http://git-wip-us.apache.org/repos/asf/hbase-site/blob/4f0b7674/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/AssignmentManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/AssignmentManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/AssignmentManager.html index 7927e94..7356e7a 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/AssignmentManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/AssignmentManager.html @@ -921,7 +921,7 @@ <span class="sourceLineNo">913</span> master.getMasterProcedureExecutor().submitProcedure(createSplitProcedure(parent, splitKey));<a name="line.913"></a> <span class="sourceLineNo">914</span><a name="line.914"></a> <span class="sourceLineNo">915</span> // If the RS is < 2.0 throw an exception to abort the operation, we are handling the split<a name="line.915"></a> -<span class="sourceLineNo">916</span> if (regionStates.getOrCreateServer(serverName).getVersionNumber() < 0x0200000) {<a name="line.916"></a> +<span class="sourceLineNo">916</span> if (master.getServerManager().getServerVersion(serverName) < 0x0200000) {<a name="line.916"></a> <span class="sourceLineNo">917</span> throw new UnsupportedOperationException(String.format(<a name="line.917"></a> <span class="sourceLineNo">918</span> "Split handled by the master: parent=%s hriA=%s hriB=%s", parent.getShortNameToLog(), hriA, hriB));<a name="line.918"></a> <span class="sourceLineNo">919</span> }<a name="line.919"></a> @@ -944,7 +944,7 @@ <span class="sourceLineNo">936</span> master.getMasterProcedureExecutor().submitProcedure(createMergeProcedure(hriA, hriB));<a name="line.936"></a> <span class="sourceLineNo">937</span><a name="line.937"></a> <span class="sourceLineNo">938</span> // If the RS is < 2.0 throw an exception to abort the operation, we are handling the merge<a name="line.938"></a> -<span class="sourceLineNo">939</span> if (regionStates.getOrCreateServer(serverName).getVersionNumber() < 0x0200000) {<a name="line.939"></a> +<span class="sourceLineNo">939</span> if (master.getServerManager().getServerVersion(serverName) < 0x0200000) {<a name="line.939"></a> <span class="sourceLineNo">940</span> throw new UnsupportedOperationException(String.format(<a name="line.940"></a> <span class="sourceLineNo">941</span> "Merge not handled yet: regionState=%s merged=%s hriA=%s hriB=%s", state, merged, hriA,<a name="line.941"></a> <span class="sourceLineNo">942</span> hriB));<a name="line.942"></a> @@ -956,13 +956,13 @@ <span class="sourceLineNo">948</span> // ============================================================================================<a name="line.948"></a> <span class="sourceLineNo">949</span> /**<a name="line.949"></a> <span class="sourceLineNo">950</span> * the master will call this method when the RS send the regionServerReport().<a name="line.950"></a> -<span class="sourceLineNo">951</span> * the report will contains the "hbase version" and the "online regions".<a name="line.951"></a> +<span class="sourceLineNo">951</span> * the report will contains the "online regions".<a name="line.951"></a> <span class="sourceLineNo">952</span> * this method will check the the online regions against the in-memory state of the AM,<a name="line.952"></a> <span class="sourceLineNo">953</span> * if there is a mismatch we will try to fence out the RS with the assumption<a name="line.953"></a> <span class="sourceLineNo">954</span> * that something went wrong on the RS side.<a name="line.954"></a> <span class="sourceLineNo">955</span> */<a name="line.955"></a> -<span class="sourceLineNo">956</span> public void reportOnlineRegions(final ServerName serverName,<a name="line.956"></a> -<span class="sourceLineNo">957</span> final int versionNumber, final Set<byte[]> regionNames) throws YouAreDeadException {<a name="line.957"></a> +<span class="sourceLineNo">956</span> public void reportOnlineRegions(final ServerName serverName, final Set<byte[]> regionNames)<a name="line.956"></a> +<span class="sourceLineNo">957</span> throws YouAreDeadException {<a name="line.957"></a> <span class="sourceLineNo">958</span> if (!isRunning()) return;<a name="line.958"></a> <span class="sourceLineNo">959</span> if (LOG.isTraceEnabled()) {<a name="line.959"></a> <span class="sourceLineNo">960</span> LOG.trace("ReportOnlineRegions " + serverName + " regionCount=" + regionNames.size() +<a name="line.960"></a> @@ -973,1021 +973,1014 @@ <span class="sourceLineNo">965</span><a name="line.965"></a> <span class="sourceLineNo">966</span> final ServerStateNode serverNode = regionStates.getOrCreateServer(serverName);<a name="line.966"></a> <span class="sourceLineNo">967</span><a name="line.967"></a> -<span class="sourceLineNo">968</span> // update the server version number. This will be used for live upgrades.<a name="line.968"></a> -<span class="sourceLineNo">969</span> synchronized (serverNode) {<a name="line.969"></a> -<span class="sourceLineNo">970</span> serverNode.setVersionNumber(versionNumber);<a name="line.970"></a> -<span class="sourceLineNo">971</span> if (serverNode.isInState(ServerState.SPLITTING, ServerState.OFFLINE)) {<a name="line.971"></a> -<span class="sourceLineNo">972</span> LOG.warn("Got a report from a server result in state " + serverNode.getState());<a name="line.972"></a> -<span class="sourceLineNo">973</span> return;<a name="line.973"></a> -<span class="sourceLineNo">974</span> }<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> if (regionNames.isEmpty()) {<a name="line.977"></a> -<span class="sourceLineNo">978</span> // nothing to do if we don't have regions<a name="line.978"></a> -<span class="sourceLineNo">979</span> LOG.trace("no online region found on " + serverName);<a name="line.979"></a> -<span class="sourceLineNo">980</span> } else if (!isMetaLoaded()) {<a name="line.980"></a> -<span class="sourceLineNo">981</span> // if we are still on startup, discard the report unless is from someone holding meta<a name="line.981"></a> -<span class="sourceLineNo">982</span> checkOnlineRegionsReportForMeta(serverNode, regionNames);<a name="line.982"></a> -<span class="sourceLineNo">983</span> } else {<a name="line.983"></a> -<span class="sourceLineNo">984</span> // The Heartbeat updates us of what regions are only. check and verify the state.<a name="line.984"></a> -<span class="sourceLineNo">985</span> checkOnlineRegionsReport(serverNode, regionNames);<a name="line.985"></a> -<span class="sourceLineNo">986</span> }<a name="line.986"></a> -<span class="sourceLineNo">987</span><a name="line.987"></a> -<span class="sourceLineNo">988</span> // wake report event<a name="line.988"></a> -<span class="sourceLineNo">989</span> wakeServerReportEvent(serverNode);<a name="line.989"></a> -<span class="sourceLineNo">990</span> }<a name="line.990"></a> -<span class="sourceLineNo">991</span><a name="line.991"></a> -<span class="sourceLineNo">992</span> void checkOnlineRegionsReportForMeta(final ServerStateNode serverNode,<a name="line.992"></a> -<span class="sourceLineNo">993</span> final Set<byte[]> regionNames) {<a name="line.993"></a> -<span class="sourceLineNo">994</span> try {<a name="line.994"></a> -<span class="sourceLineNo">995</span> for (byte[] regionName: regionNames) {<a name="line.995"></a> -<span class="sourceLineNo">996</span> final RegionInfo hri = getMetaRegionFromName(regionName);<a name="line.996"></a> -<span class="sourceLineNo">997</span> if (hri == null) {<a name="line.997"></a> -<span class="sourceLineNo">998</span> if (LOG.isTraceEnabled()) {<a name="line.998"></a> -<span class="sourceLineNo">999</span> LOG.trace("Skip online report for region=" + Bytes.toStringBinary(regionName) +<a name="line.999"></a> -<span class="sourceLineNo">1000</span> " while meta is loading");<a name="line.1000"></a> -<span class="sourceLineNo">1001</span> }<a name="line.1001"></a> -<span class="sourceLineNo">1002</span> continue;<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> final RegionStateNode regionNode = regionStates.getOrCreateRegionStateNode(hri);<a name="line.1005"></a> -<span class="sourceLineNo">1006</span> LOG.info("META REPORTED: " + regionNode);<a name="line.1006"></a> -<span class="sourceLineNo">1007</span> if (!reportTransition(regionNode, serverNode, TransitionCode.OPENED, 0)) {<a name="line.1007"></a> -<span class="sourceLineNo">1008</span> LOG.warn("META REPORTED but no procedure found (complete?); set location=" +<a name="line.1008"></a> -<span class="sourceLineNo">1009</span> serverNode.getServerName());<a name="line.1009"></a> -<span class="sourceLineNo">1010</span> regionNode.setRegionLocation(serverNode.getServerName());<a name="line.1010"></a> -<span class="sourceLineNo">1011</span> } else if (LOG.isTraceEnabled()) {<a name="line.1011"></a> -<span class="sourceLineNo">1012</span> LOG.trace("META REPORTED: " + regionNode);<a name="line.1012"></a> -<span class="sourceLineNo">1013</span> }<a name="line.1013"></a> -<span class="sourceLineNo">1014</span> }<a name="line.1014"></a> -<span class="sourceLineNo">1015</span> } catch (UnexpectedStateException e) {<a name="line.1015"></a> -<span class="sourceLineNo">1016</span> final ServerName serverName = serverNode.getServerName();<a name="line.1016"></a> -<span class="sourceLineNo">1017</span> LOG.warn("KILLING " + serverName + ": " + e.getMessage());<a name="line.1017"></a> -<span class="sourceLineNo">1018</span> killRegionServer(serverNode);<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> void checkOnlineRegionsReport(final ServerStateNode serverNode, final Set<byte[]> regionNames)<a name="line.1022"></a> -<span class="sourceLineNo">1023</span> throws YouAreDeadException {<a name="line.1023"></a> -<span class="sourceLineNo">1024</span> final ServerName serverName = serverNode.getServerName();<a name="line.1024"></a> -<span class="sourceLineNo">1025</span> try {<a name="line.1025"></a> -<span class="sourceLineNo">1026</span> for (byte[] regionName: regionNames) {<a name="line.1026"></a> -<span class="sourceLineNo">1027</span> if (!isRunning()) return;<a name="line.1027"></a> -<span class="sourceLineNo">1028</span> final RegionStateNode regionNode = regionStates.getRegionStateNodeFromName(regionName);<a name="line.1028"></a> -<span class="sourceLineNo">1029</span> if (regionNode == null) {<a name="line.1029"></a> -<span class="sourceLineNo">1030</span> throw new UnexpectedStateException("Not online: " + Bytes.toStringBinary(regionName));<a name="line.1030"></a> -<span class="sourceLineNo">1031</span> }<a name="line.1031"></a> -<span class="sourceLineNo">1032</span> synchronized (regionNode) {<a name="line.1032"></a> -<span class="sourceLineNo">1033</span> if (regionNode.isInState(State.OPENING, State.OPEN)) {<a name="line.1033"></a> -<span class="sourceLineNo">1034</span> if (!regionNode.getRegionLocation().equals(serverName)) {<a name="line.1034"></a> -<span class="sourceLineNo">1035</span> throw new UnexpectedStateException(regionNode.toString() +<a name="line.1035"></a> -<span class="sourceLineNo">1036</span> "reported OPEN on server=" + serverName +<a name="line.1036"></a> -<span class="sourceLineNo">1037</span> " but state has otherwise.");<a name="line.1037"></a> -<span class="sourceLineNo">1038</span> } else if (regionNode.isInState(State.OPENING)) {<a name="line.1038"></a> -<span class="sourceLineNo">1039</span> try {<a name="line.1039"></a> -<span class="sourceLineNo">1040</span> if (!reportTransition(regionNode, serverNode, TransitionCode.OPENED, 0)) {<a name="line.1040"></a> -<span class="sourceLineNo">1041</span> LOG.warn(regionNode.toString() + " reported OPEN on server=" + serverName +<a name="line.1041"></a> -<span class="sourceLineNo">1042</span> " but state has otherwise AND NO procedure is running");<a name="line.1042"></a> -<span class="sourceLineNo">1043</span> }<a name="line.1043"></a> -<span class="sourceLineNo">1044</span> } catch (UnexpectedStateException e) {<a name="line.1044"></a> -<span class="sourceLineNo">1045</span> LOG.warn(regionNode.toString() + " reported unexpteced OPEN: " + e.getMessage(), e);<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> } else if (!regionNode.isInState(State.CLOSING, State.SPLITTING)) {<a name="line.1048"></a> -<span class="sourceLineNo">1049</span> long diff = regionNode.getLastUpdate() - EnvironmentEdgeManager.currentTime();<a name="line.1049"></a> -<span class="sourceLineNo">1050</span> if (diff > 1000/*One Second... make configurable if an issue*/) {<a name="line.1050"></a> -<span class="sourceLineNo">1051</span> // So, we can get report that a region is CLOSED or SPLIT because a heartbeat<a name="line.1051"></a> -<span class="sourceLineNo">1052</span> // came in at about same time as a region transition. Make sure there is some<a name="line.1052"></a> -<span class="sourceLineNo">1053</span> // elapsed time between killing remote server.<a name="line.1053"></a> -<span class="sourceLineNo">1054</span> throw new UnexpectedStateException(regionNode.toString() +<a name="line.1054"></a> -<span class="sourceLineNo">1055</span> " reported an unexpected OPEN; time since last update=" + diff);<a name="line.1055"></a> -<span class="sourceLineNo">1056</span> }<a name="line.1056"></a> -<span class="sourceLineNo">1057</span> }<a name="line.1057"></a> -<span class="sourceLineNo">1058</span> }<a name="line.1058"></a> -<span class="sourceLineNo">1059</span> }<a name="line.1059"></a> -<span class="sourceLineNo">1060</span> } catch (UnexpectedStateException e) {<a name="line.1060"></a> -<span class="sourceLineNo">1061</span> LOG.warn("Killing " + serverName + ": " + e.getMessage());<a name="line.1061"></a> -<span class="sourceLineNo">1062</span> killRegionServer(serverNode);<a name="line.1062"></a> -<span class="sourceLineNo">1063</span> throw (YouAreDeadException)new YouAreDeadException(e.getMessage()).initCause(e);<a name="line.1063"></a> -<span class="sourceLineNo">1064</span> }<a name="line.1064"></a> -<span class="sourceLineNo">1065</span> }<a name="line.1065"></a> -<span class="sourceLineNo">1066</span><a name="line.1066"></a> -<span class="sourceLineNo">1067</span> protected boolean waitServerReportEvent(final ServerName serverName, final Procedure proc) {<a name="line.1067"></a> -<span class="sourceLineNo">1068</span> final ServerStateNode serverNode = regionStates.getOrCreateServer(serverName);<a name="line.1068"></a> -<span class="sourceLineNo">1069</span> if (serverNode == null) {<a name="line.1069"></a> -<span class="sourceLineNo">1070</span> LOG.warn("serverName=null; {}", proc);<a name="line.1070"></a> -<span class="sourceLineNo">1071</span> }<a name="line.1071"></a> -<span class="sourceLineNo">1072</span> return serverNode.getReportEvent().suspendIfNotReady(proc);<a name="line.1072"></a> -<span class="sourceLineNo">1073</span> }<a name="line.1073"></a> -<span class="sourceLineNo">1074</span><a name="line.1074"></a> -<span class="sourceLineNo">1075</span> protected void wakeServerReportEvent(final ServerStateNode serverNode) {<a name="line.1075"></a> -<span class="sourceLineNo">1076</span> serverNode.getReportEvent().wake(getProcedureScheduler());<a name="line.1076"></a> -<span class="sourceLineNo">1077</span> }<a name="line.1077"></a> -<span class="sourceLineNo">1078</span><a name="line.1078"></a> +<span class="sourceLineNo">968</span> synchronized (serverNode) {<a name="line.968"></a> +<span class="sourceLineNo">969</span> if (serverNode.isInState(ServerState.SPLITTING, ServerState.OFFLINE)) {<a name="line.969"></a> +<span class="sourceLineNo">970</span> LOG.warn("Got a report from a server result in state " + serverNode.getState());<a name="line.970"></a> +<span class="sourceLineNo">971</span> return;<a name="line.971"></a> +<span class="sourceLineNo">972</span> }<a name="line.972"></a> +<span class="sourceLineNo">973</span> }<a name="line.973"></a> +<span class="sourceLineNo">974</span><a name="line.974"></a> +<span class="sourceLineNo">975</span> if (regionNames.isEmpty()) {<a name="line.975"></a> +<span class="sourceLineNo">976</span> // nothing to do if we don't have regions<a name="line.976"></a> +<span class="sourceLineNo">977</span> LOG.trace("no online region found on " + serverName);<a name="line.977"></a> +<span class="sourceLineNo">978</span> } else if (!isMetaLoaded()) {<a name="line.978"></a> +<span class="sourceLineNo">979</span> // if we are still on startup, discard the report unless is from someone holding meta<a name="line.979"></a> +<span class="sourceLineNo">980</span> checkOnlineRegionsReportForMeta(serverNode, regionNames);<a name="line.980"></a> +<span class="sourceLineNo">981</span> } else {<a name="line.981"></a> +<span class="sourceLineNo">982</span> // The Heartbeat updates us of what regions are only. check and verify the state.<a name="line.982"></a> +<span class="sourceLineNo">983</span> checkOnlineRegionsReport(serverNode, regionNames);<a name="line.983"></a> +<span class="sourceLineNo">984</span> }<a name="line.984"></a> +<span class="sourceLineNo">985</span><a name="line.985"></a> +<span class="sourceLineNo">986</span> // wake report event<a name="line.986"></a> +<span class="sourceLineNo">987</span> wakeServerReportEvent(serverNode);<a name="line.987"></a> +<span class="sourceLineNo">988</span> }<a name="line.988"></a> +<span class="sourceLineNo">989</span><a name="line.989"></a> +<span class="sourceLineNo">990</span> void checkOnlineRegionsReportForMeta(final ServerStateNode serverNode,<a name="line.990"></a> +<span class="sourceLineNo">991</span> final Set<byte[]> regionNames) {<a name="line.991"></a> +<span class="sourceLineNo">992</span> try {<a name="line.992"></a> +<span class="sourceLineNo">993</span> for (byte[] regionName: regionNames) {<a name="line.993"></a> +<span class="sourceLineNo">994</span> final RegionInfo hri = getMetaRegionFromName(regionName);<a name="line.994"></a> +<span class="sourceLineNo">995</span> if (hri == null) {<a name="line.995"></a> +<span class="sourceLineNo">996</span> if (LOG.isTraceEnabled()) {<a name="line.996"></a> +<span class="sourceLineNo">997</span> LOG.trace("Skip online report for region=" + Bytes.toStringBinary(regionName) +<a name="line.997"></a> +<span class="sourceLineNo">998</span> " while meta is loading");<a name="line.998"></a> +<span class="sourceLineNo">999</span> }<a name="line.999"></a> +<span class="sourceLineNo">1000</span> continue;<a name="line.1000"></a> +<span class="sourceLineNo">1001</span> }<a name="line.1001"></a> +<span class="sourceLineNo">1002</span><a name="line.1002"></a> +<span class="sourceLineNo">1003</span> final RegionStateNode regionNode = regionStates.getOrCreateRegionStateNode(hri);<a name="line.1003"></a> +<span class="sourceLineNo">1004</span> LOG.info("META REPORTED: " + regionNode);<a name="line.1004"></a> +<span class="sourceLineNo">1005</span> if (!reportTransition(regionNode, serverNode, TransitionCode.OPENED, 0)) {<a name="line.1005"></a> +<span class="sourceLineNo">1006</span> LOG.warn("META REPORTED but no procedure found (complete?); set location=" +<a name="line.1006"></a> +<span class="sourceLineNo">1007</span> serverNode.getServerName());<a name="line.1007"></a> +<span class="sourceLineNo">1008</span> regionNode.setRegionLocation(serverNode.getServerName());<a name="line.1008"></a> +<span class="sourceLineNo">1009</span> } else if (LOG.isTraceEnabled()) {<a name="line.1009"></a> +<span class="sourceLineNo">1010</span> LOG.trace("META REPORTED: " + regionNode);<a name="line.1010"></a> +<span class="sourceLineNo">1011</span> }<a name="line.1011"></a> +<span class="sourceLineNo">1012</span> }<a name="line.1012"></a> +<span class="sourceLineNo">1013</span> } catch (UnexpectedStateException e) {<a name="line.1013"></a> +<span class="sourceLineNo">1014</span> final ServerName serverName = serverNode.getServerName();<a name="line.1014"></a> +<span class="sourceLineNo">1015</span> LOG.warn("KILLING " + serverName + ": " + e.getMessage());<a name="line.1015"></a> +<span class="sourceLineNo">1016</span> killRegionServer(serverNode);<a name="line.1016"></a> +<span class="sourceLineNo">1017</span> }<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> void checkOnlineRegionsReport(final ServerStateNode serverNode, final Set<byte[]> regionNames)<a name="line.1020"></a> +<span class="sourceLineNo">1021</span> throws YouAreDeadException {<a name="line.1021"></a> +<span class="sourceLineNo">1022</span> final ServerName serverName = serverNode.getServerName();<a name="line.1022"></a> +<span class="sourceLineNo">1023</span> try {<a name="line.1023"></a> +<span class="sourceLineNo">1024</span> for (byte[] regionName: regionNames) {<a name="line.1024"></a> +<span class="sourceLineNo">1025</span> if (!isRunning()) return;<a name="line.1025"></a> +<span class="sourceLineNo">1026</span> final RegionStateNode regionNode = regionStates.getRegionStateNodeFromName(regionName);<a name="line.1026"></a> +<span class="sourceLineNo">1027</span> if (regionNode == null) {<a name="line.1027"></a> +<span class="sourceLineNo">1028</span> throw new UnexpectedStateException("Not online: " + Bytes.toStringBinary(regionName));<a name="line.1028"></a> +<span class="sourceLineNo">1029</span> }<a name="line.1029"></a> +<span class="sourceLineNo">1030</span> synchronized (regionNode) {<a name="line.1030"></a> +<span class="sourceLineNo">1031</span> if (regionNode.isInState(State.OPENING, State.OPEN)) {<a name="line.1031"></a> +<span class="sourceLineNo">1032</span> if (!regionNode.getRegionLocation().equals(serverName)) {<a name="line.1032"></a> +<span class="sourceLineNo">1033</span> throw new UnexpectedStateException(regionNode.toString() +<a name="line.1033"></a> +<span class="sourceLineNo">1034</span> "reported OPEN on server=" + serverName +<a name="line.1034"></a> +<span class="sourceLineNo">1035</span> " but state has otherwise.");<a name="line.1035"></a> +<span class="sourceLineNo">1036</span> } else if (regionNode.isInState(State.OPENING)) {<a name="line.1036"></a> +<span class="sourceLineNo">1037</span> try {<a name="line.1037"></a> +<span class="sourceLineNo">1038</span> if (!reportTransition(regionNode, serverNode, TransitionCode.OPENED, 0)) {<a name="line.1038"></a> +<span class="sourceLineNo">1039</span> LOG.warn(regionNode.toString() + " reported OPEN on server=" + serverName +<a name="line.1039"></a> +<span class="sourceLineNo">1040</span> " but state has otherwise AND NO procedure is running");<a name="line.1040"></a> +<span class="sourceLineNo">1041</span> }<a name="line.1041"></a> +<span class="sourceLineNo">1042</span> } catch (UnexpectedStateException e) {<a name="line.1042"></a> +<span class="sourceLineNo">1043</span> LOG.warn(regionNode.toString() + " reported unexpteced OPEN: " + e.getMessage(), e);<a name="line.1043"></a> +<span class="sourceLineNo">1044</span> }<a name="line.1044"></a> +<span class="sourceLineNo">1045</span> }<a name="line.1045"></a> +<span class="sourceLineNo">1046</span> } else if (!regionNode.isInState(State.CLOSING, State.SPLITTING)) {<a name="line.1046"></a> +<span class="sourceLineNo">1047</span> long diff = regionNode.getLastUpdate() - EnvironmentEdgeManager.currentTime();<a name="line.1047"></a> +<span class="sourceLineNo">1048</span> if (diff > 1000/*One Second... make configurable if an issue*/) {<a name="line.1048"></a> +<span class="sourceLineNo">1049</span> // So, we can get report that a region is CLOSED or SPLIT because a heartbeat<a name="line.1049"></a> +<span class="sourceLineNo">1050</span> // came in at about same time as a region transition. Make sure there is some<a name="line.1050"></a> +<span class="sourceLineNo">1051</span> // elapsed time between killing remote server.<a name="line.1051"></a> +<span class="sourceLineNo">1052</span> throw new UnexpectedStateException(regionNode.toString() +<a name="line.1052"></a> +<span class="sourceLineNo">1053</span> " reported an unexpected OPEN; time since last update=" + diff);<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">1056</span> }<a name="line.1056"></a> +<span class="sourceLineNo">1057</span> }<a name="line.1057"></a> +<span class="sourceLineNo">1058</span> } catch (UnexpectedStateException e) {<a name="line.1058"></a> +<span class="sourceLineNo">1059</span> LOG.warn("Killing " + serverName + ": " + e.getMessage());<a name="line.1059"></a> +<span class="sourceLineNo">1060</span> killRegionServer(serverNode);<a name="line.1060"></a> +<span class="sourceLineNo">1061</span> throw (YouAreDeadException)new YouAreDeadException(e.getMessage()).initCause(e);<a name="line.1061"></a> +<span class="sourceLineNo">1062</span> }<a name="line.1062"></a> +<span class="sourceLineNo">1063</span> }<a name="line.1063"></a> +<span class="sourceLineNo">1064</span><a name="line.1064"></a> +<span class="sourceLineNo">1065</span> protected boolean waitServerReportEvent(final ServerName serverName, final Procedure proc) {<a name="line.1065"></a> +<span class="sourceLineNo">1066</span> final ServerStateNode serverNode = regionStates.getOrCreateServer(serverName);<a name="line.1066"></a> +<span class="sourceLineNo">1067</span> if (serverNode == null) {<a name="line.1067"></a> +<span class="sourceLineNo">1068</span> LOG.warn("serverName=null; {}", proc);<a name="line.1068"></a> +<span class="sourceLineNo">1069</span> }<a name="line.1069"></a> +<span class="sourceLineNo">1070</span> return serverNode.getReportEvent().suspendIfNotReady(proc);<a name="line.1070"></a> +<span class="sourceLineNo">1071</span> }<a name="line.1071"></a> +<span class="sourceLineNo">1072</span><a name="line.1072"></a> +<span class="sourceLineNo">1073</span> protected void wakeServerReportEvent(final ServerStateNode serverNode) {<a name="line.1073"></a> +<span class="sourceLineNo">1074</span> serverNode.getReportEvent().wake(getProcedureScheduler());<a name="line.1074"></a> +<span class="sourceLineNo">1075</span> }<a name="line.1075"></a> +<span class="sourceLineNo">1076</span><a name="line.1076"></a> +<span class="sourceLineNo">1077</span> // ============================================================================================<a name="line.1077"></a> +<span class="sourceLineNo">1078</span> // RIT chore<a name="line.1078"></a> <span class="sourceLineNo">1079</span> // ============================================================================================<a name="line.1079"></a> -<span class="sourceLineNo">1080</span> // RIT chore<a name="line.1080"></a> -<span class="sourceLineNo">1081</span> // ============================================================================================<a name="line.1081"></a> -<span class="sourceLineNo">1082</span> private static class RegionInTransitionChore extends ProcedureInMemoryChore<MasterProcedureEnv> {<a name="line.1082"></a> -<span class="sourceLineNo">1083</span> public RegionInTransitionChore(final int timeoutMsec) {<a name="line.1083"></a> -<span class="sourceLineNo">1084</span> super(timeoutMsec);<a name="line.1084"></a> -<span class="sourceLineNo">1085</span> }<a name="line.1085"></a> -<span class="sourceLineNo">1086</span><a name="line.1086"></a> -<span class="sourceLineNo">1087</span> @Override<a name="line.1087"></a> -<span class="sourceLineNo">1088</span> protected void periodicExecute(final MasterProcedureEnv env) {<a name="line.1088"></a> -<span class="sourceLineNo">1089</span> final AssignmentManager am = env.getAssignmentManager();<a name="line.1089"></a> -<span class="sourceLineNo">1090</span><a name="line.1090"></a> -<span class="sourceLineNo">1091</span> final RegionInTransitionStat ritStat = am.computeRegionInTransitionStat();<a name="line.1091"></a> -<span class="sourceLineNo">1092</span> if (ritStat.hasRegionsOverThreshold()) {<a name="line.1092"></a> -<span class="sourceLineNo">1093</span> for (RegionState hri: ritStat.getRegionOverThreshold()) {<a name="line.1093"></a> -<span class="sourceLineNo">1094</span> am.handleRegionOverStuckWarningThreshold(hri.getRegion());<a name="line.1094"></a> -<span class="sourceLineNo">1095</span> }<a name="line.1095"></a> -<span class="sourceLineNo">1096</span> }<a name="line.1096"></a> -<span class="sourceLineNo">1097</span><a name="line.1097"></a> -<span class="sourceLineNo">1098</span> // update metrics<a name="line.1098"></a> -<span class="sourceLineNo">1099</span> am.updateRegionsInTransitionMetrics(ritStat);<a name="line.1099"></a> -<span class="sourceLineNo">1100</span> }<a name="line.1100"></a> -<span class="sourceLineNo">1101</span> }<a name="line.1101"></a> -<span class="sourceLineNo">1102</span><a name="line.1102"></a> -<span class="sourceLineNo">1103</span> public RegionInTransitionStat computeRegionInTransitionStat() {<a name="line.1103"></a> -<span class="sourceLineNo">1104</span> final RegionInTransitionStat rit = new RegionInTransitionStat(getConfiguration());<a name="line.1104"></a> -<span class="sourceLineNo">1105</span> rit.update(this);<a name="line.1105"></a> -<span class="sourceLineNo">1106</span> return rit;<a name="line.1106"></a> -<span class="sourceLineNo">1107</span> }<a name="line.1107"></a> -<span class="sourceLineNo">1108</span><a name="line.1108"></a> -<span class="sourceLineNo">1109</span> public static class RegionInTransitionStat {<a name="line.1109"></a> -<span class="sourceLineNo">1110</span> private final int ritThreshold;<a name="line.1110"></a> -<span class="sourceLineNo">1111</span><a name="line.1111"></a> -<span class="sourceLineNo">1112</span> private HashMap<String, RegionState> ritsOverThreshold = null;<a name="line.1112"></a> -<span class="sourceLineNo">1113</span> private long statTimestamp;<a name="line.1113"></a> -<span class="sourceLineNo">1114</span> private long oldestRITTime = 0;<a name="line.1114"></a> -<span class="sourceLineNo">1115</span> private int totalRITsTwiceThreshold = 0;<a name="line.1115"></a> -<span class="sourceLineNo">1116</span> private int totalRITs = 0;<a name="line.1116"></a> -<span class="sourceLineNo">1117</span><a name="line.1117"></a> -<span class="sourceLineNo">1118</span> @VisibleForTesting<a name="line.1118"></a> -<span class="sourceLineNo">1119</span> public RegionInTransitionStat(final Configuration conf) {<a name="line.1119"></a> -<span class="sourceLineNo">1120</span> this.ritThreshold =<a name="line.1120"></a> -<span class="sourceLineNo">1121</span> conf.getInt(METRICS_RIT_STUCK_WARNING_THRESHOLD, DEFAULT_RIT_STUCK_WARNING_THRESHOLD);<a name="line.1121"></a> -<span class="sourceLineNo">1122</span> }<a name="line.1122"></a> -<span class="sourceLineNo">1123</span><a name="line.1123"></a> -<span class="sourceLineNo">1124</span> public int getRITThreshold() {<a name="line.1124"></a> -<span class="sourceLineNo">1125</span> return ritThreshold;<a name="line.1125"></a> -<span class="sourceLineNo">1126</span> }<a name="line.1126"></a> -<span class="sourceLineNo">1127</span><a name="line.1127"></a> -<span class="sourceLineNo">1128</span> public long getTimestamp() {<a name="line.1128"></a> -<span class="sourceLineNo">1129</span> return statTimestamp;<a name="line.1129"></a> -<span class="sourceLineNo">1130</span> }<a name="line.1130"></a> -<span class="sourceLineNo">1131</span><a name="line.1131"></a> -<span class="sourceLineNo">1132</span> public int getTotalRITs() {<a name="line.1132"></a> -<span class="sourceLineNo">1133</span> return totalRITs;<a name="line.1133"></a> -<span class="sourceLineNo">1134</span> }<a name="line.1134"></a> -<span class="sourceLineNo">1135</span><a name="line.1135"></a> -<span class="sourceLineNo">1136</span> public long getOldestRITTime() {<a name="line.1136"></a> -<span class="sourceLineNo">1137</span> return oldestRITTime;<a name="line.1137"></a> -<span class="sourceLineNo">1138</span> }<a name="line.1138"></a> -<span class="sourceLineNo">1139</span><a name="line.1139"></a> -<span class="sourceLineNo">1140</span> public int getTotalRITsOverThreshold() {<a name="line.1140"></a> -<span class="sourceLineNo">1141</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1141"></a> -<span class="sourceLineNo">1142</span> return m != null ? m.size() : 0;<a name="line.1142"></a> -<span class="sourceLineNo">1143</span> }<a name="line.1143"></a> -<span class="sourceLineNo">1144</span><a name="line.1144"></a> -<span class="sourceLineNo">1145</span> public boolean hasRegionsTwiceOverThreshold() {<a name="line.1145"></a> -<span class="sourceLineNo">1146</span> return totalRITsTwiceThreshold > 0;<a name="line.1146"></a> -<span class="sourceLineNo">1147</span> }<a name="line.1147"></a> -<span class="sourceLineNo">1148</span><a name="line.1148"></a> -<span class="sourceLineNo">1149</span> public boolean hasRegionsOverThreshold() {<a name="line.1149"></a> -<span class="sourceLineNo">1150</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1150"></a> -<span class="sourceLineNo">1151</span> return m != null && !m.isEmpty();<a name="line.1151"></a> -<span class="sourceLineNo">1152</span> }<a name="line.1152"></a> -<span class="sourceLineNo">1153</span><a name="line.1153"></a> -<span class="sourceLineNo">1154</span> public Collection<RegionState> getRegionOverThreshold() {<a name="line.1154"></a> -<span class="sourceLineNo">1155</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1155"></a> -<span class="sourceLineNo">1156</span> return m != null? m.values(): Collections.EMPTY_SET;<a name="line.1156"></a> -<span class="sourceLineNo">1157</span> }<a name="line.1157"></a> -<span class="sourceLineNo">1158</span><a name="line.1158"></a> -<span class="sourceLineNo">1159</span> public boolean isRegionOverThreshold(final RegionInfo regionInfo) {<a name="line.1159"></a> -<span class="sourceLineNo">1160</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1160"></a> -<span class="sourceLineNo">1161</span> return m != null && m.containsKey(regionInfo.getEncodedName());<a name="line.1161"></a> -<span class="sourceLineNo">1162</span> }<a name="line.1162"></a> -<span class="sourceLineNo">1163</span><a name="line.1163"></a> -<span class="sourceLineNo">1164</span> public boolean isRegionTwiceOverThreshold(final RegionInfo regionInfo) {<a name="line.1164"></a> -<span class="sourceLineNo">1165</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1165"></a> -<span class="sourceLineNo">1166</span> if (m == null) return false;<a name="line.1166"></a> -<span class="sourceLineNo">1167</span> final RegionState state = m.get(regionInfo.getEncodedName());<a name="line.1167"></a> -<span class="sourceLineNo">1168</span> if (state == null) return false;<a name="line.1168"></a> -<span class="sourceLineNo">1169</span> return (statTimestamp - state.getStamp()) > (ritThreshold * 2);<a name="line.1169"></a> -<span class="sourceLineNo">1170</span> }<a name="line.1170"></a> -<span class="sourceLineNo">1171</span><a name="line.1171"></a> -<span class="sourceLineNo">1172</span> protected void update(final AssignmentManager am) {<a name="line.1172"></a> -<span class="sourceLineNo">1173</span> final RegionStates regionStates = am.getRegionStates();<a name="line.1173"></a> -<span class="sourceLineNo">1174</span> this.statTimestamp = EnvironmentEdgeManager.currentTime();<a name="line.1174"></a> -<span class="sourceLineNo">1175</span> update(regionStates.getRegionsStateInTransition(), statTimestamp);<a name="line.1175"></a> -<span class="sourceLineNo">1176</span> update(regionStates.getRegionFailedOpen(), statTimestamp);<a name="line.1176"></a> -<span class="sourceLineNo">1177</span> }<a name="line.1177"></a> -<span class="sourceLineNo">1178</span><a name="line.1178"></a> -<span class="sourceLineNo">1179</span> private void update(final Collection<RegionState> regions, final long currentTime) {<a name="line.1179"></a> -<span class="sourceLineNo">1180</span> for (RegionState state: regions) {<a name="line.1180"></a> -<span class="sourceLineNo">1181</span> totalRITs++;<a name="line.1181"></a> -<span class="sourceLineNo">1182</span> final long ritTime = currentTime - state.getStamp();<a name="line.1182"></a> -<span class="sourceLineNo">1183</span> if (ritTime > ritThreshold) {<a name="line.1183"></a> -<span class="sourceLineNo">1184</span> if (ritsOverThreshold == null) {<a name="line.1184"></a> -<span class="sourceLineNo">1185</span> ritsOverThreshold = new HashMap<String, RegionState>();<a name="line.1185"></a> -<span class="sourceLineNo">1186</span> }<a name="line.1186"></a> -<span class="sourceLineNo">1187</span> ritsOverThreshold.put(state.getRegion().getEncodedName(), state);<a name="line.1187"></a> -<span class="sourceLineNo">1188</span> totalRITsTwiceThreshold += (ritTime > (ritThreshold * 2)) ? 1 : 0;<a name="line.1188"></a> -<span class="sourceLineNo">1189</span> }<a name="line.1189"></a> -<span class="sourceLineNo">1190</span> if (oldestRITTime < ritTime) {<a name="line.1190"></a> -<span class="sourceLineNo">1191</span> oldestRITTime = ritTime;<a name="line.1191"></a> -<span class="sourceLineNo">1192</span> }<a name="line.1192"></a> -<span class="sourceLineNo">1193</span> }<a name="line.1193"></a> -<span class="sourceLineNo">1194</span> }<a name="line.1194"></a> -<span class="sourceLineNo">1195</span> }<a name="line.1195"></a> -<span class="sourceLineNo">1196</span><a name="line.1196"></a> -<span class="sourceLineNo">1197</span> private void updateRegionsInTransitionMetrics(final RegionInTransitionStat ritStat) {<a name="line.1197"></a> -<span class="sourceLineNo">1198</span> metrics.updateRITOldestAge(ritStat.getOldestRITTime());<a name="line.1198"></a> -<span class="sourceLineNo">1199</span> metrics.updateRITCount(ritStat.getTotalRITs());<a name="line.1199"></a> -<span class="sourceLineNo">1200</span> metrics.updateRITCountOverThreshold(ritStat.getTotalRITsOverThreshold());<a name="line.1200"></a> -<span class="sourceLineNo">1201</span> }<a name="line.1201"></a> -<span class="sourceLineNo">1202</span><a name="line.1202"></a> -<span class="sourceLineNo">1203</span> private void handleRegionOverStuckWarningThreshold(final RegionInfo regionInfo) {<a name="line.1203"></a> -<span class="sourceLineNo">1204</span> final RegionStateNode regionNode = regionStates.getRegionStateNode(regionInfo);<a name="line.1204"></a> -<span class="sourceLineNo">1205</span> //if (regionNode.isStuck()) {<a name="line.1205"></a> -<span class="sourceLineNo">1206</span> LOG.warn("STUCK Region-In-Transition {}", regionNode);<a name="line.1206"></a> -<span class="sourceLineNo">1207</span> }<a name="line.1207"></a> -<span class="sourceLineNo">1208</span><a name="line.1208"></a> +<span class="sourceLineNo">1080</span> private static class RegionInTransitionChore extends ProcedureInMemoryChore<MasterProcedureEnv> {<a name="line.1080"></a> +<span class="sourceLineNo">1081</span> public RegionInTransitionChore(final int timeoutMsec) {<a name="line.1081"></a> +<span class="sourceLineNo">1082</span> super(timeoutMsec);<a name="line.1082"></a> +<span class="sourceLineNo">1083</span> }<a name="line.1083"></a> +<span class="sourceLineNo">1084</span><a name="line.1084"></a> +<span class="sourceLineNo">1085</span> @Override<a name="line.1085"></a> +<span class="sourceLineNo">1086</span> protected void periodicExecute(final MasterProcedureEnv env) {<a name="line.1086"></a> +<span class="sourceLineNo">1087</span> final AssignmentManager am = env.getAssignmentManager();<a name="line.1087"></a> +<span class="sourceLineNo">1088</span><a name="line.1088"></a> +<span class="sourceLineNo">1089</span> final RegionInTransitionStat ritStat = am.computeRegionInTransitionStat();<a name="line.1089"></a> +<span class="sourceLineNo">1090</span> if (ritStat.hasRegionsOverThreshold()) {<a name="line.1090"></a> +<span class="sourceLineNo">1091</span> for (RegionState hri: ritStat.getRegionOverThreshold()) {<a name="line.1091"></a> +<span class="sourceLineNo">1092</span> am.handleRegionOverStuckWarningThreshold(hri.getRegion());<a name="line.1092"></a> +<span class="sourceLineNo">1093</span> }<a name="line.1093"></a> +<span class="sourceLineNo">1094</span> }<a name="line.1094"></a> +<span class="sourceLineNo">1095</span><a name="line.1095"></a> +<span class="sourceLineNo">1096</span> // update metrics<a name="line.1096"></a> +<span class="sourceLineNo">1097</span> am.updateRegionsInTransitionMetrics(ritStat);<a name="line.1097"></a> +<span class="sourceLineNo">1098</span> }<a name="line.1098"></a> +<span class="sourceLineNo">1099</span> }<a name="line.1099"></a> +<span class="sourceLineNo">1100</span><a name="line.1100"></a> +<span class="sourceLineNo">1101</span> public RegionInTransitionStat computeRegionInTransitionStat() {<a name="line.1101"></a> +<span class="sourceLineNo">1102</span> final RegionInTransitionStat rit = new RegionInTransitionStat(getConfiguration());<a name="line.1102"></a> +<span class="sourceLineNo">1103</span> rit.update(this);<a name="line.1103"></a> +<span class="sourceLineNo">1104</span> return rit;<a name="line.1104"></a> +<span class="sourceLineNo">1105</span> }<a name="line.1105"></a> +<span class="sourceLineNo">1106</span><a name="line.1106"></a> +<span class="sourceLineNo">1107</span> public static class RegionInTransitionStat {<a name="line.1107"></a> +<span class="sourceLineNo">1108</span> private final int ritThreshold;<a name="line.1108"></a> +<span class="sourceLineNo">1109</span><a name="line.1109"></a> +<span class="sourceLineNo">1110</span> private HashMap<String, RegionState> ritsOverThreshold = null;<a name="line.1110"></a> +<span class="sourceLineNo">1111</span> private long statTimestamp;<a name="line.1111"></a> +<span class="sourceLineNo">1112</span> private long oldestRITTime = 0;<a name="line.1112"></a> +<span class="sourceLineNo">1113</span> private int totalRITsTwiceThreshold = 0;<a name="line.1113"></a> +<span class="sourceLineNo">1114</span> private int totalRITs = 0;<a name="line.1114"></a> +<span class="sourceLineNo">1115</span><a name="line.1115"></a> +<span class="sourceLineNo">1116</span> @VisibleForTesting<a name="line.1116"></a> +<span class="sourceLineNo">1117</span> public RegionInTransitionStat(final Configuration conf) {<a name="line.1117"></a> +<span class="sourceLineNo">1118</span> this.ritThreshold =<a name="line.1118"></a> +<span class="sourceLineNo">1119</span> conf.getInt(METRICS_RIT_STUCK_WARNING_THRESHOLD, DEFAULT_RIT_STUCK_WARNING_THRESHOLD);<a name="line.1119"></a> +<span class="sourceLineNo">1120</span> }<a name="line.1120"></a> +<span class="sourceLineNo">1121</span><a name="line.1121"></a> +<span class="sourceLineNo">1122</span> public int getRITThreshold() {<a name="line.1122"></a> +<span class="sourceLineNo">1123</span> return ritThreshold;<a name="line.1123"></a> +<span class="sourceLineNo">1124</span> }<a name="line.1124"></a> +<span class="sourceLineNo">1125</span><a name="line.1125"></a> +<span class="sourceLineNo">1126</span> public long getTimestamp() {<a name="line.1126"></a> +<span class="sourceLineNo">1127</span> return statTimestamp;<a name="line.1127"></a> +<span class="sourceLineNo">1128</span> }<a name="line.1128"></a> +<span class="sourceLineNo">1129</span><a name="line.1129"></a> +<span class="sourceLineNo">1130</span> public int getTotalRITs() {<a name="line.1130"></a> +<span class="sourceLineNo">1131</span> return totalRITs;<a name="line.1131"></a> +<span class="sourceLineNo">1132</span> }<a name="line.1132"></a> +<span class="sourceLineNo">1133</span><a name="line.1133"></a> +<span class="sourceLineNo">1134</span> public long getOldestRITTime() {<a name="line.1134"></a> +<span class="sourceLineNo">1135</span> return oldestRITTime;<a name="line.1135"></a> +<span class="sourceLineNo">1136</span> }<a name="line.1136"></a> +<span class="sourceLineNo">1137</span><a name="line.1137"></a> +<span class="sourceLineNo">1138</span> public int getTotalRITsOverThreshold() {<a name="line.1138"></a> +<span class="sourceLineNo">1139</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1139"></a> +<span class="sourceLineNo">1140</span> return m != null ? m.size() : 0;<a name="line.1140"></a> +<span class="sourceLineNo">1141</span> }<a name="line.1141"></a> +<span class="sourceLineNo">1142</span><a name="line.1142"></a> +<span class="sourceLineNo">1143</span> public boolean hasRegionsTwiceOverThreshold() {<a name="line.1143"></a> +<span class="sourceLineNo">1144</span> return totalRITsTwiceThreshold > 0;<a name="line.1144"></a> +<span class="sourceLineNo">1145</span> }<a name="line.1145"></a> +<span class="sourceLineNo">1146</span><a name="line.1146"></a> +<span class="sourceLineNo">1147</span> public boolean hasRegionsOverThreshold() {<a name="line.1147"></a> +<span class="sourceLineNo">1148</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1148"></a> +<span class="sourceLineNo">1149</span> return m != null && !m.isEmpty();<a name="line.1149"></a> +<span class="sourceLineNo">1150</span> }<a name="line.1150"></a> +<span class="sourceLineNo">1151</span><a name="line.1151"></a> +<span class="sourceLineNo">1152</span> public Collection<RegionState> getRegionOverThreshold() {<a name="line.1152"></a> +<span class="sourceLineNo">1153</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1153"></a> +<span class="sourceLineNo">1154</span> return m != null? m.values(): Collections.EMPTY_SET;<a name="line.1154"></a> +<span class="sourceLineNo">1155</span> }<a name="line.1155"></a> +<span class="sourceLineNo">1156</span><a name="line.1156"></a> +<span class="sourceLineNo">1157</span> public boolean isRegionOverThreshold(final RegionInfo regionInfo) {<a name="line.1157"></a> +<span class="sourceLineNo">1158</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1158"></a> +<span class="sourceLineNo">1159</span> return m != null && m.containsKey(regionInfo.getEncodedName());<a name="line.1159"></a> +<span class="sourceLineNo">1160</span> }<a name="line.1160"></a> +<span class="sourceLineNo">1161</span><a name="line.1161"></a> +<span class="sourceLineNo">1162</span> public boolean isRegionTwiceOverThreshold(final RegionInfo regionInfo) {<a name="line.1162"></a> +<span class="sourceLineNo">1163</span> Map<String, RegionState> m = this.ritsOverThreshold;<a name="line.1163"></a> +<span class="sourceLineNo">1164</span> if (m == null) return false;<a name="line.1164"></a> +<span class="sourceLineNo">1165</span> final RegionState state = m.get(regionInfo.getEncodedName());<a name="line.1165"></a> +<span class="sourceLineNo">1166</span> if (state == null) return false;<a name="line.1166"></a> +<span class="sourceLineNo">1167</span> return (statTimestamp - state.getStamp()) > (ritThreshold * 2);<a name="line.1167"></a> +<span class="sourceLineNo">1168</span> }<a name="line.1168"></a> +<span class="sourceLineNo">1169</span><a name="line.1169"></a> +<span class="sourceLineNo">1170</span> protected void update(final AssignmentManager am) {<a name="line.1170"></a> +<span class="sourceLineNo">1171</span> final RegionStates regionStates = am.getRegionStates();<a name="line.1171"></a> +<span class="sourceLineNo">1172</span> this.statTimestamp = EnvironmentEdgeManager.currentTime();<a name="line.1172"></a> +<span class="sourceLineNo">1173</span> update(regionStates.getRegionsStateInTransition(), statTimestamp);<a name="line.1173"></a> +<span class="sourceLineNo">1174</span> update(regionStates.getRegionFailedOpen(), statTimestamp);<a name="line.1174"></a> +<span class="sourceLineNo">1175</span> }<a name="line.1175"></a> +<span class="sourceLineNo">1176</span><a name="line.1176"></a> +<span class="sourceLineNo">1177</span> private void update(final Collection<RegionState> regions, final long currentTime) {<a name="line.1177"></a> +<span class="sourceLineNo">1178</span> for (RegionState state: regions) {<a name="line.1178"></a> +<span class="sourceLineNo">1179</span> totalRITs++;<a name="line.1179"></a> +<span class="sourceLineNo">1180</span> final long ritTime = currentTime - state.getStamp();<a name="line.1180"></a> +<span class="sourceLineNo">1181</span> if (ritTime > ritThreshold) {<a name="line.1181"></a> +<span class="sourceLineNo">1182</span> if (ritsOverThreshold == null) {<a name="line.1182"></a> +<span class="sourceLineNo">1183</span> ritsOverThreshold = new HashMap<String, RegionState>();<a name="line.1183"></a> +<span class="sourceLineNo">1184</span> }<a name="line.1184"></a> +<span class="sourceLineNo">1185</span> ritsOverThreshold.put(state.getRegion().getEncodedName(), state);<a name="line.1185"></a> +<span class="sourceLineNo">1186</span> totalRITsTwiceThreshold += (ritTime > (ritThreshold * 2)) ? 1 : 0;<a name="line.1186"></a> +<span class="sourceLineNo">1187</span> }<a name="line.1187"></a> +<span class="sourceLineNo">1188</span> if (oldestRITTime < ritTime) {<a name="line.1188"></a> +<span class="sourceLineNo">1189</span> oldestRITTime = ritTime;<a name="line.1189"></a> +<span class="sourceLineNo">1190</span> }<a name="line.1190"></a> +<span class="sourceLineNo">1191</span> }<a name="line.1191"></a> +<span class="sourceLineNo">1192</span> }<a name="line.1192"></a> +<span class="sourceLineNo">1193</span> }<a name="line.1193"></a> +<span class="sourceLineNo">1194</span><a name="line.1194"></a> +<span class="sourceLineNo">1195</span> private void updateRegionsInTransitionMetrics(final RegionInTransitionStat ritStat) {<a name="line.1195"></a> +<span class="sourceLineNo">1196</span> metrics.updateRITOldestAge(ritStat.getOldestRITTime());<a name="line.1196"></a> +<span class="sourceLineNo">1197</span> metrics.updateRITCount(ritStat.getTotalRITs());<a name="line.1197"></a> +<span class="sourceLineNo">1198</span> metrics.updateRITCountOverThreshold(ritStat.getTotalRITsOverThreshold());<a name="line.1198"></a> +<span class="sourceLineNo">1199</span> }<a name="line.1199"></a> +<span class="sourceLineNo">1200</span><a name="line.1200"></a> +<span class="sourceLineNo">1201</span> private void handleRegionOverStuckWarningThreshold(final RegionInfo regionInfo) {<a name="line.1201"></a> +<span class="sourceLineNo">1202</span> final RegionStateNode regionNode = regionStates.getRegionStateNode(regionInfo);<a name="line.1202"></a> +<span class="sourceLineNo">1203</span> //if (regionNode.isStuck()) {<a name="line.1203"></a> +<span class="sourceLineNo">1204</span> LOG.warn("STUCK Region-In-Transition {}", regionNode);<a name="line.1204"></a> +<span class="sourceLineNo">1205</span> }<a name="line.1205"></a> +<span class="sourceLineNo">1206</span><a name="line.1206"></a> +<span class="sourceLineNo">1207</span> // ============================================================================================<a name="line.1207"></a> +<span class="sourceLineNo">1208</span> // TODO: Master load/bootstrap<a name="line.1208"></a> <span class="sourceLineNo">1209</span> // ============================================================================================<a name="line.1209"></a> -<span class="sourceLineNo">1210</span> // TODO: Master load/bootstrap<a name="line.1210"></a> -<span class="sourceLineNo">1211</span> // ============================================================================================<a name="line.1211"></a> -<span class="sourceLineNo">1212</span> public void joinCluster() throws IOException {<a name="line.1212"></a> -<span class="sourceLineNo">1213</span> final long startTime = System.currentTimeMillis();<a name="line.1213"></a> -<span class="sourceLineNo">1214</span> LOG.debug("Joining cluster...");<a name="line.1214"></a> -<span class="sourceLineNo">1215</span><a name="line.1215"></a> -<span class="sourceLineNo">1216</span> // Scan hbase:meta to build list of existing regions, servers, and assignment<a name="line.1216"></a> -<span class="sourceLineNo">1217</span> // hbase:meta is online when we get to here and TableStateManager has been started.<a name="line.1217"></a> -<span class="sourceLineNo">1218</span> loadMeta();<a name="line.1218"></a> -<span class="sourceLineNo">1219</span><a name="line.1219"></a> -<span class="sourceLineNo">1220</span> for (int i = 0; master.getServerManager().countOfRegionServers() < 1; ++i) {<a name="line.1220"></a> -<span class="sourceLineNo">1221</span> LOG.info("Waiting for RegionServers to join; current count=" +<a name="line.1221"></a> -<span class="sourceLineNo">1222</span> master.getServerManager().countOfRegionServers());<a name="line.1222"></a> -<span class="sourceLineNo">1223</span> Threads.sleep(250);<a name="line.1223"></a> -<span class="sourceLineNo">1224</span> }<a name="line.1224"></a> -<span class="sourceLineNo">1225</span> LOG.info("Number of RegionServers=" + master.getServerManager().countOfRegionServers());<a name="line.1225"></a> +<span class="sourceLineNo">1210</span> public void joinCluster() throws IOException {<a name="line.1210"></a> +<span class="sourceLineNo">1211</span> final long startTime = System.currentTimeMillis();<a name="line.1211"></a> +<span class="sourceLineNo">1212</span> LOG.debug("Joining cluster...");<a name="line.1212"></a> +<span class="sourceLineNo">1213</span><a name="line.1213"></a> +<span class="sourceLineNo">1214</span> // Scan hbase:meta to build list of existing regions, servers, and assignment<a name="line.1214"></a> +<span class="sourceLineNo">1215</span> // hbase:meta is online when we get to here and TableStateManager has been started.<a name="line.1215"></a> +<span class="sourceLineNo">1216</span> loadMeta();<a name="line.1216"></a> +<span class="sourceLineNo">1217</span><a name="line.1217"></a> +<span class="sourceLineNo">1218</span> for (int i = 0; master.getServerManager().countOfRegionServers() < 1; ++i) {<a name="line.1218"></a> +<span class="sourceLineNo">1219</span> LOG.info("Waiting for RegionServers to join; current count=" +<a name="line.1219"></a> +<span class="sourceLineNo">1220</span> master.getServerManager().countOfRegionServers());<a name="line.1220"></a> +<span class="sourceLineNo">1221</span> Threads.sleep(250);<a name="line.1221"></a> +<span class="sourceLineNo">1222</span> }<a name="line.1222"></a> +<span class="sourceLineNo">1223</span> LOG.info("Number of RegionServers=" + master.getServerManager().countOfRegionServers());<a name="line.1223"></a> +<span class="sourceLineNo">1224</span><a name="line.1224"></a> +<span class="sourceLineNo">1225</span> boolean failover = processofflineServersWithOnlineRegions();<a name="line.1225"></a> <span class="sourceLineNo">1226</span><a name="line.1226"></a> -<span class="sourceLineNo">1227</span> boolean failover = processofflineServersWithOnlineRegions();<a name="line.1227"></a> -<span class="sourceLineNo">1228</span><a name="line.1228"></a> -<span class="sourceLineNo">1229</span> // Start the RIT chore<a name="line.1229"></a> -<span class="sourceLineNo">1230</span> master.getMasterProcedureExecutor().addChore(this.ritChore);<a name="line.1230"></a> -<span class="sourceLineNo">1231</span><a name="line.1231"></a> -<span class="sourceLineNo">1232</span> LOG.info(String.format("Joined the cluster in %s, failover=%s",<a name="line.1232"></a> -<span class="sourceLineNo">1233</span> StringUtils.humanTimeDiff(System.currentTimeMillis() - startTime), failover));<a name="line.1233"></a> -<span class="sourceLineNo">1234</span> }<a name="line.1234"></a> -<span class="sourceLineNo">1235</span><a name="line.1235"></a> -<span class="sourceLineNo">1236</span> private void loadMeta() throws IOException {<a name="line.1236"></a> -<span class="sourceLineNo">1237</span> // TODO: use a thread pool<a name="line.1237"></a> -<span class="sourceLineNo">1238</span> regionStateStore.visitMeta(new RegionStateStore.RegionStateVisitor() {<a name="line.1238"></a> -<span class="sourceLineNo">1239</span> @Override<a name="line.1239"></a> -<span class="sourceLineNo">1240</span> public void visitRegionState(Result result, final RegionInfo regionInfo, final State state,<a name="line.1240"></a> -<span class="sourceLineNo">1241</span> final ServerName regionLocation, final ServerName lastHost, final long openSeqNum) {<a name="line.1241"></a> -<span class="sourceLineNo">1242</span> if (state == null && regionLocation == null && lastHost == null &&<a name="line.1242"></a> -<span class="sourceLineNo">1243</span> openSeqNum == SequenceId.NO_SEQUENCE_ID) {<a name="line.1243"></a> -<span class="sourceLineNo">1244</span> // This is a row with nothing in it.<a name="line.1244"></a> -<span class="sourceLineNo">1245</span> LOG.warn("Skipping empty row={}", result);<a name="line.1245"></a> -<span class="sourceLineNo">1246</span> return;<a name="line.1246"></a> -<span class="sourceLineNo">1247</span> }<a name="line.1247"></a> -<span class="sourceLineNo">1248</span> State localState = state;<a name="line.1248"></a> -<span class="sourceLineNo">1249</span> if (localState == null) {<a name="line.1249"></a> -<span class="sourceLineNo">1250</span> // No region state column data in hbase:meta table! Are I doing a rolling upgrade from<a name="line.1250"></a> -<span class="sourceLineNo">1251</span> // hbase1 to hbase2? Am I restoring a SNAPSHOT or otherwise adding a region to hbase:meta?<a name="line.1251"></a> -<span class="sourceLineNo">1252</span> // In any of these cases, state is empty. For now, presume OFFLINE but there are probably<a name="line.1252"></a> -<span class="sourceLineNo">1253</span> // cases where we need to probe more to be sure this correct; TODO informed by experience.<a name="line.1253"></a> -<span class="sourceLineNo">1254</span> LOG.info(regionInfo.getEncodedName() + " regionState=null; presuming " + State.OFFLINE);<a name="line.1254"></a> -<span class="sourceLineNo">1255</span><a name="line.1255"></a> -<span class="sourceLineNo">1256</span> localState = State.OFFLINE;<a name="line.1256"></a> -<span class="sourceLineNo">1257</span> }<a name="line.1257"></a> -<span class="sourceLineNo">1258</span> final RegionStateNode regionNode = regionStates.getOrCreateRegionStateNode(regionInfo);<a name="line.1258"></a> -<span class="sourceLineNo">1259</span> synchronized (regionNode) {<a name="line.1259"></a> -<span class="sourceLineNo">1260</span> if (!regionNode.isInTransition()) {<a name="line.1260"></a> -<span class="sourceLineNo">1261</span> regionNode.setState(localState);<a name="line.1261"></a> -<span class="sourceLineNo">1262</span> regionNode.setLastHost(lastHost);<a name="line.1262"></a> -<span class="sourceLineNo">1263</span> regionNode.setRegionLocation(regionLocation);<a name="line.1263"></a> -<span class="sourceLineNo">1264</span> regionNode.setOpenSeqNum(openSeqNum);<a name="line.1264"></a> -<span class="sourceLineNo">1265</span><a name="line.1265"></a> -<span class="sourceLineNo">1266</span> if (localState == State.OPEN) {<a name="line.1266"></a> -<span class="sourceLineNo">1267</span> assert regionLocation != null : "found null region location for " + regionNode;<a name="line.1267"></a> -<span class="sourceLineNo">1268</span> regionStates.addRegionToServer(regionNode);<a name="line.1268"></a> -<span class="sourceLineNo">1269</span> } else if (localState == State.OFFLINE || regionInfo.isOffline()) {<a name="line.1269"></a> -<span class="sourceLineNo">1270</span> regionStates.addToOfflineRegions(regionNode);<a name="line.1270"></a> -<span class="sourceLineNo">1271</span> } else if (localState == State.CLOSED && getTableStateManager().<a name="line.1271"></a> -<span class="sourceLineNo">1272</span> isTableState(regionNode.getTable(), TableState.State.DISABLED,<a name="line.1272"></a> -<span class="sourceLineNo">1273</span> TableState.State.DISABLING)) {<a name="line.1273"></a> -<span class="sourceLineNo">1274</span> // The region is CLOSED and the table is DISABLED/ DISABLING, there is nothing to<a name="line.1274"></a> -<span class="sourceLineNo">1275</span> // schedule; the region is inert.<a name="line.1275"></a> -<span class="sourceLineNo">1276</span> } else {<a name="line.1276"></a> -<span class="sourceLineNo">1277</span> // These regions should have a procedure in replay<a name="line.1277"></a> -<span class="sourceLineNo">1278</span> regionStates.addRegionInTransition(regionNode, null);<a name="line.1278"></a> -<span class="sourceLineNo">1279</span> }<a name="line.1279"></a> -<span class="sourceLineNo">1280</span> }<a name="line.1280"></a> -<span class="sourceLineNo">1281</span> }<a name="line.1281"></a> -<span class="sourceLineNo">1282</span> }<a name="line.1282"></a> -<span class="sourceLineNo">1283</span> });<a name="line.1283"></a> -<span class="sourceLineNo">1284</span><a name="line.1284"></a> -<span class="sourceLineNo">1285</span> // every assignment is blocked until meta is loaded.<a name="line.1285"></a> -<span class="sourceLineNo">1286</span> wakeMetaLoadedEvent();<a name="line.1286"></a> -<span class="sourceLineNo">1287</span> }<a name="line.1287"></a> -<span class="sourceLineNo">1288</span><a name="line.1288"></a> -<span class="sourceLineNo">1289</span> /**<a name="line.1289"></a> -<span class="sourceLineNo">1290</span> * Look at what is in meta and the list of servers that have checked in and make reconciliation.<a name="line.1290"></a> -<span class="sourceLineNo">1291</span> * We cannot tell definitively the difference between a clean shutdown and a cluster that has<a name="line.1291"></a> -<span class="sourceLineNo">1292</span> * been crashed down. At this stage of a Master startup, they look the same: they have the<a name="line.1292"></a> -<span class="sourceLineNo">1293</span> * same state in hbase:meta. We could do detective work probing ZK and the FS for old WALs to<a name="line.1293"></a> -<span class="sourceLineNo">1294</span> * split but SCP does this already so just let it do its job.<a name="line.1294"></a> -<span class="sourceLineNo">1295</span> * <p>>The profiles of clean shutdown and cluster crash-down are the same because on clean<a name="line.1295"></a> -<span class="sourceLineNo">1296</span> * shutdown currently, we do not update hbase:meta with region close state (In AMv2, region<a name="line.1296"></a> -<span class="sourceLineNo">1297</span> * state is kept in hbse:meta). Usually the master runs all region transitions as of AMv2 but on<a name="line.1297"></a> -<span class="sourceLineNo">1298</span> * cluster controlled shutdown, the RegionServers close all their regions only reporting the<a name="line.1298"></a> -<span class="sourceLineNo">1299</span> * final change to the Master. Currently this report is ignored. Later we could take it and<a name="line.1299"></a> -<span class="sourceLineNo">1300</span> * update as many regions as we can before hbase:meta goes down or have the master run the<a name="line.1300"></a> -<span class="sourceLineNo">1301</span> * close of all regions out on the cluster but we may never be able to achieve the proper state on<a name="line.1301"></a> -<span class="sourceLineNo">1302</span> * all regions (at least not w/o lots of painful manipulations and waiting) so clean shutdown<a name="line.1302"></a> -<span class="sourceLineNo">1303</span> * might not be possible especially on big clusters.... And clean shutdown will take time. Given<a name="line.1303"></a> -<span class="sourceLineNo">1304</span> * this current state of affairs, we just run ServerCrashProcedure in both cases. It will always<a name="line.1304"></a> -<span class="sourceLineNo">1305</span> * do the right thing.<a name="line.1305"></a> -<span class="sourceLineNo">1306</span> * @return True if for sure this is a failover where a Master is starting up into an already<a name="line.1306"></a> -<span class="sourceLineNo">1307</span> * running cluster.<a name="line.1307"></a> -<span class="sourceLineNo">1308</span> */<a name="line.1308"></a> -<span class="sourceLineNo">1309</span> // The assumption here is that if RSs are crashing while we are executing this<a name="line.1309"></a> -<span class="sourceLineNo">1310</span> // they will be handled by the SSH that are put in the ServerManager deadservers "queue".<a name="line.1310"></a> -<span class="sourceLineNo">1311</span> private boolean processofflineServersWithOnlineRegions() {<a name="line.1311"></a> -<span class="sourceLineNo">1312</span> boolean deadServers = !master.getServerManager().getDeadServers().isEmpty();<a name="line.1312"></a> -<span class="sourceLineNo">1313</span> final Set<ServerName> offlineServersWithOnlineRegions = new HashSet<>();<a name="line.1313"></a> -<span class="sourceLineNo">1314</span> int size = regionStates.getRegionStateNodes().size();<a name="line.1314"></a> -<span class="sourceLineNo">1315</span> final List<RegionInfo> offlineRegionsToAssign = new ArrayList<>(size);<a name="line.1315"></a> -<span class="sourceLineNo">1316</span> // If deadservers then its a failover, else, we are not sure yet.<a name="line.1316"></a> -<span class="sourceLineNo">1317</span> boolean failover = deadServers;<a name="line.1317"></a> -<span class="sourceLineNo">1318</span> for (RegionStateNode regionNode: regionStates.getRegionStateNodes()) {<a name="line.1318"></a> -<span class="sourceLineNo">1319</span> // Region State can be OPEN even if we did controlled cluster shutdown; Master does not close<a name="line.1319"></a> -<span class="sourceLineNo">1320</span> // the regions in this case. The RegionServer does the close so hbase:meta is state in<a name="line.1320"></a> -<span class="sourceLineNo">1321</span> // hbase:meta is not updated -- Master does all updates -- and is left with OPEN as region<a name="line.1321"></a> -<span class="sourceLineNo">1322</span> // state in meta. How to tell difference between ordered shutdown and crashed-down cluster<a name="line.1322"></a> -<span class="sourceLineNo">1323</span> // then? We can't. Not currently. Perhaps if we updated hbase:meta with CLOSED on ordered<a name="line.1323"></a> -<span class="sourceLineNo">1324</span> // shutdown. This would slow shutdown though and not all edits would make it in anyways.<a name="line.1324"></a> -<span class="sourceLineNo">1325</span> // TODO: Examine.<a name="line.1325"></a> -<span class="sourceLineNo">1326</span> // Because we can't be sure it an ordered shutdown, we run ServerCrashProcedure always.<a name="line.1326"></a> -<span class="sourceLineNo">1327</span> // ServerCrashProcedure will try to retain old deploy when it goes to assign.<a name="line.1327"></a> -<span class="sourceLineNo">1328</span> if (regionNode.getState() == State.OPEN) {<a name="line.1328"></a> -<span class="sourceLineNo">1329</span> final ServerName serverName = regionNode.getRegionLocation();<a name="line.1329"></a> -<span class="sourceLineNo">1330</span> if (!master.getServerManager().isServerOnline(serverName)) {<a name="line.1330"></a> -<span class="sourceLineNo">1331</span> offlineServersWithOnlineRegions.add(serverName);<a name="line.1331"></a> -<span class="sourceLineNo">1332</span> } else {<a name="line.1332"></a> -<span class="sourceLineNo">1333</span> // Server is online. This a failover. Master is starting into already-running cluster.<a name="line.1333"></a> -<span class="sourceLineNo">1334</span> failover = true;<a name="line.1334"></a> -<span class="sourceLineNo">1335</span> }<a name="line.1335"></a> -<span class="sourceLineNo">1336</span> } else if (regionNode.getState() == State.OFFLINE) {<a name="line.1336"></a> -<span class="sourceLineNo">1337</span> if (isTableEnabled(regionNode.getTable())) {<a name="line.1337"></a> -<span class="sourceLineNo">1338</span> offlineRegionsToAssign.add(regionNode.getRegionInfo());<a name="line.1338"></a> -<span class="sourceLineNo">1339</span> }<a name="line.1339"></a> -<span class="sourceLineNo">1340</span> }<a name="line.1340"></a> -<span class="sourceLineNo">1341</span> }<a name="line.1341"></a> -<span class="sourceLineNo">1342</span> // Kill servers with online regions just-in-case. Runs ServerCrashProcedure.<a name="line.1342"></a> -<span class="sourceLineNo">1343</span> for (ServerName serverName: offlineServersWithOnlineRegions) {<a name="line.1343"></a> -<span class="sourceLineNo">1344</span> if (!master.getServerManager().isServerOnline(serverName)) {<a name="line.1344"></a> -<span class="sourceLineNo">1345</span> LOG.info("KILL RegionServer=" + serverName + " hosting regions but not online.");<a name="line.1345"></a> -<span class="sourceLineNo">1346</span> killRegionServer(serverName);<a name="line.1346"></a> -<span class="sourceLineNo">1347</span> }<a name="line.1347"></a> -<span class="sourceLineNo">1348</span> }<a name="line.1348"></a> -<span class="sourceLineNo">1349</span> setFailoverCleanupDone(true);<a name="line.1349"></a> -<span class="sourceLineNo">1350</span><a name="line.1350"></a> -<span class="sourceLineNo">1351</span> // Assign offline regions. Uses round-robin.<a name="line.1351"></a> -<span class="sourceLineNo">1352</span> if (offlineRegionsToAssign.size() > 0) {<a name="line.1352"></a> -<span class="sourceLineNo">1353</span> master.getMasterProcedureExecutor().submitProcedures(master.getAssignmentManager().<a name="line.1353"></a> -<span class="sourceLineNo">1354</span> createRoundRobinAssignProcedures(offlineRegionsToAssign));<a name="line.1354"></a> -<span class="sourceLineNo">1355</span> }<a name="line.1355"></a> -<span class="sourceLineNo">1356</span><a name="line.1356"></a> -<span class="sourceLineNo">1357</span> return failover;<a name="line.1357"></a> -<span class="sourceLineNo">1358</span> }<a name="line.1358"></a> -<span class="sourceLineNo">1359</span><a name="line.1359"></a> -<span class="sourceLineNo">1360</span> /**<a name="line.1360"></a> -<span class="sourceLineNo">1361</span> * Used by ServerCrashProcedure to make sure AssignmentManager has completed<a name="line.1361"></a> -<span class="sourceLineNo">1362</span> * the failover cleanup before re-assigning regions of dead servers. So that<a name="line.1362"></a> -<span class="sourceLineNo">1363</span> * when re-assignment happens, AssignmentManager has proper region states.<a name="line.1363"></a> -<span class="sourceLineNo">1364</span> */<a name="line.1364"></a> -<span class="sourceLineNo">1365</span> public boolean isFailoverCleanupDone() {<a name="line.1365"></a> -<span class="sourceLineNo">1366</span> return failoverCleanupDone.isReady();<a name="line.1366"></a> -<span class="sourceLineNo">1367</span> }<a name="line.1367"></a> -<span class="sourceLineNo">1368</span><a name="line.1368"></a> -<span class="sourceLineNo">1369</span> /**<a name="line.1369"></a> -<span class="sourceLineNo">1370</span> * Used by ServerCrashProcedure tests verify the ability to suspend the<a name="line.1370"></a> -<span class="sourceLineNo">1371</span> * execution of the ServerCrashProcedure.<a name="line.1371"></a> -<span class="sourceLineNo">1372</span> */<a name="line.1372"></a> -<span class="sourceLineNo">1373</span> @VisibleForTesting<a name="line.1373"></a> -<span class="sourceLineNo">1374</span> public void setFailoverCleanupDone(final boolean b) {<a name="line.1374"></a> -<span class="sourceLineNo">1375</span> master.getMasterProcedureExecutor().getEnvironment()<a name="line.1375"></a> -<span class="sourceLineNo">1376</span> .setEventReady(failoverCleanupDone, b);<a name="line.1376"></a> -<span class="sourceLineNo">1377</span> }<a name="line.1377"></a> -<span class="sourceLineNo">1378</span><a name="line.1378"></a> -<span class="sourceLineNo">1379</span> public ProcedureEvent getFailoverCleanupEvent() {<a name="line.1379"></a> -<span class="sourceLineNo">1380</span> return failoverCleanupDone;<a name="line.1380"></a> -<span class="sourceLineNo">1381</span> }<a name="line.1381"></a> -<span class="sourceLineNo">1382</span><a name="line.1382"></a> -<span class="sourceLineNo">1383</span> /**<a name="line.1383"></a> -<span class="sourceLineNo">1384</span> * Used to check if the failover cleanup is done.<a name="line.1384"></a> -<span class="sourceLineNo">1385</span> * if not we throw PleaseHoldException since we are rebuilding the RegionStates<a name="line.1385"></a> -<span class="sourceLineNo">1386</span> * @param hri region to check if it is already rebuild<a name="line.1386"></a> -<span class="sourceLineNo">1387</span> * @throws PleaseHoldException if the failover cleanup is not completed<a name="line.1387"></a> -<span class="sourceLineNo">1388</span> */<a name="line.1388"></a> -<span class="sourceLineNo">1389</span> private void checkFailoverCleanupCompleted(final RegionInfo hri) throws PleaseHoldException {<a name="line.1389"></a> -<span class="sourceLineNo">1390</span> if (!isRunning()) {<a name="line.1390"></a> -<span class="sourceLineNo">1391</span> throw new PleaseHoldException("AssignmentManager not running");<a name="line.1391"></a> -<span class="sourceLineNo">1392</span> }<a name="line.1392"></a> -<span class="sourceLineNo">1393</span><a name="line.1393"></a> -<span class="sourceLineNo">1394</span> // TODO: can we avoid throwing an exception if hri is already loaded?<a name="line.1394"></a> -<span class="sourceLineNo">1395</span> // at the moment we bypass only meta<a name="line.1395"></a> -<span class="sourceLineNo">1396</span> boolean meta = isMetaRegion(hri);<a name="line.1396"></a> -<span class="sourceLineNo">1397</span> boolean cleanup = isFailoverCleanupDone();<a name="line.1397"></a> -<span class="sourceLineNo">1398</span> if (!isMetaRegion(hri) && !isFailoverCleanupDone()) {<a name="line.1398"></a> -<span class="sourceLineNo">1399</span> String msg = "Master not fully online; hbase:meta=" + meta + ", failoverCleanup=" + cleanup;<a name="line.1399"></a> -<span class="sourceLineNo">1400</span> throw new PleaseHoldException(msg);<a name="line.1400"></a> -<span class="sourceLineNo">1401</span> }<a name="line.1401"></a> -<span class="sourceLineNo">1402</span> }<a name="line.1402"></a> -<span class="sourceLineNo">1403</span><a name="line.1403"></a> +<span class="sourceLineNo">1227</span> // Start the RIT chore<a name="line.1227"></a> +<span class="sourceLineNo">1228</span> master.getMasterProcedureExecutor().addChore(this.ritChore);<a name="line.1228"></a> +<span class="sourceLineNo">1229</span><a name="line.1229"></a> +<span class="sourceLineNo">1230</span> LOG.info(String.format("Joined the cluster in %s, failover=%s",<a name="line.1230"></a> +<span class="sourceLineNo">1231</span> StringUtils.humanTimeDiff(System.currentTimeMillis() - startTime), failover));<a name="line.1231"></a> +<span class="sourceLineNo">1232</span> }<a name="line.1232"></a> +<span class="sourceLineNo">1233</span><a name="line.1233"></a> +<span class="sourceLineNo">1234</span> private void loadMeta() throws IOException {<a name="line.1234"></a> +<span class="sourceLineNo">1235</span> // TODO: use a thread pool<a name="line.1235"></a> +<span class="sourceLineNo">1236</span> regionStateStore.visitMeta(new RegionStateStore.RegionStateVisitor() {<a name="line.1236"></a> +<span class="sourceLineNo">1237</span> @Override<a name="line.1237"></a> +<span class="sourceLineNo">1238</span> public void visitRegionState(Result result, final RegionInfo regionInfo, final State state,<a name="line.1238"></a> +<span class="sourceLineNo">1239</span> final ServerName regionLocation, final ServerName lastHost, final long openSeqNum) {<a name="line.1239"></a> +<span class="sourceLineNo">1240</span> if (state == null && regionLocation == null && lastHost == null &&<a name="line.1240"></a> +<span class="sourceLineNo">1241</span> openSeqNum == SequenceId.NO_SEQUENCE_ID) {<a name="line.1241"></a> +<span class="sourceLineNo">1242</span> // This is a row with nothing in it.<a name="line.1242"></a> +<span class="sourceLineNo">1243</span> LOG.warn("Skipping empty row={}", result);<a name="line.1243"></a> +<span class="sourceLineNo">1244</span> return;<a name="line.1244"></a> +<span class="sourceLineNo">1245</span> }<a name="line.1245"></a> +<span class="sourceLineNo">1246</span> State localState = state;<a name="line.1246"></a> +<span class="sourceLineNo">1247</span> if (localState == null) {<a name="line.1247"></a> +<span class="sourceLineNo">1248</span> // No region state column data in hbase:meta table! Are I doing a rolling upgrade from<a name="line.1248"></a> +<span class="sourceLineNo">1249</span> // hbase1 to hbase2? Am I restoring a SNAPSHOT or otherwise adding a region to hbase:meta?<a name="line.1249"></a> +<span class="sourceLineNo">1250</span> // In any of these cases, state is empty. For now, presume OFFLINE but there are probably<a name="line.1250"></a> +<span class="sourceLineNo">1251</span> // cases where we need to probe more to be sure this correct; TODO informed by experience.<a name="line.1251"></a> +<span class="sourceLineNo">1252</span> LOG.info(regionInfo.getEncodedName() + " regionState=null; presuming " + State.OFFLINE);<a name="line.1252"></a> +<span class="sourceLineNo">1253</span><a name="line.1253"></a> +<span class="sourceLineNo">1254</span> localState = State.OFFLINE;<a name="line.1254"></a> +<span class="sourceLineNo">1255</span> }<a name="line.1255"></a> +<span class="sourceLineNo">1256</span> final RegionStateNode regionNode = regionStates.getOrCreateRegionStateNode(regionInfo);<a name="line.1256"></a> +<span class="sourceLineNo">1257</span> synchronized (regionNode) {<a name="line.1257"></a> +<span class="sourceLineNo">1258</span> if (!regionNode.isInTransition()) {<a name="line.1258"></a> +<span class="sourceLineNo">1259</span> regionNode.setState(localState);<a name="line.1259"></a> +<span class="sourceLineNo">1260</span> regionNode.setLastHost(lastHost);<a name="line.1260"></a> +<span class="sourceLineNo">1261</span> regionNode.setRegionLocation(regionLocation);<a name="line.1261"></a> +<span class="sourceLineNo">1262</span> regionNode.setOpenSeqNum(openSeqNum);<a name="line.1262"></a> +<span class="sourceLineNo">1263</span><a name="line.1263"></a> +<span class="sourceLineNo">1264</span> if (localState == State.OPEN) {<a name="line.1264"></a> +<span class="sourceLineNo">1265</span> assert regionLocation != null : "found null region location for " + regionNode;<a name="line.1265"></a> +<span class="sourceLineNo">1266</span> regionStates.addRegionToServer(regionNode);<a name="line.1266"></a> +<span class="sourceLineNo">1267</span> } else if (localState == State.OFFLINE || regionInfo.isOffline()) {<a name="line.1267"></a> +<span class="sourceLineNo">1268</span> regionStates.addToOfflineRegions(regionNode);<a name="line.1268"></a> +<span class="sourceLineNo">1269</span> } else if (localState == State.CLOSED && getTableStateManager().<a name="line.1269"></a> +<span class="sourceLineNo">1270</span> isTableState(regionNode.getTable(), TableState.State.DISABLED,<a name="line.1270"></a> +<span class="sourceLineNo">1271</span> TableState.State.DISABLING)) {<a name="line.1271"></a> +<span class="sourceLineNo">1272</span> // The region is CLOSED and the table is DISABLED/ DISABLING, there is nothing to<a name="line.1272"></a> +<span class="sourceLineNo">1273</span> // schedule; the region is inert.<a name="line.1273"></a> +<span class="sourceLineNo">1274</span> } else {<a name="line.1274"></a> +<span class="sourceLineNo">1275</span> // These regions should have a procedure in replay<a name="line.1275"></a> +<span class="sourceLineNo">1276</span> regionStates.addRegionInTransition(regionNode, null);<a name="line.1276"></a> +<span class="sourceLineNo">1277</span> }<a name="line.1277"></a> +<span class="sourceLineNo">1278</span> }<a name="line.1278"></a> +<span class="sourceLineNo">1279</span> }<a name="line.1279"></a> +<span class="sourceLineNo">1280</span> }<a name="line.1280"></a> +<span class="sourceLineNo">1281</span> });<a name="line.1281"></a> +<span class="sourceLineNo">1282</span><a name="line.1282"></a> +<span class="sourceLineNo">1283</span> // every assignment is blocked until meta is loaded.<a name="line.1283"></a> +<span class="sourceLineNo">1284</span> wakeMetaLoadedEvent();<a name="line.1284"></a> +<span class="sourceLineNo">1285</span> }<a name="line.1285"></a> +<span class="sourceLineNo">1286</span><a name="line.1286"></a> +<span class="sourceLineNo">1287</span> /**<a name="line.1287"></a> +<span class="sourceLineNo">1288</span> * Look at what is in meta and the list of servers that have checked in and make reconciliation.<a name="line.1288"></a> +<span class="sourceLineNo">1289</span> * We cannot tell definitively the difference between a clean shutdown and a cluster that has<a name="line.1289"></a> +<span class="sourceLineNo">1290</span> * been crashed down. At this stage of a Master startup, they look the same: they have the<a name="line.1290"></a> +<span class="sourceLineNo">1291</span> * same state in hbase:meta. We could do detective work probing ZK and the FS for old WALs to<a name="line.1291"></a> +<span class="sourceLineNo">1292</span> * split but SCP does this already so just let it do its job.<a name="line.1292"></a> +<span class="sourceLineNo">1293</span> * <p>>The profiles of clean shutdown and cluster crash-down are the same because on clean<a name="line.1293"></a> +<span class="sourceLineNo">1294</span> * shutdown currently, we do not update hbase:meta with region close state (In AMv2, region<a name="line.1294"></a> +<span class="sourceLineNo">1295</span> * state is kept in hbse:meta). Usually the master runs all region transitions as of AMv2 but on<a name="line.1295"></a> +<span class="sourceLineNo">1296</span> * cluster controlled shutdown, the RegionServers close all their regions only reporting the<a name="line.1296"></a> +<span class="sourceLineNo">1297</span> * final change to the Master. Currently this report is ignored. Later we could take it and<a name="line.1297"></a> +<span class="sourceLineNo">1298</span> * update as many regions as we can before hbase:meta goes down or have the master run the<a name="line.1298"></a> +<span class="sourceLineNo">1299</span> * close of all regions out on the cluster but we may never be able to achieve the proper state on<a name="line.1299"></a> +<span class="sourceLineNo">1300</span> * all regions (at least not w/o lots of painful manipulations and waiting) so clean shutdown<a name="line.1300"></a> +<span class="sourceLineNo">1301</span> * might not be possible especially on big clusters.... And clean shutdown will take time. Given<a name="line.1301"></a> +<span class="sourceLineNo">1302</span> * this current state of affairs, we just run ServerCrashProcedure in both cases. It will always<a name="line.1302"></a> +<span class="sourceLineNo">1303</span> * do the right thing.<a name="line.1303"></a> +<span class="sourceLineNo">1304</span> * @return True if for sure this is a failover where a Master is starting up into an already<a name="line.1304"></a> +<span class="sourceLineNo">1305</span> * running cluster.<a name="line.1305"></a> +<span class="sourceLineNo">1306</span> */<a name="line.1306"></a> +<span class="sourceLineNo">1307</span> // The assumption here is that if RSs are crashing while we are executing this<a name="line.1307"></a> +<span class="sourceLineNo">1308</span> // they will be handled by the SSH that are put in the ServerManager deadservers "queue".<a name="line.1308"></a> +<span class="sourceLineNo">1309</span> private boolean processofflineServersWithOnlineRegions() {<a name="line.1309"></a> +<span class="sourceLineNo">1310</span> boolean deadServers = !master.getServerManager().getDeadServers().isEmpty();<a name="line.1310"></a> +<span class="sourceLineNo">1311</span> final Set<ServerName> offlineServersWithOnlineRegions = new HashSet<>();<a name="line.1311"></a> +<span class="sourceLineNo">1312</span> int size = regionStates.getRegionStateNodes().size();<a name="line.1312"></a> +<span class="sourceLineNo">1313</span> final List<RegionInfo> offlineRegionsToAssign = new ArrayList<>(size);<a name="line.1313"></a> +<span class="sourceLineNo">1314</span> // If deadservers then its a failover, else, we are not sure yet.<a name="line.1314"></a> +<span class="sourceLineNo">1315</span> boolean failover = deadServers;<a name="line.1315"></a> +<span class="sourceLineNo">1316</span> for (RegionStateNode regionNode: regionStates.getRegionStateNodes()) {<a name="line.1316"></a> +<span class="sourceLineNo">1317</span> // Region State can be OPEN even if we did controlled cluster shutdown; Master does not close<a name="line.1317"></a> +<span class="sourceLineNo">1318</span> // the regions in this case. The RegionServer does the close so hbase:meta is state in<a name="line.1318"></a> +<span class="sourceLineNo">1319</span> // hbase:meta is not updated -- Master does all updates -- a
<TRUNCATED>