals(master.getServerName())) {<a name="line.567"></a> +<span class="sourceLineNo">568</span> if (!(master.isAborted() || master.isStopped())) {<a name="line.568"></a> +<span class="sourceLineNo">569</span> master.stop("We lost our znode?");<a name="line.569"></a> +<span class="sourceLineNo">570</span> }<a name="line.570"></a> +<span class="sourceLineNo">571</span> return false;<a name="line.571"></a> +<span class="sourceLineNo">572</span> }<a name="line.572"></a> +<span class="sourceLineNo">573</span> // No SCP handling during startup.<a name="line.573"></a> +<span class="sourceLineNo">574</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.574"></a> +<span class="sourceLineNo">575</span> LOG.info("Master doesn't enable ServerShutdownHandler during initialization, "<a name="line.575"></a> +<span class="sourceLineNo">576</span> + "delay expiring server " + serverName);<a name="line.576"></a> +<span class="sourceLineNo">577</span> // Even though we delay expire of this server, we still need to handle Meta's RIT<a name="line.577"></a> +<span class="sourceLineNo">578</span> // that are against the crashed server; since when we do RecoverMetaProcedure,<a name="line.578"></a> +<span class="sourceLineNo">579</span> // the SCP is not enabled yet and Meta's RIT may be suspend forever. See HBase-19287<a name="line.579"></a> +<span class="sourceLineNo">580</span> master.getAssignmentManager().handleMetaRITOnCrashedServer(serverName);<a name="line.580"></a> +<span class="sourceLineNo">581</span> this.queuedDeadServers.add(serverName);<a name="line.581"></a> +<span class="sourceLineNo">582</span> // Return true because though on SCP queued, there will be one queued later.<a name="line.582"></a> +<span class="sourceLineNo">583</span> return true;<a name="line.583"></a> +<span class="sourceLineNo">584</span> }<a name="line.584"></a> +<span class="sourceLineNo">585</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.585"></a> +<span class="sourceLineNo">586</span> LOG.warn("Expiration called on {} but crash processing already in progress", serverName);<a name="line.586"></a> +<span class="sourceLineNo">587</span> return false;<a name="line.587"></a> +<span class="sourceLineNo">588</span> }<a name="line.588"></a> +<span class="sourceLineNo">589</span> moveFromOnlineToDeadServers(serverName);<a name="line.589"></a> +<span class="sourceLineNo">590</span><a name="line.590"></a> +<span class="sourceLineNo">591</span> // If cluster is going down, yes, servers are going to be expiring; don't<a name="line.591"></a> +<span class="sourceLineNo">592</span> // process as a dead server<a name="line.592"></a> +<span class="sourceLineNo">593</span> if (isClusterShutdown()) {<a name="line.593"></a> +<span class="sourceLineNo">594</span> LOG.info("Cluster shutdown set; " + serverName +<a name="line.594"></a> +<span class="sourceLineNo">595</span> " expired; onlineServers=" + this.onlineServers.size());<a name="line.595"></a> +<span class="sourceLineNo">596</span> if (this.onlineServers.isEmpty()) {<a name="line.596"></a> +<span class="sourceLineNo">597</span> master.stop("Cluster shutdown set; onlineServer=0");<a name="line.597"></a> +<span class="sourceLineNo">598</span> }<a name="line.598"></a> +<span class="sourceLineNo">599</span> return false;<a name="line.599"></a> +<span class="sourceLineNo">600</span> }<a name="line.600"></a> +<span class="sourceLineNo">601</span> LOG.info("Processing expiration of " + serverName + " on " + this.master.getServerName());<a name="line.601"></a> +<span class="sourceLineNo">602</span> master.getAssignmentManager().submitServerCrash(serverName, true);<a name="line.602"></a> +<span class="sourceLineNo">603</span><a name="line.603"></a> +<span class="sourceLineNo">604</span> // Tell our listeners that a server was removed<a name="line.604"></a> +<span class="sourceLineNo">605</span> if (!this.listeners.isEmpty()) {<a name="line.605"></a> +<span class="sourceLineNo">606</span> for (ServerListener listener : this.listeners) {<a name="line.606"></a> +<span class="sourceLineNo">607</span> listener.serverRemoved(serverName);<a name="line.607"></a> +<span class="sourceLineNo">608</span> }<a name="line.608"></a> +<span class="sourceLineNo">609</span> }<a name="line.609"></a> +<span class="sourceLineNo">610</span> return 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> @VisibleForTesting<a name="line.613"></a> +<span class="sourceLineNo">614</span> public void moveFromOnlineToDeadServers(final ServerName sn) {<a name="line.614"></a> +<span class="sourceLineNo">615</span> synchronized (onlineServers) {<a name="line.615"></a> +<span class="sourceLineNo">616</span> if (!this.onlineServers.containsKey(sn)) {<a name="line.616"></a> +<span class="sourceLineNo">617</span> LOG.warn("Expiration of " + sn + " but server not online");<a name="line.617"></a> +<span class="sourceLineNo">618</span> }<a name="line.618"></a> +<span class="sourceLineNo">619</span> // Remove the server from the known servers lists and update load info BUT<a name="line.619"></a> +<span class="sourceLineNo">620</span> // add to deadservers first; do this so it'll show in dead servers list if<a name="line.620"></a> +<span class="sourceLineNo">621</span> // not in online servers list.<a name="line.621"></a> +<span class="sourceLineNo">622</span> this.deadservers.add(sn);<a name="line.622"></a> +<span class="sourceLineNo">623</span> this.onlineServers.remove(sn);<a name="line.623"></a> +<span class="sourceLineNo">624</span> onlineServers.notifyAll();<a name="line.624"></a> +<span class="sourceLineNo">625</span> }<a name="line.625"></a> +<span class="sourceLineNo">626</span> this.rsAdmins.remove(sn);<a name="line.626"></a> +<span class="sourceLineNo">627</span> }<a name="line.627"></a> +<span class="sourceLineNo">628</span><a name="line.628"></a> +<span class="sourceLineNo">629</span> public synchronized void processDeadServer(final ServerName serverName, boolean shouldSplitWal) {<a name="line.629"></a> +<span class="sourceLineNo">630</span> // When assignment manager is cleaning up the zookeeper nodes and rebuilding the<a name="line.630"></a> +<span class="sourceLineNo">631</span> // in-memory region states, region servers could be down. Meta table can and<a name="line.631"></a> +<span class="sourceLineNo">632</span> // should be re-assigned, log splitting can be done too. However, it is better to<a name="line.632"></a> +<span class="sourceLineNo">633</span> // wait till the cleanup is done before re-assigning user regions.<a name="line.633"></a> +<span class="sourceLineNo">634</span> //<a name="line.634"></a> +<span class="sourceLineNo">635</span> // We should not wait in the server shutdown handler thread since it can clog<a name="line.635"></a> +<span class="sourceLineNo">636</span> // the handler threads and meta table could not be re-assigned in case<a name="line.636"></a> +<span class="sourceLineNo">637</span> // the corresponding server is down. So we queue them up here instead.<a name="line.637"></a> +<span class="sourceLineNo">638</span> if (!master.getAssignmentManager().isFailoverCleanupDone()) {<a name="line.638"></a> +<span class="sourceLineNo">639</span> requeuedDeadServers.put(serverName, shouldSplitWal);<a name="line.639"></a> +<span class="sourceLineNo">640</span> return;<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> this.deadservers.add(serverName);<a name="line.643"></a> +<span class="sourceLineNo">644</span> master.getAssignmentManager().submitServerCrash(serverName, shouldSplitWal);<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> /**<a name="line.647"></a> +<span class="sourceLineNo">648</span> * Process the servers which died during master's initialization. It will be<a name="line.648"></a> +<span class="sourceLineNo">649</span> * called after HMaster#assignMeta and AssignmentManager#joinCluster.<a name="line.649"></a> +<span class="sourceLineNo">650</span> * */<a name="line.650"></a> +<span class="sourceLineNo">651</span> synchronized void processQueuedDeadServers() {<a name="line.651"></a> +<span class="sourceLineNo">652</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.652"></a> +<span class="sourceLineNo">653</span> LOG.info("Master hasn't enabled ServerShutdownHandler");<a name="line.653"></a> +<span class="sourceLineNo">654</span> }<a name="line.654"></a> +<span class="sourceLineNo">655</span> Iterator<ServerName> serverIterator = queuedDeadServers.iterator();<a name="line.655"></a> +<span class="sourceLineNo">656</span> while (serverIterator.hasNext()) {<a name="line.656"></a> +<span class="sourceLineNo">657</span> ServerName tmpServerName = serverIterator.next();<a name="line.657"></a> +<span class="sourceLineNo">658</span> expireServer(tmpServerName);<a name="line.658"></a> +<span class="sourceLineNo">659</span> serverIterator.remove();<a name="line.659"></a> +<span class="sourceLineNo">660</span> requeuedDeadServers.remove(tmpServerName);<a name="line.660"></a> <span class="sourceLineNo">661</span> }<a name="line.661"></a> <span class="sourceLineNo">662</span><a name="line.662"></a> -<span class="sourceLineNo">663</span> for (Map.Entry<ServerName, Boolean> entry : requeuedDeadServers.entrySet()) {<a name="line.663"></a> -<span class="sourceLineNo">664</span> processDeadServer(entry.getKey(), entry.getValue());<a name="line.664"></a> -<span class="sourceLineNo">665</span> }<a name="line.665"></a> -<span class="sourceLineNo">666</span> requeuedDeadServers.clear();<a name="line.666"></a> -<span class="sourceLineNo">667</span> }<a name="line.667"></a> +<span class="sourceLineNo">663</span> if (!master.getAssignmentManager().isFailoverCleanupDone()) {<a name="line.663"></a> +<span class="sourceLineNo">664</span> if (LOG.isTraceEnabled()) {<a name="line.664"></a> +<span class="sourceLineNo">665</span> LOG.trace("AssignmentManager failover cleanup not done.");<a name="line.665"></a> +<span class="sourceLineNo">666</span> }<a name="line.666"></a> +<span class="sourceLineNo">667</span> }<a name="line.667"></a> <span class="sourceLineNo">668</span><a name="line.668"></a> -<span class="sourceLineNo">669</span> /*<a name="line.669"></a> -<span class="sourceLineNo">670</span> * Remove the server from the drain list.<a name="line.670"></a> -<span class="sourceLineNo">671</span> */<a name="line.671"></a> -<span class="sourceLineNo">672</span> public synchronized boolean removeServerFromDrainList(final ServerName sn) {<a name="line.672"></a> -<span class="sourceLineNo">673</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.673"></a> -<span class="sourceLineNo">674</span> // <hostname> , <port> , <startcode><a name="line.674"></a> -<span class="sourceLineNo">675</span><a name="line.675"></a> -<span class="sourceLineNo">676</span> if (!this.isServerOnline(sn)) {<a name="line.676"></a> -<span class="sourceLineNo">677</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.677"></a> -<span class="sourceLineNo">678</span> "Removing from draining list anyway, as requested.");<a name="line.678"></a> -<span class="sourceLineNo">679</span> }<a name="line.679"></a> -<span class="sourceLineNo">680</span> // Remove the server from the draining servers lists.<a name="line.680"></a> -<span class="sourceLineNo">681</span> return this.drainingServers.remove(sn);<a name="line.681"></a> -<span class="sourceLineNo">682</span> }<a name="line.682"></a> -<span class="sourceLineNo">683</span><a name="line.683"></a> -<span class="sourceLineNo">684</span> /**<a name="line.684"></a> -<span class="sourceLineNo">685</span> * Add the server to the drain list.<a name="line.685"></a> -<span class="sourceLineNo">686</span> * @param sn<a name="line.686"></a> -<span class="sourceLineNo">687</span> * @return True if the server is added or the server is already on the drain list.<a name="line.687"></a> -<span class="sourceLineNo">688</span> */<a name="line.688"></a> -<span class="sourceLineNo">689</span> public synchronized boolean addServerToDrainList(final ServerName sn) {<a name="line.689"></a> -<span class="sourceLineNo">690</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.690"></a> -<span class="sourceLineNo">691</span> // <hostname> , <port> , <startcode><a name="line.691"></a> -<span class="sourceLineNo">692</span><a name="line.692"></a> -<span class="sourceLineNo">693</span> if (!this.isServerOnline(sn)) {<a name="line.693"></a> -<span class="sourceLineNo">694</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.694"></a> -<span class="sourceLineNo">695</span> "Ignoring request to add it to draining list.");<a name="line.695"></a> -<span class="sourceLineNo">696</span> return false;<a name="line.696"></a> -<span class="sourceLineNo">697</span> }<a name="line.697"></a> -<span class="sourceLineNo">698</span> // Add the server to the draining servers lists, if it's not already in<a name="line.698"></a> -<span class="sourceLineNo">699</span> // it.<a name="line.699"></a> -<span class="sourceLineNo">700</span> if (this.drainingServers.contains(sn)) {<a name="line.700"></a> -<span class="sourceLineNo">701</span> LOG.warn("Server " + sn + " is already in the draining server list." +<a name="line.701"></a> -<span class="sourceLineNo">702</span> "Ignoring request to add it again.");<a name="line.702"></a> -<span class="sourceLineNo">703</span> return true;<a name="line.703"></a> -<span class="sourceLineNo">704</span> }<a name="line.704"></a> -<span class="sourceLineNo">705</span> LOG.info("Server " + sn + " added to draining server list.");<a name="line.705"></a> -<span class="sourceLineNo">706</span> return this.drainingServers.add(sn);<a name="line.706"></a> -<span class="sourceLineNo">707</span> }<a name="line.707"></a> -<span class="sourceLineNo">708</span><a name="line.708"></a> -<span class="sourceLineNo">709</span> // RPC methods to region servers<a name="line.709"></a> -<span class="sourceLineNo">710</span><a name="line.710"></a> -<span class="sourceLineNo">711</span> private HBaseRpcController newRpcController() {<a name="line.711"></a> -<span class="sourceLineNo">712</span> return rpcControllerFactory == null ? null : rpcControllerFactory.newController();<a name="line.712"></a> +<span class="sourceLineNo">669</span> for (Map.Entry<ServerName, Boolean> entry : requeuedDeadServers.entrySet()) {<a name="line.669"></a> +<span class="sourceLineNo">670</span> processDeadServer(entry.getKey(), entry.getValue());<a name="line.670"></a> +<span class="sourceLineNo">671</span> }<a name="line.671"></a> +<span class="sourceLineNo">672</span> requeuedDeadServers.clear();<a name="line.672"></a> +<span class="sourceLineNo">673</span> }<a name="line.673"></a> +<span class="sourceLineNo">674</span><a name="line.674"></a> +<span class="sourceLineNo">675</span> /*<a name="line.675"></a> +<span class="sourceLineNo">676</span> * Remove the server from the drain list.<a name="line.676"></a> +<span class="sourceLineNo">677</span> */<a name="line.677"></a> +<span class="sourceLineNo">678</span> public synchronized boolean removeServerFromDrainList(final ServerName sn) {<a name="line.678"></a> +<span class="sourceLineNo">679</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.679"></a> +<span class="sourceLineNo">680</span> // <hostname> , <port> , <startcode><a name="line.680"></a> +<span class="sourceLineNo">681</span><a name="line.681"></a> +<span class="sourceLineNo">682</span> if (!this.isServerOnline(sn)) {<a name="line.682"></a> +<span class="sourceLineNo">683</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.683"></a> +<span class="sourceLineNo">684</span> "Removing from draining list anyway, as requested.");<a name="line.684"></a> +<span class="sourceLineNo">685</span> }<a name="line.685"></a> +<span class="sourceLineNo">686</span> // Remove the server from the draining servers lists.<a name="line.686"></a> +<span class="sourceLineNo">687</span> return this.drainingServers.remove(sn);<a name="line.687"></a> +<span class="sourceLineNo">688</span> }<a name="line.688"></a> +<span class="sourceLineNo">689</span><a name="line.689"></a> +<span class="sourceLineNo">690</span> /**<a name="line.690"></a> +<span class="sourceLineNo">691</span> * Add the server to the drain list.<a name="line.691"></a> +<span class="sourceLineNo">692</span> * @param sn<a name="line.692"></a> +<span class="sourceLineNo">693</span> * @return True if the server is added or the server is already on the drain list.<a name="line.693"></a> +<span class="sourceLineNo">694</span> */<a name="line.694"></a> +<span class="sourceLineNo">695</span> public synchronized boolean addServerToDrainList(final ServerName sn) {<a name="line.695"></a> +<span class="sourceLineNo">696</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.696"></a> +<span class="sourceLineNo">697</span> // <hostname> , <port> , <startcode><a name="line.697"></a> +<span class="sourceLineNo">698</span><a name="line.698"></a> +<span class="sourceLineNo">699</span> if (!this.isServerOnline(sn)) {<a name="line.699"></a> +<span class="sourceLineNo">700</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.700"></a> +<span class="sourceLineNo">701</span> "Ignoring request to add it to draining list.");<a name="line.701"></a> +<span class="sourceLineNo">702</span> return false;<a name="line.702"></a> +<span class="sourceLineNo">703</span> }<a name="line.703"></a> +<span class="sourceLineNo">704</span> // Add the server to the draining servers lists, if it's not already in<a name="line.704"></a> +<span class="sourceLineNo">705</span> // it.<a name="line.705"></a> +<span class="sourceLineNo">706</span> if (this.drainingServers.contains(sn)) {<a name="line.706"></a> +<span class="sourceLineNo">707</span> LOG.warn("Server " + sn + " is already in the draining server list." +<a name="line.707"></a> +<span class="sourceLineNo">708</span> "Ignoring request to add it again.");<a name="line.708"></a> +<span class="sourceLineNo">709</span> return true;<a name="line.709"></a> +<span class="sourceLineNo">710</span> }<a name="line.710"></a> +<span class="sourceLineNo">711</span> LOG.info("Server " + sn + " added to draining server list.");<a name="line.711"></a> +<span class="sourceLineNo">712</span> return this.drainingServers.add(sn);<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> /**<a name="line.715"></a> -<span class="sourceLineNo">716</span> * Sends a WARMUP RPC to the specified server to warmup the specified region.<a name="line.716"></a> -<span class="sourceLineNo">717</span> * <p><a name="line.717"></a> -<span class="sourceLineNo">718</span> * A region server could reject the close request because it either does not<a name="line.718"></a> -<span class="sourceLineNo">719</span> * have the specified region or the region is being split.<a name="line.719"></a> -<span class="sourceLineNo">720</span> * @param server server to warmup a region<a name="line.720"></a> -<span class="sourceLineNo">721</span> * @param region region to warmup<a name="line.721"></a> -<span class="sourceLineNo">722</span> */<a name="line.722"></a> -<span class="sourceLineNo">723</span> public void sendRegionWarmup(ServerName server,<a name="line.723"></a> -<span class="sourceLineNo">724</span> RegionInfo region) {<a name="line.724"></a> -<span class="sourceLineNo">725</span> if (server == null) return;<a name="line.725"></a> -<span class="sourceLineNo">726</span> try {<a name="line.726"></a> -<span class="sourceLineNo">727</span> AdminService.BlockingInterface admin = getRsAdmin(server);<a name="line.727"></a> -<span class="sourceLineNo">728</span> HBaseRpcController controller = newRpcController();<a name="line.728"></a> -<span class="sourceLineNo">729</span> ProtobufUtil.warmupRegion(controller, admin, region);<a name="line.729"></a> -<span class="sourceLineNo">730</span> } catch (IOException e) {<a name="line.730"></a> -<span class="sourceLineNo">731</span> LOG.error("Received exception in RPC for warmup server:" +<a name="line.731"></a> -<span class="sourceLineNo">732</span> server + "region: " + region +<a name="line.732"></a> -<span class="sourceLineNo">733</span> "exception: " + e);<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><a name="line.736"></a> -<span class="sourceLineNo">737</span> /**<a name="line.737"></a> -<span class="sourceLineNo">738</span> * Contacts a region server and waits up to timeout ms<a name="line.738"></a> -<span class="sourceLineNo">739</span> * to close the region. This bypasses the active hmaster.<a name="line.739"></a> -<span class="sourceLineNo">740</span> */<a name="line.740"></a> -<span class="sourceLineNo">741</span> public static void closeRegionSilentlyAndWait(ClusterConnection connection,<a name="line.741"></a> -<span class="sourceLineNo">742</span> ServerName server, RegionInfo region, long timeout) throws IOException, InterruptedException {<a name="line.742"></a> -<span class="sourceLineNo">743</span> AdminService.BlockingInterface rs = connection.getAdmin(server);<a name="line.743"></a> -<span class="sourceLineNo">744</span> HBaseRpcController controller = connection.getRpcControllerFactory().newController();<a name="line.744"></a> -<span class="sourceLineNo">745</span> try {<a name="line.745"></a> -<span class="sourceLineNo">746</span> ProtobufUtil.closeRegion(controller, rs, server, region.getRegionName());<a name="line.746"></a> -<span class="sourceLineNo">747</span> } catch (IOException e) {<a name="line.747"></a> -<span class="sourceLineNo">748</span> LOG.warn("Exception when closing region: " + region.getRegionNameAsString(), e);<a name="line.748"></a> -<span class="sourceLineNo">749</span> }<a name="line.749"></a> -<span class="sourceLineNo">750</span> long expiration = timeout + System.currentTimeMillis();<a name="line.750"></a> -<span class="sourceLineNo">751</span> while (System.currentTimeMillis() < expiration) {<a name="line.751"></a> -<span class="sourceLineNo">752</span> controller.reset();<a name="line.752"></a> -<span class="sourceLineNo">753</span> try {<a name="line.753"></a> -<span class="sourceLineNo">754</span> RegionInfo rsRegion =<a name="line.754"></a> -<span class="sourceLineNo">755</span> ProtobufUtil.getRegionInfo(controller, rs, region.getRegionName());<a name="line.755"></a> -<span class="sourceLineNo">756</span> if (rsRegion == null) return;<a name="line.756"></a> -<span class="sourceLineNo">757</span> } catch (IOException ioe) {<a name="line.757"></a> -<span class="sourceLineNo">758</span> if (ioe instanceof NotServingRegionException) // no need to retry again<a name="line.758"></a> -<span class="sourceLineNo">759</span> return;<a name="line.759"></a> -<span class="sourceLineNo">760</span> LOG.warn("Exception when retrieving regioninfo from: "<a name="line.760"></a> -<span class="sourceLineNo">761</span> + region.getRegionNameAsString(), ioe);<a name="line.761"></a> -<span class="sourceLineNo">762</span> }<a name="line.762"></a> -<span class="sourceLineNo">763</span> Thread.sleep(1000);<a name="line.763"></a> -<span class="sourceLineNo">764</span> }<a name="line.764"></a> -<span class="sourceLineNo">765</span> throw new IOException("Region " + region + " failed to close within"<a name="line.765"></a> -<span class="sourceLineNo">766</span> + " timeout " + timeout);<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> * @param sn<a name="line.770"></a> -<span class="sourceLineNo">771</span> * @return Admin interface for the remote regionserver named <code>sn</code><a name="line.771"></a> -<span class="sourceLineNo">772</span> * @throws IOException<a name="line.772"></a> -<span class="sourceLineNo">773</span> * @throws RetriesExhaustedException wrapping a ConnectException if failed<a name="line.773"></a> -<span class="sourceLineNo">774</span> */<a name="line.774"></a> -<span class="sourceLineNo">775</span> public AdminService.BlockingInterface getRsAdmin(final ServerName sn)<a name="line.775"></a> -<span class="sourceLineNo">776</span> throws IOException {<a name="line.776"></a> -<span class="sourceLineNo">777</span> AdminService.BlockingInterface admin = this.rsAdmins.get(sn);<a name="line.777"></a> -<span class="sourceLineNo">778</span> if (admin == null) {<a name="line.778"></a> -<span class="sourceLineNo">779</span> LOG.debug("New admin connection to " + sn.toString());<a name="line.779"></a> -<span class="sourceLineNo">780</span> if (sn.equals(master.getServerName()) && master instanceof HRegionServer) {<a name="line.780"></a> -<span class="sourceLineNo">781</span> // A master is also a region server now, see HBASE-10569 for details<a name="line.781"></a> -<span class="sourceLineNo">782</span> admin = ((HRegionServer)master).getRSRpcServices();<a name="line.782"></a> -<span class="sourceLineNo">783</span> } else {<a name="line.783"></a> -<span class="sourceLineNo">784</span> admin = this.connection.getAdmin(sn);<a name="line.784"></a> -<span class="sourceLineNo">785</span> }<a name="line.785"></a> -<span class="sourceLineNo">786</span> this.rsAdmins.put(sn, admin);<a name="line.786"></a> -<span class="sourceLineNo">787</span> }<a name="line.787"></a> -<span class="sourceLineNo">788</span> return admin;<a name="line.788"></a> -<span class="sourceLineNo">789</span> }<a name="line.789"></a> -<span class="sourceLineNo">790</span><a name="line.790"></a> -<span class="sourceLineNo">791</span> /**<a name="line.791"></a> -<span class="sourceLineNo">792</span> * Calculate min necessary to start. This is not an absolute. It is just<a name="line.792"></a> -<span class="sourceLineNo">793</span> * a friction that will cause us hang around a bit longer waiting on<a name="line.793"></a> -<span class="sourceLineNo">794</span> * RegionServers to check-in.<a name="line.794"></a> -<span class="sourceLineNo">795</span> */<a name="line.795"></a> -<span class="sourceLineNo">796</span> private int getMinToStart() {<a name="line.796"></a> -<span class="sourceLineNo">797</span> // One server should be enough to get us off the ground.<a name="line.797"></a> -<span class="sourceLineNo">798</span> int requiredMinToStart = 1;<a name="line.798"></a> -<span class="sourceLineNo">799</span> if (LoadBalancer.isTablesOnMaster(master.getConfiguration())) {<a name="line.799"></a> -<span class="sourceLineNo">800</span> if (LoadBalancer.isSystemTablesOnlyOnMaster(master.getConfiguration())) {<a name="line.800"></a> -<span class="sourceLineNo">801</span> // If Master is carrying regions but NOT user-space regions, it<a name="line.801"></a> -<span class="sourceLineNo">802</span> // still shows as a 'server'. We need at least one more server to check<a name="line.802"></a> -<span class="sourceLineNo">803</span> // in before we can start up so set defaultMinToStart to 2.<a name="line.803"></a> -<span class="sourceLineNo">804</span> requiredMinToStart = requiredMinToStart + 1;<a name="line.804"></a> -<span class="sourceLineNo">805</span> }<a name="line.805"></a> -<span class="sourceLineNo">806</span> }<a name="line.806"></a> -<span class="sourceLineNo">807</span> int minToStart = this.master.getConfiguration().getInt(WAIT_ON_REGIONSERVERS_MINTOSTART, -1);<a name="line.807"></a> -<span class="sourceLineNo">808</span> // Ensure we are never less than requiredMinToStart else stuff won't work.<a name="line.808"></a> -<span class="sourceLineNo">809</span> return minToStart == -1 || minToStart < requiredMinToStart? requiredMinToStart: minToStart;<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> /**<a name="line.812"></a> -<span class="sourceLineNo">813</span> * Wait for the region servers to report in.<a name="line.813"></a> -<span class="sourceLineNo">814</span> * We will wait until one of this condition is met:<a name="line.814"></a> -<span class="sourceLineNo">815</span> * - the master is stopped<a name="line.815"></a> -<span class="sourceLineNo">816</span> * - the 'hbase.master.wait.on.regionservers.maxtostart' number of<a name="line.816"></a> -<span class="sourceLineNo">817</span> * region servers is reached<a name="line.817"></a> -<span class="sourceLineNo">818</span> * - the 'hbase.master.wait.on.regionservers.mintostart' is reached AND<a name="line.818"></a> -<span class="sourceLineNo">819</span> * there have been no new region server in for<a name="line.819"></a> -<span class="sourceLineNo">820</span> * 'hbase.master.wait.on.regionservers.interval' time AND<a name="line.820"></a> -<span class="sourceLineNo">821</span> * the 'hbase.master.wait.on.regionservers.timeout' is reached<a name="line.821"></a> -<span class="sourceLineNo">822</span> *<a name="line.822"></a> -<span class="sourceLineNo">823</span> * @throws InterruptedException<a name="line.823"></a> -<span class="sourceLineNo">824</span> */<a name="line.824"></a> -<span class="sourceLineNo">825</span> public void waitForRegionServers(MonitoredTask status) throws InterruptedException {<a name="line.825"></a> -<span class="sourceLineNo">826</span> final long interval = this.master.getConfiguration().<a name="line.826"></a> -<span class="sourceLineNo">827</span> getLong(WAIT_ON_REGIONSERVERS_INTERVAL, 1500);<a name="line.827"></a> -<span class="sourceLineNo">828</span> final long timeout = this.master.getConfiguration().<a name="line.828"></a> -<span class="sourceLineNo">829</span> getLong(WAIT_ON_REGIONSERVERS_TIMEOUT, 4500);<a name="line.829"></a> -<span class="sourceLineNo">830</span> // Min is not an absolute; just a friction making us wait longer on server checkin.<a name="line.830"></a> -<span class="sourceLineNo">831</span> int minToStart = getMinToStart();<a name="line.831"></a> -<span class="sourceLineNo">832</span> int maxToStart = this.master.getConfiguration().<a name="line.832"></a> -<span class="sourceLineNo">833</span> getInt(WAIT_ON_REGIONSERVERS_MAXTOSTART, Integer.MAX_VALUE);<a name="line.833"></a> -<span class="sourceLineNo">834</span> if (maxToStart < minToStart) {<a name="line.834"></a> -<span class="sourceLineNo">835</span> LOG.warn(String.format("The value of '%s' (%d) is set less than '%s' (%d), ignoring.",<a name="line.835"></a> -<span class="sourceLineNo">836</span> WAIT_ON_REGIONSERVERS_MAXTOSTART, maxToStart,<a name="line.836"></a> -<span class="sourceLineNo">837</span> WAIT_ON_REGIONSERVERS_MINTOSTART, minToStart));<a name="line.837"></a> -<span class="sourceLineNo">838</span> maxToStart = Integer.MAX_VALUE;<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> long now = System.currentTimeMillis();<a name="line.841"></a> -<span class="sourceLineNo">842</span> final long startTime = now;<a name="line.842"></a> -<span class="sourceLineNo">843</span> long slept = 0;<a name="line.843"></a> -<span class="sourceLineNo">844</span> long lastLogTime = 0;<a name="line.844"></a> -<span class="sourceLineNo">845</span> long lastCountChange = startTime;<a name="line.845"></a> -<span class="sourceLineNo">846</span> int count = countOfRegionServers();<a name="line.846"></a> -<span class="sourceLineNo">847</span> int oldCount = 0;<a name="line.847"></a> -<span class="sourceLineNo">848</span> // This while test is a little hard to read. We try to comment it in below but in essence:<a name="line.848"></a> -<span class="sourceLineNo">849</span> // Wait if Master is not stopped and the number of regionservers that have checked-in is<a name="line.849"></a> -<span class="sourceLineNo">850</span> // less than the maxToStart. Both of these conditions will be true near universally.<a name="line.850"></a> -<span class="sourceLineNo">851</span> // Next, we will keep cycling if ANY of the following three conditions are true:<a name="line.851"></a> -<span class="sourceLineNo">852</span> // 1. The time since a regionserver registered is < interval (means servers are actively checking in).<a name="line.852"></a> -<span class="sourceLineNo">853</span> // 2. We are under the total timeout.<a name="line.853"></a> -<span class="sourceLineNo">854</span> // 3. The count of servers is < minimum.<a name="line.854"></a> -<span class="sourceLineNo">855</span> for (ServerListener listener: this.listeners) {<a name="line.855"></a> -<span class="sourceLineNo">856</span> listener.waiting();<a name="line.856"></a> -<span class="sourceLineNo">857</span> }<a name="line.857"></a> -<span class="sourceLineNo">858</span> while (!this.master.isStopped() && !isClusterShutdown() && count < maxToStart &&<a name="line.858"></a> -<span class="sourceLineNo">859</span> ((lastCountChange + interval) > now || timeout > slept || count < minToStart)) {<a name="line.859"></a> -<span class="sourceLineNo">860</span> // Log some info at every interval time or if there is a change<a name="line.860"></a> -<span class="sourceLineNo">861</span> if (oldCount != count || lastLogTime + interval < now) {<a name="line.861"></a> -<span class="sourceLineNo">862</span> lastLogTime = now;<a name="line.862"></a> -<span class="sourceLineNo">863</span> String msg =<a name="line.863"></a> -<span class="sourceLineNo">864</span> "Waiting on regionserver count=" + count + "; waited="+<a name="line.864"></a> -<span class="sourceLineNo">865</span> slept + "ms, expecting min=" + minToStart + " server(s), max="+ getStrForMax(maxToStart) +<a name="line.865"></a> -<span class="sourceLineNo">866</span> " server(s), " + "timeout=" + timeout + "ms, lastChange=" + (lastCountChange - now) + "ms";<a name="line.866"></a> -<span class="sourceLineNo">867</span> LOG.info(msg);<a name="line.867"></a> -<span class="sourceLineNo">868</span> status.setStatus(msg);<a name="line.868"></a> -<span class="sourceLineNo">869</span> }<a name="line.869"></a> -<span class="sourceLineNo">870</span><a name="line.870"></a> -<span class="sourceLineNo">871</span> // We sleep for some time<a name="line.871"></a> -<span class="sourceLineNo">872</span> final long sleepTime = 50;<a name="line.872"></a> -<span class="sourceLineNo">873</span> Thread.sleep(sleepTime);<a name="line.873"></a> -<span class="sourceLineNo">874</span> now = System.currentTimeMillis();<a name="line.874"></a> -<span class="sourceLineNo">875</span> slept = now - startTime;<a name="line.875"></a> +<span class="sourceLineNo">715</span> // RPC methods to region servers<a name="line.715"></a> +<span class="sourceLineNo">716</span><a name="line.716"></a> +<span class="sourceLineNo">717</span> private HBaseRpcController newRpcController() {<a name="line.717"></a> +<span class="sourceLineNo">718</span> return rpcControllerFactory == null ? null : rpcControllerFactory.newController();<a name="line.718"></a> +<span class="sourceLineNo">719</span> }<a name="line.719"></a> +<span class="sourceLineNo">720</span><a name="line.720"></a> +<span class="sourceLineNo">721</span> /**<a name="line.721"></a> +<span class="sourceLineNo">722</span> * Sends a WARMUP RPC to the specified server to warmup the specified region.<a name="line.722"></a> +<span class="sourceLineNo">723</span> * <p><a name="line.723"></a> +<span class="sourceLineNo">724</span> * A region server could reject the close request because it either does not<a name="line.724"></a> +<span class="sourceLineNo">725</span> * have the specified region or the region is being split.<a name="line.725"></a> +<span class="sourceLineNo">726</span> * @param server server to warmup a region<a name="line.726"></a> +<span class="sourceLineNo">727</span> * @param region region to warmup<a name="line.727"></a> +<span class="sourceLineNo">728</span> */<a name="line.728"></a> +<span class="sourceLineNo">729</span> public void sendRegionWarmup(ServerName server,<a name="line.729"></a> +<span class="sourceLineNo">730</span> RegionInfo region) {<a name="line.730"></a> +<span class="sourceLineNo">731</span> if (server == null) return;<a name="line.731"></a> +<span class="sourceLineNo">732</span> try {<a name="line.732"></a> +<span class="sourceLineNo">733</span> AdminService.BlockingInterface admin = getRsAdmin(server);<a name="line.733"></a> +<span class="sourceLineNo">734</span> HBaseRpcController controller = newRpcController();<a name="line.734"></a> +<span class="sourceLineNo">735</span> ProtobufUtil.warmupRegion(controller, admin, region);<a name="line.735"></a> +<span class="sourceLineNo">736</span> } catch (IOException e) {<a name="line.736"></a> +<span class="sourceLineNo">737</span> LOG.error("Received exception in RPC for warmup server:" +<a name="line.737"></a> +<span class="sourceLineNo">738</span> server + "region: " + region +<a name="line.738"></a> +<span class="sourceLineNo">739</span> "exception: " + e);<a name="line.739"></a> +<span class="sourceLineNo">740</span> }<a name="line.740"></a> +<span class="sourceLineNo">741</span> }<a name="line.741"></a> +<span class="sourceLineNo">742</span><a name="line.742"></a> +<span class="sourceLineNo">743</span> /**<a name="line.743"></a> +<span class="sourceLineNo">744</span> * Contacts a region server and waits up to timeout ms<a name="line.744"></a> +<span class="sourceLineNo">745</span> * to close the region. This bypasses the active hmaster.<a name="line.745"></a> +<span class="sourceLineNo">746</span> */<a name="line.746"></a> +<span class="sourceLineNo">747</span> public static void closeRegionSilentlyAndWait(ClusterConnection connection,<a name="line.747"></a> +<span class="sourceLineNo">748</span> ServerName server, RegionInfo region, long timeout) throws IOException, InterruptedException {<a name="line.748"></a> +<span class="sourceLineNo">749</span> AdminService.BlockingInterface rs = connection.getAdmin(server);<a name="line.749"></a> +<span class="sourceLineNo">750</span> HBaseRpcController controller = connection.getRpcControllerFactory().newController();<a name="line.750"></a> +<span class="sourceLineNo">751</span> try {<a name="line.751"></a> +<span class="sourceLineNo">752</span> ProtobufUtil.closeRegion(controller, rs, server, region.getRegionName());<a name="line.752"></a> +<span class="sourceLineNo">753</span> } catch (IOException e) {<a name="line.753"></a> +<span class="sourceLineNo">754</span> LOG.warn("Exception when closing region: " + region.getRegionNameAsString(), e);<a name="line.754"></a> +<span class="sourceLineNo">755</span> }<a name="line.755"></a> +<span class="sourceLineNo">756</span> long expiration = timeout + System.currentTimeMillis();<a name="line.756"></a> +<span class="sourceLineNo">757</span> while (System.currentTimeMillis() < expiration) {<a name="line.757"></a> +<span class="sourceLineNo">758</span> controller.reset();<a name="line.758"></a> +<span class="sourceLineNo">759</span> try {<a name="line.759"></a> +<span class="sourceLineNo">760</span> RegionInfo rsRegion =<a name="line.760"></a> +<span class="sourceLineNo">761</span> ProtobufUtil.getRegionInfo(controller, rs, region.getRegionName());<a name="line.761"></a> +<span class="sourceLineNo">762</span> if (rsRegion == null) return;<a name="line.762"></a> +<span class="sourceLineNo">763</span> } catch (IOException ioe) {<a name="line.763"></a> +<span class="sourceLineNo">764</span> if (ioe instanceof NotServingRegionException) // no need to retry again<a name="line.764"></a> +<span class="sourceLineNo">765</span> return;<a name="line.765"></a> +<span class="sourceLineNo">766</span> LOG.warn("Exception when retrieving regioninfo from: "<a name="line.766"></a> +<span class="sourceLineNo">767</span> + region.getRegionNameAsString(), ioe);<a name="line.767"></a> +<span class="sourceLineNo">768</span> }<a name="line.768"></a> +<span class="sourceLineNo">769</span> Thread.sleep(1000);<a name="line.769"></a> +<span class="sourceLineNo">770</span> }<a name="line.770"></a> +<span class="sourceLineNo">771</span> throw new IOException("Region " + region + " failed to close within"<a name="line.771"></a> +<span class="sourceLineNo">772</span> + " timeout " + timeout);<a name="line.772"></a> +<span class="sourceLineNo">773</span> }<a name="line.773"></a> +<span class="sourceLineNo">774</span><a name="line.774"></a> +<span class="sourceLineNo">775</span> /**<a name="line.775"></a> +<span class="sourceLineNo">776</span> * @param sn<a name="line.776"></a> +<span class="sourceLineNo">777</span> * @return Admin interface for the remote regionserver named <code>sn</code><a name="line.777"></a> +<span class="sourceLineNo">778</span> * @throws IOException<a name="line.778"></a> +<span class="sourceLineNo">779</span> * @throws RetriesExhaustedException wrapping a ConnectException if failed<a name="line.779"></a> +<span class="sourceLineNo">780</span> */<a name="line.780"></a> +<span class="sourceLineNo">781</span> public AdminService.BlockingInterface getRsAdmin(final ServerName sn)<a name="line.781"></a> +<span class="sourceLineNo">782</span> throws IOException {<a name="line.782"></a> +<span class="sourceLineNo">783</span> AdminService.BlockingInterface admin = this.rsAdmins.get(sn);<a name="line.783"></a> +<span class="sourceLineNo">784</span> if (admin == null) {<a name="line.784"></a> +<span class="sourceLineNo">785</span> LOG.debug("New admin connection to " + sn.toString());<a name="line.785"></a> +<span class="sourceLineNo">786</span> if (sn.equals(master.getServerName()) && master instanceof HRegionServer) {<a name="line.786"></a> +<span class="sourceLineNo">787</span> // A master is also a region server now, see HBASE-10569 for details<a name="line.787"></a> +<span class="sourceLineNo">788</span> admin = ((HRegionServer)master).getRSRpcServices();<a name="line.788"></a> +<span class="sourceLineNo">789</span> } else {<a name="line.789"></a> +<span class="sourceLineNo">790</span> admin = this.connection.getAdmin(sn);<a name="line.790"></a> +<span class="sourceLineNo">791</span> }<a name="line.791"></a> +<span class="sourceLineNo">792</span> this.rsAdmins.put(sn, admin);<a name="line.792"></a> +<span class="sourceLineNo">793</span> }<a name="line.793"></a> +<span class="sourceLineNo">794</span> return admin;<a name="line.794"></a> +<span class="sourceLineNo">795</span> }<a name="line.795"></a> +<span class="sourceLineNo">796</span><a name="line.796"></a> +<span class="sourceLineNo">797</span> /**<a name="line.797"></a> +<span class="sourceLineNo">798</span> * Calculate min necessary to start. This is not an absolute. It is just<a name="line.798"></a> +<span class="sourceLineNo">799</span> * a friction that will cause us hang around a bit longer waiting on<a name="line.799"></a> +<span class="sourceLineNo">800</span> * RegionServers to check-in.<a name="line.800"></a> +<span class="sourceLineNo">801</span> */<a name="line.801"></a> +<span class="sourceLineNo">802</span> private int getMinToStart() {<a name="line.802"></a> +<span class="sourceLineNo">803</span> // One server should be enough to get us off the ground.<a name="line.803"></a> +<span class="sourceLineNo">804</span> int requiredMinToStart = 1;<a name="line.804"></a> +<span class="sourceLineNo">805</span> if (LoadBalancer.isTablesOnMaster(master.getConfiguration())) {<a name="line.805"></a> +<span class="sourceLineNo">806</span> if (LoadBalancer.isSystemTablesOnlyOnMaster(master.getConfiguration())) {<a name="line.806"></a> +<span class="sourceLineNo">807</span> // If Master is carrying regions but NOT user-space regions, it<a name="line.807"></a> +<span class="sourceLineNo">808</span> // still shows as a 'server'. We need at least one more server to check<a name="line.808"></a> +<span class="sourceLineNo">809</span> // in before we can start up so set defaultMinToStart to 2.<a name="line.809"></a> +<span class="sourceLineNo">810</span> requiredMinToStart = requiredMinToStart + 1;<a name="line.810"></a> +<span class="sourceLineNo">811</span> }<a name="line.811"></a> +<span class="sourceLineNo">812</span> }<a name="line.812"></a> +<span class="sourceLineNo">813</span> int minToStart = this.master.getConfiguration().getInt(WAIT_ON_REGIONSERVERS_MINTOSTART, -1);<a name="line.813"></a> +<span class="sourceLineNo">814</span> // Ensure we are never less than requiredMinToStart else stuff won't work.<a name="line.814"></a> +<span class="sourceLineNo">815</span> return minToStart == -1 || minToStart < requiredMinToStart? requiredMinToStart: minToStart;<a name="line.815"></a> +<span class="sourceLineNo">816</span> }<a name="line.816"></a> +<span class="sourceLineNo">817</span><a name="line.817"></a> +<span class="sourceLineNo">818</span> /**<a name="line.818"></a> +<span class="sourceLineNo">819</span> * Wait for the region servers to report in.<a name="line.819"></a> +<span class="sourceLineNo">820</span> * We will wait until one of this condition is met:<a name="line.820"></a> +<span class="sourceLineNo">821</span> * - the master is stopped<a name="line.821"></a> +<span class="sourceLineNo">822</span> * - the 'hbase.master.wait.on.regionservers.maxtostart' number of<a name="line.822"></a> +<span class="sourceLineNo">823</span> * region servers is reached<a name="line.823"></a> +<span class="sourceLineNo">824</span> * - the 'hbase.master.wait.on.regionservers.mintostart' is reached AND<a name="line.824"></a> +<span class="sourceLineNo">825</span> * there have been no new region server in for<a name="line.825"></a> +<span class="sourceLineNo">826</span> * 'hbase.master.wait.on.regionservers.interval' time AND<a name="line.826"></a> +<span class="sourceLineNo">827</span> * the 'hbase.master.wait.on.regionservers.timeout' is reached<a name="line.827"></a> +<span class="sourceLineNo">828</span> *<a name="line.828"></a> +<span class="sourceLineNo">829</span> * @throws InterruptedException<a name="line.829"></a> +<span class="sourceLineNo">830</span> */<a name="line.830"></a> +<span class="sourceLineNo">831</span> public void waitForRegionServers(MonitoredTask status) throws InterruptedException {<a name="line.831"></a> +<span class="sourceLineNo">832</span> final long interval = this.master.getConfiguration().<a name="line.832"></a> +<span class="sourceLineNo">833</span> getLong(WAIT_ON_REGIONSERVERS_INTERVAL, 1500);<a name="line.833"></a> +<span class="sourceLineNo">834</span> final long timeout = this.master.getConfiguration().<a name="line.834"></a> +<span class="sourceLineNo">835</span> getLong(WAIT_ON_REGIONSERVERS_TIMEOUT, 4500);<a name="line.835"></a> +<span class="sourceLineNo">836</span> // Min is not an absolute; just a friction making us wait longer on server checkin.<a name="line.836"></a> +<span class="sourceLineNo">837</span> int minToStart = getMinToStart();<a name="line.837"></a> +<span class="sourceLineNo">838</span> int maxToStart = this.master.getConfiguration().<a name="line.838"></a> +<span class="sourceLineNo">839</span> getInt(WAIT_ON_REGIONSERVERS_MAXTOSTART, Integer.MAX_VALUE);<a name="line.839"></a> +<span class="sourceLineNo">840</span> if (maxToStart < minToStart) {<a name="line.840"></a> +<span class="sourceLineNo">841</span> LOG.warn(String.format("The value of '%s' (%d) is set less than '%s' (%d), ignoring.",<a name="line.841"></a> +<span class="sourceLineNo">842</span> WAIT_ON_REGIONSERVERS_MAXTOSTART, maxToStart,<a name="line.842"></a> +<span class="sourceLineNo">843</span> WAIT_ON_REGIONSERVERS_MINTOSTART, minToStart));<a name="line.843"></a> +<span class="sourceLineNo">844</span> maxToStart = Integer.MAX_VALUE;<a name="line.844"></a> +<span class="sourceLineNo">845</span> }<a name="line.845"></a> +<span class="sourceLineNo">846</span><a name="line.846"></a> +<span class="sourceLineNo">847</span> long now = System.currentTimeMillis();<a name="line.847"></a> +<span class="sourceLineNo">848</span> final long startTime = now;<a name="line.848"></a> +<span class="sourceLineNo">849</span> long slept = 0;<a name="line.849"></a> +<span class="sourceLineNo">850</span> long lastLogTime = 0;<a name="line.850"></a> +<span class="sourceLineNo">851</span> long lastCountChange = startTime;<a name="line.851"></a> +<span class="sourceLineNo">852</span> int count = countOfRegionServers();<a name="line.852"></a> +<span class="sourceLineNo">853</span> int oldCount = 0;<a name="line.853"></a> +<span class="sourceLineNo">854</span> // This while test is a little hard to read. We try to comment it in below but in essence:<a name="line.854"></a> +<span class="sourceLineNo">855</span> // Wait if Master is not stopped and the number of regionservers that have checked-in is<a name="line.855"></a> +<span class="sourceLineNo">856</span> // less than the maxToStart. Both of these conditions will be true near universally.<a name="line.856"></a> +<span class="sourceLineNo">857</span> // Next, we will keep cycling if ANY of the following three conditions are true:<a name="line.857"></a> +<span class="sourceLineNo">858</span> // 1. The time since a regionserver registered is < interval (means servers are actively checking in).<a name="line.858"></a> +<span class="sourceLineNo">859</span> // 2. We are under the total timeout.<a name="line.859"></a> +<span class="sourceLineNo">860</span> // 3. The count of servers is < minimum.<a name="line.860"></a> +<span class="sourceLineNo">861</span> for (ServerListener listener: this.listeners) {<a name="line.861"></a> +<span class="sourceLineNo">862</span> listener.waiting();<a name="line.862"></a> +<span class="sourceLineNo">863</span> }<a name="line.863"></a> +<span class="sourceLineNo">864</span> while (!this.master.isStopped() && !isClusterShutdown() && count < maxToStart &&<a name="line.864"></a> +<span class="sourceLineNo">865</span> ((lastCountChange + interval) > now || timeout > slept || count < minToStart)) {<a name="line.865"></a> +<span class="sourceLineNo">866</span> // Log some info at every interval time or if there is a change<a name="line.866"></a> +<span class="sourceLineNo">867</span> if (oldCount != count || lastLogTime + interval < now) {<a name="line.867"></a> +<span class="sourceLineNo">868</span> lastLogTime = now;<a name="line.868"></a> +<span class="sourceLineNo">869</span> String msg =<a name="line.869"></a> +<span class="sourceLineNo">870</span> "Waiting on regionserver count=" + count + "; waited="+<a name="line.870"></a> +<span class="sourceLineNo">871</span> slept + "ms, expecting min=" + minToStart + " server(s), max="+ getStrForMax(maxToStart) +<a name="line.871"></a> +<span class="sourceLineNo">872</span> " server(s), " + "timeout=" + timeout + "ms, lastChange=" + (lastCountChange - now) + "ms";<a name="line.872"></a> +<span class="sourceLineNo">873</span> LOG.info(msg);<a name="line.873"></a> +<span class="sourceLineNo">874</span> status.setStatus(msg);<a name="line.874"></a> +<span class="sourceLineNo">875</span> }<a name="line.875"></a> <span class="sourceLineNo">876</span><a name="line.876"></a> -<span class="sourceLineNo">877</span> oldCount = count;<a name="line.877"></a> -<span class="sourceLineNo">878</span> count = countOfRegionServers();<a name="line.878"></a> -<span class="sourceLineNo">879</span> if (count != oldCount) {<a name="line.879"></a> -<span class="sourceLineNo">880</span> lastCountChange = now;<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> // Did we exit the loop because cluster is going down?<a name="line.883"></a> -<span class="sourceLineNo">884</span> if (isClusterShutdown()) {<a name="line.884"></a> -<span class="sourceLineNo">885</span> this.master.stop("Cluster shutdown");<a name="line.885"></a> -<span class="sourceLineNo">886</span> }<a name="line.886"></a> -<span class="sourceLineNo">887</span> LOG.info("Finished waiting on RegionServer count=" + count + "; waited=" + slept + "ms," +<a name="line.887"></a> -<span class="sourceLineNo">888</span> " expected min=" + minToStart + " server(s), max=" + getStrForMax(maxToStart) + " server(s),"+<a name="line.888"></a> -<span class="sourceLineNo">889</span> " master is "+ (this.master.isStopped() ? "stopped.": "running"));<a name="line.889"></a> -<span class="sourceLineNo">890</span> }<a name="line.890"></a> -<span class="sourceLineNo">891</span><a name="line.891"></a> -<span class="sourceLineNo">892</span> private String getStrForMax(final int max) {<a name="line.892"></a> -<span class="sourceLineNo">893</span> return max == Integer.MAX_VALUE? "NO_LIMIT": Integer.toString(max);<a name="line.893"></a> -<span class="sourceLineNo">894</span> }<a name="line.894"></a> -<span class="sourceLineNo">895</span><a name="line.895"></a> -<span class="sourceLineNo">896</span> /**<a name="line.896"></a> -<span class="sourceLineNo">897</span> * @return A copy of the internal list of online servers.<a name="line.897"></a> -<span class="sourceLineNo">898</span> */<a name="line.898"></a> -<span class="sourceLineNo">899</span> public List<ServerName> getOnlineServersList() {<a name="line.899"></a> -<span class="sourceLineNo">900</span> // TODO: optimize the load balancer call so we don't need to make a new list<a name="line.900"></a> -<span class="sourceLineNo">901</span> // TODO: FIX. THIS IS POPULAR CALL.<a name="line.901"></a> -<span class="sourceLineNo">902</span> return new ArrayList<>(this.onlineServers.keySet());<a name="line.902"></a> -<span class="sourceLineNo">903</span> }<a name="line.903"></a> -<span class="sourceLineNo">904</span><a name="line.904"></a> -<span class="sourceLineNo">905</span> /**<a name="line.905"></a> -<span class="sourceLineNo">906</span> * @param keys The target server name<a name="line.906"></a> -<span class="sourceLineNo">907</span> * @param idleServerPredicator Evaluates the server on the given load<a name="line.907"></a> -<span class="sourceLineNo">908</span> * @return A copy of the internal list of online servers matched by the predicator<a name="line.908"></a> -<span class="sourceLineNo">909</span> */<a name="line.909"></a> -<span class="sourceLineNo">910</span> public List<ServerName> getOnlineServersListWithPredicator(List<ServerName> keys,<a name="line.910"></a> -<span class="sourceLineNo">911</span> Predicate<ServerMetrics> idleServerPredicator) {<a name="line.911"></a> -<span class="sourceLineNo">912</span> List<ServerName> names = new ArrayList<>();<a name="line.912"></a> -<span class="sourceLineNo">913</span> if (keys != null && idleServerPredicator != null) {<a name="line.913"></a> -<span class="sourceLineNo">914</span> keys.forEach(name -> {<a name="line.914"></a> -<span class="sourceLineNo">915</span> ServerMetrics load = onlineServers.get(name);<a name="line.915"></a> -<span class="sourceLineNo">916</span> if (load != null) {<a name="line.916"></a> -<span class="sourceLineNo">917</span> if (idleServerPredicator.test(load)) {<a name="line.917"></a> -<span class="sourceLineNo">918</span> names.add(name);<a name="line.918"></a> -<span class="sourceLineNo">919</span> }<a name="line.919"></a> -<span class="sourceLineNo">920</span> }<a name="line.920"></a> -<span class="sourceLineNo">921</span> });<a name="line.921"></a> -<span class="sourceLineNo">922</span> }<a name="line.922"></a> -<span class="sourceLineNo">923</span> return names;<a name="line.923"></a> -<span class="sourceLineNo">924</span> }<a name="line.924"></a> -<span class="sourceLineNo">925</span><a name="line.925"></a> -<span class="sourceLineNo">926</span> /**<a name="line.926"></a> -<span class="sourceLineNo">927</span> * @return A copy of the internal list of draining servers.<a name="line.927"></a> -<span class="sourceLineNo">928</span> */<a name="line.928"></a> -<span class="sourceLineNo">929</span> public List<ServerName> getDrainingServersList() {<a name="line.929"></a> -<span class="sourceLineNo">930</span> return new ArrayList<>(this.drainingServers);<a name="line.930"></a> -<span class="sourceLineNo">931</span> }<a name="line.931"></a> -<span class="sourceLineNo">932</span><a name="line.932"></a> -<span class="sourceLineNo">933</span> /**<a name="line.933"></a> -<span class="sourceLineNo">934</span> * @return A copy of the internal set of deadNotExpired servers.<a name="line.934"></a> -<span class="sourceLineNo">935</span> */<a name="line.935"></a> -<span class="sourceLineNo">936</span> Set<ServerName> getDeadNotExpiredServers() {<a name="line.936"></a> -<span class="sourceLineNo">937</span> return new HashSet<>(this.queuedDeadServers);<a name="line.937"></a> -<span class="sourceLineNo">938</span> }<a name="line.938"></a> -<span class="sourceLineNo">939</span><a name="line.939"></a> -<span class="sourceLineNo">940</span> public boolean isServerOnline(ServerName serverName) {<a name="line.940"></a> -<span class="sourceLineNo">941</span> return serverName != null && onlineServers.containsKey(serverName);<a name="line.941"></a> -<span class="sourceLineNo">942</span> }<a name="line.942"></a> -<span class="sourceLineNo">943</span><a name="line.943"></a> -<span class="sourceLineNo">944</span> /**<a name="line.944"></a> -<span class="sourceLineNo">945</span> * Check if a server is known to be dead. A server can be online,<a name="line.945"></a> -<span class="sourceLineNo">946</span> * or known to be dead, or unknown to this manager (i.e, not online,<a name="line.946"></a> -<span class="sourceLineNo">947</span> * not known to be dead either. it is simply not tracked by the<a name="line.947"></a> -<span class="sourceLineNo">948</span> * master any more, for example, a very old previous instance).<a name="line.948"></a> -<span class="sourceLineNo">949</span> */<a name="line.949"></a> -<span class="sourceLineNo">950</span> public synchronized boolean isServerDead(ServerName serverName) {<a name="line.950"></a> -<span class="sourceLineNo">951</span> return serverName == null || deadservers.isDeadServer(serverName)<a name="line.951"></a> -<span class="sourceLineNo">952</span> || queuedDeadServers.contains(serverName)<a name="line.952"></a> -<span class="sourceLineNo">953</span> || requeuedDeadServers.containsKey(serverName);<a name="line.953"></a> -<span class="sourceLineNo">954</span> }<a name="line.954"></a> -<span class="sourceLineNo">955</span><a name="line.955"></a> -<span class="sourceLineNo">956</span> public void shutdownCluster() {<a name="line.956"></a> -<span class="sourceLineNo">957</span> String statusStr = "Cluster shutdown requested of master=" + this.master.getServerName();<a name="line.957"></a> -<span class="sourceLineNo">958</span> LOG.info(statusStr);<a name="line.958"></a> -<span class="sourceLineNo">959</span> this.clusterShutdown.set(true);<a name="line.959"></a> -<span class="sourceLineNo">960</span> if (onlineServers.isEmpty()) {<a name="line.960"></a> -<span class="sourceLineNo">961</span> // we do not synchronize here so this may cause a double stop, but not a big deal<a name="line.961"></a> -<span class="sourceLineNo">962</span> master.stop("OnlineServer=0 right after cluster shutdown set");<a name="line.962"></a> -<span class="sourceLineNo">963</span> }<a name="line.963"></a> -<span class="sourceLineNo">964</span> }<a name="line.964"></a> -<span class="sourceLineNo">965</span><a name="line.965"></a> -<span class="sourceLineNo">966</span> boolean isClusterShutdown() {<a name="line.966"></a> -<span class="sourceLineNo">967</span> return this.clusterShutdown.get();<a name="line.967"></a> -<span class="sourceLineNo">968</span> }<a name="line.968"></a> -<span class="sourceLineNo">969</span><a name="line.969"></a> -<span class="sourceLineNo">970</span> /**<a name="line.970"></a> -<span class="sourceLineNo">971</span> * Stop the ServerManager.<a name="line.971"></a> -<span class="sourceLineNo">972</span> */<a name="line.972"></a> -<span class="sourceLineNo">973</span> public void stop() {<a name="line.973"></a> -<span class="sourceLineNo">974</span> // Nothing to do.<a name="line.974"></a> -<span class="sourceLineNo">975</span> }<a name="line.975"></a> -<span class="sourceLineNo">976</span><a name="line.976"></a> -<span class="sourceLineNo">977</span> /**<a name="line.977"></a> -<span class="sourceLineNo">978</span> * Creates a list of possible destinations for a region. It contains the online servers, but not<a name="line.978"></a> -<span class="sourceLineNo">979</span> * the draining or dying servers.<a name="line.979"></a> -<span class="sourceLineNo">980</span> * @param serversToExclude can be null if there is no server to exclude<a name="line.980"></a> -<span class="sourceLineNo">981</span> */<a name="line.981"></a> -<span class="sourceLineNo">982</span> public List<ServerName> createDestinationServersList(final List<ServerName> serversToExclude){<a name="line.982"></a> -<span class="sourceLineNo">983</span> final List<ServerName> destServers = getOnlineServersList();<a name="line.983"></a> -<span class="sourceLineNo">984</span><a name="line.984"></a> -<span class="sourceLineNo">985</span> if (serversToExclude != null) {<a name="line.985"></a> -<span class="sourceLineNo">986</span> destServers.removeAll(serversToExclude);<a name="line.986"></a> -<span class="sourceLineNo">987</span> }<a name="line.987"></a> -<span class="sourceLineNo">988</span><a name="line.988"></a> -<span class="sourceLineNo">989</span> // Loop through the draining server list and remove them from the server list<a name="line.989"></a> -<span class="sourceLineNo">990</span> final List<ServerName> drainingServersCopy = getDrainingServersList();<a name="line.990"></a> -<span class="sourceLineNo">991</span> destServers.removeAll(drainingServersCopy);<a name="line.991"></a> -<span class="sourceLineNo">992</span><a name="line.992"></a> -<span class="sourceLineNo">993</span> // Remove the deadNotExpired servers from the server list.<a name="line.993"></a> -<span class="sourceLineNo">994</span> removeDeadNotExpiredServers(destServers);<a name="line.994"></a> -<span class="sourceLineNo">995</span> return destServers;<a name="line.995"></a> -<span class="sourceLineNo">996</span> }<a name="line.996"></a> -<span class="sourceLineNo">997</span><a name="line.997"></a> -<span class="sourceLineNo">998</span> /**<a name="line.998"></a> -<span class="sourceLineNo">999</span> * Calls {@link #createDestinationServersList} without server to exclude.<a name="line.999"></a> -<span class="sourceLineNo">1000</span> */<a name="line.1000"></a> -<span class="sourceLineNo">1001</span> public List<ServerName> createDestinationServersList(){<a name="line.1001"></a> -<span class="sourceLineNo">1002</span> return createDestinationServersList(null);<a name="line.1002"></a> -<span class="sourceLineNo">1003</span> }<a name="line.1003"></a> -<span class="sourceLineNo">1004</span><a name="line.1004"></a> -<span class="sourceLineNo">1005</span> /**<a name="line.1005"></a> -<span class="sourceLineNo">1006</span> * Loop through the deadNotExpired server list and remove them from the<a name="line.1006"></a> -<span class="sourceLineNo">1007</span> * servers.<a name="line.1007"></a> -<span class="sourceLineNo">1008</span> * This function should be used carefully outside of this class. You should use a high level<a name="line.1008"></a> -<span class="sourceLineNo">1009</span> * method such as {@link #createDestinationServersList()} instead of managing you own list.<a name="line.1009"></a> -<span class="sourceLineNo">1010</span> */<a name="line.1010"></a> -<span class="sourceLineNo">1011</span> void removeDeadNotExpiredServers(List<ServerName> servers) {<a name="line.1011"></a> -<span class="sourceLineNo">1012</span> Set<ServerName> deadNotExpiredServersCopy = this.getDeadNotExpiredServers();<a name="line.1012"></a> -<span class="sourceLineNo">1013</span> if (!deadNotExpiredServersCopy.isEmpty()) {<a name="line.1013"></a> -<span class="sourceLineNo">1014</span> for (ServerName server : deadNotExpiredServersCopy) {<a name="line.1014"></a> -<span class="sourceLineNo">1015</span> LOG.debug("Removing dead but not expired server: " + server<a name="line.1015"></a> -<span class="sourceLineNo">1016</span> + " from eligible server pool.");<a name="line.1016"></a> -<span class="sourceLineNo">1017</span> servers.remove(server);<a name="line.1017"></a> -<span class="sourceLineNo">1018</span> }<a name="line.1018"></a> -<span class="sourceLineNo">1019</span> }<a name="line.1019"></a> -<span class="sourceLineNo">1020</span> }<a name="line.1020"></a> -<span class="sourceLineNo">1021</span><a name="line.1021"></a> -<span class="sourceLineNo">1022</span> /**<a name="line.1022"></a> -<span class="sourceLineNo">1023</span> * To clear any dead server with same host name and port of any online server<a name="line.1023"></a> -<span class="sourceLineNo">1024</span> */<a name="line.1024"></a> -<span class="sourceLineNo">1025</span> void clearDeadServersWithSameHostNameAndPortOfOnlineServer() {<a name="line.1025"></a> -<span class="sourceLineNo">1026</span> for (ServerName serverName : getOnlineServersList()) {<a name="line.1026"></a> -<span class="sourceLineNo">1027</span> deadservers.cleanAllPreviousInstances(serverName);<a name="line.1027"></a> -<span class="sourceLineNo">1028</span> }<a name="line.1028"></a> -<span class="sourceLineNo">1029</span> }<a name="line.1029"></a> -<span class="sourceLineNo">1030</span><a name="line.1030"></a> -<span class="sourceLineNo">1031</span> /**<a name="line.1031"></a> -<span class="sourceLineNo">1032</span> * Called by delete table and similar to notify the ServerManager that a region was removed.<a name="line.1032"></a> -<span class="sourceLineNo">1033</span> */<a name="line.1033"></a> -<span class="sourceLineNo">1034</span> public void removeRegion(final RegionInfo regionInfo) {<a name="line.1034"></a> -<span class="sourceLineNo">1035</span> final byte[] encodedName = regionInfo.getEncodedNameAsBytes();<a name="line.1035"></a> -<span class="sourceLineNo">1036</span> storeFlushedSequenceIdsByRegion.remove(encodedName);<a name="line.1036"></a> -<span class="sourceLineNo">1037</span> flushedSequenceIdByRegion.remove(encodedName);<a name="line.1037"></a> -<span class="sourceLineNo">1038</span> }<a name="line.1038"></a> -<span class="sourceLineNo">1039</span><a name="line.1039"></a> -<span class="sourceLineNo">1040</span> @VisibleForTesting<a name="line.1040"></a> -<span class="sourceLineNo">1041</span> public boolean isRegionInServerManagerStates(final RegionInfo hri) {<a name="line.1041"></a> -<span class="sourceLineNo">1042</span> final byte[] encodedName = hri.getEncodedNameAsBytes();<a name="line.1042"></a> -<span class="sourceLineNo">1043</span> return (storeFlushedSequenceIdsByRegion.containsKey(encodedName)<a name="line.1043"></a> -<span class="sourceLineNo">1044</span> || flushedSequenceIdByRegion.containsKey(encodedName));<a name="line.1044"></a> -<span class="sourceLineNo">1045</span> }<a name="line.1045"></a> -<span class="sourceLineNo">1046</span><a name="line.1046"></a> -<span class="sourceLineNo">1047</span> /**<a name="line.1047"></a> -<span class="sourceLineNo">1048</span> * Called by delete table and similar to notify the ServerManager that a region was removed.<a name="line.1048"></a> -<span class="sourceLineNo">1049</span> */<a name="line.1049"></a> -<span class="sourceLineNo">1050</span> public void removeRegions(final List<RegionInfo> regions) {<a name="line.1050"></a> -<span class="sourceLineNo">1051</span> for (RegionInfo hri: regions) {<a name="line.1051"></a> -<span class="sourceLineNo">1052</span> removeRegion(hri);<a name="line.1052"></a> -<span class="sourceLineNo">1053</span> }<a name="line.1053"></a> -<span class="sourceLineNo">1054</span> }<a name="line.1054"></a> -<span class="sourceLineNo">1055</span>}<a name="line.1055"></a> +<span class="sourceLineNo">877</span> // We sleep for some time<a name="line.877"></a> +<span class="sourceLineNo">878</span> final long sleepTime = 50;<a name="line.878"></a> +<span class="sourceLineNo">879</span> Thread.sleep(sleepTime);<a name="line.879"></a> +<span class="sourceLineNo">880</span> now = System.currentTimeMillis();<a name="line.880"></a> +<span class="sourceLineNo">881</span> slept = now - startTime;<a name="line.881"></a> +<span class="sourceLineNo">882</span><a name="line.882"></a> +<span class="sourceLineNo">883</span> oldCount = count;<a name="line.883"></a> +<span class="sourceLineNo">884</span> count = countOfRegionServers();<a name="line.884"></a> +<span class="sourceLineNo">885</span> if (count != oldCount) {<a name="line.885"></a> +<span class="sourceLineNo">886</span> lastCountChange = now;<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> // Did we exit the loop because cluster is going down?<a name="line.889"></a> +<span class="sourceLineNo">890</span> if (isClusterShutdown()) {<a name="line.890"></a> +<span class="sourceLineNo">891</span> this.master.stop("Cluster shutdown");<a name="line.891"></a> +<span class="sourceLineNo">892</span> }<a name="line.892"></a> +<span class="sourceLineNo">893</span> LOG.info("Finished waiting on RegionServer count=" + count + "; waited=" + slept + "ms," +<a name="line.893"></a> +<span class="sourceLineNo">894</span> " expected min=" + minToStart + " server(s), max=" + getStrForMax(maxToStart) + " server(s),"+<a name="line.894"></a> +<span class="sourceLineNo">895</span> " master is "+ (this.master.isStopped() ? "stopped.": "running"));<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">898</span> private String getStrForMax(final int max) {<a name="line.898"></a> +<span class="sourceLineNo">899</span> return max == Integer.MAX_VALUE? "NO_LIMIT": Integer.toString(max);<a name="line.899"></a> +<span class="sourceLineNo">900</span> }<a name="line.900"></a> +<span class="sourceLineNo">901</span><a name="line.901"></a> +<span class="sourceLineNo">902</span> /**<a name="line.902"></a> +<span class="sourceLineNo">903</span> * @return A copy of the internal list of online servers.<a name="line.903"></a> +<span class="sourceLineNo">904</span> */<a name="line.904"></a> +<span class="sourceLineNo">905</span> public List<ServerName> getOnlineServersList() {<a name="line.905"></a> +<span class="sourceLineNo">906</span> // TODO: optimize the load balancer call so we don't need to make a new list<a name="line.906"></a> +<span class="sourceLineNo">907</span> // TODO: FIX. THIS IS POPULAR CALL.<a name="line.907"></a> +<span class="sourceLineNo">908</span> return new ArrayList<>(this.onlineServers.keySet());<a name="line.908"></a> +<span class="sourceLineNo">909</span> }<a name="line.909"></a> +<span class="sourceLineNo">910</span><a name="line.910"></a> +<span class="sourceLineNo">911</span> /**<a name="line.911"></a> +<span class="sourceLineNo">912</span> * @param keys The target server name<a name="line.912"></a> +<span class="sourceLineNo">913</span> * @param idleServerPredicator Evaluates the server on the given load<a name="line.913"></a> +<span class="sourceLineNo">914</span> * @return A copy of the internal list of online servers matched by the predicator<a name="line.914"></a> +<span class="sourceLineNo">915</span> */<a name="line.915"></a> +<span class="sourceLineNo">916</span> public List<ServerName> getOnlineServersListWithPredicator(List<ServerName> keys,<a name="line.916"></a> +<span class="sourceLineNo">917</span> Predicate<ServerMetrics> idleServerPredicator) {<a name="line.917"></a> +<span class="sourceLineNo">918</span> List<ServerName> names = new ArrayList<>();<a name="line.918"></a> +<span class="sourceLineNo">919</span> if (keys != null && idleServerPredicator != null) {<a name="line.919"></a> +<span class="sourceLineNo">920</span> keys.forEach(name -> {<a name="line.920"></a> +<span class="sourceLineNo">921</span> ServerMetrics load = onlineServers.get(name);<a name="line.921"></a> +<span class="sourceLineNo">922</span> if (load != null) {<a name="line.922"></a> +<span class="sourceLineNo">923</span> if (idleServerPredicator.test(load)) {<a name="line.923"></a> +<span class="sourceLineNo">924</span> names.add(name);<a name="line.924"></a> +<span class="sourceLineNo">925</span> }<a name="line.925"></a> +<span class="sourceLineNo">926</span> }<a name="line.926"></a> +<span class="sourceLineNo">927</span> });<a name="line.927"></a> +<span class="sourceLineNo">928</span> }<a name="line.928"></a> +<span class="sourceLineNo">929</span> return names;<a name="line.929"></a> +<span class="sourceLineNo">930</span> }<a name="line.930"></a> +<span class="sourceLineNo">931</span><a name="line.931"></a> +<span class="sourceLineNo">932</span> /**<a name="line.932"></a> +<span class="sourceLineNo">933</span> * @return A copy of the internal list of draining servers.<a name="line.933"></a> +<span class="sourceLineNo">934</span> */<a name="line.934"></a> +<span class="sourceLineNo">935</span> public List<ServerName> getDrainingServersList() {<a name="line.935"></a> +<span class="sourceLineNo">936</span> return new ArrayList<>(this.drainingServers);<a name="line.936"></a> +<span class="sourceLineNo">937</span> }<a name="line.937"></a> +<span class="sourceLineNo">938</span><a name="line.938"></a> +<span class="sourceLineNo">939</span> /**<a name="line.939"></a> +<span class="sourceLineNo">940</span> * @return A copy of the internal set of deadNotExpired servers.<a name="line.940"></a> +<span c
<TRUNCATED>
