http://git-wip-us.apache.org/repos/asf/hbase-site/blob/3649682c/devapidocs/src-html/org/apache/hadoop/hbase/regionserver/HRegionServer.MovedRegionsCleaner.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/regionserver/HRegionServer.MovedRegionsCleaner.html b/devapidocs/src-html/org/apache/hadoop/hbase/regionserver/HRegionServer.MovedRegionsCleaner.html index 90087b4..b2a9771 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/regionserver/HRegionServer.MovedRegionsCleaner.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/regionserver/HRegionServer.MovedRegionsCleaner.html @@ -956,2857 +956,2858 @@ <span class="sourceLineNo">948</span> // Try and register with the Master; tell it we are here. Break if server is stopped or the<a name="line.948"></a> <span class="sourceLineNo">949</span> // clusterup flag is down or hdfs went wacky. Once registered successfully, go ahead and start<a name="line.949"></a> <span class="sourceLineNo">950</span> // up all Services. Use RetryCounter to get backoff in case Master is struggling to come up.<a name="line.950"></a> -<span class="sourceLineNo">951</span> RetryCounterFactory rcf = new RetryCounterFactory(Integer.MAX_VALUE,<a name="line.951"></a> -<span class="sourceLineNo">952</span> this.sleeper.getPeriod(), 1000 * 60 * 5);<a name="line.952"></a> -<span class="sourceLineNo">953</span> RetryCounter rc = rcf.create();<a name="line.953"></a> -<span class="sourceLineNo">954</span> while (keepLooping()) {<a name="line.954"></a> -<span class="sourceLineNo">955</span> RegionServerStartupResponse w = reportForDuty();<a name="line.955"></a> -<span class="sourceLineNo">956</span> if (w == null) {<a name="line.956"></a> -<span class="sourceLineNo">957</span> long sleepTime = rc.getBackoffTimeAndIncrementAttempts();<a name="line.957"></a> -<span class="sourceLineNo">958</span> LOG.warn("reportForDuty failed; sleeping {} ms and then retrying.", sleepTime);<a name="line.958"></a> -<span class="sourceLineNo">959</span> this.sleeper.sleep(sleepTime);<a name="line.959"></a> -<span class="sourceLineNo">960</span> } else {<a name="line.960"></a> -<span class="sourceLineNo">961</span> handleReportForDutyResponse(w);<a name="line.961"></a> -<span class="sourceLineNo">962</span> break;<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> if (!isStopped() && isHealthy()) {<a name="line.966"></a> -<span class="sourceLineNo">967</span> // start the snapshot handler and other procedure handlers,<a name="line.967"></a> -<span class="sourceLineNo">968</span> // since the server is ready to run<a name="line.968"></a> -<span class="sourceLineNo">969</span> if (this.rspmHost != null) {<a name="line.969"></a> -<span class="sourceLineNo">970</span> this.rspmHost.start();<a name="line.970"></a> -<span class="sourceLineNo">971</span> }<a name="line.971"></a> -<span class="sourceLineNo">972</span> // Start the Quota Manager<a name="line.972"></a> -<span class="sourceLineNo">973</span> if (this.rsQuotaManager != null) {<a name="line.973"></a> -<span class="sourceLineNo">974</span> rsQuotaManager.start(getRpcServer().getScheduler());<a name="line.974"></a> -<span class="sourceLineNo">975</span> }<a name="line.975"></a> -<span class="sourceLineNo">976</span> if (this.rsSpaceQuotaManager != null) {<a name="line.976"></a> -<span class="sourceLineNo">977</span> this.rsSpaceQuotaManager.start();<a name="line.977"></a> -<span class="sourceLineNo">978</span> }<a name="line.978"></a> -<span class="sourceLineNo">979</span> }<a name="line.979"></a> -<span class="sourceLineNo">980</span><a name="line.980"></a> -<span class="sourceLineNo">981</span> // We registered with the Master. Go into run mode.<a name="line.981"></a> -<span class="sourceLineNo">982</span> long lastMsg = System.currentTimeMillis();<a name="line.982"></a> -<span class="sourceLineNo">983</span> long oldRequestCount = -1;<a name="line.983"></a> -<span class="sourceLineNo">984</span> // The main run loop.<a name="line.984"></a> -<span class="sourceLineNo">985</span> while (!isStopped() && isHealthy()) {<a name="line.985"></a> -<span class="sourceLineNo">986</span> if (!isClusterUp()) {<a name="line.986"></a> -<span class="sourceLineNo">987</span> if (isOnlineRegionsEmpty()) {<a name="line.987"></a> -<span class="sourceLineNo">988</span> stop("Exiting; cluster shutdown set and not carrying any regions");<a name="line.988"></a> -<span class="sourceLineNo">989</span> } else if (!this.stopping) {<a name="line.989"></a> -<span class="sourceLineNo">990</span> this.stopping = true;<a name="line.990"></a> -<span class="sourceLineNo">991</span> LOG.info("Closing user regions");<a name="line.991"></a> -<span class="sourceLineNo">992</span> closeUserRegions(this.abortRequested);<a name="line.992"></a> -<span class="sourceLineNo">993</span> } else if (this.stopping) {<a name="line.993"></a> -<span class="sourceLineNo">994</span> boolean allUserRegionsOffline = areAllUserRegionsOffline();<a name="line.994"></a> -<span class="sourceLineNo">995</span> if (allUserRegionsOffline) {<a name="line.995"></a> -<span class="sourceLineNo">996</span> // Set stopped if no more write requests tp meta tables<a name="line.996"></a> -<span class="sourceLineNo">997</span> // since last time we went around the loop. Any open<a name="line.997"></a> -<span class="sourceLineNo">998</span> // meta regions will be closed on our way out.<a name="line.998"></a> -<span class="sourceLineNo">999</span> if (oldRequestCount == getWriteRequestCount()) {<a name="line.999"></a> -<span class="sourceLineNo">1000</span> stop("Stopped; only catalog regions remaining online");<a name="line.1000"></a> -<span class="sourceLineNo">1001</span> break;<a name="line.1001"></a> -<span class="sourceLineNo">1002</span> }<a name="line.1002"></a> -<span class="sourceLineNo">1003</span> oldRequestCount = getWriteRequestCount();<a name="line.1003"></a> -<span class="sourceLineNo">1004</span> } else {<a name="line.1004"></a> -<span class="sourceLineNo">1005</span> // Make sure all regions have been closed -- some regions may<a name="line.1005"></a> -<span class="sourceLineNo">1006</span> // have not got it because we were splitting at the time of<a name="line.1006"></a> -<span class="sourceLineNo">1007</span> // the call to closeUserRegions.<a name="line.1007"></a> -<span class="sourceLineNo">1008</span> closeUserRegions(this.abortRequested);<a name="line.1008"></a> -<span class="sourceLineNo">1009</span> }<a name="line.1009"></a> -<span class="sourceLineNo">1010</span> LOG.debug("Waiting on " + getOnlineRegionsAsPrintableString());<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> long now = System.currentTimeMillis();<a name="line.1013"></a> -<span class="sourceLineNo">1014</span> if ((now - lastMsg) >= msgInterval) {<a name="line.1014"></a> -<span class="sourceLineNo">1015</span> tryRegionServerReport(lastMsg, now);<a name="line.1015"></a> -<span class="sourceLineNo">1016</span> lastMsg = System.currentTimeMillis();<a name="line.1016"></a> -<span class="sourceLineNo">1017</span> }<a name="line.1017"></a> -<span class="sourceLineNo">1018</span> if (!isStopped() && !isAborted()) {<a name="line.1018"></a> -<span class="sourceLineNo">1019</span> this.sleeper.sleep();<a name="line.1019"></a> -<span class="sourceLineNo">1020</span> }<a name="line.1020"></a> -<span class="sourceLineNo">1021</span> } // for<a name="line.1021"></a> -<span class="sourceLineNo">1022</span> } catch (Throwable t) {<a name="line.1022"></a> -<span class="sourceLineNo">1023</span> if (!rpcServices.checkOOME(t)) {<a name="line.1023"></a> -<span class="sourceLineNo">1024</span> String prefix = t instanceof YouAreDeadException? "": "Unhandled: ";<a name="line.1024"></a> -<span class="sourceLineNo">1025</span> abort(prefix + t.getMessage(), t);<a name="line.1025"></a> -<span class="sourceLineNo">1026</span> }<a name="line.1026"></a> -<span class="sourceLineNo">1027</span> }<a name="line.1027"></a> -<span class="sourceLineNo">1028</span> if (this.leases != null) {<a name="line.1028"></a> -<span class="sourceLineNo">1029</span> this.leases.closeAfterLeasesExpire();<a name="line.1029"></a> -<span class="sourceLineNo">1030</span> }<a name="line.1030"></a> -<span class="sourceLineNo">1031</span> if (this.splitLogWorker != null) {<a name="line.1031"></a> -<span class="sourceLineNo">1032</span> splitLogWorker.stop();<a name="line.1032"></a> -<span class="sourceLineNo">1033</span> }<a name="line.1033"></a> -<span class="sourceLineNo">1034</span> if (this.infoServer != null) {<a name="line.1034"></a> -<span class="sourceLineNo">1035</span> LOG.info("Stopping infoServer");<a name="line.1035"></a> -<span class="sourceLineNo">1036</span> try {<a name="line.1036"></a> -<span class="sourceLineNo">1037</span> this.infoServer.stop();<a name="line.1037"></a> -<span class="sourceLineNo">1038</span> } catch (Exception e) {<a name="line.1038"></a> -<span class="sourceLineNo">1039</span> LOG.error("Failed to stop infoServer", e);<a name="line.1039"></a> -<span class="sourceLineNo">1040</span> }<a name="line.1040"></a> -<span class="sourceLineNo">1041</span> }<a name="line.1041"></a> -<span class="sourceLineNo">1042</span> // Send cache a shutdown.<a name="line.1042"></a> -<span class="sourceLineNo">1043</span> if (cacheConfig != null && cacheConfig.isBlockCacheEnabled()) {<a name="line.1043"></a> -<span class="sourceLineNo">1044</span> cacheConfig.getBlockCache().shutdown();<a name="line.1044"></a> -<span class="sourceLineNo">1045</span> }<a name="line.1045"></a> -<span class="sourceLineNo">1046</span> mobCacheConfig.getMobFileCache().shutdown();<a name="line.1046"></a> -<span class="sourceLineNo">1047</span><a name="line.1047"></a> -<span class="sourceLineNo">1048</span> if (movedRegionsCleaner != null) {<a name="line.1048"></a> -<span class="sourceLineNo">1049</span> movedRegionsCleaner.stop("Region Server stopping");<a name="line.1049"></a> -<span class="sourceLineNo">1050</span> }<a name="line.1050"></a> -<span class="sourceLineNo">1051</span><a name="line.1051"></a> -<span class="sourceLineNo">1052</span> // Send interrupts to wake up threads if sleeping so they notice shutdown.<a name="line.1052"></a> -<span class="sourceLineNo">1053</span> // TODO: Should we check they are alive? If OOME could have exited already<a name="line.1053"></a> -<span class="sourceLineNo">1054</span> if (this.hMemManager != null) this.hMemManager.stop();<a name="line.1054"></a> -<span class="sourceLineNo">1055</span> if (this.cacheFlusher != null) this.cacheFlusher.interruptIfNecessary();<a name="line.1055"></a> -<span class="sourceLineNo">1056</span> if (this.compactSplitThread != null) this.compactSplitThread.interruptIfNecessary();<a name="line.1056"></a> -<span class="sourceLineNo">1057</span> sendShutdownInterrupt();<a name="line.1057"></a> -<span class="sourceLineNo">1058</span><a name="line.1058"></a> -<span class="sourceLineNo">1059</span> // Stop the snapshot and other procedure handlers, forcefully killing all running tasks<a name="line.1059"></a> -<span class="sourceLineNo">1060</span> if (rspmHost != null) {<a name="line.1060"></a> -<span class="sourceLineNo">1061</span> rspmHost.stop(this.abortRequested || this.killed);<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> if (this.killed) {<a name="line.1064"></a> -<span class="sourceLineNo">1065</span> // Just skip out w/o closing regions. Used when testing.<a name="line.1065"></a> -<span class="sourceLineNo">1066</span> } else if (abortRequested) {<a name="line.1066"></a> -<span class="sourceLineNo">1067</span> if (this.fsOk) {<a name="line.1067"></a> -<span class="sourceLineNo">1068</span> closeUserRegions(abortRequested); // Don't leave any open file handles<a name="line.1068"></a> -<span class="sourceLineNo">1069</span> }<a name="line.1069"></a> -<span class="sourceLineNo">1070</span> LOG.info("aborting server " + this.serverName);<a name="line.1070"></a> -<span class="sourceLineNo">1071</span> } else {<a name="line.1071"></a> -<span class="sourceLineNo">1072</span> closeUserRegions(abortRequested);<a name="line.1072"></a> -<span class="sourceLineNo">1073</span> LOG.info("stopping server " + this.serverName);<a name="line.1073"></a> -<span class="sourceLineNo">1074</span> }<a name="line.1074"></a> -<span class="sourceLineNo">1075</span><a name="line.1075"></a> -<span class="sourceLineNo">1076</span> // so callers waiting for meta without timeout can stop<a name="line.1076"></a> -<span class="sourceLineNo">1077</span> if (this.metaTableLocator != null) this.metaTableLocator.stop();<a name="line.1077"></a> -<span class="sourceLineNo">1078</span> if (this.clusterConnection != null && !clusterConnection.isClosed()) {<a name="line.1078"></a> -<span class="sourceLineNo">1079</span> try {<a name="line.1079"></a> -<span class="sourceLineNo">1080</span> this.clusterConnection.close();<a name="line.1080"></a> -<span class="sourceLineNo">1081</span> } catch (IOException e) {<a name="line.1081"></a> -<span class="sourceLineNo">1082</span> // Although the {@link Closeable} interface throws an {@link<a name="line.1082"></a> -<span class="sourceLineNo">1083</span> // IOException}, in reality, the implementation would never do that.<a name="line.1083"></a> -<span class="sourceLineNo">1084</span> LOG.warn("Attempt to close server's short circuit ClusterConnection failed.", e);<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><a name="line.1087"></a> -<span class="sourceLineNo">1088</span> // Closing the compactSplit thread before closing meta regions<a name="line.1088"></a> -<span class="sourceLineNo">1089</span> if (!this.killed && containsMetaTableRegions()) {<a name="line.1089"></a> -<span class="sourceLineNo">1090</span> if (!abortRequested || this.fsOk) {<a name="line.1090"></a> -<span class="sourceLineNo">1091</span> if (this.compactSplitThread != null) {<a name="line.1091"></a> -<span class="sourceLineNo">1092</span> this.compactSplitThread.join();<a name="line.1092"></a> -<span class="sourceLineNo">1093</span> this.compactSplitThread = null;<a name="line.1093"></a> -<span class="sourceLineNo">1094</span> }<a name="line.1094"></a> -<span class="sourceLineNo">1095</span> closeMetaTableRegions(abortRequested);<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><a name="line.1098"></a> -<span class="sourceLineNo">1099</span> if (!this.killed && this.fsOk) {<a name="line.1099"></a> -<span class="sourceLineNo">1100</span> waitOnAllRegionsToClose(abortRequested);<a name="line.1100"></a> -<span class="sourceLineNo">1101</span> LOG.info("stopping server " + this.serverName + "; all regions closed.");<a name="line.1101"></a> -<span class="sourceLineNo">1102</span> }<a name="line.1102"></a> -<span class="sourceLineNo">1103</span><a name="line.1103"></a> -<span class="sourceLineNo">1104</span> // Stop the quota manager<a name="line.1104"></a> -<span class="sourceLineNo">1105</span> if (rsQuotaManager != null) {<a name="line.1105"></a> -<span class="sourceLineNo">1106</span> rsQuotaManager.stop();<a name="line.1106"></a> -<span class="sourceLineNo">1107</span> }<a name="line.1107"></a> -<span class="sourceLineNo">1108</span> if (rsSpaceQuotaManager != null) {<a name="line.1108"></a> -<span class="sourceLineNo">1109</span> rsSpaceQuotaManager.stop();<a name="line.1109"></a> -<span class="sourceLineNo">1110</span> rsSpaceQuotaManager = null;<a name="line.1110"></a> -<span class="sourceLineNo">1111</span> }<a name="line.1111"></a> -<span class="sourceLineNo">1112</span><a name="line.1112"></a> -<span class="sourceLineNo">1113</span> //fsOk flag may be changed when closing regions throws exception.<a name="line.1113"></a> -<span class="sourceLineNo">1114</span> if (this.fsOk) {<a name="line.1114"></a> -<span class="sourceLineNo">1115</span> shutdownWAL(!abortRequested);<a name="line.1115"></a> -<span class="sourceLineNo">1116</span> }<a name="line.1116"></a> -<span class="sourceLineNo">1117</span><a name="line.1117"></a> -<span class="sourceLineNo">1118</span> // Make sure the proxy is down.<a name="line.1118"></a> -<span class="sourceLineNo">1119</span> if (this.rssStub != null) {<a name="line.1119"></a> -<span class="sourceLineNo">1120</span> this.rssStub = null;<a name="line.1120"></a> -<span class="sourceLineNo">1121</span> }<a name="line.1121"></a> -<span class="sourceLineNo">1122</span> if (this.lockStub != null) {<a name="line.1122"></a> -<span class="sourceLineNo">1123</span> this.lockStub = null;<a name="line.1123"></a> -<span class="sourceLineNo">1124</span> }<a name="line.1124"></a> -<span class="sourceLineNo">1125</span> if (this.rpcClient != null) {<a name="line.1125"></a> -<span class="sourceLineNo">1126</span> this.rpcClient.close();<a name="line.1126"></a> -<span class="sourceLineNo">1127</span> }<a name="line.1127"></a> -<span class="sourceLineNo">1128</span> if (this.leases != null) {<a name="line.1128"></a> -<span class="sourceLineNo">1129</span> this.leases.close();<a name="line.1129"></a> -<span class="sourceLineNo">1130</span> }<a name="line.1130"></a> -<span class="sourceLineNo">1131</span> if (this.pauseMonitor != null) {<a name="line.1131"></a> -<span class="sourceLineNo">1132</span> this.pauseMonitor.stop();<a name="line.1132"></a> -<span class="sourceLineNo">1133</span> }<a name="line.1133"></a> -<span class="sourceLineNo">1134</span><a name="line.1134"></a> -<span class="sourceLineNo">1135</span> if (!killed) {<a name="line.1135"></a> -<span class="sourceLineNo">1136</span> stopServiceThreads();<a name="line.1136"></a> -<span class="sourceLineNo">1137</span> }<a name="line.1137"></a> -<span class="sourceLineNo">1138</span><a name="line.1138"></a> -<span class="sourceLineNo">1139</span> if (this.rpcServices != null) {<a name="line.1139"></a> -<span class="sourceLineNo">1140</span> this.rpcServices.stop();<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> try {<a name="line.1143"></a> -<span class="sourceLineNo">1144</span> deleteMyEphemeralNode();<a name="line.1144"></a> -<span class="sourceLineNo">1145</span> } catch (KeeperException.NoNodeException nn) {<a name="line.1145"></a> -<span class="sourceLineNo">1146</span> } catch (KeeperException e) {<a name="line.1146"></a> -<span class="sourceLineNo">1147</span> LOG.warn("Failed deleting my ephemeral node", e);<a name="line.1147"></a> -<span class="sourceLineNo">1148</span> }<a name="line.1148"></a> -<span class="sourceLineNo">1149</span> // We may have failed to delete the znode at the previous step, but<a name="line.1149"></a> -<span class="sourceLineNo">1150</span> // we delete the file anyway: a second attempt to delete the znode is likely to fail again.<a name="line.1150"></a> -<span class="sourceLineNo">1151</span> ZNodeClearer.deleteMyEphemeralNodeOnDisk();<a name="line.1151"></a> -<span class="sourceLineNo">1152</span><a name="line.1152"></a> -<span class="sourceLineNo">1153</span> if (this.zooKeeper != null) {<a name="line.1153"></a> -<span class="sourceLineNo">1154</span> this.zooKeeper.close();<a name="line.1154"></a> -<span class="sourceLineNo">1155</span> }<a name="line.1155"></a> -<span class="sourceLineNo">1156</span> this.shutDown = true;<a name="line.1156"></a> -<span class="sourceLineNo">1157</span> LOG.info("Exiting; stopping=" + this.serverName + "; zookeeper connection closed.");<a name="line.1157"></a> -<span class="sourceLineNo">1158</span> }<a name="line.1158"></a> -<span class="sourceLineNo">1159</span><a name="line.1159"></a> -<span class="sourceLineNo">1160</span> private boolean containsMetaTableRegions() {<a name="line.1160"></a> -<span class="sourceLineNo">1161</span> return onlineRegions.containsKey(RegionInfoBuilder.FIRST_META_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> private boolean areAllUserRegionsOffline() {<a name="line.1164"></a> -<span class="sourceLineNo">1165</span> if (getNumberOfOnlineRegions() > 2) return false;<a name="line.1165"></a> -<span class="sourceLineNo">1166</span> boolean allUserRegionsOffline = true;<a name="line.1166"></a> -<span class="sourceLineNo">1167</span> for (Map.Entry<String, HRegion> e: this.onlineRegions.entrySet()) {<a name="line.1167"></a> -<span class="sourceLineNo">1168</span> if (!e.getValue().getRegionInfo().isMetaRegion()) {<a name="line.1168"></a> -<span class="sourceLineNo">1169</span> allUserRegionsOffline = false;<a name="line.1169"></a> -<span class="sourceLineNo">1170</span> break;<a name="line.1170"></a> -<span class="sourceLineNo">1171</span> }<a name="line.1171"></a> -<span class="sourceLineNo">1172</span> }<a name="line.1172"></a> -<span class="sourceLineNo">1173</span> return allUserRegionsOffline;<a name="line.1173"></a> -<span class="sourceLineNo">1174</span> }<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> * @return Current write count for all online regions.<a name="line.1177"></a> -<span class="sourceLineNo">1178</span> */<a name="line.1178"></a> -<span class="sourceLineNo">1179</span> private long getWriteRequestCount() {<a name="line.1179"></a> -<span class="sourceLineNo">1180</span> long writeCount = 0;<a name="line.1180"></a> -<span class="sourceLineNo">1181</span> for (Map.Entry<String, HRegion> e: this.onlineRegions.entrySet()) {<a name="line.1181"></a> -<span class="sourceLineNo">1182</span> writeCount += e.getValue().getWriteRequestsCount();<a name="line.1182"></a> -<span class="sourceLineNo">1183</span> }<a name="line.1183"></a> -<span class="sourceLineNo">1184</span> return writeCount;<a name="line.1184"></a> -<span class="sourceLineNo">1185</span> }<a name="line.1185"></a> -<span class="sourceLineNo">1186</span><a name="line.1186"></a> -<span class="sourceLineNo">1187</span> @VisibleForTesting<a name="line.1187"></a> -<span class="sourceLineNo">1188</span> protected void tryRegionServerReport(long reportStartTime, long reportEndTime)<a name="line.1188"></a> -<span class="sourceLineNo">1189</span> throws IOException {<a name="line.1189"></a> -<span class="sourceLineNo">1190</span> RegionServerStatusService.BlockingInterface rss = rssStub;<a name="line.1190"></a> -<span class="sourceLineNo">1191</span> if (rss == null) {<a name="line.1191"></a> -<span class="sourceLineNo">1192</span> // the current server could be stopping.<a name="line.1192"></a> -<span class="sourceLineNo">1193</span> return;<a name="line.1193"></a> -<span class="sourceLineNo">1194</span> }<a name="line.1194"></a> -<span class="sourceLineNo">1195</span> ClusterStatusProtos.ServerLoad sl = buildServerLoad(reportStartTime, reportEndTime);<a name="line.1195"></a> -<span class="sourceLineNo">1196</span> try {<a name="line.1196"></a> -<span class="sourceLineNo">1197</span> RegionServerReportRequest.Builder request = RegionServerReportRequest.newBuilder();<a name="line.1197"></a> -<span class="sourceLineNo">1198</span> request.setServer(ProtobufUtil.toServerName(this.serverName));<a name="line.1198"></a> -<span class="sourceLineNo">1199</span> request.setLoad(sl);<a name="line.1199"></a> -<span class="sourceLineNo">1200</span> rss.regionServerReport(null, request.build());<a name="line.1200"></a> -<span class="sourceLineNo">1201</span> } catch (ServiceException se) {<a name="line.1201"></a> -<span class="sourceLineNo">1202</span> IOException ioe = ProtobufUtil.getRemoteException(se);<a name="line.1202"></a> -<span class="sourceLineNo">1203</span> if (ioe instanceof YouAreDeadException) {<a name="line.1203"></a> -<span class="sourceLineNo">1204</span> // This will be caught and handled as a fatal error in run()<a name="line.1204"></a> -<span class="sourceLineNo">1205</span> throw ioe;<a name="line.1205"></a> -<span class="sourceLineNo">1206</span> }<a name="line.1206"></a> -<span class="sourceLineNo">1207</span> if (rssStub == rss) {<a name="line.1207"></a> -<span class="sourceLineNo">1208</span> rssStub = null;<a name="line.1208"></a> -<span class="sourceLineNo">1209</span> }<a name="line.1209"></a> -<span class="sourceLineNo">1210</span> // Couldn't connect to the master, get location from zk and reconnect<a name="line.1210"></a> -<span class="sourceLineNo">1211</span> // Method blocks until new master is found or we are stopped<a name="line.1211"></a> -<span class="sourceLineNo">1212</span> createRegionServerStatusStub(true);<a name="line.1212"></a> -<span class="sourceLineNo">1213</span> }<a name="line.1213"></a> -<span class="sourceLineNo">1214</span> }<a name="line.1214"></a> -<span class="sourceLineNo">1215</span><a name="line.1215"></a> -<span class="sourceLineNo">1216</span> /**<a name="line.1216"></a> -<span class="sourceLineNo">1217</span> * Reports the given map of Regions and their size on the filesystem to the active Master.<a name="line.1217"></a> -<span class="sourceLineNo">1218</span> *<a name="line.1218"></a> -<span class="sourceLineNo">1219</span> * @param regionSizeStore The store containing region sizes<a name="line.1219"></a> -<span class="sourceLineNo">1220</span> * @return false if FileSystemUtilizationChore should pause reporting to master. true otherwise<a name="line.1220"></a> -<span class="sourceLineNo">1221</span> */<a name="line.1221"></a> -<span class="sourceLineNo">1222</span> public boolean reportRegionSizesForQuotas(RegionSizeStore regionSizeStore) {<a name="line.1222"></a> -<span class="sourceLineNo">1223</span> RegionServerStatusService.BlockingInterface rss = rssStub;<a name="line.1223"></a> -<span class="sourceLineNo">1224</span> if (rss == null) {<a name="line.1224"></a> -<span class="sourceLineNo">1225</span> // the current server could be stopping.<a name="line.1225"></a> -<span class="sourceLineNo">1226</span> LOG.trace("Skipping Region size report to HMaster as stub is null");<a name="line.1226"></a> -<span class="sourceLineNo">1227</span> return true;<a name="line.1227"></a> -<span class="sourceLineNo">1228</span> }<a name="line.1228"></a> -<span class="sourceLineNo">1229</span> try {<a name="line.1229"></a> -<span class="sourceLineNo">1230</span> buildReportAndSend(rss, regionSizeStore);<a name="line.1230"></a> -<span class="sourceLineNo">1231</span> } catch (ServiceException se) {<a name="line.1231"></a> -<span class="sourceLineNo">1232</span> IOException ioe = ProtobufUtil.getRemoteException(se);<a name="line.1232"></a> -<span class="sourceLineNo">1233</span> if (ioe instanceof PleaseHoldException) {<a name="line.1233"></a> -<span class="sourceLineNo">1234</span> LOG.trace("Failed to report region sizes to Master because it is initializing."<a name="line.1234"></a> -<span class="sourceLineNo">1235</span> + " This will be retried.", ioe);<a name="line.1235"></a> -<span class="sourceLineNo">1236</span> // The Master is coming up. Will retry the report later. Avoid re-creating the stub.<a name="line.1236"></a> -<span class="sourceLineNo">1237</span> return true;<a name="line.1237"></a> -<span class="sourceLineNo">1238</span> }<a name="line.1238"></a> -<span class="sourceLineNo">1239</span> if (rssStub == rss) {<a name="line.1239"></a> -<span class="sourceLineNo">1240</span> rssStub = null;<a name="line.1240"></a> -<span class="sourceLineNo">1241</span> }<a name="line.1241"></a> -<span class="sourceLineNo">1242</span> createRegionServerStatusStub(true);<a name="line.1242"></a> -<span class="sourceLineNo">1243</span> if (ioe instanceof DoNotRetryIOException) {<a name="line.1243"></a> -<span class="sourceLineNo">1244</span> DoNotRetryIOException doNotRetryEx = (DoNotRetryIOException) ioe;<a name="line.1244"></a> -<span class="sourceLineNo">1245</span> if (doNotRetryEx.getCause() != null) {<a name="line.1245"></a> -<span class="sourceLineNo">1246</span> Throwable t = doNotRetryEx.getCause();<a name="line.1246"></a> -<span class="sourceLineNo">1247</span> if (t instanceof UnsupportedOperationException) {<a name="line.1247"></a> -<span class="sourceLineNo">1248</span> LOG.debug("master doesn't support ReportRegionSpaceUse, pause before retrying");<a name="line.1248"></a> -<span class="sourceLineNo">1249</span> return false;<a name="line.1249"></a> -<span class="sourceLineNo">1250</span> }<a name="line.1250"></a> -<span class="sourceLineNo">1251</span> }<a name="line.1251"></a> -<span class="sourceLineNo">1252</span> }<a name="line.1252"></a> -<span class="sourceLineNo">1253</span> LOG.debug("Failed to report region sizes to Master. This will be retried.", ioe);<a name="line.1253"></a> -<span class="sourceLineNo">1254</span> }<a name="line.1254"></a> -<span class="sourceLineNo">1255</span> return true;<a name="line.1255"></a> -<span class="sourceLineNo">1256</span> }<a name="line.1256"></a> -<span class="sourceLineNo">1257</span><a name="line.1257"></a> -<span class="sourceLineNo">1258</span> /**<a name="line.1258"></a> -<span class="sourceLineNo">1259</span> * Builds the region size report and sends it to the master. Upon successful sending of the<a name="line.1259"></a> -<span class="sourceLineNo">1260</span> * report, the region sizes that were sent are marked as sent.<a name="line.1260"></a> -<span class="sourceLineNo">1261</span> *<a name="line.1261"></a> -<span class="sourceLineNo">1262</span> * @param rss The stub to send to the Master<a name="line.1262"></a> -<span class="sourceLineNo">1263</span> * @param regionSizeStore The store containing region sizes<a name="line.1263"></a> -<span class="sourceLineNo">1264</span> */<a name="line.1264"></a> -<span class="sourceLineNo">1265</span> void buildReportAndSend(RegionServerStatusService.BlockingInterface rss,<a name="line.1265"></a> -<span class="sourceLineNo">1266</span> RegionSizeStore regionSizeStore) throws ServiceException {<a name="line.1266"></a> -<span class="sourceLineNo">1267</span> RegionSpaceUseReportRequest request =<a name="line.1267"></a> -<span class="sourceLineNo">1268</span> buildRegionSpaceUseReportRequest(Objects.requireNonNull(regionSizeStore));<a name="line.1268"></a> -<span class="sourceLineNo">1269</span> rss.reportRegionSpaceUse(null, request);<a name="line.1269"></a> -<span class="sourceLineNo">1270</span> // Record the number of size reports sent<a name="line.1270"></a> -<span class="sourceLineNo">1271</span> if (metricsRegionServer != null) {<a name="line.1271"></a> -<span class="sourceLineNo">1272</span> metricsRegionServer.incrementNumRegionSizeReportsSent(regionSizeStore.size());<a name="line.1272"></a> -<span class="sourceLineNo">1273</span> }<a name="line.1273"></a> -<span class="sourceLineNo">1274</span> }<a name="line.1274"></a> -<span class="sourceLineNo">1275</span><a name="line.1275"></a> -<span class="sourceLineNo">1276</span> /**<a name="line.1276"></a> -<span class="sourceLineNo">1277</span> * Builds a {@link RegionSpaceUseReportRequest} protobuf message from the region size map.<a name="line.1277"></a> -<span class="sourceLineNo">1278</span> *<a name="line.1278"></a> -<span class="sourceLineNo">1279</span> * @param regionSizes The size in bytes of regions<a name="line.1279"></a> -<span class="sourceLineNo">1280</span> * @return The corresponding protocol buffer message.<a name="line.1280"></a> -<span class="sourceLineNo">1281</span> */<a name="line.1281"></a> -<span class="sourceLineNo">1282</span> RegionSpaceUseReportRequest buildRegionSpaceUseReportRequest(RegionSizeStore regionSizes) {<a name="line.1282"></a> -<span class="sourceLineNo">1283</span> RegionSpaceUseReportRequest.Builder request = RegionSpaceUseReportRequest.newBuilder();<a name="line.1283"></a> -<span class="sourceLineNo">1284</span> for (Entry<RegionInfo, RegionSize> entry : regionSizes) {<a name="line.1284"></a> -<span class="sourceLineNo">1285</span> request.addSpaceUse(convertRegionSize(entry.getKey(), entry.getValue().getSize()));<a name="line.1285"></a> -<span class="sourceLineNo">1286</span> }<a name="line.1286"></a> -<span class="sourceLineNo">1287</span> return request.build();<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> /**<a name="line.1290"></a> -<span class="sourceLineNo">1291</span> * Converts a pair of {@link RegionInfo} and {@code long} into a {@link RegionSpaceUse}<a name="line.1291"></a> -<span class="sourceLineNo">1292</span> * protobuf message.<a name="line.1292"></a> -<span class="sourceLineNo">1293</span> *<a name="line.1293"></a> -<span class="sourceLineNo">1294</span> * @param regionInfo The RegionInfo<a name="line.1294"></a> -<span class="sourceLineNo">1295</span> * @param sizeInBytes The size in bytes of the Region<a name="line.1295"></a> -<span class="sourceLineNo">1296</span> * @return The protocol buffer<a name="line.1296"></a> -<span class="sourceLineNo">1297</span> */<a name="line.1297"></a> -<span class="sourceLineNo">1298</span> RegionSpaceUse convertRegionSize(RegionInfo regionInfo, Long sizeInBytes) {<a name="line.1298"></a> -<span class="sourceLineNo">1299</span> return RegionSpaceUse.newBuilder()<a name="line.1299"></a> -<span class="sourceLineNo">1300</span> .setRegionInfo(ProtobufUtil.toRegionInfo(Objects.requireNonNull(regionInfo)))<a name="line.1300"></a> -<span class="sourceLineNo">1301</span> .setRegionSize(Objects.requireNonNull(sizeInBytes))<a name="line.1301"></a> -<span class="sourceLineNo">1302</span> .build();<a name="line.1302"></a> -<span class="sourceLineNo">1303</span> }<a name="line.1303"></a> -<span class="sourceLineNo">1304</span><a name="line.1304"></a> -<span class="sourceLineNo">1305</span> ClusterStatusProtos.ServerLoad buildServerLoad(long reportStartTime, long reportEndTime)<a name="line.1305"></a> -<span class="sourceLineNo">1306</span> throws IOException {<a name="line.1306"></a> -<span class="sourceLineNo">1307</span> // We're getting the MetricsRegionServerWrapper here because the wrapper computes requests<a name="line.1307"></a> -<span class="sourceLineNo">1308</span> // per second, and other metrics As long as metrics are part of ServerLoad it's best to use<a name="line.1308"></a> -<span class="sourceLineNo">1309</span> // the wrapper to compute those numbers in one place.<a name="line.1309"></a> -<span class="sourceLineNo">1310</span> // In the long term most of these should be moved off of ServerLoad and the heart beat.<a name="line.1310"></a> -<span class="sourceLineNo">1311</span> // Instead they should be stored in an HBase table so that external visibility into HBase is<a name="line.1311"></a> -<span class="sourceLineNo">1312</span> // improved; Additionally the load balancer will be able to take advantage of a more complete<a name="line.1312"></a> -<span class="sourceLineNo">1313</span> // history.<a name="line.1313"></a> -<span class="sourceLineNo">1314</span> MetricsRegionServerWrapper regionServerWrapper = metricsRegionServer.getRegionServerWrapper();<a name="line.1314"></a> -<span class="sourceLineNo">1315</span> Collection<HRegion> regions = getOnlineRegionsLocalContext();<a name="line.1315"></a> -<span class="sourceLineNo">1316</span> long usedMemory = -1L;<a name="line.1316"></a> -<span class="sourceLineNo">1317</span> long maxMemory = -1L;<a name="line.1317"></a> -<span class="sourceLineNo">1318</span> final MemoryUsage usage = MemorySizeUtil.safeGetHeapMemoryUsage();<a name="line.1318"></a> -<span class="sourceLineNo">1319</span> if (usage != null) {<a name="line.1319"></a> -<span class="sourceLineNo">1320</span> usedMemory = usage.getUsed();<a name="line.1320"></a> -<span class="sourceLineNo">1321</span> maxMemory = usage.getMax();<a name="line.1321"></a> -<span class="sourceLineNo">1322</span> }<a name="line.1322"></a> -<span class="sourceLineNo">1323</span><a name="line.1323"></a> -<span class="sourceLineNo">1324</span> ClusterStatusProtos.ServerLoad.Builder serverLoad = ClusterStatusProtos.ServerLoad.newBuilder();<a name="line.1324"></a> -<span class="sourceLineNo">1325</span> serverLoad.setNumberOfRequests((int) regionServerWrapper.getRequestsPerSecond());<a name="line.1325"></a> -<span class="sourceLineNo">1326</span> serverLoad.setTotalNumberOfRequests(regionServerWrapper.getTotalRequestCount());<a name="line.1326"></a> -<span class="sourceLineNo">1327</span> serverLoad.setUsedHeapMB((int)(usedMemory / 1024 / 1024));<a name="line.1327"></a> -<span class="sourceLineNo">1328</span> serverLoad.setMaxHeapMB((int) (maxMemory / 1024 / 1024));<a name="line.1328"></a> -<span class="sourceLineNo">1329</span> Set<String> coprocessors = getWAL(null).getCoprocessorHost().getCoprocessors();<a name="line.1329"></a> -<span class="sourceLineNo">1330</span> Builder coprocessorBuilder = Coprocessor.newBuilder();<a name="line.1330"></a> -<span class="sourceLineNo">1331</span> for (String coprocessor : coprocessors) {<a name="line.1331"></a> -<span class="sourceLineNo">1332</span> serverLoad.addCoprocessors(coprocessorBuilder.setName(coprocessor).build());<a name="line.1332"></a> -<span class="sourceLineNo">1333</span> }<a name="line.1333"></a> -<span class="sourceLineNo">1334</span> RegionLoad.Builder regionLoadBldr = RegionLoad.newBuilder();<a name="line.1334"></a> -<span class="sourceLineNo">1335</span> RegionSpecifier.Builder regionSpecifier = RegionSpecifier.newBuilder();<a name="line.1335"></a> -<span class="sourceLineNo">1336</span> for (HRegion region : regions) {<a name="line.1336"></a> -<span class="sourceLineNo">1337</span> if (region.getCoprocessorHost() != null) {<a name="line.1337"></a> -<span class="sourceLineNo">1338</span> Set<String> regionCoprocessors = region.getCoprocessorHost().getCoprocessors();<a name="line.1338"></a> -<span class="sourceLineNo">1339</span> Iterator<String> iterator = regionCoprocessors.iterator();<a name="line.1339"></a> -<span class="sourceLineNo">1340</span> while (iterator.hasNext()) {<a name="line.1340"></a> -<span class="sourceLineNo">1341</span> serverLoad.addCoprocessors(coprocessorBuilder.setName(iterator.next()).build());<a name="line.1341"></a> -<span class="sourceLineNo">1342</span> }<a name="line.1342"></a> -<span class="sourceLineNo">1343</span> }<a name="line.1343"></a> -<span class="sourceLineNo">1344</span> serverLoad.addRegionLoads(createRegionLoad(region, regionLoadBldr, regionSpecifier));<a name="line.1344"></a> -<span class="sourceLineNo">1345</span> for (String coprocessor : getWAL(region.getRegionInfo()).getCoprocessorHost()<a name="line.1345"></a> -<span class="sourceLineNo">1346</span> .getCoprocessors()) {<a name="line.1346"></a> -<span class="sourceLineNo">1347</span> serverLoad.addCoprocessors(coprocessorBuilder.setName(coprocessor).build());<a name="line.1347"></a> -<span class="sourceLineNo">1348</span> }<a name="line.1348"></a> -<span class="sourceLineNo">1349</span> }<a name="line.1349"></a> -<span class="sourceLineNo">1350</span> serverLoad.setReportStartTime(reportStartTime);<a name="line.1350"></a> -<span class="sourceLineNo">1351</span> serverLoad.setReportEndTime(reportEndTime);<a name="line.1351"></a> -<span class="sourceLineNo">1352</span> if (this.infoServer != null) {<a name="line.1352"></a> -<span class="sourceLineNo">1353</span> serverLoad.setInfoServerPort(this.infoServer.getPort());<a name="line.1353"></a> -<span class="sourceLineNo">1354</span> } else {<a name="line.1354"></a> -<span class="sourceLineNo">1355</span> serverLoad.setInfoServerPort(-1);<a name="line.1355"></a> -<span class="sourceLineNo">1356</span> }<a name="line.1356"></a> -<span class="sourceLineNo">1357</span><a name="line.1357"></a> -<span class="sourceLineNo">1358</span> // for the replicationLoad purpose. Only need to get from one executorService<a name="line.1358"></a> -<span class="sourceLineNo">1359</span> // either source or sink will get the same info<a name="line.1359"></a> -<span class="sourceLineNo">1360</span> ReplicationSourceService rsources = getReplicationSourceService();<a name="line.1360"></a> -<span class="sourceLineNo">1361</span><a name="line.1361"></a> -<span class="sourceLineNo">1362</span> if (rsources != null) {<a name="line.1362"></a> -<span class="sourceLineNo">1363</span> // always refresh first to get the latest value<a name="line.1363"></a> -<span class="sourceLineNo">1364</span> ReplicationLoad rLoad = rsources.refreshAndGetReplicationLoad();<a name="line.1364"></a> -<span class="sourceLineNo">1365</span> if (rLoad != null) {<a name="line.1365"></a> -<span class="sourceLineNo">1366</span> serverLoad.setReplLoadSink(rLoad.getReplicationLoadSink());<a name="line.1366"></a> -<span class="sourceLineNo">1367</span> for (ClusterStatusProtos.ReplicationLoadSource rLS : rLoad.getReplicationLoadSourceList()) {<a name="line.1367"></a> -<span class="sourceLineNo">1368</span> serverLoad.addReplLoadSource(rLS);<a name="line.1368"></a> -<span class="sourceLineNo">1369</span> }<a name="line.1369"></a> -<span class="sourceLineNo">1370</span> }<a name="line.1370"></a> -<span class="sourceLineNo">1371</span> }<a name="line.1371"></a> -<span class="sourceLineNo">1372</span><a name="line.1372"></a> -<span class="sourceLineNo">1373</span> return serverLoad.build();<a name="line.1373"></a> -<span class="sourceLineNo">1374</span> }<a name="line.1374"></a> -<span class="sourceLineNo">1375</span><a name="line.1375"></a> -<span class="sourceLineNo">1376</span> String getOnlineRegionsAsPrintableString() {<a name="line.1376"></a> -<span class="sourceLineNo">1377</span> StringBuilder sb = new StringBuilder();<a name="line.1377"></a> -<span class="sourceLineNo">1378</span> for (Region r: this.onlineRegions.values()) {<a name="line.1378"></a> -<span class="sourceLineNo">1379</span> if (sb.length() > 0) sb.append(", ");<a name="line.1379"></a> -<span class="sourceLineNo">1380</span> sb.append(r.getRegionInfo().getEncodedName());<a name="line.1380"></a> -<span class="sourceLineNo">1381</span> }<a name="line.1381"></a> -<span class="sourceLineNo">1382</span> return sb.toString();<a name="line.1382"></a> -<span class="sourceLineNo">1383</span> }<a name="line.1383"></a> -<span class="sourceLineNo">1384</span><a name="line.1384"></a> -<span class="sourceLineNo">1385</span> /**<a name="line.1385"></a> -<span class="sourceLineNo">1386</span> * Wait on regions close.<a name="line.1386"></a> -<span class="sourceLineNo">1387</span> */<a name="line.1387"></a> -<span class="sourceLineNo">1388</span> private void waitOnAllRegionsToClose(final boolean abort) {<a name="line.1388"></a> -<span class="sourceLineNo">1389</span> // Wait till all regions are closed before going out.<a name="line.1389"></a> -<span class="sourceLineNo">1390</span> int lastCount = -1;<a name="line.1390"></a> -<span class="sourceLineNo">1391</span> long previousLogTime = 0;<a name="line.1391"></a> -<span class="sourceLineNo">1392</span> Set<String> closedRegions = new HashSet<>();<a name="line.1392"></a> -<span class="sourceLineNo">1393</span> boolean interrupted = false;<a name="line.1393"></a> -<span class="sourceLineNo">1394</span> try {<a name="line.1394"></a> -<span class="sourceLineNo">1395</span> while (!isOnlineRegionsEmpty()) {<a name="line.1395"></a> -<span class="sourceLineNo">1396</span> int count = getNumberOfOnlineRegions();<a name="line.1396"></a> -<span class="sourceLineNo">1397</span> // Only print a message if the count of regions has changed.<a name="line.1397"></a> -<span class="sourceLineNo">1398</span> if (count != lastCount) {<a name="line.1398"></a> -<span class="sourceLineNo">1399</span> // Log every second at most<a name="line.1399"></a> -<span class="sourceLineNo">1400</span> if (System.currentTimeMillis() > (previousLogTime + 1000)) {<a name="line.1400"></a> -<span class="sourceLineNo">1401</span> previousLogTime = System.currentTimeMillis();<a name="line.1401"></a> -<span class="sourceLineNo">1402</span> lastCount = count;<a name="line.1402"></a> -<span class="sourceLineNo">1403</span> LOG.info("Waiting on " + count + " regions to close");<a name="line.1403"></a> -<span class="sourceLineNo">1404</span> // Only print out regions still closing if a small number else will<a name="line.1404"></a> -<span class="sourceLineNo">1405</span> // swamp the log.<a name="line.1405"></a> -<span class="sourceLineNo">1406</span> if (count < 10 && LOG.isDebugEnabled()) {<a name="line.1406"></a> -<span class="sourceLineNo">1407</span> LOG.debug("Online Regions=" + this.onlineRegions);<a name="line.1407"></a> -<span class="sourceLineNo">1408</span> }<a name="line.1408"></a> -<span class="sourceLineNo">1409</span> }<a name="line.1409"></a> -<span class="sourceLineNo">1410</span> }<a name="line.1410"></a> -<span class="sourceLineNo">1411</span> // Ensure all user regions have been sent a close. Use this to<a name="line.1411"></a> -<span class="sourceLineNo">1412</span> // protect against the case where an open comes in after we start the<a name="line.1412"></a> -<span class="sourceLineNo">1413</span> // iterator of onlineRegions to close all user regions.<a name="line.1413"></a> -<span class="sourceLineNo">1414</span> for (Map.Entry<String, HRegion> e : this.onlineRegions.entrySet()) {<a name="line.1414"></a> -<span class="sourceLineNo">1415</span> RegionInfo hri = e.getValue().getRegionInfo();<a name="line.1415"></a> -<span class="sourceLineNo">1416</span> if (!this.regionsInTransitionInRS.containsKey(hri.getEncodedNameAsBytes()) &&<a name="line.1416"></a> -<span class="sourceLineNo">1417</span> !closedRegions.contains(hri.getEncodedName())) {<a name="line.1417"></a> -<span class="sourceLineNo">1418</span> closedRegions.add(hri.getEncodedName());<a name="line.1418"></a> -<span class="sourceLineNo">1419</span> // Don't update zk with this close transition; pass false.<a name="line.1419"></a> -<span class="sourceLineNo">1420</span> closeRegionIgnoreErrors(hri, abort);<a name="line.1420"></a> -<span class="sourceLineNo">1421</span> }<a name="line.1421"></a> -<span class="sourceLineNo">1422</span> }<a name="line.1422"></a> -<span class="sourceLineNo">1423</span> // No regions in RIT, we could stop waiting now.<a name="line.1423"></a> -<span class="sourceLineNo">1424</span> if (this.regionsInTransitionInRS.isEmpty()) {<a name="line.1424"></a> -<span class="sourceLineNo">1425</span> if (!isOnlineRegionsEmpty()) {<a name="line.1425"></a> -<span class="sourceLineNo">1426</span> LOG.info("We were exiting though online regions are not empty," +<a name="line.1426"></a> -<span class="sourceLineNo">1427</span> " because some regions failed closing");<a name="line.1427"></a> -<span class="sourceLineNo">1428</span> }<a name="line.1428"></a> -<span class="sourceLineNo">1429</span> break;<a name="line.1429"></a> -<span class="sourceLineNo">1430</span> }<a name="line.1430"></a> -<span class="sourceLineNo">1431</span> if (sleep(200)) {<a name="line.1431"></a> -<span class="sourceLineNo">1432</span> interrupted = true;<a name="line.1432"></a> -<span class="sourceLineNo">1433</span> }<a name="line.1433"></a> -<span class="sourceLineNo">1434</span> }<a name="line.1434"></a> -<span class="sourceLineNo">1435</span> } finally {<a name="line.1435"></a> -<span class="sourceLineNo">1436</span> if (interrupted) {<a name="line.1436"></a> -<span class="sourceLineNo">1437</span> Thread.currentThread().interrupt();<a name="line.1437"></a> -<span class="sourceLineNo">1438</span> }<a name="line.1438"></a> -<span class="sourceLineNo">1439</span> }<a name="line.1439"></a> -<span class="sourceLineNo">1440</span> }<a name="line.1440"></a> -<span class="sourceLineNo">1441</span><a name="line.1441"></a> -<span class="sourceLineNo">1442</span> private boolean sleep(long millis) {<a name="line.1442"></a> -<span class="sourceLineNo">1443</span> boolean interrupted = false;<a name="line.1443"></a> -<span class="sourceLineNo">1444</span> try {<a name="line.1444"></a> -<span class="sourceLineNo">1445</span> Thread.sleep(millis);<a name="line.1445"></a> -<span class="sourceLineNo">1446</span> } catch (InterruptedException e) {<a name="line.1446"></a> -<span class="sourceLineNo">1447</span> LOG.warn("Interrupted while sleeping");<a name="line.1447"></a> -<span class="sourceLineNo">1448</span> interrupted = true;<a name="line.1448"></a> -<span class="sourceLineNo">1449</span> }<a name="line.1449"></a> -<span class="sourceLineNo">1450</span> return interrupted;<a name="line.1450"></a> -<span class="sourceLineNo">1451</span> }<a name="line.1451"></a> -<span class="sourceLineNo">1452</span><a name="line.1452"></a> -<span class="sourceLineNo">1453</span> private void shutdownWAL(final boolean close) {<a name="line.1453"></a> -<span class="sourceLineNo">1454</span> if (this.walFactory != null) {<a name="line.1454"></a> -<span class="sourceLineNo">1455</span> try {<a name="line.1455"></a> -<span class="sourceLineNo">1456</span> if (close) {<a name="line.1456"></a> -<span class="sourceLineNo">1457</span> walFactory.close();<a name="line.1457"></a> -<span class="sourceLineNo">1458</span> } else {<a name="line.1458"></a> -<span class="sourceLineNo">1459</span> walFactory.shutdown();<a name="line.1459"></a> -<span class="sourceLineNo">1460</span> }<a name="line.1460"></a> -<span class="sourceLineNo">1461</span> } catch (Throwable e) {<a name="line.1461"></a> -<span class="sourceLineNo">1462</span> e = e instanceof RemoteException ? ((RemoteException) e).unwrapRemoteException() : e;<a name="line.1462"></a> -<span class="sourceLineNo">1463</span> LOG.error("Shutdown / close of WAL failed: " + e);<a name="line.1463"></a> -<span class="sourceLineNo">1464</span> LOG.debug("Shutdown / close exception details:", e);<a name="line.1464"></a> -<span class="sourceLineNo">1465</span> }<a name="line.1465"></a> -<span class="sourceLineNo">1466</span> }<a name="line.1466"></a> -<span class="sourceLineNo">1467</span> }<a name="line.1467"></a> -<span class="sourceLineNo">1468</span><a name="line.1468"></a> -<span class="sourceLineNo">1469</span> /*<a name="line.1469"></a> -<span class="sourceLineNo">1470</span> * Run init. Sets up wal and starts up all server threads.<a name="line.1470"></a> -<span class="sourceLineNo">1471</span> *<a name="line.1471"></a> -<span class="sourceLineNo">1472</span> * @param c Extra configuration.<a name="line.1472"></a> -<span class="sourceLineNo">1473</span> */<a name="line.1473"></a> -<span class="sourceLineNo">1474</span> protected void handleReportForDutyResponse(final RegionServerStartupResponse c)<a name="line.1474"></a> -<span class="sourceLineNo">1475</span> throws IOException {<a name="line.1475"></a> -<span class="sourceLineNo">1476</span> try {<a name="line.1476"></a> -<span class="sourceLineNo">1477</span> boolean updateRootDir = false;<a name="line.1477"></a> -<span class="sourceLineNo">1478</span> for (NameStringPair e : c.getMapEntriesList()) {<a name="line.1478"></a> -<span class="sourceLineNo">1479</span> String key = e.getName();<a name="line.1479"></a> -<span class="sourceLineNo">1480</span> // The hostname the master sees us as.<a name="line.1480"></a> -<span class="sourceLineNo">1481</span> if (key.equals(HConstants.KEY_FOR_HOSTNAME_SEEN_BY_MASTER)) {<a name="line.1481"></a> -<span class="sourceLineNo">1482</span> String hostnameFromMasterPOV = e.getValue();<a name="line.1482"></a> -<span class="sourceLineNo">1483</span> this.serverName = ServerName.valueOf(hostnameFromMasterPOV, rpcServices.isa.getPort(),<a name="line.1483"></a> -<span class="sourceLineNo">1484</span> this.startcode);<a name="line.1484"></a> -<span class="sourceLineNo">1485</span> if (!StringUtils.isBlank(useThisHostnameInstead) &&<a name="line.1485"></a> -<span class="sourceLineNo">1486</span> !hostnameFromMasterPOV.equals(useThisHostnameInstead)) {<a name="line.1486"></a> -<span class="sourceLineNo">1487</span> String msg = "Master passed us a different hostname to use; was=" +<a name="line.1487"></a> -<span class="sourceLineNo">1488</span> this.useThisHostnameInstead + ", but now=" + hostnameFromMasterPOV;<a name="line.1488"></a> -<span class="sourceLineNo">1489</span> LOG.error(msg);<a name="line.1489"></a> -<span class="sourceLineNo">1490</span> throw new IOException(msg);<a name="line.1490"></a> -<span class="sourceLineNo">1491</span> }<a name="line.1491"></a> -<span class="sourceLineNo">1492</span> if (StringUtils.isBlank(useThisHostnameInstead) &&<a name="line.1492"></a> -<span class="sourceLineNo">1493</span> !hostnameFromMasterPOV.equals(rpcServices.isa.getHostName())) {<a name="line.1493"></a> -<span class="sourceLineNo">1494</span> String msg = "Master passed us a different hostname to use; was=" +<a name="line.1494"></a> -<span class="sourceLineNo">1495</span> rpcServices.isa.getHostName() + ", but now=" + hostnameFromMasterPOV;<a name="line.1495"></a> -<span class="sourceLineNo">1496</span> LOG.error(msg);<a name="line.1496"></a> -<span class="sourceLineNo">1497</span> }<a name="line.1497"></a> -<span class="sourceLineNo">1498</span> continue;<a name="line.1498"></a> -<span class="sourceLineNo">1499</span> }<a name="line.1499"></a> -<span class="sourceLineNo">1500</span><a name="line.1500"></a> -<span class="sourceLineNo">1501</span> String value = e.getValue();<a name="line.1501"></a> -<span class="sourceLineNo">1502</span> if (key.equals(HConstants.HBASE_DIR)) {<a name="line.1502"></a> -<span class="sourceLineNo">1503</span> if (value != null && !value.equals(conf.get(HConstants.HBASE_DIR))) {<a name="line.1503"></a> -<span class="sourceLineNo">1504</span> updateRootDir = true;<a name="line.1504"></a> -<span class="sourceLineNo">1505</span> }<a name="line.1505"></a> -<span class="sourceLineNo">1506</span> }<a name="line.1506"></a> -<span class="sourceLineNo">1507</span><a name="line.1507"></a> -<span class="sourceLineNo">1508</span> if (LOG.isDebugEnabled()) {<a name="line.1508"></a> -<span class="sourceLineNo">1509</span> LOG.debug("Config from master: " + key + "=" + value);<a name="line.1509"></a> -<span class="sourceLineNo">1510</span> }<a name="line.1510"></a> -<span class="sourceLineNo">1511</span> this.conf.set(key, value);<a name="line.1511"></a> -<span class="sourceLineNo">1512</span> }<a name="line.1512"></a> -<span class="sourceLineNo">1513</span> // Set our ephemeral znode up in zookeeper now we have a name.<a name="line.1513"></a> -<span class="sourceLineNo">1514</span> createMyEphemeralNode();<a name="line.1514"></a> -<span class="sourceLineNo">1515</span><a name="line.1515"></a> -<span class="sourceLineNo">1516</span> if (updateRootDir) {<a name="line.1516"></a> -<span class="sourceLineNo">1517</span> // initialize file system by the config fs.defaultFS and hbase.rootdir from master<a name="line.1517"></a> -<span class="sourceLineNo">1518</span> initializeFileSystem();<a name="line.1518"></a> -<span class="sourceLineNo">1519</span> }<a name="line.1519"></a> -<span class="sourceLineNo">1520</span><a name="line.1520"></a> -<span class="sourceLineNo">1521</span> // hack! Maps DFSClient => RegionServer for logs. HDFS made this<a name="line.1521"></a> -<span class="sourceLineNo">1522</span> // config param for task trackers, but we can piggyback off of it.<a name="line.1522"></a> -<span class="sourceLineNo">1523</span> if (this.conf.get("mapreduce.task.attempt.id") == null) {<a name="line.1523"></a> -<span class="sourceLineNo">1524</span> this.conf.set("mapreduce.task.attempt.id", "hb_rs_" + this.serverName.toString());<a name="line.1524"></a> -<span class="sourceLineNo">1525</span> }<a name="line.1525"></a> -<span class="sourceLineNo">1526</span><a name="line.1526"></a> -<span class="sourceLineNo">1527</span> // Save it in a file, this will allow to see if we crash<a name="line.1527"></a> -<span class="sourceLineNo">1528</span> ZNodeClearer.writeMyEphemeralNodeOnDisk(getMyEphemeralNodePath());<a name="line.1528"></a> -<span class="sourceLineNo">1529</span><a name="line.1529"></a> -<span class="sourceLineNo">1530</span> // This call sets up an initialized replication and WAL. Later we start it up.<a name="line.1530"></a> -<span class="sourceLineNo">1531</span> setupWALAndReplication();<a name="line.1531"></a> -<span class="sourceLineNo">1532</span> // Init in here rather than in constructor after thread name has been set<a name="line.1532"></a> -<span class="sourceLineNo">1533</span> this.metricsTable = new MetricsTable(new MetricsTableWrapperAggregateImpl(this));<a name="line.1533"></a> -<span class="sourceLineNo">1534</span> this.metricsRegionServer = new MetricsRegionServer(<a name="line.1534"></a> -<span class="sourceLineNo">1535</span> new MetricsRegionServerWrapperImpl(this), conf, metricsTable);<a name="line.1535"></a> -<span class="sourceLineNo">1536</span> // Now that we have a metrics source, start the pause monitor<a name="line.1536"></a> -<span class="sourceLineNo">1537</span> this.pauseMonitor = new JvmPauseMonitor(conf, getMetrics().getMetricsSource());<a name="line.1537"></a> -<span class="sourceLineNo">1538</span> pauseMonitor.start();<a name="line.1538"></a> -<span class="sourceLineNo">1539</span><a name="line.1539"></a> -<span class="sourceLineNo">1540</span> // There is a rare case where we do NOT want services to start. Check config.<a name="line.1540"></a> -<span class="sourceLineNo">1541</span> if (getConfiguration().getBoolean("hbase.regionserver.workers", true)) {<a name="line.1541"></a> -<span class="sourceLineNo">1542</span> startServices();<a name="line.1542"></a> -<span class="sourceLineNo">1543</span> }<a name="line.1543"></a> -<span class="sourceLineNo">1544</span> // In here we start up the replication Service. Above we initialized it. TODO. Reconcile.<a name="line.1544"></a> -<span class="sourceLineNo">1545</span> // or make sense of it.<a name="line.1545"></a> -<span class="sourceLineNo">1546</span> startReplicationService();<a name="line.1546"></a> -<span class="sourceLineNo">1547</span><a name="line.1547"></a> +<span class="sourceLineNo">951</span> LOG.debug("About to register with Master.");<a name="line.951"></a> +<span class="sourceLineNo">952</span> RetryCounterFactory rcf = new RetryCounterFactory(Integer.MAX_VALUE,<a name="line.952"></a> +<span class="sourceLineNo">953</span> this.sleeper.getPeriod(), 1000 * 60 * 5);<a name="line.953"></a> +<span class="sourceLineNo">954</span> RetryCounter rc = rcf.create();<a name="line.954"></a> +<span class="sourceLineNo">955</span> while (keepLooping()) {<a name="line.955"></a> +<span class="sourceLineNo">956</span> RegionServerStartupResponse w = reportForDuty();<a name="line.956"></a> +<span class="sourceLineNo">957</span> if (w == null) {<a name="line.957"></a> +<span class="sourceLineNo">958</span> long sleepTime = rc.getBackoffTimeAndIncrementAttempts();<a name="line.958"></a> +<span class="sourceLineNo">959</span> LOG.warn("reportForDuty failed; sleeping {} ms and then retrying.", sleepTime);<a name="line.959"></a> +<span class="sourceLineNo">960</span> this.sleeper.sleep(sleepTime);<a name="line.960"></a> +<span class="sourceLineNo">961</span> } else {<a name="line.961"></a> +<span class="sourceLineNo">962</span> handleReportForDutyResponse(w);<a name="line.962"></a> +<span class="sourceLineNo">963</span> break;<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><a name="line.966"></a> +<span class="sourceLineNo">967</span> if (!isStopped() && isHealthy()) {<a name="line.967"></a> +<span class="sourceLineNo">968</span> // start the snapshot handler and other procedure handlers,<a name="line.968"></a> +<span class="sourceLineNo">969</span> // since the server is ready to run<a name="line.969"></a> +<span class="sourceLineNo">970</span> if (this.rspmHost != null) {<a name="line.970"></a> +<span class="sourceLineNo">971</span> this.rspmHost.start();<a name="line.971"></a> +<span class="sourceLineNo">972</span> }<a name="line.972"></a> +<span class="sourceLineNo">973</span> // Start the Quota Manager<a name="line.973"></a> +<span class="sourceLineNo">974</span> if (this.rsQuotaManager != null) {<a name="line.974"></a> +<span class="sourceLineNo">975</span> rsQuotaManager.start(getRpcServer().getScheduler());<a name="line.975"></a> +<span class="sourceLineNo">976</span> }<a name="line.976"></a> +<span class="sourceLineNo">977</span> if (this.rsSpaceQuotaManager != null) {<a name="line.977"></a> +<span class="sourceLineNo">978</span> this.rsSpaceQuotaManager.start();<a name="line.978"></a> +<span class="sourceLineNo">979</span> }<a name="line.979"></a> +<span class="sourceLineNo">980</span> }<a name="line.980"></a> +<span class="sourceLineNo">981</span><a name="line.981"></a> +<span class="sourceLineNo">982</span> // We registered with the Master. Go into run mode.<a name="line.982"></a> +<span class="sourceLineNo">983</span> long lastMsg = System.currentTimeMillis();<a name="line.983"></a> +<span class="sourceLineNo">984</span> long oldRequestCount = -1;<a name="line.984"></a> +<span class="sourceLineNo">985</span> // The main run loop.<a name="line.985"></a> +<span class="sourceLineNo">986</span> while (!isStopped() && isHealthy()) {<a name="line.986"></a> +<span class="sourceLineNo">987</span> if (!isClusterUp()) {<a name="line.987"></a> +<span class="sourceLineNo">988</span> if (isOnlineRegionsEmpty()) {<a name="line.988"></a> +<span class="sourceLineNo">989</span> stop("Exiting; cluster shutdown set and not carrying any regions");<a name="line.989"></a> +<span class="sourceLineNo">990</span> } else if (!this.stopping) {<a name="line.990"></a> +<span class="sourceLineNo">991</span> this.stopping = true;<a name="line.991"></a> +<span class="sourceLineNo">992</span> LOG.info("Closing user regions");<a name="line.992"></a> +<span class="sourceLineNo">993</span> closeUserRegions(this.abortRequested);<a name="line.993"></a> +<span class="sourceLineNo">994</span> } else if (this.stopping) {<a name="line.994"></a> +<span class="sourceLineNo">995</span> boolean allUserRegionsOffline = areAllUserRegionsOffline();<a name="line.995"></a> +<span class="sourceLineNo">996</span> if (allUserRegionsOffline) {<a name="line.996"></a> +<span class="sourceLineNo">997</span> // Set stopped if no more write requests tp meta tables<a name="line.997"></a> +<span class="sourceLineNo">998</span> // since last time we went around the loop. Any open<a name="line.998"></a> +<span class="sourceLineNo">999</span> // meta regions will be closed on our way out.<a name="line.999"></a> +<span class="sourceLineNo">1000</span> if (oldRequestCount == getWriteRequestCount()) {<a name="line.1000"></a> +<span class="sourceLineNo">1001</span> stop("Stopped; only catalog regions remaining online");<a name="line.1001"></a> +<span class="sourceLineNo">1002</span> break;<a name="line.1002"></a> +<span class="sourceLineNo">1003</span> }<a name="line.1003"></a> +<span class="sourceLineNo">1004</span> oldRequestCount = getWriteRequestCount();<a name="line.1004"></a> +<span class="sourceLineNo">1005</span> } else {<a name="line.1005"></a> +<span class="sourceLineNo">1006</span> // Make sure all regions have been closed -- some regions may<a name="line.1006"></a> +<span class="sourceLineNo">1007</span> // have not got it because we were splitting at the time of<a name="line.1007"></a> +<span class="sourceLineNo">1008</span> // the call to closeUserRegions.<a name="line.1008"></a> +<span class="sourceLineNo">1009</span> closeUserRegions(this.abortRequested);<a name="line.1009"></a> +<span class="sourceLineNo">1010</span> }<a name="line.1010"></a> +<span class="sourceLineNo">1011</span> LOG.debug("Waiting on " + getOnlineRegionsAsPrintableString());<a name="line.1011"></a> +<span class="sourceLineNo">1012</span> }<a name="line.1012"></a> +<span class="sourceLineNo">1013</span> }<a name="line.1013"></a> +<span class="sourceLineNo">1014</span> long now = System.currentTimeMillis();<a name="line.1014"></a> +<span class="sourceLineNo">1015</span> if ((now - lastMsg) >= msgInterval) {<a name="line.1015"></a> +<span class="sourceLineNo">1016</span> tryRegionServerReport(lastMsg, now);<a name="line.1016"></a> +<span class="sourceLineNo">1017</span> lastMsg = System.currentTimeMillis();<a name="line.1017"></a> +<span class="sourceLineNo">1018</span> }<a name="line.1018"></a> +<span class="sourceLineNo">1019</span> if (!isStopped() && !isAborted()) {<a name="line.1019"></a> +<span class="sourceLineNo">1020</span> this.sleeper.sleep();<a name="line.1020"></a> +<span class="sourceLineNo">1021</span> }<a name="line.1021"></a> +<span class="sourceLineNo">1022</span> } // for<a name="line.1022"></a> +<span class="sourceLineNo">1023</span> } catch (Throwable t) {<a name="line.1023"></a> +<span class="sourceLineNo">1024</span> if (!rpcServices.checkOOME(t)) {<a name="line.1024"></a> +<span class="sourceLineNo">1025</span> String prefix = t instanceof YouAreDeadException? "": "Unhandled: ";<a name="line.1025"></a> +<span class="sourceLineNo">1026</span> abort(prefix + t.getMessage(), t);<a name="line.1026"></a> +<span class="sourceLineNo">1027</span> }<a name="line.1027"></a> +<span class="sourceLineNo">1028</span> }<a name="line.1028"></a> +<span class="sourceLineNo">1029</span> if (this.leases != null) {<a name="line.1029"></a> +<span class="sourceLineNo">1030</span> this.leases.closeAfterLeasesExpire();<a name="line.1030"></a> +<span class="sourceLineNo">1031</span> }<a name="line.1031"></a> +<span class="sourceLineNo">1032</span> if (this.splitLogWorker != null) {<a name="line.1032"></a> +<span class="sourceLineNo">1033</span> splitLogWorker.stop();<a name="line.1033"></a> +<span class="sourceLineNo">1034</span> }<a name="line.1034"></a> +<span class="sourceLineNo">1035</span> if (this.infoServer != null) {<a name="line.1035"></a> +<span class="sourceLineNo">1036</span> LOG.info("Stopping infoServer");<a name="line.1036"></a> +<span class="sourceLineNo">1037</span> try {<a name="line.1037"></a> +<span class="sourceLineNo">1038</span> this.infoServer.stop();<a name="line.1038"></a> +<span class="sourceLineNo">1039</span> } catch (Exception e) {<a name="line.1039"></a> +<span class="sourceLineNo">1040</span> LOG.error("Failed to stop infoServer", e);<a name="line.1040"></a> +<span class="sourceLineNo">1041</span> }<a name="line.1041"></a> +<span class="sourceLineNo">1042</span> }<a name="line.1042"></a> +<span class="sourceLineNo">1043</span> // Send cache a shutdown.<a name="line.1043"></a> +<span class="sourceLineNo">1044</span> if (cacheConfig != null && cacheConfig.isBlockCacheEnabled()) {<a name="line.1044"></a> +<span class="sourceLineNo">1045</span> cacheConfig.getBlockCache().shutdown();<a name="line.1045"></a> +<span class="sourceLineNo">1046</span> }<a name="line.1046"></a> +<span class="sourceLineNo">1047</span> mobCacheConfig.getMobFileCache().shutdown();<a name="line.1047"></a> +<span class="sourceLineNo">1048</span><a name="line.1048"></a> +<span class="sourceLineNo">1049</span> if (movedRegionsCleaner != null) {<a name="line.1049"></a> +<span class="sourceLineNo">1050</span> movedRegionsCleaner.stop("Region Server stopping");<a name="line.1050"></a> +<span class="sourceLineNo">1051</span> }<a name="line.1051"></a> +<span class="sourceLineNo">1052</span><a name="line.1052"></a> +<span class="sourceLineNo">1053</span> // Send interrupts to wake up threads if sleeping so they notice shutdown.<a name="line.1053"></a> +<span class="sourceLineNo">1054</span> // TODO: Should we check they are alive? If OOME could have exited already<a name="line.1054"></a> +<span class="sourceLineNo">1055</span> if (this.hMemManager != null) this.hMemManager.stop();<a name="line.1055"></a> +<span class="sourceLineNo">1056</span> if (this.cacheFlusher != null) this.cacheFlusher.interruptIfNecessary();<a name="line.1056"></a> +<span class="sourceLineNo">1057</span> if (this.compactSplitThread != null) this.compactSplitThread.interruptIfNecessary();<a name="line.1057"></a> +<span class="sourceLineNo">1058</span> sendShutdownInterrupt();<a name="line.1058"></a> +<span class="sourceLineNo">1059</span><a name="line.1059"></a> +<span class="sourceLineNo">1060</span> // Stop the snapshot and other procedure handlers, forcefully killing all running tasks<a name="line.1060"></a> +<span class="sourceLineNo">1061</span> if (rspmHost != null) {<a name="line.1061"></a> +<span class="sourceLineNo">1062</span> rspmHost.stop(this.abortRequested || this.killed);<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> if (this.killed) {<a name="line.1065"></a> +<span class="sourceLineNo">1066</span> // Just skip out w/o closing regions. Used when testing.<a name="line.1066"></a> +<span class="sourceLineNo">1067</span> } else if (abortRequested) {<a name="line.1067"></a> +<span class="sourceLineNo">1068</span> if (this.fsOk) {<a name="line.1068"></a> +<span class="sourceLineNo">1069</span> closeUserRegions(abortRequested); // Don't leave any open file handles<a name="line.1069"></a> +<span class="sourceLineNo">1070</span> }<a name="line.1070"></a> +<span class="sourceLineNo">1071</span> LOG.info("aborting server " + this.serverName);<a name="line.1071"></a> +<span class="sourceLineNo">1072</span> } else {<a name="line.1072"></a> +<span class="sourceLineNo">1073</span> closeUserRegions(abortRequested);<a name="line.1073"></a> +<span class="sourceLineNo">1074</span> LOG.info("stopping server " + this.serverName);<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> // so callers waiting for meta without timeout can stop<a name="line.1077"></a> +<span class="sourceLineNo">1078</span> if (this.metaTableLocator != null) this.metaTableLocator.stop();<a name="line.1078"></a> +<span class="sourceLineNo">1079</span> if (this.clusterConnection != null && !clusterConnection.isClosed()) {<a name="line.1079"></a> +<span class="sourceLineNo">1080</span> try {<a name="line.1080"></a> +<span class="sourceLineNo">1081</span> this.clusterConnection.close();<a name="line.1081"></a> +<span class="sourceLineNo">1082</span> } catch (IOException e) {<a name="line.1082"></a> +<span class="sourceLineNo">1083</span> // Although the {@link Closeable} interface throws an {@link<a name="line.1083"></a> +<span class="sourceLineNo">1084</span> // IOException}, in reality, the implementation would never do that.<a name="line.1084"></a> +<span class="sourceLineNo">1085</span> LOG.warn("Attempt to close server's short circuit ClusterConnection failed.", e);<a name="line.1085"></a> +<span class="sourceLineNo">1086</span> }<a name="line.1086"></a> +<span class="sourceLineNo">1087</span> }<a name="line.1087"></a> +<span class="sourceLineNo">1088</span><a name="line.1088"></a> +<span class="sourceLineNo">1089</span> // Closing the compactSplit thread before closing meta regions<a name="line.1089"></a> +<span class="sourceLineNo">1090</span> if (!this.killed && containsMetaTableRegions()) {<a name="line.1090"></a> +<span class="sourceLineNo">1091</span> if (!abortRequested || this.fsOk) {<a name="line.1091"></a> +<span class="sourceLineNo">1092</span> if (this.compactSplitThread != null) {<a name="line.1092"></a> +<span class="sourceLineNo">1093</span> this.compactSplitThread.join();<a name="line.1093"></a> +<span class="sourceLineNo">1094</span> this.compactSplitThread = null;<a name="line.1094"></a> +<span class="sourceLineNo">1095</span> }<a name="line.1095"></a> +<span class="sourceLineNo">1096</span> closeMetaTableRegions(abortRequested);<a name="line.1096"></a> +<span class="sourceLineNo">1097</span> }<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> if (!this.killed && this.fsOk) {<a name="line.1100"></a> +<span class="sourceLineNo">1101</span> waitOnAllRegionsToClose(abortRequested);<a name="line.1101"></a> +<span class="sourceLineNo">1102</span> LOG.info("stopping server " + this.serverName + "; all regions closed.");<a name="line.1102"></a> +<span class="sourceLineNo">1103</span> }<a name="line.1103"></a> +<span class="sourceLineNo">1104</span><a name="line.1104"></a> +<span class="sourceLineNo">1105</span> // Stop the quota manager<a name="line.1105"></a> +<span class="sourceLineNo">1106</span> if (rsQuotaManager != null) {<a name="line.1106"></a> +<span class="sourceLineNo">1107</span> rsQuotaManager.stop();<a name="line.1107"></a> +<span class="sourceLineNo">1108</span> }<a name="line.1108"></a> +<span class="sourceLineNo">1109</span> if (rsSpaceQuotaManager != null) {<a name="line.1109"></a> +<span class="sourceLineNo">1110</span> rsSpaceQuotaManager.stop();<a name="line.1110"></a> +<span class="sourceLineNo">1111</span> rsSpaceQuotaManager = null;<a name="line.1111"></a> +<span class="sourceLineNo">1112</span> }<a name="line.1112"></a> +<span class="sourceLineNo">1113</span><a name="line.1113"></a> +<span class="sourceLineNo">1114</span> //fsOk flag may be changed when closing regions throws exception.<a name="line.1114"></a> +<span class="sourceLineNo">1115</span> if (this.fsOk) {<a name="line.1115"></a> +<span class="sourceLineNo">1116</span> shutdownWAL(!abortRequested);<a name="line.1116"></a> +<span class="sourceLineNo">1117</span> }<a name="line.1117"></a> +<span class="sourceLineNo">1118</span><a name="line.1118"></a> +<span class="sourceLineNo">1119</span> // Make sure the proxy is down.<a name="line.1119"></a> +<span class="sourceLineNo">1120</span> if (this.rssStub != null) {<a name="line.1120"></a> +<span class="sourceLineNo">1121</span> this.rssStub = null;<a name="line.1121"></a> +<span class="sourceLineNo">1122</span> }<a name="line.1122"></a> +<span class="sourceLineNo">1123</span> if (this.lockStub != null) {<a name="line.1123"></a> +<span class="sourceLineNo">1124</span> this.lockStub = null;<a name="line.1124"></a> +<span class="sourceLineNo">1125</span> }<a name="line.1125"></a> +<span class="sourceLineNo">1126</span> if (this.rpcClient != null) {<a name="line.1126"></a> +<span class="sourceLineNo">1127</span> this.rpcClient.close();<a name="line.1127"></a> +<span class="sourceLineNo">1128</span> }<a name="line.1128"></a> +<span class="sourceLineNo">1129</span> if (this.leases != null) {<a name="line.1129"></a> +<span class="sourceLineNo">1130</span> this.leases.close();<a name="line.1130"></a> +<span class="sourceLineNo">1131</span> }<a name="line.1131"></a> +<span class="sourceLineNo">1132</span> if (this.pauseMonitor != null) {<a name="line.1132"></a> +<span class="sourceLineNo">1133</span> this.pauseMonitor.stop();<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> if (!killed) {<a name="line.1136"></a> +<span class="sourceLineNo">1137</span> stopServiceThreads();<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> if (this.rpcServices != null) {<a name="line.1140"></a> +<span class="sourceLineNo">1141</span> this.rpcServices.stop();<a name="line.1141"></a> +<span class="sourceLineNo">1142</span> }<a name="line.1142"></a> +<span class="sourceLineNo">1143</span><a name="line.1143"></a> +<span class="sourceLineNo">1144</span> try {<a name="line.1144"></a> +<span class="sourceLineNo">1145</span> deleteMyEphemeralNode();<a name="line.1145"></a> +<span class="sourceLineNo">1146</span> } catch (KeeperException.NoNodeException nn) {<a name="line.1146"></a> +<span class="sourceLineNo">1147</span> } catch (KeeperException e) {<a name="line.1147"></a> +<span class="sourceLineNo">1148</span> LOG.warn("Failed deleting my ephemeral node", e);<a name="line.1148"></a> +<span class="sourceLineNo">1149</span> }<a name="line.1149"></a> +<span class="sourceLineNo">1150</span> // We may have failed to delete the znode at the previous step, but<a name="line.1150"></a> +<span class="sourceLineNo">1151</span> // we delete the file anyway: a second attempt to delete the znode is likely to fail again.<a name="line.1151"></a> +<span class="sourceLineNo">1152</span> ZNodeClearer.deleteMyEphemeralNodeOnDisk();<a name="line.1152"></a> +<span class="sourceLineNo">1153</span><a name="line.1153"></a> +<span class="sourceLineNo">1154</span> if (this.zooKeeper != null) {<a name="line.1154"></a> +<span class="sourceLineNo">1155</span> this.zooKeeper.close();<a name="line.1155"></a> +<span class="sourceLineNo">1156</span> }<a name="line.1156"></a> +<span class="sourceLineNo">1157</span> this.shutDown = true;<a name="line.1157"></a> +<span class="sourceLineNo">1158</span> LOG.info("Exiting; stopping=" + this.serverName + "; zookeeper connection closed.");<a name="line.1158"></a> +<span class="sourceLineNo">1159</span> }<a name="line.1159"></a> +<span class="sourceLineNo">1160</span><a name="line.1160"></a> +<span class="sourceLineNo">1161</span> private boolean containsMetaTableRegions() {<a name="line.1161"></a> +<span class="sourceLineNo">1162</span> return onlineRegions.containsKey(RegionInfoBuilder.FIRST_META_REGIONINFO.getEncodedName());<a name="line.1162"></a> +<span class="sourceLineNo">1163</span> }<a name="line.1163"></a> +<span class="sourceLineNo">1164</span><a name="line.1164"></a> +<span class="sourceLineNo">1165</span> private boolean areAllUserRegionsOffline() {<a name="line.1165"></a> +<span class="sourceLineNo">1166</span> if (getNumberOfOnlineRegions() > 2) return false;<a name="line.1166"></a> +<span class="sourceLineNo">1167</span> boolean allUserRegionsOffline = true;<a name="line.1167"></a> +<span class="sourceLineNo">1168</span> for (Map.Entry<String, HRegion> e: this.onlineRegions.entrySet()) {<a name="line.1168"></a> +<span class="sourceLineNo">1169</span> if (!e.getValue().getRegionInfo().isMetaRegion()) {<a name="line.1169"></a> +<span class="sourceLineNo">1170</span> allUserRegionsOffline = false;<a name="line.1170"></a> +<span class="sourceLineNo">1171</span> break;<a name="line.1171"></a> +<span class="sourceLineNo">1172</span> }<a name="line.1172"></a> +<span class="sourceLineNo">1173</span> }<a name="line.1173"></a> +<span class="sourceLineNo">1174</span> return allUserRegionsOffline;<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> /**<a name="line.1177"></a> +<span class="sourceLineNo">1178</span> * @return Current write count for all online regions.<a name="line.1178"></a> +<span class="sourceLineNo">1179</span> */<a name="line.1179"></a> +<span class="sourceLineNo">1180</span> private long getWriteRequestCount() {<a name="line.1180"></a> +<span class="sourceLineNo">1181</span> long writeCount = 0;<a name="line.1181"></a> +<span class="sourceLineNo">1182</span> for (Map.Entry<String, HRegion> e: this.onlineRegions.entrySet()) {<a name="line.1182"></a> +<span class="sourceLineNo">1183</span> writeCount += e.getValue().getWriteRequestsCount();<a name="line.1183"></a> +<span class="sourceLineNo">1184</span> }<a name="line.1184"></a> +<span class="sourceLineNo">1185</span> return writeCount;<a name="line.1185"></a> +<span class="sourceLineNo">1186</span> }<a name="line.1186"></a> +<span class="sourceLineNo">1187</span><a name="line.1187"></a> +<span class="sourceLineNo">1188</span> @VisibleForTesting<a name="line.1188"></a> +<span class="sourceLineNo">1189</span> protected void tryRegionServerReport(long reportStartTime, long reportEndTime)<a name="line.1189"></a> +<span class="sourceLineNo">1190</span> throws IOException {<a name="line.1190"></a> +<span class="sourceLineNo">1191</span> RegionServerStatusService.BlockingInterface rss = rssStub;<a name="line.1191"></a> +<span class="sourceLineNo">1192</span> if (rss == null) {<a name="line.1192"></a> +<span class="sourceLineNo">1193</span> // the current server could be stopping.<a name="line.1193"></a> +<span class="sourceLineNo">1194</span> return;<a name="line.1194"></a> +<span class="sourceLineNo">1195</span> }<a name="line.1195"></a> +<span class="sourceLineNo">1196</span> ClusterStatusProtos.ServerLoad sl = buildServerLoad(reportStartTime, reportEndTime);<a name="line.1196"></a> +<span class="sourceLineNo">1197</span> try {<a name="line.1197"></a> +<span class="sourceLineNo">1198</span> RegionServerReportRequest.Builder request = RegionServerReportRequest.newBuilder();<a name="line.1198"></a> +<span class="sourceLineNo">1199</span> request.setServer(ProtobufUtil.toServerName(this.serverName));<a name="line.1199"></a> +<span class="sourceLineNo">1200</span> request.setLoad(sl);<a name="line.1200"></a> +<span class="sourceLineNo">1201</span> rss.regionServerReport(null, request.build());<a name="line.1201"></a> +<span class="sourceLineNo">1202</span> } catch (ServiceException se) {<a name="line.1202"></a> +<span class="sourceLineNo">1203</span> IOException ioe = ProtobufUtil.getRemoteException(se);<a name="line.1203"></a> +<span class="sourceLineNo">1204</span> if (ioe instanceof YouAreDeadException) {<a name="line.1204"></a> +<span class="sourceLineNo">1205</span> // This will be caught and handled as a fatal error in run()<a name="line.1205"></a> +<span class="sourceLineNo">1206</span> throw ioe;<a name="line.1206"></a> +<span class="sourceLineNo">1207</span> }<a name="line.1207"></a> +<span class="sourceLineNo">1208</span> if (rssStub == rss) {<a name="line.1208"></a> +<span class="sourceLineNo">1209</span> rssStub = null;<a name="line.1209"></a> +<span class="sourceLineNo">1210</span> }<a name="line.1210"></a> +<span class="sourceLineNo">1211</span> // Couldn't connect to the master, get location from zk and reconnect<a name="line.1211"></a> +<span class="sourceLineNo">1212</span> // Method blocks until new master is found or we are stopped<a name="line.1212"></a> +<span class="sourceLineNo">1213</span> createRegionServerStatusStub(true);<a name="line.1213"></a> +<span class="sourceLineNo">1214</span> }<a name="line.1214"></a> +<span class="sourceLineNo">1215</span> }<a name="line.1215"></a> +<span class="sourceLineNo">1216</span><a name="line.1216"></a> +<span class="sourceLineNo">1217</span> /**<a name="line.1217"></a> +<span class="sourceLineNo">1218</span> * Reports the given map of Regions and their size on the filesystem to the active Master.<a name="line.1218"></a> +<span class="sourceLineNo">1219</span> *<a name="line.1219"></a> +<span class="sourceLineNo">1220</span> * @param regionSizeStore The store containing region sizes<a name="line.1220"></a> +<span class="sourceLineNo">1221</span> * @return false if FileSystemUtilizationChore should pause reporting to master. true otherwise<a name="line.1221"></a> +<span class="sourceLineNo">1222</span> */<a name="line.1222"></a> +<span class="sourceLineNo">1223</span> public boolean reportRegionSizesForQuotas(RegionSizeStore regionSizeStore) {<a name="line.1223"></a> +<span class="sourceLineNo">1224</span> RegionServerStatusService.BlockingInterface rss = rssStub;<a name="line.1224"></a> +<span class="sourceLineNo">1225</span> if (rss == null) {<a name="line.1225"></a> +<span class="sourceLineNo">1226</span> // the current server could be stopping.<a name="line.1226"></a> +<span class="sourceLineNo">1227</span> LOG.trace("Skipping Region size report to HMaster as stub is null");<a name="line.1227"></a> +<span class="sourceLineNo">1228</span> return true;<a name="line.1228"></a> +<span class="sourceLineNo">1229</span> }<a name="line.1229"></a> +<span class="sourceLineNo">1230</span> try {<a name="line.1230"></a> +<span class="sourceLineNo">1231</span> buildReportAndSend(rss, regionSizeStore);<a name="line.1231"></a> +<span class="sourceLineNo">1232</span> } catch (ServiceException se) {<a name="line.1232">
<TRUNCATED>
