| <!DOCTYPE HTML> |
| <html lang="en"> |
| <head> |
| <!-- Generated by javadoc (17) --> |
| <title>Source code</title> |
| <meta name="viewport" content="width=device-width, initial-scale=1"> |
| <meta name="description" content="source: package: org.apache.hadoop.hbase.master.assignment, class: RegionRemoteProcedureBase"> |
| <meta name="generator" content="javadoc/SourceToHTMLConverter"> |
| <link rel="stylesheet" type="text/css" href="../../../../../../../stylesheet.css" title="Style"> |
| </head> |
| <body class="source-page"> |
| <main role="main"> |
| <div class="source-container"> |
| <pre><span class="source-line-no">001</span><span id="line-1">/*</span> |
| <span class="source-line-no">002</span><span id="line-2"> * Licensed to the Apache Software Foundation (ASF) under one</span> |
| <span class="source-line-no">003</span><span id="line-3"> * or more contributor license agreements. See the NOTICE file</span> |
| <span class="source-line-no">004</span><span id="line-4"> * distributed with this work for additional information</span> |
| <span class="source-line-no">005</span><span id="line-5"> * regarding copyright ownership. The ASF licenses this file</span> |
| <span class="source-line-no">006</span><span id="line-6"> * to you under the Apache License, Version 2.0 (the</span> |
| <span class="source-line-no">007</span><span id="line-7"> * "License"); you may not use this file except in compliance</span> |
| <span class="source-line-no">008</span><span id="line-8"> * with the License. You may obtain a copy of the License at</span> |
| <span class="source-line-no">009</span><span id="line-9"> *</span> |
| <span class="source-line-no">010</span><span id="line-10"> * http://www.apache.org/licenses/LICENSE-2.0</span> |
| <span class="source-line-no">011</span><span id="line-11"> *</span> |
| <span class="source-line-no">012</span><span id="line-12"> * Unless required by applicable law or agreed to in writing, software</span> |
| <span class="source-line-no">013</span><span id="line-13"> * distributed under the License is distributed on an "AS IS" BASIS,</span> |
| <span class="source-line-no">014</span><span id="line-14"> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span> |
| <span class="source-line-no">015</span><span id="line-15"> * See the License for the specific language governing permissions and</span> |
| <span class="source-line-no">016</span><span id="line-16"> * limitations under the License.</span> |
| <span class="source-line-no">017</span><span id="line-17"> */</span> |
| <span class="source-line-no">018</span><span id="line-18">package org.apache.hadoop.hbase.master.assignment;</span> |
| <span class="source-line-no">019</span><span id="line-19"></span> |
| <span class="source-line-no">020</span><span id="line-20">import java.io.IOException;</span> |
| <span class="source-line-no">021</span><span id="line-21">import java.util.Optional;</span> |
| <span class="source-line-no">022</span><span id="line-22">import java.util.concurrent.CompletableFuture;</span> |
| <span class="source-line-no">023</span><span id="line-23">import org.apache.hadoop.hbase.HConstants;</span> |
| <span class="source-line-no">024</span><span id="line-24">import org.apache.hadoop.hbase.ServerName;</span> |
| <span class="source-line-no">025</span><span id="line-25">import org.apache.hadoop.hbase.TableName;</span> |
| <span class="source-line-no">026</span><span id="line-26">import org.apache.hadoop.hbase.client.RegionInfo;</span> |
| <span class="source-line-no">027</span><span id="line-27">import org.apache.hadoop.hbase.exceptions.UnexpectedStateException;</span> |
| <span class="source-line-no">028</span><span id="line-28">import org.apache.hadoop.hbase.master.procedure.MasterProcedureEnv;</span> |
| <span class="source-line-no">029</span><span id="line-29">import org.apache.hadoop.hbase.master.procedure.TableProcedureInterface;</span> |
| <span class="source-line-no">030</span><span id="line-30">import org.apache.hadoop.hbase.procedure2.FailedRemoteDispatchException;</span> |
| <span class="source-line-no">031</span><span id="line-31">import org.apache.hadoop.hbase.procedure2.Procedure;</span> |
| <span class="source-line-no">032</span><span id="line-32">import org.apache.hadoop.hbase.procedure2.ProcedureEvent;</span> |
| <span class="source-line-no">033</span><span id="line-33">import org.apache.hadoop.hbase.procedure2.ProcedureFutureUtil;</span> |
| <span class="source-line-no">034</span><span id="line-34">import org.apache.hadoop.hbase.procedure2.ProcedureStateSerializer;</span> |
| <span class="source-line-no">035</span><span id="line-35">import org.apache.hadoop.hbase.procedure2.ProcedureSuspendedException;</span> |
| <span class="source-line-no">036</span><span id="line-36">import org.apache.hadoop.hbase.procedure2.ProcedureUtil;</span> |
| <span class="source-line-no">037</span><span id="line-37">import org.apache.hadoop.hbase.procedure2.ProcedureYieldException;</span> |
| <span class="source-line-no">038</span><span id="line-38">import org.apache.hadoop.hbase.procedure2.RemoteProcedureDispatcher;</span> |
| <span class="source-line-no">039</span><span id="line-39">import org.apache.hadoop.hbase.procedure2.RemoteProcedureDispatcher.RemoteProcedure;</span> |
| <span class="source-line-no">040</span><span id="line-40">import org.apache.hadoop.hbase.procedure2.RemoteProcedureException;</span> |
| <span class="source-line-no">041</span><span id="line-41">import org.apache.hadoop.hbase.util.RetryCounter;</span> |
| <span class="source-line-no">042</span><span id="line-42">import org.apache.yetus.audience.InterfaceAudience;</span> |
| <span class="source-line-no">043</span><span id="line-43">import org.slf4j.Logger;</span> |
| <span class="source-line-no">044</span><span id="line-44">import org.slf4j.LoggerFactory;</span> |
| <span class="source-line-no">045</span><span id="line-45"></span> |
| <span class="source-line-no">046</span><span id="line-46">import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;</span> |
| <span class="source-line-no">047</span><span id="line-47">import org.apache.hadoop.hbase.shaded.protobuf.generated.MasterProcedureProtos.RegionRemoteProcedureBaseState;</span> |
| <span class="source-line-no">048</span><span id="line-48">import org.apache.hadoop.hbase.shaded.protobuf.generated.MasterProcedureProtos.RegionRemoteProcedureBaseStateData;</span> |
| <span class="source-line-no">049</span><span id="line-49">import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos;</span> |
| <span class="source-line-no">050</span><span id="line-50">import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureState;</span> |
| <span class="source-line-no">051</span><span id="line-51">import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionStateTransition.TransitionCode;</span> |
| <span class="source-line-no">052</span><span id="line-52"></span> |
| <span class="source-line-no">053</span><span id="line-53">/**</span> |
| <span class="source-line-no">054</span><span id="line-54"> * The base class for the remote procedures used to open/close a region.</span> |
| <span class="source-line-no">055</span><span id="line-55"> * <p/></span> |
| <span class="source-line-no">056</span><span id="line-56"> * Notice that here we do not care about the result of the remote call, if the remote call is</span> |
| <span class="source-line-no">057</span><span id="line-57"> * finished, either succeeded or not, we will always finish the procedure. The parent procedure</span> |
| <span class="source-line-no">058</span><span id="line-58"> * should take care of the result and try to reschedule if the result is not good.</span> |
| <span class="source-line-no">059</span><span id="line-59"> */</span> |
| <span class="source-line-no">060</span><span id="line-60">@InterfaceAudience.Private</span> |
| <span class="source-line-no">061</span><span id="line-61">public abstract class RegionRemoteProcedureBase extends Procedure<MasterProcedureEnv></span> |
| <span class="source-line-no">062</span><span id="line-62"> implements TableProcedureInterface, RemoteProcedure<MasterProcedureEnv, ServerName> {</span> |
| <span class="source-line-no">063</span><span id="line-63"></span> |
| <span class="source-line-no">064</span><span id="line-64"> private static final Logger LOG = LoggerFactory.getLogger(RegionRemoteProcedureBase.class);</span> |
| <span class="source-line-no">065</span><span id="line-65"></span> |
| <span class="source-line-no">066</span><span id="line-66"> protected RegionInfo region;</span> |
| <span class="source-line-no">067</span><span id="line-67"></span> |
| <span class="source-line-no">068</span><span id="line-68"> protected ServerName targetServer;</span> |
| <span class="source-line-no">069</span><span id="line-69"></span> |
| <span class="source-line-no">070</span><span id="line-70"> private RegionRemoteProcedureBaseState state =</span> |
| <span class="source-line-no">071</span><span id="line-71"> RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_DISPATCH;</span> |
| <span class="source-line-no">072</span><span id="line-72"></span> |
| <span class="source-line-no">073</span><span id="line-73"> private TransitionCode transitionCode;</span> |
| <span class="source-line-no">074</span><span id="line-74"></span> |
| <span class="source-line-no">075</span><span id="line-75"> private long seqId;</span> |
| <span class="source-line-no">076</span><span id="line-76"></span> |
| <span class="source-line-no">077</span><span id="line-77"> private RetryCounter retryCounter;</span> |
| <span class="source-line-no">078</span><span id="line-78"></span> |
| <span class="source-line-no">079</span><span id="line-79"> private CompletableFuture<Void> future;</span> |
| <span class="source-line-no">080</span><span id="line-80"></span> |
| <span class="source-line-no">081</span><span id="line-81"> protected RegionRemoteProcedureBase() {</span> |
| <span class="source-line-no">082</span><span id="line-82"> }</span> |
| <span class="source-line-no">083</span><span id="line-83"></span> |
| <span class="source-line-no">084</span><span id="line-84"> protected RegionRemoteProcedureBase(TransitRegionStateProcedure parent, RegionInfo region,</span> |
| <span class="source-line-no">085</span><span id="line-85"> ServerName targetServer) {</span> |
| <span class="source-line-no">086</span><span id="line-86"> this.region = region;</span> |
| <span class="source-line-no">087</span><span id="line-87"> this.targetServer = targetServer;</span> |
| <span class="source-line-no">088</span><span id="line-88"> parent.attachRemoteProc(this);</span> |
| <span class="source-line-no">089</span><span id="line-89"> }</span> |
| <span class="source-line-no">090</span><span id="line-90"></span> |
| <span class="source-line-no">091</span><span id="line-91"> @Override</span> |
| <span class="source-line-no">092</span><span id="line-92"> public Optional<RemoteProcedureDispatcher.RemoteOperation> remoteCallBuild(MasterProcedureEnv env,</span> |
| <span class="source-line-no">093</span><span id="line-93"> ServerName remote) {</span> |
| <span class="source-line-no">094</span><span id="line-94"> // REPORT_SUCCEED means that this remote open/close request already executed in RegionServer.</span> |
| <span class="source-line-no">095</span><span id="line-95"> // So return empty operation and RSProcedureDispatcher no need to send it again.</span> |
| <span class="source-line-no">096</span><span id="line-96"> if (state == RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_REPORT_SUCCEED) {</span> |
| <span class="source-line-no">097</span><span id="line-97"> return Optional.empty();</span> |
| <span class="source-line-no">098</span><span id="line-98"> }</span> |
| <span class="source-line-no">099</span><span id="line-99"> return Optional.of(newRemoteOperation(env));</span> |
| <span class="source-line-no">100</span><span id="line-100"> }</span> |
| <span class="source-line-no">101</span><span id="line-101"></span> |
| <span class="source-line-no">102</span><span id="line-102"> protected abstract RemoteProcedureDispatcher.RemoteOperation</span> |
| <span class="source-line-no">103</span><span id="line-103"> newRemoteOperation(MasterProcedureEnv env);</span> |
| <span class="source-line-no">104</span><span id="line-104"></span> |
| <span class="source-line-no">105</span><span id="line-105"> @Override</span> |
| <span class="source-line-no">106</span><span id="line-106"> public void remoteOperationCompleted(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">107</span><span id="line-107"> // should not be called since we use reportRegionStateTransition to report the result</span> |
| <span class="source-line-no">108</span><span id="line-108"> throw new UnsupportedOperationException();</span> |
| <span class="source-line-no">109</span><span id="line-109"> }</span> |
| <span class="source-line-no">110</span><span id="line-110"></span> |
| <span class="source-line-no">111</span><span id="line-111"> @Override</span> |
| <span class="source-line-no">112</span><span id="line-112"> public void remoteOperationFailed(MasterProcedureEnv env, RemoteProcedureException error) {</span> |
| <span class="source-line-no">113</span><span id="line-113"> // should not be called since we use reportRegionStateTransition to report the result</span> |
| <span class="source-line-no">114</span><span id="line-114"> throw new UnsupportedOperationException();</span> |
| <span class="source-line-no">115</span><span id="line-115"> }</span> |
| <span class="source-line-no">116</span><span id="line-116"></span> |
| <span class="source-line-no">117</span><span id="line-117"> private RegionStateNode getRegionNode(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">118</span><span id="line-118"> return env.getAssignmentManager().getRegionStates().getRegionStateNode(region);</span> |
| <span class="source-line-no">119</span><span id="line-119"> }</span> |
| <span class="source-line-no">120</span><span id="line-120"></span> |
| <span class="source-line-no">121</span><span id="line-121"> @Override</span> |
| <span class="source-line-no">122</span><span id="line-122"> public void remoteCallFailed(MasterProcedureEnv env, ServerName remote, IOException exception) {</span> |
| <span class="source-line-no">123</span><span id="line-123"> RegionStateNode regionNode = getRegionNode(env);</span> |
| <span class="source-line-no">124</span><span id="line-124"> regionNode.lock();</span> |
| <span class="source-line-no">125</span><span id="line-125"> try {</span> |
| <span class="source-line-no">126</span><span id="line-126"> if (!env.getMasterServices().getServerManager().isServerOnline(remote)) {</span> |
| <span class="source-line-no">127</span><span id="line-127"> // the SCP will interrupt us, give up</span> |
| <span class="source-line-no">128</span><span id="line-128"> LOG.debug("{} for region {}, targetServer {} is dead, SCP will interrupt us, give up", this,</span> |
| <span class="source-line-no">129</span><span id="line-129"> regionNode, remote);</span> |
| <span class="source-line-no">130</span><span id="line-130"> return;</span> |
| <span class="source-line-no">131</span><span id="line-131"> }</span> |
| <span class="source-line-no">132</span><span id="line-132"> if (state != RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_DISPATCH) {</span> |
| <span class="source-line-no">133</span><span id="line-133"> // not sure how can this happen but anyway let's add a check here to avoid waking the wrong</span> |
| <span class="source-line-no">134</span><span id="line-134"> // procedure...</span> |
| <span class="source-line-no">135</span><span id="line-135"> LOG.warn("{} for region {}, targetServer={} has already been woken up, ignore", this,</span> |
| <span class="source-line-no">136</span><span id="line-136"> regionNode, remote);</span> |
| <span class="source-line-no">137</span><span id="line-137"> return;</span> |
| <span class="source-line-no">138</span><span id="line-138"> }</span> |
| <span class="source-line-no">139</span><span id="line-139"> LOG.warn("The remote operation {} for region {} to server {} failed", this, regionNode,</span> |
| <span class="source-line-no">140</span><span id="line-140"> remote, exception);</span> |
| <span class="source-line-no">141</span><span id="line-141"> // It is OK to not persist the state here, as we do not need to change the region state if the</span> |
| <span class="source-line-no">142</span><span id="line-142"> // remote call is failed. If the master crashed before we actually execute the procedure and</span> |
| <span class="source-line-no">143</span><span id="line-143"> // persist the new state, it is fine to retry on the same target server again.</span> |
| <span class="source-line-no">144</span><span id="line-144"> state = RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_DISPATCH_FAIL;</span> |
| <span class="source-line-no">145</span><span id="line-145"> regionNode.getProcedureEvent().wake(env.getProcedureScheduler());</span> |
| <span class="source-line-no">146</span><span id="line-146"> } finally {</span> |
| <span class="source-line-no">147</span><span id="line-147"> regionNode.unlock();</span> |
| <span class="source-line-no">148</span><span id="line-148"> }</span> |
| <span class="source-line-no">149</span><span id="line-149"> }</span> |
| <span class="source-line-no">150</span><span id="line-150"></span> |
| <span class="source-line-no">151</span><span id="line-151"> @Override</span> |
| <span class="source-line-no">152</span><span id="line-152"> public TableName getTableName() {</span> |
| <span class="source-line-no">153</span><span id="line-153"> return region.getTable();</span> |
| <span class="source-line-no">154</span><span id="line-154"> }</span> |
| <span class="source-line-no">155</span><span id="line-155"></span> |
| <span class="source-line-no">156</span><span id="line-156"> @Override</span> |
| <span class="source-line-no">157</span><span id="line-157"> protected boolean waitInitialized(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">158</span><span id="line-158"> if (TableName.isMetaTableName(getTableName())) {</span> |
| <span class="source-line-no">159</span><span id="line-159"> return false;</span> |
| <span class="source-line-no">160</span><span id="line-160"> }</span> |
| <span class="source-line-no">161</span><span id="line-161"> // First we need meta to be loaded, and second, if meta is not online then we will likely to</span> |
| <span class="source-line-no">162</span><span id="line-162"> // fail when updating meta so we wait until it is assigned.</span> |
| <span class="source-line-no">163</span><span id="line-163"> AssignmentManager am = env.getAssignmentManager();</span> |
| <span class="source-line-no">164</span><span id="line-164"> return am.waitMetaLoaded(this) || am.waitMetaAssigned(this, region);</span> |
| <span class="source-line-no">165</span><span id="line-165"> }</span> |
| <span class="source-line-no">166</span><span id="line-166"></span> |
| <span class="source-line-no">167</span><span id="line-167"> @Override</span> |
| <span class="source-line-no">168</span><span id="line-168"> protected void rollback(MasterProcedureEnv env) throws IOException, InterruptedException {</span> |
| <span class="source-line-no">169</span><span id="line-169"> throw new UnsupportedOperationException();</span> |
| <span class="source-line-no">170</span><span id="line-170"> }</span> |
| <span class="source-line-no">171</span><span id="line-171"></span> |
| <span class="source-line-no">172</span><span id="line-172"> @Override</span> |
| <span class="source-line-no">173</span><span id="line-173"> protected boolean abort(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">174</span><span id="line-174"> return false;</span> |
| <span class="source-line-no">175</span><span id="line-175"> }</span> |
| <span class="source-line-no">176</span><span id="line-176"></span> |
| <span class="source-line-no">177</span><span id="line-177"> // do some checks to see if the report is valid</span> |
| <span class="source-line-no">178</span><span id="line-178"> protected abstract void checkTransition(RegionStateNode regionNode, TransitionCode transitionCode,</span> |
| <span class="source-line-no">179</span><span id="line-179"> long seqId) throws UnexpectedStateException;</span> |
| <span class="source-line-no">180</span><span id="line-180"></span> |
| <span class="source-line-no">181</span><span id="line-181"> // change the in memory state of the regionNode, but do not update meta.</span> |
| <span class="source-line-no">182</span><span id="line-182"> protected abstract void updateTransitionWithoutPersistingToMeta(MasterProcedureEnv env,</span> |
| <span class="source-line-no">183</span><span id="line-183"> RegionStateNode regionNode, TransitionCode transitionCode, long seqId) throws IOException;</span> |
| <span class="source-line-no">184</span><span id="line-184"></span> |
| <span class="source-line-no">185</span><span id="line-185"> // A bit strange but the procedure store will throw RuntimeException if we can not persist the</span> |
| <span class="source-line-no">186</span><span id="line-186"> // state, so upper layer should take care of this...</span> |
| <span class="source-line-no">187</span><span id="line-187"> private void persistAndWake(MasterProcedureEnv env, RegionStateNode regionNode) {</span> |
| <span class="source-line-no">188</span><span id="line-188"> // The synchronization here is to guard with ProcedureExecutor.executeRollback, as here we will</span> |
| <span class="source-line-no">189</span><span id="line-189"> // not hold the procedure execution lock, but we should not persist a procedure in ROLLEDBACK</span> |
| <span class="source-line-no">190</span><span id="line-190"> // state to the procedure store.</span> |
| <span class="source-line-no">191</span><span id="line-191"> // The ProcedureStore.update must be inside the lock, so here the check for procedure state and</span> |
| <span class="source-line-no">192</span><span id="line-192"> // update could be atomic. In ProcedureExecutor.cleanupAfterRollbackOneStep, we will set the</span> |
| <span class="source-line-no">193</span><span id="line-193"> // state to ROLLEDBACK, which will hold the same lock too as the Procedure.setState method is</span> |
| <span class="source-line-no">194</span><span id="line-194"> // synchronized. This is the key to keep us safe.</span> |
| <span class="source-line-no">195</span><span id="line-195"> synchronized (this) {</span> |
| <span class="source-line-no">196</span><span id="line-196"> if (getState() == ProcedureState.ROLLEDBACK) {</span> |
| <span class="source-line-no">197</span><span id="line-197"> LOG.warn("Procedure {} has already been rolled back, skip persistent", this);</span> |
| <span class="source-line-no">198</span><span id="line-198"> return;</span> |
| <span class="source-line-no">199</span><span id="line-199"> }</span> |
| <span class="source-line-no">200</span><span id="line-200"> env.getMasterServices().getMasterProcedureExecutor().getStore().update(this);</span> |
| <span class="source-line-no">201</span><span id="line-201"> }</span> |
| <span class="source-line-no">202</span><span id="line-202"> regionNode.getProcedureEvent().wake(env.getProcedureScheduler());</span> |
| <span class="source-line-no">203</span><span id="line-203"> }</span> |
| <span class="source-line-no">204</span><span id="line-204"></span> |
| <span class="source-line-no">205</span><span id="line-205"> // should be called with RegionStateNode locked, to avoid race with the execute method below</span> |
| <span class="source-line-no">206</span><span id="line-206"> void reportTransition(MasterProcedureEnv env, RegionStateNode regionNode, ServerName serverName,</span> |
| <span class="source-line-no">207</span><span id="line-207"> TransitionCode transitionCode, long seqId) throws IOException {</span> |
| <span class="source-line-no">208</span><span id="line-208"> if (state != RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_DISPATCH) {</span> |
| <span class="source-line-no">209</span><span id="line-209"> // should be a retry</span> |
| <span class="source-line-no">210</span><span id="line-210"> return;</span> |
| <span class="source-line-no">211</span><span id="line-211"> }</span> |
| <span class="source-line-no">212</span><span id="line-212"> if (!targetServer.equals(serverName)) {</span> |
| <span class="source-line-no">213</span><span id="line-213"> throw new UnexpectedStateException("Received report from " + serverName + ", expected "</span> |
| <span class="source-line-no">214</span><span id="line-214"> + targetServer + ", " + regionNode + ", proc=" + this);</span> |
| <span class="source-line-no">215</span><span id="line-215"> }</span> |
| <span class="source-line-no">216</span><span id="line-216"> checkTransition(regionNode, transitionCode, seqId);</span> |
| <span class="source-line-no">217</span><span id="line-217"> // this state means we have received the report from RS, does not mean the result is fine, as we</span> |
| <span class="source-line-no">218</span><span id="line-218"> // may received a FAILED_OPEN.</span> |
| <span class="source-line-no">219</span><span id="line-219"> this.state = RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_REPORT_SUCCEED;</span> |
| <span class="source-line-no">220</span><span id="line-220"> this.transitionCode = transitionCode;</span> |
| <span class="source-line-no">221</span><span id="line-221"> this.seqId = seqId;</span> |
| <span class="source-line-no">222</span><span id="line-222"> // Persist the transition code and openSeqNum(if provided).</span> |
| <span class="source-line-no">223</span><span id="line-223"> // We should not update the hbase:meta directly as this may cause races when master restarts,</span> |
| <span class="source-line-no">224</span><span id="line-224"> // as the old active master may incorrectly report back to RS and cause the new master to hang</span> |
| <span class="source-line-no">225</span><span id="line-225"> // on a OpenRegionProcedure forever. See HBASE-22060 and HBASE-22074 for more details.</span> |
| <span class="source-line-no">226</span><span id="line-226"> boolean succ = false;</span> |
| <span class="source-line-no">227</span><span id="line-227"> try {</span> |
| <span class="source-line-no">228</span><span id="line-228"> persistAndWake(env, regionNode);</span> |
| <span class="source-line-no">229</span><span id="line-229"> succ = true;</span> |
| <span class="source-line-no">230</span><span id="line-230"> } finally {</span> |
| <span class="source-line-no">231</span><span id="line-231"> if (!succ) {</span> |
| <span class="source-line-no">232</span><span id="line-232"> this.state = RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_DISPATCH;</span> |
| <span class="source-line-no">233</span><span id="line-233"> this.transitionCode = null;</span> |
| <span class="source-line-no">234</span><span id="line-234"> this.seqId = HConstants.NO_SEQNUM;</span> |
| <span class="source-line-no">235</span><span id="line-235"> }</span> |
| <span class="source-line-no">236</span><span id="line-236"> }</span> |
| <span class="source-line-no">237</span><span id="line-237"> try {</span> |
| <span class="source-line-no">238</span><span id="line-238"> updateTransitionWithoutPersistingToMeta(env, regionNode, transitionCode, seqId);</span> |
| <span class="source-line-no">239</span><span id="line-239"> } catch (IOException e) {</span> |
| <span class="source-line-no">240</span><span id="line-240"> throw new AssertionError("should not happen", e);</span> |
| <span class="source-line-no">241</span><span id="line-241"> }</span> |
| <span class="source-line-no">242</span><span id="line-242"> }</span> |
| <span class="source-line-no">243</span><span id="line-243"></span> |
| <span class="source-line-no">244</span><span id="line-244"> void serverCrashed(MasterProcedureEnv env, RegionStateNode regionNode, ServerName serverName) {</span> |
| <span class="source-line-no">245</span><span id="line-245"> if (state == RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_SERVER_CRASH) {</span> |
| <span class="source-line-no">246</span><span id="line-246"> // should be a retry</span> |
| <span class="source-line-no">247</span><span id="line-247"> return;</span> |
| <span class="source-line-no">248</span><span id="line-248"> }</span> |
| <span class="source-line-no">249</span><span id="line-249"> RegionRemoteProcedureBaseState oldState = state;</span> |
| <span class="source-line-no">250</span><span id="line-250"> // it is possible that the state is in REGION_REMOTE_PROCEDURE_SERVER_CRASH, think of this</span> |
| <span class="source-line-no">251</span><span id="line-251"> // sequence</span> |
| <span class="source-line-no">252</span><span id="line-252"> // 1. region is open on the target server and the above reportTransition call is succeeded</span> |
| <span class="source-line-no">253</span><span id="line-253"> // 2. before we are woken up and update the meta, the target server crashes, and then we arrive</span> |
| <span class="source-line-no">254</span><span id="line-254"> // here</span> |
| <span class="source-line-no">255</span><span id="line-255"> this.state = RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_SERVER_CRASH;</span> |
| <span class="source-line-no">256</span><span id="line-256"> boolean succ = false;</span> |
| <span class="source-line-no">257</span><span id="line-257"> try {</span> |
| <span class="source-line-no">258</span><span id="line-258"> persistAndWake(env, regionNode);</span> |
| <span class="source-line-no">259</span><span id="line-259"> succ = true;</span> |
| <span class="source-line-no">260</span><span id="line-260"> } finally {</span> |
| <span class="source-line-no">261</span><span id="line-261"> if (!succ) {</span> |
| <span class="source-line-no">262</span><span id="line-262"> this.state = oldState;</span> |
| <span class="source-line-no">263</span><span id="line-263"> }</span> |
| <span class="source-line-no">264</span><span id="line-264"> }</span> |
| <span class="source-line-no">265</span><span id="line-265"> }</span> |
| <span class="source-line-no">266</span><span id="line-266"></span> |
| <span class="source-line-no">267</span><span id="line-267"> protected abstract void restoreSucceedState(AssignmentManager am, RegionStateNode regionNode,</span> |
| <span class="source-line-no">268</span><span id="line-268"> long seqId) throws IOException;</span> |
| <span class="source-line-no">269</span><span id="line-269"></span> |
| <span class="source-line-no">270</span><span id="line-270"> void stateLoaded(AssignmentManager am, RegionStateNode regionNode) {</span> |
| <span class="source-line-no">271</span><span id="line-271"> if (state == RegionRemoteProcedureBaseState.REGION_REMOTE_PROCEDURE_REPORT_SUCCEED) {</span> |
| <span class="source-line-no">272</span><span id="line-272"> try {</span> |
| <span class="source-line-no">273</span><span id="line-273"> restoreSucceedState(am, regionNode, seqId);</span> |
| <span class="source-line-no">274</span><span id="line-274"> } catch (IOException e) {</span> |
| <span class="source-line-no">275</span><span id="line-275"> // should not happen as we are just restoring the state</span> |
| <span class="source-line-no">276</span><span id="line-276"> throw new AssertionError(e);</span> |
| <span class="source-line-no">277</span><span id="line-277"> }</span> |
| <span class="source-line-no">278</span><span id="line-278"> }</span> |
| <span class="source-line-no">279</span><span id="line-279"> }</span> |
| <span class="source-line-no">280</span><span id="line-280"></span> |
| <span class="source-line-no">281</span><span id="line-281"> private TransitRegionStateProcedure getParent(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">282</span><span id="line-282"> return (TransitRegionStateProcedure) env.getMasterServices().getMasterProcedureExecutor()</span> |
| <span class="source-line-no">283</span><span id="line-283"> .getProcedure(getParentProcId());</span> |
| <span class="source-line-no">284</span><span id="line-284"> }</span> |
| <span class="source-line-no">285</span><span id="line-285"></span> |
| <span class="source-line-no">286</span><span id="line-286"> private void unattach(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">287</span><span id="line-287"> getParent(env).unattachRemoteProc(this);</span> |
| <span class="source-line-no">288</span><span id="line-288"> }</span> |
| <span class="source-line-no">289</span><span id="line-289"></span> |
| <span class="source-line-no">290</span><span id="line-290"> private CompletableFuture<Void> getFuture() {</span> |
| <span class="source-line-no">291</span><span id="line-291"> return future;</span> |
| <span class="source-line-no">292</span><span id="line-292"> }</span> |
| <span class="source-line-no">293</span><span id="line-293"></span> |
| <span class="source-line-no">294</span><span id="line-294"> private void setFuture(CompletableFuture<Void> f) {</span> |
| <span class="source-line-no">295</span><span id="line-295"> future = f;</span> |
| <span class="source-line-no">296</span><span id="line-296"> }</span> |
| <span class="source-line-no">297</span><span id="line-297"></span> |
| <span class="source-line-no">298</span><span id="line-298"> @Override</span> |
| <span class="source-line-no">299</span><span id="line-299"> protected Procedure<MasterProcedureEnv>[] execute(MasterProcedureEnv env)</span> |
| <span class="source-line-no">300</span><span id="line-300"> throws ProcedureYieldException, ProcedureSuspendedException, InterruptedException {</span> |
| <span class="source-line-no">301</span><span id="line-301"> RegionStateNode regionNode = getRegionNode(env);</span> |
| <span class="source-line-no">302</span><span id="line-302"> if (!regionNode.isLockedBy(this)) {</span> |
| <span class="source-line-no">303</span><span id="line-303"> regionNode.lock(this, () -> ProcedureFutureUtil.wakeUp(this, env));</span> |
| <span class="source-line-no">304</span><span id="line-304"> }</span> |
| <span class="source-line-no">305</span><span id="line-305"> try {</span> |
| <span class="source-line-no">306</span><span id="line-306"> switch (state) {</span> |
| <span class="source-line-no">307</span><span id="line-307"> case REGION_REMOTE_PROCEDURE_DISPATCH: {</span> |
| <span class="source-line-no">308</span><span id="line-308"> // The code which wakes us up also needs to lock the RSN so here we do not need to</span> |
| <span class="source-line-no">309</span><span id="line-309"> // synchronize on the event.</span> |
| <span class="source-line-no">310</span><span id="line-310"> ProcedureEvent<?> event = regionNode.getProcedureEvent();</span> |
| <span class="source-line-no">311</span><span id="line-311"> try {</span> |
| <span class="source-line-no">312</span><span id="line-312"> env.getRemoteDispatcher().addOperationToNode(targetServer, this);</span> |
| <span class="source-line-no">313</span><span id="line-313"> } catch (FailedRemoteDispatchException e) {</span> |
| <span class="source-line-no">314</span><span id="line-314"> LOG.warn("Can not add remote operation {} for region {} to server {}, this usually "</span> |
| <span class="source-line-no">315</span><span id="line-315"> + "because the server is alread dead, give up and mark the procedure as complete, "</span> |
| <span class="source-line-no">316</span><span id="line-316"> + "the parent procedure will take care of this.", this, region, targetServer, e);</span> |
| <span class="source-line-no">317</span><span id="line-317"> unattach(env);</span> |
| <span class="source-line-no">318</span><span id="line-318"> return null;</span> |
| <span class="source-line-no">319</span><span id="line-319"> }</span> |
| <span class="source-line-no">320</span><span id="line-320"> event.suspend();</span> |
| <span class="source-line-no">321</span><span id="line-321"> event.suspendIfNotReady(this);</span> |
| <span class="source-line-no">322</span><span id="line-322"> throw new ProcedureSuspendedException();</span> |
| <span class="source-line-no">323</span><span id="line-323"> }</span> |
| <span class="source-line-no">324</span><span id="line-324"> case REGION_REMOTE_PROCEDURE_REPORT_SUCCEED:</span> |
| <span class="source-line-no">325</span><span id="line-325"> if (</span> |
| <span class="source-line-no">326</span><span id="line-326"> ProcedureFutureUtil.checkFuture(this, this::getFuture, this::setFuture,</span> |
| <span class="source-line-no">327</span><span id="line-327"> () -> unattach(env))</span> |
| <span class="source-line-no">328</span><span id="line-328"> ) {</span> |
| <span class="source-line-no">329</span><span id="line-329"> return null;</span> |
| <span class="source-line-no">330</span><span id="line-330"> }</span> |
| <span class="source-line-no">331</span><span id="line-331"> ProcedureFutureUtil.suspendIfNecessary(this, this::setFuture,</span> |
| <span class="source-line-no">332</span><span id="line-332"> env.getAssignmentManager().persistToMeta(regionNode), env, () -> unattach(env));</span> |
| <span class="source-line-no">333</span><span id="line-333"> return null;</span> |
| <span class="source-line-no">334</span><span id="line-334"> case REGION_REMOTE_PROCEDURE_DISPATCH_FAIL:</span> |
| <span class="source-line-no">335</span><span id="line-335"> // the remote call is failed so we do not need to change the region state, just return.</span> |
| <span class="source-line-no">336</span><span id="line-336"> unattach(env);</span> |
| <span class="source-line-no">337</span><span id="line-337"> return null;</span> |
| <span class="source-line-no">338</span><span id="line-338"> case REGION_REMOTE_PROCEDURE_SERVER_CRASH:</span> |
| <span class="source-line-no">339</span><span id="line-339"> if (</span> |
| <span class="source-line-no">340</span><span id="line-340"> ProcedureFutureUtil.checkFuture(this, this::getFuture, this::setFuture,</span> |
| <span class="source-line-no">341</span><span id="line-341"> () -> unattach(env))</span> |
| <span class="source-line-no">342</span><span id="line-342"> ) {</span> |
| <span class="source-line-no">343</span><span id="line-343"> return null;</span> |
| <span class="source-line-no">344</span><span id="line-344"> }</span> |
| <span class="source-line-no">345</span><span id="line-345"> ProcedureFutureUtil.suspendIfNecessary(this, this::setFuture,</span> |
| <span class="source-line-no">346</span><span id="line-346"> env.getAssignmentManager().regionClosedAbnormally(regionNode), env,</span> |
| <span class="source-line-no">347</span><span id="line-347"> () -> unattach(env));</span> |
| <span class="source-line-no">348</span><span id="line-348"> return null;</span> |
| <span class="source-line-no">349</span><span id="line-349"> default:</span> |
| <span class="source-line-no">350</span><span id="line-350"> throw new IllegalStateException("Unknown state: " + state);</span> |
| <span class="source-line-no">351</span><span id="line-351"> }</span> |
| <span class="source-line-no">352</span><span id="line-352"> } catch (IOException e) {</span> |
| <span class="source-line-no">353</span><span id="line-353"> if (retryCounter == null) {</span> |
| <span class="source-line-no">354</span><span id="line-354"> retryCounter = ProcedureUtil.createRetryCounter(env.getMasterConfiguration());</span> |
| <span class="source-line-no">355</span><span id="line-355"> }</span> |
| <span class="source-line-no">356</span><span id="line-356"> long backoff = retryCounter.getBackoffTimeAndIncrementAttempts();</span> |
| <span class="source-line-no">357</span><span id="line-357"> LOG.warn("Failed updating meta, suspend {}secs {}; {};", backoff / 1000, this, regionNode, e);</span> |
| <span class="source-line-no">358</span><span id="line-358"> throw suspend(Math.toIntExact(backoff), true);</span> |
| <span class="source-line-no">359</span><span id="line-359"> } finally {</span> |
| <span class="source-line-no">360</span><span id="line-360"> if (future == null) {</span> |
| <span class="source-line-no">361</span><span id="line-361"> regionNode.unlock(this);</span> |
| <span class="source-line-no">362</span><span id="line-362"> }</span> |
| <span class="source-line-no">363</span><span id="line-363"> }</span> |
| <span class="source-line-no">364</span><span id="line-364"> }</span> |
| <span class="source-line-no">365</span><span id="line-365"></span> |
| <span class="source-line-no">366</span><span id="line-366"> @Override</span> |
| <span class="source-line-no">367</span><span id="line-367"> protected synchronized boolean setTimeoutFailure(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">368</span><span id="line-368"> setState(ProcedureProtos.ProcedureState.RUNNABLE);</span> |
| <span class="source-line-no">369</span><span id="line-369"> env.getProcedureScheduler().addFront(this);</span> |
| <span class="source-line-no">370</span><span id="line-370"> return false; // 'false' means that this procedure handled the timeout</span> |
| <span class="source-line-no">371</span><span id="line-371"> }</span> |
| <span class="source-line-no">372</span><span id="line-372"></span> |
| <span class="source-line-no">373</span><span id="line-373"> @Override</span> |
| <span class="source-line-no">374</span><span id="line-374"> public boolean storeInDispatchedQueue() {</span> |
| <span class="source-line-no">375</span><span id="line-375"> return false;</span> |
| <span class="source-line-no">376</span><span id="line-376"> }</span> |
| <span class="source-line-no">377</span><span id="line-377"></span> |
| <span class="source-line-no">378</span><span id="line-378"> @Override</span> |
| <span class="source-line-no">379</span><span id="line-379"> protected void serializeStateData(ProcedureStateSerializer serializer) throws IOException {</span> |
| <span class="source-line-no">380</span><span id="line-380"> RegionRemoteProcedureBaseStateData.Builder builder =</span> |
| <span class="source-line-no">381</span><span id="line-381"> RegionRemoteProcedureBaseStateData.newBuilder().setRegion(ProtobufUtil.toRegionInfo(region))</span> |
| <span class="source-line-no">382</span><span id="line-382"> .setTargetServer(ProtobufUtil.toServerName(targetServer)).setState(state);</span> |
| <span class="source-line-no">383</span><span id="line-383"> if (transitionCode != null) {</span> |
| <span class="source-line-no">384</span><span id="line-384"> builder.setTransitionCode(transitionCode);</span> |
| <span class="source-line-no">385</span><span id="line-385"> builder.setSeqId(seqId);</span> |
| <span class="source-line-no">386</span><span id="line-386"> }</span> |
| <span class="source-line-no">387</span><span id="line-387"> serializer.serialize(builder.build());</span> |
| <span class="source-line-no">388</span><span id="line-388"> }</span> |
| <span class="source-line-no">389</span><span id="line-389"></span> |
| <span class="source-line-no">390</span><span id="line-390"> @Override</span> |
| <span class="source-line-no">391</span><span id="line-391"> protected void deserializeStateData(ProcedureStateSerializer serializer) throws IOException {</span> |
| <span class="source-line-no">392</span><span id="line-392"> RegionRemoteProcedureBaseStateData data =</span> |
| <span class="source-line-no">393</span><span id="line-393"> serializer.deserialize(RegionRemoteProcedureBaseStateData.class);</span> |
| <span class="source-line-no">394</span><span id="line-394"> region = ProtobufUtil.toRegionInfo(data.getRegion());</span> |
| <span class="source-line-no">395</span><span id="line-395"> targetServer = ProtobufUtil.toServerName(data.getTargetServer());</span> |
| <span class="source-line-no">396</span><span id="line-396"> // 'state' may not be present if we are reading an 'old' form of this pb Message.</span> |
| <span class="source-line-no">397</span><span id="line-397"> if (data.hasState()) {</span> |
| <span class="source-line-no">398</span><span id="line-398"> state = data.getState();</span> |
| <span class="source-line-no">399</span><span id="line-399"> }</span> |
| <span class="source-line-no">400</span><span id="line-400"> if (data.hasTransitionCode()) {</span> |
| <span class="source-line-no">401</span><span id="line-401"> transitionCode = data.getTransitionCode();</span> |
| <span class="source-line-no">402</span><span id="line-402"> seqId = data.getSeqId();</span> |
| <span class="source-line-no">403</span><span id="line-403"> }</span> |
| <span class="source-line-no">404</span><span id="line-404"> }</span> |
| <span class="source-line-no">405</span><span id="line-405"></span> |
| <span class="source-line-no">406</span><span id="line-406"> @Override</span> |
| <span class="source-line-no">407</span><span id="line-407"> protected void afterReplay(MasterProcedureEnv env) {</span> |
| <span class="source-line-no">408</span><span id="line-408"> getParent(env).attachRemoteProc(this);</span> |
| <span class="source-line-no">409</span><span id="line-409"> }</span> |
| <span class="source-line-no">410</span><span id="line-410"></span> |
| <span class="source-line-no">411</span><span id="line-411"> @Override</span> |
| <span class="source-line-no">412</span><span id="line-412"> public String getProcName() {</span> |
| <span class="source-line-no">413</span><span id="line-413"> return getClass().getSimpleName() + " " + region.getEncodedName();</span> |
| <span class="source-line-no">414</span><span id="line-414"> }</span> |
| <span class="source-line-no">415</span><span id="line-415"></span> |
| <span class="source-line-no">416</span><span id="line-416"> @Override</span> |
| <span class="source-line-no">417</span><span id="line-417"> protected void toStringClassDetails(StringBuilder builder) {</span> |
| <span class="source-line-no">418</span><span id="line-418"> builder.append(getProcName());</span> |
| <span class="source-line-no">419</span><span id="line-419"> if (targetServer != null) {</span> |
| <span class="source-line-no">420</span><span id="line-420"> builder.append(", server=");</span> |
| <span class="source-line-no">421</span><span id="line-421"> builder.append(this.targetServer);</span> |
| <span class="source-line-no">422</span><span id="line-422"> }</span> |
| <span class="source-line-no">423</span><span id="line-423"> if (this.retryCounter != null) {</span> |
| <span class="source-line-no">424</span><span id="line-424"> builder.append(", retry=");</span> |
| <span class="source-line-no">425</span><span id="line-425"> builder.append(this.retryCounter);</span> |
| <span class="source-line-no">426</span><span id="line-426"> }</span> |
| <span class="source-line-no">427</span><span id="line-427"> }</span> |
| <span class="source-line-no">428</span><span id="line-428">}</span> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </main> |
| </body> |
| </html> |
