| <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"> |
| <html lang="en"> |
| <head> |
| <title>Source code</title> |
| <link rel="stylesheet" type="text/css" href="../../../../../../../stylesheet.css" title="Style"> |
| </head> |
| <body> |
| <div class="sourceContainer"> |
| <pre><span class="sourceLineNo">001</span>/*<a name="line.1"></a> |
| <span class="sourceLineNo">002</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a> |
| <span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a> |
| <span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a> |
| <span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a> |
| <span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a> |
| <span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a> |
| <span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a> |
| <span class="sourceLineNo">009</span> *<a name="line.9"></a> |
| <span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a> |
| <span class="sourceLineNo">011</span> *<a name="line.11"></a> |
| <span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a> |
| <span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a> |
| <span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a> |
| <span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a> |
| <span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a> |
| <span class="sourceLineNo">017</span> */<a name="line.17"></a> |
| <span class="sourceLineNo">018</span>package org.apache.hadoop.hbase.master.procedure;<a name="line.18"></a> |
| <span class="sourceLineNo">019</span>import static org.apache.hadoop.hbase.HConstants.DEFAULT_HBASE_SPLIT_COORDINATED_BY_ZK;<a name="line.19"></a> |
| <span class="sourceLineNo">020</span>import static org.apache.hadoop.hbase.HConstants.HBASE_SPLIT_WAL_COORDINATED_BY_ZK;<a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import java.io.IOException;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span>import java.util.ArrayList;<a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.util.Arrays;<a name="line.23"></a> |
| <span class="sourceLineNo">024</span>import java.util.List;<a name="line.24"></a> |
| <span class="sourceLineNo">025</span>import org.apache.hadoop.hbase.DoNotRetryIOException;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.ServerName;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.client.RegionInfoBuilder;<a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.client.RegionReplicaUtil;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.client.TableState;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.master.MasterServices;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.master.MasterWalManager;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.master.SplitWALManager;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.master.assignment.AssignmentManager;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.master.assignment.RegionStateNode;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.master.assignment.TransitRegionStateProcedure;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.monitoring.MonitoredTask;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.monitoring.TaskMonitor;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.procedure2.Procedure;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.procedure2.ProcedureMetrics;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.procedure2.ProcedureStateSerializer;<a name="line.41"></a> |
| <span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.procedure2.ProcedureSuspendedException;<a name="line.42"></a> |
| <span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.procedure2.ProcedureYieldException;<a name="line.43"></a> |
| <span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.procedure2.StateMachineProcedure;<a name="line.44"></a> |
| <span class="sourceLineNo">045</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.45"></a> |
| <span class="sourceLineNo">046</span>import org.slf4j.Logger;<a name="line.46"></a> |
| <span class="sourceLineNo">047</span>import org.slf4j.LoggerFactory;<a name="line.47"></a> |
| <span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.48"></a> |
| <span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.MasterProcedureProtos;<a name="line.49"></a> |
| <span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.MasterProcedureProtos.ServerCrashState;<a name="line.50"></a> |
| <span class="sourceLineNo">051</span><a name="line.51"></a> |
| <span class="sourceLineNo">052</span>/**<a name="line.52"></a> |
| <span class="sourceLineNo">053</span> * Handle crashed server. This is a port to ProcedureV2 of what used to be euphemistically called<a name="line.53"></a> |
| <span class="sourceLineNo">054</span> * ServerShutdownHandler.<a name="line.54"></a> |
| <span class="sourceLineNo">055</span> *<a name="line.55"></a> |
| <span class="sourceLineNo">056</span> * <p>The procedure flow varies dependent on whether meta is assigned and if we are to split logs.<a name="line.56"></a> |
| <span class="sourceLineNo">057</span> *<a name="line.57"></a> |
| <span class="sourceLineNo">058</span> * <p>We come in here after ServerManager has noticed a server has expired. Procedures<a name="line.58"></a> |
| <span class="sourceLineNo">059</span> * queued on the rpc should have been notified about fail and should be concurrently<a name="line.59"></a> |
| <span class="sourceLineNo">060</span> * getting themselves ready to assign elsewhere.<a name="line.60"></a> |
| <span class="sourceLineNo">061</span> */<a name="line.61"></a> |
| <span class="sourceLineNo">062</span>@InterfaceAudience.Private<a name="line.62"></a> |
| <span class="sourceLineNo">063</span>public class ServerCrashProcedure<a name="line.63"></a> |
| <span class="sourceLineNo">064</span> extends StateMachineProcedure<MasterProcedureEnv, ServerCrashState><a name="line.64"></a> |
| <span class="sourceLineNo">065</span> implements ServerProcedureInterface {<a name="line.65"></a> |
| <span class="sourceLineNo">066</span> private static final Logger LOG = LoggerFactory.getLogger(ServerCrashProcedure.class);<a name="line.66"></a> |
| <span class="sourceLineNo">067</span><a name="line.67"></a> |
| <span class="sourceLineNo">068</span> /**<a name="line.68"></a> |
| <span class="sourceLineNo">069</span> * Configuration parameter to enable/disable the retain region assignment during<a name="line.69"></a> |
| <span class="sourceLineNo">070</span> * ServerCrashProcedure.<a name="line.70"></a> |
| <span class="sourceLineNo">071</span> * <p><a name="line.71"></a> |
| <span class="sourceLineNo">072</span> * By default retain assignment is disabled which makes the failover faster and improve the<a name="line.72"></a> |
| <span class="sourceLineNo">073</span> * availability; useful for cloud scenario where region block locality is not important. Enable<a name="line.73"></a> |
| <span class="sourceLineNo">074</span> * this when RegionServers are deployed on same host where Datanode are running, this will improve<a name="line.74"></a> |
| <span class="sourceLineNo">075</span> * read performance due to local read.<a name="line.75"></a> |
| <span class="sourceLineNo">076</span> * <p><a name="line.76"></a> |
| <span class="sourceLineNo">077</span> * see HBASE-24900 for more details.<a name="line.77"></a> |
| <span class="sourceLineNo">078</span> */<a name="line.78"></a> |
| <span class="sourceLineNo">079</span> public static final String MASTER_SCP_RETAIN_ASSIGNMENT = "hbase.master.scp.retain.assignment";<a name="line.79"></a> |
| <span class="sourceLineNo">080</span> /** Default value of {@link #MASTER_SCP_RETAIN_ASSIGNMENT} */<a name="line.80"></a> |
| <span class="sourceLineNo">081</span> public static final boolean DEFAULT_MASTER_SCP_RETAIN_ASSIGNMENT = false;<a name="line.81"></a> |
| <span class="sourceLineNo">082</span><a name="line.82"></a> |
| <span class="sourceLineNo">083</span> /**<a name="line.83"></a> |
| <span class="sourceLineNo">084</span> * Name of the crashed server to process.<a name="line.84"></a> |
| <span class="sourceLineNo">085</span> */<a name="line.85"></a> |
| <span class="sourceLineNo">086</span> private ServerName serverName;<a name="line.86"></a> |
| <span class="sourceLineNo">087</span><a name="line.87"></a> |
| <span class="sourceLineNo">088</span> /**<a name="line.88"></a> |
| <span class="sourceLineNo">089</span> * Whether DeadServer knows that we are processing it.<a name="line.89"></a> |
| <span class="sourceLineNo">090</span> */<a name="line.90"></a> |
| <span class="sourceLineNo">091</span> private boolean notifiedDeadServer = false;<a name="line.91"></a> |
| <span class="sourceLineNo">092</span><a name="line.92"></a> |
| <span class="sourceLineNo">093</span> /**<a name="line.93"></a> |
| <span class="sourceLineNo">094</span> * Regions that were on the crashed server.<a name="line.94"></a> |
| <span class="sourceLineNo">095</span> */<a name="line.95"></a> |
| <span class="sourceLineNo">096</span> private List<RegionInfo> regionsOnCrashedServer;<a name="line.96"></a> |
| <span class="sourceLineNo">097</span><a name="line.97"></a> |
| <span class="sourceLineNo">098</span> private boolean carryingMeta = false;<a name="line.98"></a> |
| <span class="sourceLineNo">099</span> private boolean shouldSplitWal;<a name="line.99"></a> |
| <span class="sourceLineNo">100</span> private MonitoredTask status;<a name="line.100"></a> |
| <span class="sourceLineNo">101</span> // currentRunningState is updated when ServerCrashProcedure get scheduled, child procedures update<a name="line.101"></a> |
| <span class="sourceLineNo">102</span> // progress will not update the state because the actual state is overwritten by its next state<a name="line.102"></a> |
| <span class="sourceLineNo">103</span> private ServerCrashState currentRunningState = getInitialState();<a name="line.103"></a> |
| <span class="sourceLineNo">104</span><a name="line.104"></a> |
| <span class="sourceLineNo">105</span> /**<a name="line.105"></a> |
| <span class="sourceLineNo">106</span> * Call this constructor queuing up a Procedure.<a name="line.106"></a> |
| <span class="sourceLineNo">107</span> * @param serverName Name of the crashed server.<a name="line.107"></a> |
| <span class="sourceLineNo">108</span> * @param shouldSplitWal True if we should split WALs as part of crashed server processing.<a name="line.108"></a> |
| <span class="sourceLineNo">109</span> * @param carryingMeta True if carrying hbase:meta table region.<a name="line.109"></a> |
| <span class="sourceLineNo">110</span> */<a name="line.110"></a> |
| <span class="sourceLineNo">111</span> public ServerCrashProcedure(final MasterProcedureEnv env, final ServerName serverName,<a name="line.111"></a> |
| <span class="sourceLineNo">112</span> final boolean shouldSplitWal, final boolean carryingMeta) {<a name="line.112"></a> |
| <span class="sourceLineNo">113</span> this.serverName = serverName;<a name="line.113"></a> |
| <span class="sourceLineNo">114</span> this.shouldSplitWal = shouldSplitWal;<a name="line.114"></a> |
| <span class="sourceLineNo">115</span> this.carryingMeta = carryingMeta;<a name="line.115"></a> |
| <span class="sourceLineNo">116</span> this.setOwner(env.getRequestUser());<a name="line.116"></a> |
| <span class="sourceLineNo">117</span> }<a name="line.117"></a> |
| <span class="sourceLineNo">118</span><a name="line.118"></a> |
| <span class="sourceLineNo">119</span> /**<a name="line.119"></a> |
| <span class="sourceLineNo">120</span> * Used when deserializing from a procedure store; we'll construct one of these then call<a name="line.120"></a> |
| <span class="sourceLineNo">121</span> * #deserializeStateData(InputStream). Do not use directly.<a name="line.121"></a> |
| <span class="sourceLineNo">122</span> */<a name="line.122"></a> |
| <span class="sourceLineNo">123</span> public ServerCrashProcedure() {<a name="line.123"></a> |
| <span class="sourceLineNo">124</span> }<a name="line.124"></a> |
| <span class="sourceLineNo">125</span><a name="line.125"></a> |
| <span class="sourceLineNo">126</span> public boolean isInRecoverMetaState() {<a name="line.126"></a> |
| <span class="sourceLineNo">127</span> return getCurrentState() == ServerCrashState.SERVER_CRASH_PROCESS_META;<a name="line.127"></a> |
| <span class="sourceLineNo">128</span> }<a name="line.128"></a> |
| <span class="sourceLineNo">129</span><a name="line.129"></a> |
| <span class="sourceLineNo">130</span> @Override<a name="line.130"></a> |
| <span class="sourceLineNo">131</span> protected Flow executeFromState(MasterProcedureEnv env, ServerCrashState state)<a name="line.131"></a> |
| <span class="sourceLineNo">132</span> throws ProcedureSuspendedException, ProcedureYieldException {<a name="line.132"></a> |
| <span class="sourceLineNo">133</span> final MasterServices services = env.getMasterServices();<a name="line.133"></a> |
| <span class="sourceLineNo">134</span> final AssignmentManager am = env.getAssignmentManager();<a name="line.134"></a> |
| <span class="sourceLineNo">135</span> updateProgress(true);<a name="line.135"></a> |
| <span class="sourceLineNo">136</span> // HBASE-14802 If we have not yet notified that we are processing a dead server, do so now.<a name="line.136"></a> |
| <span class="sourceLineNo">137</span> // This adds server to the DeadServer processing list but not to the DeadServers list.<a name="line.137"></a> |
| <span class="sourceLineNo">138</span> // Server gets removed from processing list below on procedure successful finish.<a name="line.138"></a> |
| <span class="sourceLineNo">139</span> if (!notifiedDeadServer) {<a name="line.139"></a> |
| <span class="sourceLineNo">140</span> notifiedDeadServer = true;<a name="line.140"></a> |
| <span class="sourceLineNo">141</span> }<a name="line.141"></a> |
| <span class="sourceLineNo">142</span><a name="line.142"></a> |
| <span class="sourceLineNo">143</span> switch (state) {<a name="line.143"></a> |
| <span class="sourceLineNo">144</span> case SERVER_CRASH_START:<a name="line.144"></a> |
| <span class="sourceLineNo">145</span> case SERVER_CRASH_SPLIT_META_LOGS:<a name="line.145"></a> |
| <span class="sourceLineNo">146</span> case SERVER_CRASH_DELETE_SPLIT_META_WALS_DIR:<a name="line.146"></a> |
| <span class="sourceLineNo">147</span> case SERVER_CRASH_ASSIGN_META:<a name="line.147"></a> |
| <span class="sourceLineNo">148</span> break;<a name="line.148"></a> |
| <span class="sourceLineNo">149</span> default:<a name="line.149"></a> |
| <span class="sourceLineNo">150</span> // If hbase:meta is not assigned, yield.<a name="line.150"></a> |
| <span class="sourceLineNo">151</span> if (env.getAssignmentManager().waitMetaLoaded(this)) {<a name="line.151"></a> |
| <span class="sourceLineNo">152</span> throw new ProcedureSuspendedException();<a name="line.152"></a> |
| <span class="sourceLineNo">153</span> }<a name="line.153"></a> |
| <span class="sourceLineNo">154</span> }<a name="line.154"></a> |
| <span class="sourceLineNo">155</span> try {<a name="line.155"></a> |
| <span class="sourceLineNo">156</span> switch (state) {<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> case SERVER_CRASH_START:<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> LOG.info("Start " + this);<a name="line.158"></a> |
| <span class="sourceLineNo">159</span> // If carrying meta, process it first. Else, get list of regions on crashed server.<a name="line.159"></a> |
| <span class="sourceLineNo">160</span> if (this.carryingMeta) {<a name="line.160"></a> |
| <span class="sourceLineNo">161</span> setNextState(ServerCrashState.SERVER_CRASH_SPLIT_META_LOGS);<a name="line.161"></a> |
| <span class="sourceLineNo">162</span> } else {<a name="line.162"></a> |
| <span class="sourceLineNo">163</span> setNextState(ServerCrashState.SERVER_CRASH_GET_REGIONS);<a name="line.163"></a> |
| <span class="sourceLineNo">164</span> }<a name="line.164"></a> |
| <span class="sourceLineNo">165</span> break;<a name="line.165"></a> |
| <span class="sourceLineNo">166</span> case SERVER_CRASH_SPLIT_META_LOGS:<a name="line.166"></a> |
| <span class="sourceLineNo">167</span> if (env.getMasterConfiguration().getBoolean(HBASE_SPLIT_WAL_COORDINATED_BY_ZK,<a name="line.167"></a> |
| <span class="sourceLineNo">168</span> DEFAULT_HBASE_SPLIT_COORDINATED_BY_ZK)) {<a name="line.168"></a> |
| <span class="sourceLineNo">169</span> zkCoordinatedSplitMetaLogs(env);<a name="line.169"></a> |
| <span class="sourceLineNo">170</span> setNextState(ServerCrashState.SERVER_CRASH_ASSIGN_META);<a name="line.170"></a> |
| <span class="sourceLineNo">171</span> } else {<a name="line.171"></a> |
| <span class="sourceLineNo">172</span> am.getRegionStates().metaLogSplitting(serverName);<a name="line.172"></a> |
| <span class="sourceLineNo">173</span> addChildProcedure(createSplittingWalProcedures(env, true));<a name="line.173"></a> |
| <span class="sourceLineNo">174</span> setNextState(ServerCrashState.SERVER_CRASH_DELETE_SPLIT_META_WALS_DIR);<a name="line.174"></a> |
| <span class="sourceLineNo">175</span> }<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> break;<a name="line.176"></a> |
| <span class="sourceLineNo">177</span> case SERVER_CRASH_DELETE_SPLIT_META_WALS_DIR:<a name="line.177"></a> |
| <span class="sourceLineNo">178</span> if (isSplittingDone(env, true)) {<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> setNextState(ServerCrashState.SERVER_CRASH_ASSIGN_META);<a name="line.179"></a> |
| <span class="sourceLineNo">180</span> am.getRegionStates().metaLogSplit(serverName);<a name="line.180"></a> |
| <span class="sourceLineNo">181</span> } else {<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> setNextState(ServerCrashState.SERVER_CRASH_SPLIT_META_LOGS);<a name="line.182"></a> |
| <span class="sourceLineNo">183</span> }<a name="line.183"></a> |
| <span class="sourceLineNo">184</span> break;<a name="line.184"></a> |
| <span class="sourceLineNo">185</span> case SERVER_CRASH_ASSIGN_META:<a name="line.185"></a> |
| <span class="sourceLineNo">186</span> assignRegions(env, Arrays.asList(RegionInfoBuilder.FIRST_META_REGIONINFO));<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> setNextState(ServerCrashState.SERVER_CRASH_GET_REGIONS);<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> break;<a name="line.188"></a> |
| <span class="sourceLineNo">189</span> case SERVER_CRASH_GET_REGIONS:<a name="line.189"></a> |
| <span class="sourceLineNo">190</span> this.regionsOnCrashedServer = getRegionsOnCrashedServer(env);<a name="line.190"></a> |
| <span class="sourceLineNo">191</span> // Where to go next? Depends on whether we should split logs at all or<a name="line.191"></a> |
| <span class="sourceLineNo">192</span> // if we should do distributed log splitting.<a name="line.192"></a> |
| <span class="sourceLineNo">193</span> if (regionsOnCrashedServer != null) {<a name="line.193"></a> |
| <span class="sourceLineNo">194</span> LOG.info("{} had {} regions", serverName, regionsOnCrashedServer.size());<a name="line.194"></a> |
| <span class="sourceLineNo">195</span> if (LOG.isTraceEnabled()) {<a name="line.195"></a> |
| <span class="sourceLineNo">196</span> this.regionsOnCrashedServer.stream().forEach(ri -> LOG.trace(ri.getShortNameToLog()));<a name="line.196"></a> |
| <span class="sourceLineNo">197</span> }<a name="line.197"></a> |
| <span class="sourceLineNo">198</span> }<a name="line.198"></a> |
| <span class="sourceLineNo">199</span> if (!this.shouldSplitWal) {<a name="line.199"></a> |
| <span class="sourceLineNo">200</span> setNextState(ServerCrashState.SERVER_CRASH_ASSIGN);<a name="line.200"></a> |
| <span class="sourceLineNo">201</span> } else {<a name="line.201"></a> |
| <span class="sourceLineNo">202</span> setNextState(ServerCrashState.SERVER_CRASH_SPLIT_LOGS);<a name="line.202"></a> |
| <span class="sourceLineNo">203</span> }<a name="line.203"></a> |
| <span class="sourceLineNo">204</span> break;<a name="line.204"></a> |
| <span class="sourceLineNo">205</span> case SERVER_CRASH_SPLIT_LOGS:<a name="line.205"></a> |
| <span class="sourceLineNo">206</span> if (env.getMasterConfiguration().getBoolean(HBASE_SPLIT_WAL_COORDINATED_BY_ZK,<a name="line.206"></a> |
| <span class="sourceLineNo">207</span> DEFAULT_HBASE_SPLIT_COORDINATED_BY_ZK)) {<a name="line.207"></a> |
| <span class="sourceLineNo">208</span> zkCoordinatedSplitLogs(env);<a name="line.208"></a> |
| <span class="sourceLineNo">209</span> setNextState(ServerCrashState.SERVER_CRASH_ASSIGN);<a name="line.209"></a> |
| <span class="sourceLineNo">210</span> } else {<a name="line.210"></a> |
| <span class="sourceLineNo">211</span> am.getRegionStates().logSplitting(this.serverName);<a name="line.211"></a> |
| <span class="sourceLineNo">212</span> addChildProcedure(createSplittingWalProcedures(env, false));<a name="line.212"></a> |
| <span class="sourceLineNo">213</span> setNextState(ServerCrashState.SERVER_CRASH_DELETE_SPLIT_WALS_DIR);<a name="line.213"></a> |
| <span class="sourceLineNo">214</span> }<a name="line.214"></a> |
| <span class="sourceLineNo">215</span> break;<a name="line.215"></a> |
| <span class="sourceLineNo">216</span> case SERVER_CRASH_DELETE_SPLIT_WALS_DIR:<a name="line.216"></a> |
| <span class="sourceLineNo">217</span> if (isSplittingDone(env, false)) {<a name="line.217"></a> |
| <span class="sourceLineNo">218</span> cleanupSplitDir(env);<a name="line.218"></a> |
| <span class="sourceLineNo">219</span> setNextState(ServerCrashState.SERVER_CRASH_ASSIGN);<a name="line.219"></a> |
| <span class="sourceLineNo">220</span> am.getRegionStates().logSplit(this.serverName);<a name="line.220"></a> |
| <span class="sourceLineNo">221</span> } else {<a name="line.221"></a> |
| <span class="sourceLineNo">222</span> setNextState(ServerCrashState.SERVER_CRASH_SPLIT_LOGS);<a name="line.222"></a> |
| <span class="sourceLineNo">223</span> }<a name="line.223"></a> |
| <span class="sourceLineNo">224</span> break;<a name="line.224"></a> |
| <span class="sourceLineNo">225</span> case SERVER_CRASH_ASSIGN:<a name="line.225"></a> |
| <span class="sourceLineNo">226</span> // If no regions to assign, skip assign and skip to the finish.<a name="line.226"></a> |
| <span class="sourceLineNo">227</span> // Filter out meta regions. Those are handled elsewhere in this procedure.<a name="line.227"></a> |
| <span class="sourceLineNo">228</span> // Filter changes this.regionsOnCrashedServer.<a name="line.228"></a> |
| <span class="sourceLineNo">229</span> if (filterDefaultMetaRegions()) {<a name="line.229"></a> |
| <span class="sourceLineNo">230</span> if (LOG.isTraceEnabled()) {<a name="line.230"></a> |
| <span class="sourceLineNo">231</span> LOG<a name="line.231"></a> |
| <span class="sourceLineNo">232</span> .trace("Assigning regions " + RegionInfo.getShortNameToLog(regionsOnCrashedServer) +<a name="line.232"></a> |
| <span class="sourceLineNo">233</span> ", " + this + "; cycles=" + getCycles());<a name="line.233"></a> |
| <span class="sourceLineNo">234</span> }<a name="line.234"></a> |
| <span class="sourceLineNo">235</span> assignRegions(env, regionsOnCrashedServer);<a name="line.235"></a> |
| <span class="sourceLineNo">236</span> }<a name="line.236"></a> |
| <span class="sourceLineNo">237</span> setNextState(ServerCrashState.SERVER_CRASH_FINISH);<a name="line.237"></a> |
| <span class="sourceLineNo">238</span> break;<a name="line.238"></a> |
| <span class="sourceLineNo">239</span> case SERVER_CRASH_HANDLE_RIT2:<a name="line.239"></a> |
| <span class="sourceLineNo">240</span> // Noop. Left in place because we used to call handleRIT here for a second time<a name="line.240"></a> |
| <span class="sourceLineNo">241</span> // but no longer necessary since HBASE-20634.<a name="line.241"></a> |
| <span class="sourceLineNo">242</span> setNextState(ServerCrashState.SERVER_CRASH_FINISH);<a name="line.242"></a> |
| <span class="sourceLineNo">243</span> break;<a name="line.243"></a> |
| <span class="sourceLineNo">244</span> case SERVER_CRASH_FINISH:<a name="line.244"></a> |
| <span class="sourceLineNo">245</span> LOG.info("removed crashed server {} after splitting done", serverName);<a name="line.245"></a> |
| <span class="sourceLineNo">246</span> services.getAssignmentManager().getRegionStates().removeServer(serverName);<a name="line.246"></a> |
| <span class="sourceLineNo">247</span> updateProgress(true);<a name="line.247"></a> |
| <span class="sourceLineNo">248</span> return Flow.NO_MORE_STATE;<a name="line.248"></a> |
| <span class="sourceLineNo">249</span> default:<a name="line.249"></a> |
| <span class="sourceLineNo">250</span> throw new UnsupportedOperationException("unhandled state=" + state);<a name="line.250"></a> |
| <span class="sourceLineNo">251</span> }<a name="line.251"></a> |
| <span class="sourceLineNo">252</span> } catch (IOException e) {<a name="line.252"></a> |
| <span class="sourceLineNo">253</span> LOG.warn("Failed state=" + state + ", retry " + this + "; cycles=" + getCycles(), e);<a name="line.253"></a> |
| <span class="sourceLineNo">254</span> }<a name="line.254"></a> |
| <span class="sourceLineNo">255</span> return Flow.HAS_MORE_STATE;<a name="line.255"></a> |
| <span class="sourceLineNo">256</span> }<a name="line.256"></a> |
| <span class="sourceLineNo">257</span><a name="line.257"></a> |
| <span class="sourceLineNo">258</span> /**<a name="line.258"></a> |
| <span class="sourceLineNo">259</span> * @return List of Regions on crashed server.<a name="line.259"></a> |
| <span class="sourceLineNo">260</span> */<a name="line.260"></a> |
| <span class="sourceLineNo">261</span> List<RegionInfo> getRegionsOnCrashedServer(MasterProcedureEnv env) {<a name="line.261"></a> |
| <span class="sourceLineNo">262</span> return env.getMasterServices().getAssignmentManager().getRegionsOnServer(serverName);<a name="line.262"></a> |
| <span class="sourceLineNo">263</span> }<a name="line.263"></a> |
| <span class="sourceLineNo">264</span><a name="line.264"></a> |
| <span class="sourceLineNo">265</span> private void cleanupSplitDir(MasterProcedureEnv env) {<a name="line.265"></a> |
| <span class="sourceLineNo">266</span> SplitWALManager splitWALManager = env.getMasterServices().getSplitWALManager();<a name="line.266"></a> |
| <span class="sourceLineNo">267</span> try {<a name="line.267"></a> |
| <span class="sourceLineNo">268</span> if (!this.carryingMeta) {<a name="line.268"></a> |
| <span class="sourceLineNo">269</span> // If we are NOT carrying hbase:meta, check if any left-over hbase:meta WAL files from an<a name="line.269"></a> |
| <span class="sourceLineNo">270</span> // old hbase:meta tenancy on this server; clean these up if any before trying to remove the<a name="line.270"></a> |
| <span class="sourceLineNo">271</span> // WAL directory of this server or we will fail. See archiveMetaLog comment for more details<a name="line.271"></a> |
| <span class="sourceLineNo">272</span> // on this condition.<a name="line.272"></a> |
| <span class="sourceLineNo">273</span> env.getMasterServices().getMasterWalManager().archiveMetaLog(this.serverName);<a name="line.273"></a> |
| <span class="sourceLineNo">274</span> }<a name="line.274"></a> |
| <span class="sourceLineNo">275</span> splitWALManager.deleteWALDir(serverName);<a name="line.275"></a> |
| <span class="sourceLineNo">276</span> } catch (IOException e) {<a name="line.276"></a> |
| <span class="sourceLineNo">277</span> LOG.warn("Remove WAL directory for {} failed, ignore...{}", serverName, e.getMessage());<a name="line.277"></a> |
| <span class="sourceLineNo">278</span> }<a name="line.278"></a> |
| <span class="sourceLineNo">279</span> }<a name="line.279"></a> |
| <span class="sourceLineNo">280</span><a name="line.280"></a> |
| <span class="sourceLineNo">281</span> private boolean isSplittingDone(MasterProcedureEnv env, boolean splitMeta) {<a name="line.281"></a> |
| <span class="sourceLineNo">282</span> SplitWALManager splitWALManager = env.getMasterServices().getSplitWALManager();<a name="line.282"></a> |
| <span class="sourceLineNo">283</span> try {<a name="line.283"></a> |
| <span class="sourceLineNo">284</span> int wals = splitWALManager.getWALsToSplit(serverName, splitMeta).size();<a name="line.284"></a> |
| <span class="sourceLineNo">285</span> LOG.debug("Check if {} WAL splitting is done? wals={}, meta={}", serverName, wals, splitMeta);<a name="line.285"></a> |
| <span class="sourceLineNo">286</span> return wals == 0;<a name="line.286"></a> |
| <span class="sourceLineNo">287</span> } catch (IOException e) {<a name="line.287"></a> |
| <span class="sourceLineNo">288</span> LOG.warn("Get WALs of {} failed, retry...", serverName, e);<a name="line.288"></a> |
| <span class="sourceLineNo">289</span> return false;<a name="line.289"></a> |
| <span class="sourceLineNo">290</span> }<a name="line.290"></a> |
| <span class="sourceLineNo">291</span> }<a name="line.291"></a> |
| <span class="sourceLineNo">292</span><a name="line.292"></a> |
| <span class="sourceLineNo">293</span> private Procedure[] createSplittingWalProcedures(MasterProcedureEnv env, boolean splitMeta)<a name="line.293"></a> |
| <span class="sourceLineNo">294</span> throws IOException {<a name="line.294"></a> |
| <span class="sourceLineNo">295</span> LOG.info("Splitting WALs {}, isMeta: {}", this, splitMeta);<a name="line.295"></a> |
| <span class="sourceLineNo">296</span> SplitWALManager splitWALManager = env.getMasterServices().getSplitWALManager();<a name="line.296"></a> |
| <span class="sourceLineNo">297</span> List<Procedure> procedures = splitWALManager.splitWALs(serverName, splitMeta);<a name="line.297"></a> |
| <span class="sourceLineNo">298</span> return procedures.toArray(new Procedure[procedures.size()]);<a name="line.298"></a> |
| <span class="sourceLineNo">299</span> }<a name="line.299"></a> |
| <span class="sourceLineNo">300</span><a name="line.300"></a> |
| <span class="sourceLineNo">301</span> private boolean filterDefaultMetaRegions() {<a name="line.301"></a> |
| <span class="sourceLineNo">302</span> if (regionsOnCrashedServer == null) {<a name="line.302"></a> |
| <span class="sourceLineNo">303</span> return false;<a name="line.303"></a> |
| <span class="sourceLineNo">304</span> }<a name="line.304"></a> |
| <span class="sourceLineNo">305</span> regionsOnCrashedServer.removeIf(this::isDefaultMetaRegion);<a name="line.305"></a> |
| <span class="sourceLineNo">306</span> return !regionsOnCrashedServer.isEmpty();<a name="line.306"></a> |
| <span class="sourceLineNo">307</span> }<a name="line.307"></a> |
| <span class="sourceLineNo">308</span><a name="line.308"></a> |
| <span class="sourceLineNo">309</span> private boolean isDefaultMetaRegion(RegionInfo hri) {<a name="line.309"></a> |
| <span class="sourceLineNo">310</span> return hri.isMetaRegion() && RegionReplicaUtil.isDefaultReplica(hri);<a name="line.310"></a> |
| <span class="sourceLineNo">311</span> }<a name="line.311"></a> |
| <span class="sourceLineNo">312</span><a name="line.312"></a> |
| <span class="sourceLineNo">313</span> /**<a name="line.313"></a> |
| <span class="sourceLineNo">314</span> * Split hbase:meta logs using 'classic' zk-based coordination.<a name="line.314"></a> |
| <span class="sourceLineNo">315</span> * Superceded by procedure-based WAL splitting.<a name="line.315"></a> |
| <span class="sourceLineNo">316</span> * @see #createSplittingWalProcedures(MasterProcedureEnv, boolean)<a name="line.316"></a> |
| <span class="sourceLineNo">317</span> */<a name="line.317"></a> |
| <span class="sourceLineNo">318</span> private void zkCoordinatedSplitMetaLogs(MasterProcedureEnv env) throws IOException {<a name="line.318"></a> |
| <span class="sourceLineNo">319</span> LOG.debug("Splitting meta WALs {}", this);<a name="line.319"></a> |
| <span class="sourceLineNo">320</span> MasterWalManager mwm = env.getMasterServices().getMasterWalManager();<a name="line.320"></a> |
| <span class="sourceLineNo">321</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.321"></a> |
| <span class="sourceLineNo">322</span> am.getRegionStates().metaLogSplitting(serverName);<a name="line.322"></a> |
| <span class="sourceLineNo">323</span> mwm.splitMetaLog(serverName);<a name="line.323"></a> |
| <span class="sourceLineNo">324</span> am.getRegionStates().metaLogSplit(serverName);<a name="line.324"></a> |
| <span class="sourceLineNo">325</span> LOG.debug("Done splitting meta WALs {}", this);<a name="line.325"></a> |
| <span class="sourceLineNo">326</span> }<a name="line.326"></a> |
| <span class="sourceLineNo">327</span><a name="line.327"></a> |
| <span class="sourceLineNo">328</span> /**<a name="line.328"></a> |
| <span class="sourceLineNo">329</span> * Split logs using 'classic' zk-based coordination.<a name="line.329"></a> |
| <span class="sourceLineNo">330</span> * Superceded by procedure-based WAL splitting.<a name="line.330"></a> |
| <span class="sourceLineNo">331</span> * @see #createSplittingWalProcedures(MasterProcedureEnv, boolean)<a name="line.331"></a> |
| <span class="sourceLineNo">332</span> */<a name="line.332"></a> |
| <span class="sourceLineNo">333</span> private void zkCoordinatedSplitLogs(final MasterProcedureEnv env) throws IOException {<a name="line.333"></a> |
| <span class="sourceLineNo">334</span> LOG.debug("Splitting WALs {}", this);<a name="line.334"></a> |
| <span class="sourceLineNo">335</span> MasterWalManager mwm = env.getMasterServices().getMasterWalManager();<a name="line.335"></a> |
| <span class="sourceLineNo">336</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.336"></a> |
| <span class="sourceLineNo">337</span> // TODO: For Matteo. Below BLOCKs!!!! Redo so can relinquish executor while it is running.<a name="line.337"></a> |
| <span class="sourceLineNo">338</span> // PROBLEM!!! WE BLOCK HERE. Can block for hours if hundreds of WALs to split and hundreds<a name="line.338"></a> |
| <span class="sourceLineNo">339</span> // of SCPs running because big cluster crashed down.<a name="line.339"></a> |
| <span class="sourceLineNo">340</span> am.getRegionStates().logSplitting(this.serverName);<a name="line.340"></a> |
| <span class="sourceLineNo">341</span> mwm.splitLog(this.serverName);<a name="line.341"></a> |
| <span class="sourceLineNo">342</span> if (!carryingMeta) {<a name="line.342"></a> |
| <span class="sourceLineNo">343</span> mwm.archiveMetaLog(this.serverName);<a name="line.343"></a> |
| <span class="sourceLineNo">344</span> }<a name="line.344"></a> |
| <span class="sourceLineNo">345</span> am.getRegionStates().logSplit(this.serverName);<a name="line.345"></a> |
| <span class="sourceLineNo">346</span> LOG.debug("Done splitting WALs {}", this);<a name="line.346"></a> |
| <span class="sourceLineNo">347</span> }<a name="line.347"></a> |
| <span class="sourceLineNo">348</span><a name="line.348"></a> |
| <span class="sourceLineNo">349</span> void updateProgress(boolean updateState) {<a name="line.349"></a> |
| <span class="sourceLineNo">350</span> String msg = "Processing ServerCrashProcedure of " + serverName;<a name="line.350"></a> |
| <span class="sourceLineNo">351</span> if (status == null) {<a name="line.351"></a> |
| <span class="sourceLineNo">352</span> status = TaskMonitor.get().createStatus(msg);<a name="line.352"></a> |
| <span class="sourceLineNo">353</span> return;<a name="line.353"></a> |
| <span class="sourceLineNo">354</span> }<a name="line.354"></a> |
| <span class="sourceLineNo">355</span> if (currentRunningState == ServerCrashState.SERVER_CRASH_FINISH) {<a name="line.355"></a> |
| <span class="sourceLineNo">356</span> status.markComplete(msg + " done");<a name="line.356"></a> |
| <span class="sourceLineNo">357</span> return;<a name="line.357"></a> |
| <span class="sourceLineNo">358</span> }<a name="line.358"></a> |
| <span class="sourceLineNo">359</span> if (updateState) {<a name="line.359"></a> |
| <span class="sourceLineNo">360</span> currentRunningState = getCurrentState();<a name="line.360"></a> |
| <span class="sourceLineNo">361</span> }<a name="line.361"></a> |
| <span class="sourceLineNo">362</span> int childrenLatch = getChildrenLatch();<a name="line.362"></a> |
| <span class="sourceLineNo">363</span> status.setStatus(msg + " current State " + currentRunningState + (childrenLatch > 0?<a name="line.363"></a> |
| <span class="sourceLineNo">364</span> "; remaining num of running child procedures = " + childrenLatch: ""));<a name="line.364"></a> |
| <span class="sourceLineNo">365</span> }<a name="line.365"></a> |
| <span class="sourceLineNo">366</span><a name="line.366"></a> |
| <span class="sourceLineNo">367</span> @Override<a name="line.367"></a> |
| <span class="sourceLineNo">368</span> protected void rollbackState(MasterProcedureEnv env, ServerCrashState state) throws IOException {<a name="line.368"></a> |
| <span class="sourceLineNo">369</span> // Can't rollback.<a name="line.369"></a> |
| <span class="sourceLineNo">370</span> throw new UnsupportedOperationException("unhandled state=" + state);<a name="line.370"></a> |
| <span class="sourceLineNo">371</span> }<a name="line.371"></a> |
| <span class="sourceLineNo">372</span><a name="line.372"></a> |
| <span class="sourceLineNo">373</span> @Override<a name="line.373"></a> |
| <span class="sourceLineNo">374</span> protected ServerCrashState getState(int stateId) {<a name="line.374"></a> |
| <span class="sourceLineNo">375</span> return ServerCrashState.forNumber(stateId);<a name="line.375"></a> |
| <span class="sourceLineNo">376</span> }<a name="line.376"></a> |
| <span class="sourceLineNo">377</span><a name="line.377"></a> |
| <span class="sourceLineNo">378</span> @Override<a name="line.378"></a> |
| <span class="sourceLineNo">379</span> protected int getStateId(ServerCrashState state) {<a name="line.379"></a> |
| <span class="sourceLineNo">380</span> return state.getNumber();<a name="line.380"></a> |
| <span class="sourceLineNo">381</span> }<a name="line.381"></a> |
| <span class="sourceLineNo">382</span><a name="line.382"></a> |
| <span class="sourceLineNo">383</span> @Override<a name="line.383"></a> |
| <span class="sourceLineNo">384</span> protected ServerCrashState getInitialState() {<a name="line.384"></a> |
| <span class="sourceLineNo">385</span> return ServerCrashState.SERVER_CRASH_START;<a name="line.385"></a> |
| <span class="sourceLineNo">386</span> }<a name="line.386"></a> |
| <span class="sourceLineNo">387</span><a name="line.387"></a> |
| <span class="sourceLineNo">388</span> @Override<a name="line.388"></a> |
| <span class="sourceLineNo">389</span> protected boolean abort(MasterProcedureEnv env) {<a name="line.389"></a> |
| <span class="sourceLineNo">390</span> // TODO<a name="line.390"></a> |
| <span class="sourceLineNo">391</span> return false;<a name="line.391"></a> |
| <span class="sourceLineNo">392</span> }<a name="line.392"></a> |
| <span class="sourceLineNo">393</span><a name="line.393"></a> |
| <span class="sourceLineNo">394</span> @Override<a name="line.394"></a> |
| <span class="sourceLineNo">395</span> protected LockState acquireLock(final MasterProcedureEnv env) {<a name="line.395"></a> |
| <span class="sourceLineNo">396</span> if (env.getProcedureScheduler().waitServerExclusiveLock(this, getServerName())) {<a name="line.396"></a> |
| <span class="sourceLineNo">397</span> return LockState.LOCK_EVENT_WAIT;<a name="line.397"></a> |
| <span class="sourceLineNo">398</span> }<a name="line.398"></a> |
| <span class="sourceLineNo">399</span> return LockState.LOCK_ACQUIRED;<a name="line.399"></a> |
| <span class="sourceLineNo">400</span> }<a name="line.400"></a> |
| <span class="sourceLineNo">401</span><a name="line.401"></a> |
| <span class="sourceLineNo">402</span> @Override<a name="line.402"></a> |
| <span class="sourceLineNo">403</span> protected void releaseLock(final MasterProcedureEnv env) {<a name="line.403"></a> |
| <span class="sourceLineNo">404</span> env.getProcedureScheduler().wakeServerExclusiveLock(this, getServerName());<a name="line.404"></a> |
| <span class="sourceLineNo">405</span> }<a name="line.405"></a> |
| <span class="sourceLineNo">406</span><a name="line.406"></a> |
| <span class="sourceLineNo">407</span> @Override<a name="line.407"></a> |
| <span class="sourceLineNo">408</span> public void toStringClassDetails(StringBuilder sb) {<a name="line.408"></a> |
| <span class="sourceLineNo">409</span> sb.append(getProcName());<a name="line.409"></a> |
| <span class="sourceLineNo">410</span> sb.append(", splitWal=");<a name="line.410"></a> |
| <span class="sourceLineNo">411</span> sb.append(shouldSplitWal);<a name="line.411"></a> |
| <span class="sourceLineNo">412</span> sb.append(", meta=");<a name="line.412"></a> |
| <span class="sourceLineNo">413</span> sb.append(carryingMeta);<a name="line.413"></a> |
| <span class="sourceLineNo">414</span> }<a name="line.414"></a> |
| <span class="sourceLineNo">415</span><a name="line.415"></a> |
| <span class="sourceLineNo">416</span> @Override public String getProcName() {<a name="line.416"></a> |
| <span class="sourceLineNo">417</span> return getClass().getSimpleName() + " " + this.serverName;<a name="line.417"></a> |
| <span class="sourceLineNo">418</span> }<a name="line.418"></a> |
| <span class="sourceLineNo">419</span><a name="line.419"></a> |
| <span class="sourceLineNo">420</span> @Override<a name="line.420"></a> |
| <span class="sourceLineNo">421</span> protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.421"></a> |
| <span class="sourceLineNo">422</span> throws IOException {<a name="line.422"></a> |
| <span class="sourceLineNo">423</span> super.serializeStateData(serializer);<a name="line.423"></a> |
| <span class="sourceLineNo">424</span><a name="line.424"></a> |
| <span class="sourceLineNo">425</span> MasterProcedureProtos.ServerCrashStateData.Builder state =<a name="line.425"></a> |
| <span class="sourceLineNo">426</span> MasterProcedureProtos.ServerCrashStateData.newBuilder().<a name="line.426"></a> |
| <span class="sourceLineNo">427</span> setServerName(ProtobufUtil.toServerName(this.serverName)).<a name="line.427"></a> |
| <span class="sourceLineNo">428</span> setCarryingMeta(this.carryingMeta).<a name="line.428"></a> |
| <span class="sourceLineNo">429</span> setShouldSplitWal(this.shouldSplitWal);<a name="line.429"></a> |
| <span class="sourceLineNo">430</span> if (this.regionsOnCrashedServer != null && !this.regionsOnCrashedServer.isEmpty()) {<a name="line.430"></a> |
| <span class="sourceLineNo">431</span> for (RegionInfo hri: this.regionsOnCrashedServer) {<a name="line.431"></a> |
| <span class="sourceLineNo">432</span> state.addRegionsOnCrashedServer(ProtobufUtil.toRegionInfo(hri));<a name="line.432"></a> |
| <span class="sourceLineNo">433</span> }<a name="line.433"></a> |
| <span class="sourceLineNo">434</span> }<a name="line.434"></a> |
| <span class="sourceLineNo">435</span> serializer.serialize(state.build());<a name="line.435"></a> |
| <span class="sourceLineNo">436</span> }<a name="line.436"></a> |
| <span class="sourceLineNo">437</span><a name="line.437"></a> |
| <span class="sourceLineNo">438</span> @Override<a name="line.438"></a> |
| <span class="sourceLineNo">439</span> protected void deserializeStateData(ProcedureStateSerializer serializer)<a name="line.439"></a> |
| <span class="sourceLineNo">440</span> throws IOException {<a name="line.440"></a> |
| <span class="sourceLineNo">441</span> super.deserializeStateData(serializer);<a name="line.441"></a> |
| <span class="sourceLineNo">442</span><a name="line.442"></a> |
| <span class="sourceLineNo">443</span> MasterProcedureProtos.ServerCrashStateData state =<a name="line.443"></a> |
| <span class="sourceLineNo">444</span> serializer.deserialize(MasterProcedureProtos.ServerCrashStateData.class);<a name="line.444"></a> |
| <span class="sourceLineNo">445</span> this.serverName = ProtobufUtil.toServerName(state.getServerName());<a name="line.445"></a> |
| <span class="sourceLineNo">446</span> this.carryingMeta = state.hasCarryingMeta()? state.getCarryingMeta(): false;<a name="line.446"></a> |
| <span class="sourceLineNo">447</span> // shouldSplitWAL has a default over in pb so this invocation will always work.<a name="line.447"></a> |
| <span class="sourceLineNo">448</span> this.shouldSplitWal = state.getShouldSplitWal();<a name="line.448"></a> |
| <span class="sourceLineNo">449</span> int size = state.getRegionsOnCrashedServerCount();<a name="line.449"></a> |
| <span class="sourceLineNo">450</span> if (size > 0) {<a name="line.450"></a> |
| <span class="sourceLineNo">451</span> this.regionsOnCrashedServer = new ArrayList<>(size);<a name="line.451"></a> |
| <span class="sourceLineNo">452</span> for (org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionInfo ri:<a name="line.452"></a> |
| <span class="sourceLineNo">453</span> state.getRegionsOnCrashedServerList()) {<a name="line.453"></a> |
| <span class="sourceLineNo">454</span> this.regionsOnCrashedServer.add(ProtobufUtil.toRegionInfo(ri));<a name="line.454"></a> |
| <span class="sourceLineNo">455</span> }<a name="line.455"></a> |
| <span class="sourceLineNo">456</span> }<a name="line.456"></a> |
| <span class="sourceLineNo">457</span> updateProgress(false);<a name="line.457"></a> |
| <span class="sourceLineNo">458</span> }<a name="line.458"></a> |
| <span class="sourceLineNo">459</span><a name="line.459"></a> |
| <span class="sourceLineNo">460</span> @Override<a name="line.460"></a> |
| <span class="sourceLineNo">461</span> public ServerName getServerName() {<a name="line.461"></a> |
| <span class="sourceLineNo">462</span> return this.serverName;<a name="line.462"></a> |
| <span class="sourceLineNo">463</span> }<a name="line.463"></a> |
| <span class="sourceLineNo">464</span><a name="line.464"></a> |
| <span class="sourceLineNo">465</span> @Override<a name="line.465"></a> |
| <span class="sourceLineNo">466</span> public boolean hasMetaTableRegion() {<a name="line.466"></a> |
| <span class="sourceLineNo">467</span> return this.carryingMeta;<a name="line.467"></a> |
| <span class="sourceLineNo">468</span> }<a name="line.468"></a> |
| <span class="sourceLineNo">469</span><a name="line.469"></a> |
| <span class="sourceLineNo">470</span> @Override<a name="line.470"></a> |
| <span class="sourceLineNo">471</span> public ServerOperationType getServerOperationType() {<a name="line.471"></a> |
| <span class="sourceLineNo">472</span> return ServerOperationType.CRASH_HANDLER;<a name="line.472"></a> |
| <span class="sourceLineNo">473</span> }<a name="line.473"></a> |
| <span class="sourceLineNo">474</span><a name="line.474"></a> |
| <span class="sourceLineNo">475</span><a name="line.475"></a> |
| <span class="sourceLineNo">476</span> @Override<a name="line.476"></a> |
| <span class="sourceLineNo">477</span> protected boolean shouldWaitClientAck(MasterProcedureEnv env) {<a name="line.477"></a> |
| <span class="sourceLineNo">478</span> // The operation is triggered internally on the server<a name="line.478"></a> |
| <span class="sourceLineNo">479</span> // the client does not know about this procedure.<a name="line.479"></a> |
| <span class="sourceLineNo">480</span> return false;<a name="line.480"></a> |
| <span class="sourceLineNo">481</span> }<a name="line.481"></a> |
| <span class="sourceLineNo">482</span><a name="line.482"></a> |
| <span class="sourceLineNo">483</span> /**<a name="line.483"></a> |
| <span class="sourceLineNo">484</span> * Moved out here so can be overridden by the HBCK fix-up SCP to be less strict about what<a name="line.484"></a> |
| <span class="sourceLineNo">485</span> * it will tolerate as a 'match'.<a name="line.485"></a> |
| <span class="sourceLineNo">486</span> * @return True if the region location in <code>rsn</code> matches that of this crashed server.<a name="line.486"></a> |
| <span class="sourceLineNo">487</span> */<a name="line.487"></a> |
| <span class="sourceLineNo">488</span> protected boolean isMatchingRegionLocation(RegionStateNode rsn) {<a name="line.488"></a> |
| <span class="sourceLineNo">489</span> return this.serverName.equals(rsn.getRegionLocation());<a name="line.489"></a> |
| <span class="sourceLineNo">490</span> }<a name="line.490"></a> |
| <span class="sourceLineNo">491</span><a name="line.491"></a> |
| <span class="sourceLineNo">492</span> /**<a name="line.492"></a> |
| <span class="sourceLineNo">493</span> * Assign the regions on the crashed RS to other Rses.<a name="line.493"></a> |
| <span class="sourceLineNo">494</span> * <p/><a name="line.494"></a> |
| <span class="sourceLineNo">495</span> * In this method we will go through all the RegionStateNodes of the give regions to find out<a name="line.495"></a> |
| <span class="sourceLineNo">496</span> * whether there is already an TRSP for the region, if so we interrupt it and let it retry on<a name="line.496"></a> |
| <span class="sourceLineNo">497</span> * other server, otherwise we will schedule a TRSP to bring the region online.<a name="line.497"></a> |
| <span class="sourceLineNo">498</span> * <p/><a name="line.498"></a> |
| <span class="sourceLineNo">499</span> * We will also check whether the table for a region is enabled, if not, we will skip assigning<a name="line.499"></a> |
| <span class="sourceLineNo">500</span> * it.<a name="line.500"></a> |
| <span class="sourceLineNo">501</span> */<a name="line.501"></a> |
| <span class="sourceLineNo">502</span> private void assignRegions(MasterProcedureEnv env, List<RegionInfo> regions) throws IOException {<a name="line.502"></a> |
| <span class="sourceLineNo">503</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.503"></a> |
| <span class="sourceLineNo">504</span> boolean retainAssignment = env.getMasterConfiguration().getBoolean(MASTER_SCP_RETAIN_ASSIGNMENT,<a name="line.504"></a> |
| <span class="sourceLineNo">505</span> DEFAULT_MASTER_SCP_RETAIN_ASSIGNMENT);<a name="line.505"></a> |
| <span class="sourceLineNo">506</span> for (RegionInfo region : regions) {<a name="line.506"></a> |
| <span class="sourceLineNo">507</span> RegionStateNode regionNode = am.getRegionStates().getOrCreateRegionStateNode(region);<a name="line.507"></a> |
| <span class="sourceLineNo">508</span> regionNode.lock();<a name="line.508"></a> |
| <span class="sourceLineNo">509</span> try {<a name="line.509"></a> |
| <span class="sourceLineNo">510</span> // This is possible, as when a server is dead, TRSP will fail to schedule a RemoteProcedure<a name="line.510"></a> |
| <span class="sourceLineNo">511</span> // and then try to assign the region to a new RS. And before it has updated the region<a name="line.511"></a> |
| <span class="sourceLineNo">512</span> // location to the new RS, we may have already called the am.getRegionsOnServer so we will<a name="line.512"></a> |
| <span class="sourceLineNo">513</span> // consider the region is still on this crashed server. Then before we arrive here, the<a name="line.513"></a> |
| <span class="sourceLineNo">514</span> // TRSP could have updated the region location, or even finished itself, so the region is<a name="line.514"></a> |
| <span class="sourceLineNo">515</span> // no longer on this crashed server any more. We should not try to assign it again. Please<a name="line.515"></a> |
| <span class="sourceLineNo">516</span> // see HBASE-23594 for more details.<a name="line.516"></a> |
| <span class="sourceLineNo">517</span> // UPDATE: HBCKServerCrashProcedure overrides isMatchingRegionLocation; this check can get<a name="line.517"></a> |
| <span class="sourceLineNo">518</span> // in the way of our clearing out 'Unknown Servers'.<a name="line.518"></a> |
| <span class="sourceLineNo">519</span> if (!isMatchingRegionLocation(regionNode)) {<a name="line.519"></a> |
| <span class="sourceLineNo">520</span> // See HBASE-24117, though we have already changed the shutdown order, it is still worth<a name="line.520"></a> |
| <span class="sourceLineNo">521</span> // double checking here to confirm that we do not skip assignment incorrectly.<a name="line.521"></a> |
| <span class="sourceLineNo">522</span> if (!am.isRunning()) {<a name="line.522"></a> |
| <span class="sourceLineNo">523</span> throw new DoNotRetryIOException(<a name="line.523"></a> |
| <span class="sourceLineNo">524</span> "AssignmentManager has been stopped, can not process assignment any more");<a name="line.524"></a> |
| <span class="sourceLineNo">525</span> }<a name="line.525"></a> |
| <span class="sourceLineNo">526</span> LOG.info("{} found {} whose regionLocation no longer matches {}, skipping assign...",<a name="line.526"></a> |
| <span class="sourceLineNo">527</span> this, regionNode, serverName);<a name="line.527"></a> |
| <span class="sourceLineNo">528</span> continue;<a name="line.528"></a> |
| <span class="sourceLineNo">529</span> }<a name="line.529"></a> |
| <span class="sourceLineNo">530</span> if (regionNode.getProcedure() != null) {<a name="line.530"></a> |
| <span class="sourceLineNo">531</span> LOG.info("{} found RIT {}; {}", this, regionNode.getProcedure(), regionNode);<a name="line.531"></a> |
| <span class="sourceLineNo">532</span> regionNode.getProcedure().serverCrashed(env, regionNode, getServerName(),<a name="line.532"></a> |
| <span class="sourceLineNo">533</span> !retainAssignment);<a name="line.533"></a> |
| <span class="sourceLineNo">534</span> continue;<a name="line.534"></a> |
| <span class="sourceLineNo">535</span> }<a name="line.535"></a> |
| <span class="sourceLineNo">536</span> if (env.getMasterServices().getTableStateManager()<a name="line.536"></a> |
| <span class="sourceLineNo">537</span> .isTableState(regionNode.getTable(), TableState.State.DISABLING)) {<a name="line.537"></a> |
| <span class="sourceLineNo">538</span> // We need to change the state here otherwise the TRSP scheduled by DTP will try to<a name="line.538"></a> |
| <span class="sourceLineNo">539</span> // close the region from a dead server and will never succeed. Please see HBASE-23636<a name="line.539"></a> |
| <span class="sourceLineNo">540</span> // for more details.<a name="line.540"></a> |
| <span class="sourceLineNo">541</span> env.getAssignmentManager().regionClosedAbnormally(regionNode);<a name="line.541"></a> |
| <span class="sourceLineNo">542</span> LOG.info("{} found table disabling for region {}, set it state to ABNORMALLY_CLOSED.",<a name="line.542"></a> |
| <span class="sourceLineNo">543</span> this, regionNode);<a name="line.543"></a> |
| <span class="sourceLineNo">544</span> continue;<a name="line.544"></a> |
| <span class="sourceLineNo">545</span> }<a name="line.545"></a> |
| <span class="sourceLineNo">546</span> if (env.getMasterServices().getTableStateManager()<a name="line.546"></a> |
| <span class="sourceLineNo">547</span> .isTableState(regionNode.getTable(), TableState.State.DISABLED)) {<a name="line.547"></a> |
| <span class="sourceLineNo">548</span> // This should not happen, table disabled but has regions on server.<a name="line.548"></a> |
| <span class="sourceLineNo">549</span> LOG.warn("Found table disabled for region {}, procDetails: {}", regionNode, this);<a name="line.549"></a> |
| <span class="sourceLineNo">550</span> continue;<a name="line.550"></a> |
| <span class="sourceLineNo">551</span> }<a name="line.551"></a> |
| <span class="sourceLineNo">552</span> TransitRegionStateProcedure proc =<a name="line.552"></a> |
| <span class="sourceLineNo">553</span> TransitRegionStateProcedure.assign(env, region, !retainAssignment, null);<a name="line.553"></a> |
| <span class="sourceLineNo">554</span> regionNode.setProcedure(proc);<a name="line.554"></a> |
| <span class="sourceLineNo">555</span> addChildProcedure(proc);<a name="line.555"></a> |
| <span class="sourceLineNo">556</span> } finally {<a name="line.556"></a> |
| <span class="sourceLineNo">557</span> regionNode.unlock();<a name="line.557"></a> |
| <span class="sourceLineNo">558</span> }<a name="line.558"></a> |
| <span class="sourceLineNo">559</span> }<a name="line.559"></a> |
| <span class="sourceLineNo">560</span> }<a name="line.560"></a> |
| <span class="sourceLineNo">561</span><a name="line.561"></a> |
| <span class="sourceLineNo">562</span> @Override<a name="line.562"></a> |
| <span class="sourceLineNo">563</span> protected ProcedureMetrics getProcedureMetrics(MasterProcedureEnv env) {<a name="line.563"></a> |
| <span class="sourceLineNo">564</span> return env.getMasterServices().getMasterMetrics().getServerCrashProcMetrics();<a name="line.564"></a> |
| <span class="sourceLineNo">565</span> }<a name="line.565"></a> |
| <span class="sourceLineNo">566</span><a name="line.566"></a> |
| <span class="sourceLineNo">567</span> @Override<a name="line.567"></a> |
| <span class="sourceLineNo">568</span> protected boolean holdLock(MasterProcedureEnv env) {<a name="line.568"></a> |
| <span class="sourceLineNo">569</span> return true;<a name="line.569"></a> |
| <span class="sourceLineNo">570</span> }<a name="line.570"></a> |
| <span class="sourceLineNo">571</span><a name="line.571"></a> |
| <span class="sourceLineNo">572</span> public static void updateProgress(MasterProcedureEnv env, long parentId) {<a name="line.572"></a> |
| <span class="sourceLineNo">573</span> if (parentId == NO_PROC_ID) {<a name="line.573"></a> |
| <span class="sourceLineNo">574</span> return;<a name="line.574"></a> |
| <span class="sourceLineNo">575</span> }<a name="line.575"></a> |
| <span class="sourceLineNo">576</span> Procedure parentProcedure =<a name="line.576"></a> |
| <span class="sourceLineNo">577</span> env.getMasterServices().getMasterProcedureExecutor().getProcedure(parentId);<a name="line.577"></a> |
| <span class="sourceLineNo">578</span> if (parentProcedure != null && parentProcedure instanceof ServerCrashProcedure) {<a name="line.578"></a> |
| <span class="sourceLineNo">579</span> ((ServerCrashProcedure) parentProcedure).updateProgress(false);<a name="line.579"></a> |
| <span class="sourceLineNo">580</span> }<a name="line.580"></a> |
| <span class="sourceLineNo">581</span> }<a name="line.581"></a> |
| <span class="sourceLineNo">582</span>}<a name="line.582"></a> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |