| <!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.procedure2;<a name="line.18"></a> |
| <span class="sourceLineNo">019</span><a name="line.19"></a> |
| <span class="sourceLineNo">020</span>import java.io.IOException;<a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import java.io.UncheckedIOException;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span>import java.util.ArrayDeque;<a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.util.ArrayList;<a name="line.23"></a> |
| <span class="sourceLineNo">024</span>import java.util.Arrays;<a name="line.24"></a> |
| <span class="sourceLineNo">025</span>import java.util.Collection;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import java.util.Deque;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import java.util.HashSet;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span>import java.util.List;<a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import java.util.Set;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import java.util.concurrent.CopyOnWriteArrayList;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import java.util.concurrent.Executor;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import java.util.concurrent.Executors;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import java.util.concurrent.TimeUnit;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import java.util.concurrent.atomic.AtomicInteger;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import java.util.stream.Collectors;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import java.util.stream.Stream;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import org.apache.hadoop.conf.Configuration;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.HConstants;<a name="line.41"></a> |
| <span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.exceptions.IllegalArgumentIOException;<a name="line.42"></a> |
| <span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.log.HBaseMarkers;<a name="line.43"></a> |
| <span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.procedure2.Procedure.LockState;<a name="line.44"></a> |
| <span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStore;<a name="line.45"></a> |
| <span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;<a name="line.46"></a> |
| <span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureStoreListener;<a name="line.47"></a> |
| <span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.procedure2.util.StringUtils;<a name="line.48"></a> |
| <span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.security.User;<a name="line.49"></a> |
| <span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.50"></a> |
| <span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.util.IdLock;<a name="line.51"></a> |
| <span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.util.NonceKey;<a name="line.52"></a> |
| <span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.53"></a> |
| <span class="sourceLineNo">054</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.54"></a> |
| <span class="sourceLineNo">055</span>import org.slf4j.Logger;<a name="line.55"></a> |
| <span class="sourceLineNo">056</span>import org.slf4j.LoggerFactory;<a name="line.56"></a> |
| <span class="sourceLineNo">057</span><a name="line.57"></a> |
| <span class="sourceLineNo">058</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.58"></a> |
| <span class="sourceLineNo">059</span>import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;<a name="line.59"></a> |
| <span class="sourceLineNo">060</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.60"></a> |
| <span class="sourceLineNo">061</span><a name="line.61"></a> |
| <span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureState;<a name="line.62"></a> |
| <span class="sourceLineNo">063</span><a name="line.63"></a> |
| <span class="sourceLineNo">064</span>/**<a name="line.64"></a> |
| <span class="sourceLineNo">065</span> * Thread Pool that executes the submitted procedures.<a name="line.65"></a> |
| <span class="sourceLineNo">066</span> * The executor has a ProcedureStore associated.<a name="line.66"></a> |
| <span class="sourceLineNo">067</span> * Each operation is logged and on restart the pending procedures are resumed.<a name="line.67"></a> |
| <span class="sourceLineNo">068</span> *<a name="line.68"></a> |
| <span class="sourceLineNo">069</span> * Unless the Procedure code throws an error (e.g. invalid user input)<a name="line.69"></a> |
| <span class="sourceLineNo">070</span> * the procedure will complete (at some point in time), On restart the pending<a name="line.70"></a> |
| <span class="sourceLineNo">071</span> * procedures are resumed and the once failed will be rolledback.<a name="line.71"></a> |
| <span class="sourceLineNo">072</span> *<a name="line.72"></a> |
| <span class="sourceLineNo">073</span> * The user can add procedures to the executor via submitProcedure(proc)<a name="line.73"></a> |
| <span class="sourceLineNo">074</span> * check for the finished state via isFinished(procId)<a name="line.74"></a> |
| <span class="sourceLineNo">075</span> * and get the result via getResult(procId)<a name="line.75"></a> |
| <span class="sourceLineNo">076</span> */<a name="line.76"></a> |
| <span class="sourceLineNo">077</span>@InterfaceAudience.Private<a name="line.77"></a> |
| <span class="sourceLineNo">078</span>public class ProcedureExecutor<TEnvironment> {<a name="line.78"></a> |
| <span class="sourceLineNo">079</span> private static final Logger LOG = LoggerFactory.getLogger(ProcedureExecutor.class);<a name="line.79"></a> |
| <span class="sourceLineNo">080</span><a name="line.80"></a> |
| <span class="sourceLineNo">081</span> public static final String CHECK_OWNER_SET_CONF_KEY = "hbase.procedure.check.owner.set";<a name="line.81"></a> |
| <span class="sourceLineNo">082</span> private static final boolean DEFAULT_CHECK_OWNER_SET = false;<a name="line.82"></a> |
| <span class="sourceLineNo">083</span><a name="line.83"></a> |
| <span class="sourceLineNo">084</span> public static final String WORKER_KEEP_ALIVE_TIME_CONF_KEY =<a name="line.84"></a> |
| <span class="sourceLineNo">085</span> "hbase.procedure.worker.keep.alive.time.msec";<a name="line.85"></a> |
| <span class="sourceLineNo">086</span> private static final long DEFAULT_WORKER_KEEP_ALIVE_TIME = TimeUnit.MINUTES.toMillis(1);<a name="line.86"></a> |
| <span class="sourceLineNo">087</span><a name="line.87"></a> |
| <span class="sourceLineNo">088</span> public static final String EVICT_TTL_CONF_KEY = "hbase.procedure.cleaner.evict.ttl";<a name="line.88"></a> |
| <span class="sourceLineNo">089</span> static final int DEFAULT_EVICT_TTL = 15 * 60000; // 15min<a name="line.89"></a> |
| <span class="sourceLineNo">090</span><a name="line.90"></a> |
| <span class="sourceLineNo">091</span> public static final String EVICT_ACKED_TTL_CONF_KEY ="hbase.procedure.cleaner.acked.evict.ttl";<a name="line.91"></a> |
| <span class="sourceLineNo">092</span> static final int DEFAULT_ACKED_EVICT_TTL = 5 * 60000; // 5min<a name="line.92"></a> |
| <span class="sourceLineNo">093</span><a name="line.93"></a> |
| <span class="sourceLineNo">094</span> /**<a name="line.94"></a> |
| <span class="sourceLineNo">095</span> * {@link #testing} is non-null when ProcedureExecutor is being tested. Tests will try to<a name="line.95"></a> |
| <span class="sourceLineNo">096</span> * break PE having it fail at various junctures. When non-null, testing is set to an instance of<a name="line.96"></a> |
| <span class="sourceLineNo">097</span> * the below internal {@link Testing} class with flags set for the particular test.<a name="line.97"></a> |
| <span class="sourceLineNo">098</span> */<a name="line.98"></a> |
| <span class="sourceLineNo">099</span> volatile Testing testing = null;<a name="line.99"></a> |
| <span class="sourceLineNo">100</span><a name="line.100"></a> |
| <span class="sourceLineNo">101</span> /**<a name="line.101"></a> |
| <span class="sourceLineNo">102</span> * Class with parameters describing how to fail/die when in testing-context.<a name="line.102"></a> |
| <span class="sourceLineNo">103</span> */<a name="line.103"></a> |
| <span class="sourceLineNo">104</span> public static class Testing {<a name="line.104"></a> |
| <span class="sourceLineNo">105</span> protected volatile boolean killIfHasParent = true;<a name="line.105"></a> |
| <span class="sourceLineNo">106</span> protected volatile boolean killIfSuspended = false;<a name="line.106"></a> |
| <span class="sourceLineNo">107</span><a name="line.107"></a> |
| <span class="sourceLineNo">108</span> /**<a name="line.108"></a> |
| <span class="sourceLineNo">109</span> * Kill the PE BEFORE we store state to the WAL. Good for figuring out if a Procedure is<a name="line.109"></a> |
| <span class="sourceLineNo">110</span> * persisting all the state it needs to recover after a crash.<a name="line.110"></a> |
| <span class="sourceLineNo">111</span> */<a name="line.111"></a> |
| <span class="sourceLineNo">112</span> protected volatile boolean killBeforeStoreUpdate = false;<a name="line.112"></a> |
| <span class="sourceLineNo">113</span> protected volatile boolean toggleKillBeforeStoreUpdate = false;<a name="line.113"></a> |
| <span class="sourceLineNo">114</span><a name="line.114"></a> |
| <span class="sourceLineNo">115</span> /**<a name="line.115"></a> |
| <span class="sourceLineNo">116</span> * Set when we want to fail AFTER state has been stored into the WAL. Rarely used. HBASE-20978<a name="line.116"></a> |
| <span class="sourceLineNo">117</span> * is about a case where memory-state was being set after store to WAL where a crash could<a name="line.117"></a> |
| <span class="sourceLineNo">118</span> * cause us to get stuck. This flag allows killing at what was a vulnerable time.<a name="line.118"></a> |
| <span class="sourceLineNo">119</span> */<a name="line.119"></a> |
| <span class="sourceLineNo">120</span> protected volatile boolean killAfterStoreUpdate = false;<a name="line.120"></a> |
| <span class="sourceLineNo">121</span> protected volatile boolean toggleKillAfterStoreUpdate = false;<a name="line.121"></a> |
| <span class="sourceLineNo">122</span><a name="line.122"></a> |
| <span class="sourceLineNo">123</span> protected boolean shouldKillBeforeStoreUpdate() {<a name="line.123"></a> |
| <span class="sourceLineNo">124</span> final boolean kill = this.killBeforeStoreUpdate;<a name="line.124"></a> |
| <span class="sourceLineNo">125</span> if (this.toggleKillBeforeStoreUpdate) {<a name="line.125"></a> |
| <span class="sourceLineNo">126</span> this.killBeforeStoreUpdate = !kill;<a name="line.126"></a> |
| <span class="sourceLineNo">127</span> LOG.warn("Toggle KILL before store update to: " + this.killBeforeStoreUpdate);<a name="line.127"></a> |
| <span class="sourceLineNo">128</span> }<a name="line.128"></a> |
| <span class="sourceLineNo">129</span> return kill;<a name="line.129"></a> |
| <span class="sourceLineNo">130</span> }<a name="line.130"></a> |
| <span class="sourceLineNo">131</span><a name="line.131"></a> |
| <span class="sourceLineNo">132</span> protected boolean shouldKillBeforeStoreUpdate(boolean isSuspended, boolean hasParent) {<a name="line.132"></a> |
| <span class="sourceLineNo">133</span> if (isSuspended && !killIfSuspended) {<a name="line.133"></a> |
| <span class="sourceLineNo">134</span> return false;<a name="line.134"></a> |
| <span class="sourceLineNo">135</span> }<a name="line.135"></a> |
| <span class="sourceLineNo">136</span> if (hasParent && !killIfHasParent) {<a name="line.136"></a> |
| <span class="sourceLineNo">137</span> return false;<a name="line.137"></a> |
| <span class="sourceLineNo">138</span> }<a name="line.138"></a> |
| <span class="sourceLineNo">139</span> return shouldKillBeforeStoreUpdate();<a name="line.139"></a> |
| <span class="sourceLineNo">140</span> }<a name="line.140"></a> |
| <span class="sourceLineNo">141</span><a name="line.141"></a> |
| <span class="sourceLineNo">142</span> protected boolean shouldKillAfterStoreUpdate() {<a name="line.142"></a> |
| <span class="sourceLineNo">143</span> final boolean kill = this.killAfterStoreUpdate;<a name="line.143"></a> |
| <span class="sourceLineNo">144</span> if (this.toggleKillAfterStoreUpdate) {<a name="line.144"></a> |
| <span class="sourceLineNo">145</span> this.killAfterStoreUpdate = !kill;<a name="line.145"></a> |
| <span class="sourceLineNo">146</span> LOG.warn("Toggle KILL after store update to: " + this.killAfterStoreUpdate);<a name="line.146"></a> |
| <span class="sourceLineNo">147</span> }<a name="line.147"></a> |
| <span class="sourceLineNo">148</span> return kill;<a name="line.148"></a> |
| <span class="sourceLineNo">149</span> }<a name="line.149"></a> |
| <span class="sourceLineNo">150</span><a name="line.150"></a> |
| <span class="sourceLineNo">151</span> protected boolean shouldKillAfterStoreUpdate(final boolean isSuspended) {<a name="line.151"></a> |
| <span class="sourceLineNo">152</span> return (isSuspended && !killIfSuspended) ? false : shouldKillAfterStoreUpdate();<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><a name="line.155"></a> |
| <span class="sourceLineNo">156</span> public interface ProcedureExecutorListener {<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> void procedureLoaded(long procId);<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> void procedureAdded(long procId);<a name="line.158"></a> |
| <span class="sourceLineNo">159</span> void procedureFinished(long procId);<a name="line.159"></a> |
| <span class="sourceLineNo">160</span> }<a name="line.160"></a> |
| <span class="sourceLineNo">161</span><a name="line.161"></a> |
| <span class="sourceLineNo">162</span> /**<a name="line.162"></a> |
| <span class="sourceLineNo">163</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the Procedure.<a name="line.163"></a> |
| <span class="sourceLineNo">164</span> * Once a Root-Procedure completes (success or failure), the result will be added to this map.<a name="line.164"></a> |
| <span class="sourceLineNo">165</span> * The user of ProcedureExecutor should call getResult(procId) to get the result.<a name="line.165"></a> |
| <span class="sourceLineNo">166</span> */<a name="line.166"></a> |
| <span class="sourceLineNo">167</span> private final ConcurrentHashMap<Long, CompletedProcedureRetainer<TEnvironment>> completed =<a name="line.167"></a> |
| <span class="sourceLineNo">168</span> new ConcurrentHashMap<>();<a name="line.168"></a> |
| <span class="sourceLineNo">169</span><a name="line.169"></a> |
| <span class="sourceLineNo">170</span> /**<a name="line.170"></a> |
| <span class="sourceLineNo">171</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the RootProcedureState.<a name="line.171"></a> |
| <span class="sourceLineNo">172</span> * The RootProcedureState contains the execution stack of the Root-Procedure,<a name="line.172"></a> |
| <span class="sourceLineNo">173</span> * It is added to the map by submitProcedure() and removed on procedure completion.<a name="line.173"></a> |
| <span class="sourceLineNo">174</span> */<a name="line.174"></a> |
| <span class="sourceLineNo">175</span> private final ConcurrentHashMap<Long, RootProcedureState<TEnvironment>> rollbackStack =<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> new ConcurrentHashMap<>();<a name="line.176"></a> |
| <span class="sourceLineNo">177</span><a name="line.177"></a> |
| <span class="sourceLineNo">178</span> /**<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> * Helper map to lookup the live procedures by ID.<a name="line.179"></a> |
| <span class="sourceLineNo">180</span> * This map contains every procedure. root-procedures and subprocedures.<a name="line.180"></a> |
| <span class="sourceLineNo">181</span> */<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> private final ConcurrentHashMap<Long, Procedure<TEnvironment>> procedures =<a name="line.182"></a> |
| <span class="sourceLineNo">183</span> new ConcurrentHashMap<>();<a name="line.183"></a> |
| <span class="sourceLineNo">184</span><a name="line.184"></a> |
| <span class="sourceLineNo">185</span> /**<a name="line.185"></a> |
| <span class="sourceLineNo">186</span> * Helper map to lookup whether the procedure already issued from the same client. This map<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> * contains every root procedure.<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> */<a name="line.188"></a> |
| <span class="sourceLineNo">189</span> private final ConcurrentHashMap<NonceKey, Long> nonceKeysToProcIdsMap = new ConcurrentHashMap<>();<a name="line.189"></a> |
| <span class="sourceLineNo">190</span><a name="line.190"></a> |
| <span class="sourceLineNo">191</span> private final CopyOnWriteArrayList<ProcedureExecutorListener> listeners =<a name="line.191"></a> |
| <span class="sourceLineNo">192</span> new CopyOnWriteArrayList<>();<a name="line.192"></a> |
| <span class="sourceLineNo">193</span><a name="line.193"></a> |
| <span class="sourceLineNo">194</span> private Configuration conf;<a name="line.194"></a> |
| <span class="sourceLineNo">195</span><a name="line.195"></a> |
| <span class="sourceLineNo">196</span> /**<a name="line.196"></a> |
| <span class="sourceLineNo">197</span> * Created in the {@link #init(int, boolean)} method. Destroyed in {@link #join()} (FIX! Doing<a name="line.197"></a> |
| <span class="sourceLineNo">198</span> * resource handling rather than observing in a #join is unexpected).<a name="line.198"></a> |
| <span class="sourceLineNo">199</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.199"></a> |
| <span class="sourceLineNo">200</span> * (Should be ok).<a name="line.200"></a> |
| <span class="sourceLineNo">201</span> */<a name="line.201"></a> |
| <span class="sourceLineNo">202</span> private ThreadGroup threadGroup;<a name="line.202"></a> |
| <span class="sourceLineNo">203</span><a name="line.203"></a> |
| <span class="sourceLineNo">204</span> /**<a name="line.204"></a> |
| <span class="sourceLineNo">205</span> * Created in the {@link #init(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.205"></a> |
| <span class="sourceLineNo">206</span> * resource handling rather than observing in a #join is unexpected).<a name="line.206"></a> |
| <span class="sourceLineNo">207</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.207"></a> |
| <span class="sourceLineNo">208</span> * (Should be ok).<a name="line.208"></a> |
| <span class="sourceLineNo">209</span> */<a name="line.209"></a> |
| <span class="sourceLineNo">210</span> private CopyOnWriteArrayList<WorkerThread> workerThreads;<a name="line.210"></a> |
| <span class="sourceLineNo">211</span><a name="line.211"></a> |
| <span class="sourceLineNo">212</span> /**<a name="line.212"></a> |
| <span class="sourceLineNo">213</span> * Created in the {@link #init(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.213"></a> |
| <span class="sourceLineNo">214</span> * resource handling rather than observing in a #join is unexpected).<a name="line.214"></a> |
| <span class="sourceLineNo">215</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.215"></a> |
| <span class="sourceLineNo">216</span> * (Should be ok).<a name="line.216"></a> |
| <span class="sourceLineNo">217</span> */<a name="line.217"></a> |
| <span class="sourceLineNo">218</span> private TimeoutExecutorThread<TEnvironment> timeoutExecutor;<a name="line.218"></a> |
| <span class="sourceLineNo">219</span><a name="line.219"></a> |
| <span class="sourceLineNo">220</span> private int corePoolSize;<a name="line.220"></a> |
| <span class="sourceLineNo">221</span> private int maxPoolSize;<a name="line.221"></a> |
| <span class="sourceLineNo">222</span><a name="line.222"></a> |
| <span class="sourceLineNo">223</span> private volatile long keepAliveTime;<a name="line.223"></a> |
| <span class="sourceLineNo">224</span><a name="line.224"></a> |
| <span class="sourceLineNo">225</span> /**<a name="line.225"></a> |
| <span class="sourceLineNo">226</span> * Scheduler/Queue that contains runnable procedures.<a name="line.226"></a> |
| <span class="sourceLineNo">227</span> */<a name="line.227"></a> |
| <span class="sourceLineNo">228</span> private final ProcedureScheduler scheduler;<a name="line.228"></a> |
| <span class="sourceLineNo">229</span><a name="line.229"></a> |
| <span class="sourceLineNo">230</span> private final Executor forceUpdateExecutor = Executors.newSingleThreadExecutor(<a name="line.230"></a> |
| <span class="sourceLineNo">231</span> new ThreadFactoryBuilder().setDaemon(true).setNameFormat("Force-Update-PEWorker-%d").build());<a name="line.231"></a> |
| <span class="sourceLineNo">232</span><a name="line.232"></a> |
| <span class="sourceLineNo">233</span> private final AtomicLong lastProcId = new AtomicLong(-1);<a name="line.233"></a> |
| <span class="sourceLineNo">234</span> private final AtomicLong workerId = new AtomicLong(0);<a name="line.234"></a> |
| <span class="sourceLineNo">235</span> private final AtomicInteger activeExecutorCount = new AtomicInteger(0);<a name="line.235"></a> |
| <span class="sourceLineNo">236</span> private final AtomicBoolean running = new AtomicBoolean(false);<a name="line.236"></a> |
| <span class="sourceLineNo">237</span> private final TEnvironment environment;<a name="line.237"></a> |
| <span class="sourceLineNo">238</span> private final ProcedureStore store;<a name="line.238"></a> |
| <span class="sourceLineNo">239</span><a name="line.239"></a> |
| <span class="sourceLineNo">240</span> private final boolean checkOwnerSet;<a name="line.240"></a> |
| <span class="sourceLineNo">241</span><a name="line.241"></a> |
| <span class="sourceLineNo">242</span> // To prevent concurrent execution of the same procedure.<a name="line.242"></a> |
| <span class="sourceLineNo">243</span> // For some rare cases, especially if the procedure uses ProcedureEvent, it is possible that the<a name="line.243"></a> |
| <span class="sourceLineNo">244</span> // procedure is woken up before we finish the suspend which causes the same procedures to be<a name="line.244"></a> |
| <span class="sourceLineNo">245</span> // executed in parallel. This does lead to some problems, see HBASE-20939&HBASE-20949, and is also<a name="line.245"></a> |
| <span class="sourceLineNo">246</span> // a bit confusing to the developers. So here we introduce this lock to prevent the concurrent<a name="line.246"></a> |
| <span class="sourceLineNo">247</span> // execution of the same procedure.<a name="line.247"></a> |
| <span class="sourceLineNo">248</span> private final IdLock procExecutionLock = new IdLock();<a name="line.248"></a> |
| <span class="sourceLineNo">249</span><a name="line.249"></a> |
| <span class="sourceLineNo">250</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.250"></a> |
| <span class="sourceLineNo">251</span> final ProcedureStore store) {<a name="line.251"></a> |
| <span class="sourceLineNo">252</span> this(conf, environment, store, new SimpleProcedureScheduler());<a name="line.252"></a> |
| <span class="sourceLineNo">253</span> }<a name="line.253"></a> |
| <span class="sourceLineNo">254</span><a name="line.254"></a> |
| <span class="sourceLineNo">255</span> private boolean isRootFinished(Procedure<?> proc) {<a name="line.255"></a> |
| <span class="sourceLineNo">256</span> Procedure<?> rootProc = procedures.get(proc.getRootProcId());<a name="line.256"></a> |
| <span class="sourceLineNo">257</span> return rootProc == null || rootProc.isFinished();<a name="line.257"></a> |
| <span class="sourceLineNo">258</span> }<a name="line.258"></a> |
| <span class="sourceLineNo">259</span><a name="line.259"></a> |
| <span class="sourceLineNo">260</span> private void forceUpdateProcedure(long procId) throws IOException {<a name="line.260"></a> |
| <span class="sourceLineNo">261</span> IdLock.Entry lockEntry = procExecutionLock.getLockEntry(procId);<a name="line.261"></a> |
| <span class="sourceLineNo">262</span> try {<a name="line.262"></a> |
| <span class="sourceLineNo">263</span> Procedure<TEnvironment> proc = procedures.get(procId);<a name="line.263"></a> |
| <span class="sourceLineNo">264</span> if (proc != null) {<a name="line.264"></a> |
| <span class="sourceLineNo">265</span> if (proc.isFinished() && proc.hasParent() && isRootFinished(proc)) {<a name="line.265"></a> |
| <span class="sourceLineNo">266</span> LOG.debug("Procedure {} has already been finished and parent is succeeded," +<a name="line.266"></a> |
| <span class="sourceLineNo">267</span> " skip force updating", proc);<a name="line.267"></a> |
| <span class="sourceLineNo">268</span> return;<a name="line.268"></a> |
| <span class="sourceLineNo">269</span> }<a name="line.269"></a> |
| <span class="sourceLineNo">270</span> } else {<a name="line.270"></a> |
| <span class="sourceLineNo">271</span> CompletedProcedureRetainer<TEnvironment> retainer = completed.get(procId);<a name="line.271"></a> |
| <span class="sourceLineNo">272</span> if (retainer == null || retainer.getProcedure() instanceof FailedProcedure) {<a name="line.272"></a> |
| <span class="sourceLineNo">273</span> LOG.debug("No pending procedure with id = {}, skip force updating.", procId);<a name="line.273"></a> |
| <span class="sourceLineNo">274</span> return;<a name="line.274"></a> |
| <span class="sourceLineNo">275</span> }<a name="line.275"></a> |
| <span class="sourceLineNo">276</span> long evictTtl = conf.getInt(EVICT_TTL_CONF_KEY, DEFAULT_EVICT_TTL);<a name="line.276"></a> |
| <span class="sourceLineNo">277</span> long evictAckTtl = conf.getInt(EVICT_ACKED_TTL_CONF_KEY, DEFAULT_ACKED_EVICT_TTL);<a name="line.277"></a> |
| <span class="sourceLineNo">278</span> if (retainer.isExpired(System.currentTimeMillis(), evictTtl, evictAckTtl)) {<a name="line.278"></a> |
| <span class="sourceLineNo">279</span> LOG.debug("Procedure {} has already been finished and expired, skip force updating",<a name="line.279"></a> |
| <span class="sourceLineNo">280</span> procId);<a name="line.280"></a> |
| <span class="sourceLineNo">281</span> return;<a name="line.281"></a> |
| <span class="sourceLineNo">282</span> }<a name="line.282"></a> |
| <span class="sourceLineNo">283</span> proc = retainer.getProcedure();<a name="line.283"></a> |
| <span class="sourceLineNo">284</span> }<a name="line.284"></a> |
| <span class="sourceLineNo">285</span> LOG.debug("Force update procedure {}", proc);<a name="line.285"></a> |
| <span class="sourceLineNo">286</span> store.update(proc);<a name="line.286"></a> |
| <span class="sourceLineNo">287</span> } finally {<a name="line.287"></a> |
| <span class="sourceLineNo">288</span> procExecutionLock.releaseLockEntry(lockEntry);<a name="line.288"></a> |
| <span class="sourceLineNo">289</span> }<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> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.292"></a> |
| <span class="sourceLineNo">293</span> final ProcedureStore store, final ProcedureScheduler scheduler) {<a name="line.293"></a> |
| <span class="sourceLineNo">294</span> this.environment = environment;<a name="line.294"></a> |
| <span class="sourceLineNo">295</span> this.scheduler = scheduler;<a name="line.295"></a> |
| <span class="sourceLineNo">296</span> this.store = store;<a name="line.296"></a> |
| <span class="sourceLineNo">297</span> this.conf = conf;<a name="line.297"></a> |
| <span class="sourceLineNo">298</span> this.checkOwnerSet = conf.getBoolean(CHECK_OWNER_SET_CONF_KEY, DEFAULT_CHECK_OWNER_SET);<a name="line.298"></a> |
| <span class="sourceLineNo">299</span> refreshConfiguration(conf);<a name="line.299"></a> |
| <span class="sourceLineNo">300</span> store.registerListener(new ProcedureStoreListener() {<a name="line.300"></a> |
| <span class="sourceLineNo">301</span><a name="line.301"></a> |
| <span class="sourceLineNo">302</span> @Override<a name="line.302"></a> |
| <span class="sourceLineNo">303</span> public void forceUpdate(long[] procIds) {<a name="line.303"></a> |
| <span class="sourceLineNo">304</span> Arrays.stream(procIds).forEach(procId -> forceUpdateExecutor.execute(() -> {<a name="line.304"></a> |
| <span class="sourceLineNo">305</span> try {<a name="line.305"></a> |
| <span class="sourceLineNo">306</span> forceUpdateProcedure(procId);<a name="line.306"></a> |
| <span class="sourceLineNo">307</span> } catch (IOException e) {<a name="line.307"></a> |
| <span class="sourceLineNo">308</span> LOG.warn("Failed to force update procedure with pid={}", procId);<a name="line.308"></a> |
| <span class="sourceLineNo">309</span> }<a name="line.309"></a> |
| <span class="sourceLineNo">310</span> }));<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><a name="line.314"></a> |
| <span class="sourceLineNo">315</span> private void load(final boolean abortOnCorruption) throws IOException {<a name="line.315"></a> |
| <span class="sourceLineNo">316</span> Preconditions.checkArgument(completed.isEmpty(), "completed not empty");<a name="line.316"></a> |
| <span class="sourceLineNo">317</span> Preconditions.checkArgument(rollbackStack.isEmpty(), "rollback state not empty");<a name="line.317"></a> |
| <span class="sourceLineNo">318</span> Preconditions.checkArgument(procedures.isEmpty(), "procedure map not empty");<a name="line.318"></a> |
| <span class="sourceLineNo">319</span> Preconditions.checkArgument(scheduler.size() == 0, "run queue not empty");<a name="line.319"></a> |
| <span class="sourceLineNo">320</span><a name="line.320"></a> |
| <span class="sourceLineNo">321</span> store.load(new ProcedureStore.ProcedureLoader() {<a name="line.321"></a> |
| <span class="sourceLineNo">322</span> @Override<a name="line.322"></a> |
| <span class="sourceLineNo">323</span> public void setMaxProcId(long maxProcId) {<a name="line.323"></a> |
| <span class="sourceLineNo">324</span> assert lastProcId.get() < 0 : "expected only one call to setMaxProcId()";<a name="line.324"></a> |
| <span class="sourceLineNo">325</span> lastProcId.set(maxProcId);<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> @Override<a name="line.328"></a> |
| <span class="sourceLineNo">329</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.329"></a> |
| <span class="sourceLineNo">330</span> loadProcedures(procIter, abortOnCorruption);<a name="line.330"></a> |
| <span class="sourceLineNo">331</span> }<a name="line.331"></a> |
| <span class="sourceLineNo">332</span><a name="line.332"></a> |
| <span class="sourceLineNo">333</span> @Override<a name="line.333"></a> |
| <span class="sourceLineNo">334</span> public void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.334"></a> |
| <span class="sourceLineNo">335</span> int corruptedCount = 0;<a name="line.335"></a> |
| <span class="sourceLineNo">336</span> while (procIter.hasNext()) {<a name="line.336"></a> |
| <span class="sourceLineNo">337</span> Procedure<?> proc = procIter.next();<a name="line.337"></a> |
| <span class="sourceLineNo">338</span> LOG.error("Corrupt " + proc);<a name="line.338"></a> |
| <span class="sourceLineNo">339</span> corruptedCount++;<a name="line.339"></a> |
| <span class="sourceLineNo">340</span> }<a name="line.340"></a> |
| <span class="sourceLineNo">341</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.341"></a> |
| <span class="sourceLineNo">342</span> throw new IOException("found " + corruptedCount + " corrupted procedure(s) on replay");<a name="line.342"></a> |
| <span class="sourceLineNo">343</span> }<a name="line.343"></a> |
| <span class="sourceLineNo">344</span> }<a name="line.344"></a> |
| <span class="sourceLineNo">345</span> });<a name="line.345"></a> |
| <span class="sourceLineNo">346</span> }<a name="line.346"></a> |
| <span class="sourceLineNo">347</span><a name="line.347"></a> |
| <span class="sourceLineNo">348</span> private void restoreLock(Procedure<TEnvironment> proc, Set<Long> restored) {<a name="line.348"></a> |
| <span class="sourceLineNo">349</span> proc.restoreLock(getEnvironment());<a name="line.349"></a> |
| <span class="sourceLineNo">350</span> restored.add(proc.getProcId());<a name="line.350"></a> |
| <span class="sourceLineNo">351</span> }<a name="line.351"></a> |
| <span class="sourceLineNo">352</span><a name="line.352"></a> |
| <span class="sourceLineNo">353</span> private void restoreLocks(Deque<Procedure<TEnvironment>> stack, Set<Long> restored) {<a name="line.353"></a> |
| <span class="sourceLineNo">354</span> while (!stack.isEmpty()) {<a name="line.354"></a> |
| <span class="sourceLineNo">355</span> restoreLock(stack.pop(), restored);<a name="line.355"></a> |
| <span class="sourceLineNo">356</span> }<a name="line.356"></a> |
| <span class="sourceLineNo">357</span> }<a name="line.357"></a> |
| <span class="sourceLineNo">358</span><a name="line.358"></a> |
| <span class="sourceLineNo">359</span> // Restore the locks for all the procedures.<a name="line.359"></a> |
| <span class="sourceLineNo">360</span> // Notice that we need to restore the locks starting from the root proc, otherwise there will be<a name="line.360"></a> |
| <span class="sourceLineNo">361</span> // problem that a sub procedure may hold the exclusive lock first and then we are stuck when<a name="line.361"></a> |
| <span class="sourceLineNo">362</span> // calling the acquireLock method for the parent procedure.<a name="line.362"></a> |
| <span class="sourceLineNo">363</span> // The algorithm is straight-forward:<a name="line.363"></a> |
| <span class="sourceLineNo">364</span> // 1. Use a set to record the procedures which locks have already been restored.<a name="line.364"></a> |
| <span class="sourceLineNo">365</span> // 2. Use a stack to store the hierarchy of the procedures<a name="line.365"></a> |
| <span class="sourceLineNo">366</span> // 3. For all the procedure, we will first try to find its parent and push it into the stack,<a name="line.366"></a> |
| <span class="sourceLineNo">367</span> // unless<a name="line.367"></a> |
| <span class="sourceLineNo">368</span> // a. We have no parent, i.e, we are the root procedure<a name="line.368"></a> |
| <span class="sourceLineNo">369</span> // b. The lock has already been restored(by checking the set introduced in #1)<a name="line.369"></a> |
| <span class="sourceLineNo">370</span> // then we start to pop the stack and call acquireLock for each procedure.<a name="line.370"></a> |
| <span class="sourceLineNo">371</span> // Notice that this should be done for all procedures, not only the ones in runnableList.<a name="line.371"></a> |
| <span class="sourceLineNo">372</span> private void restoreLocks() {<a name="line.372"></a> |
| <span class="sourceLineNo">373</span> Set<Long> restored = new HashSet<>();<a name="line.373"></a> |
| <span class="sourceLineNo">374</span> Deque<Procedure<TEnvironment>> stack = new ArrayDeque<>();<a name="line.374"></a> |
| <span class="sourceLineNo">375</span> procedures.values().forEach(proc -> {<a name="line.375"></a> |
| <span class="sourceLineNo">376</span> for (;;) {<a name="line.376"></a> |
| <span class="sourceLineNo">377</span> if (restored.contains(proc.getProcId())) {<a name="line.377"></a> |
| <span class="sourceLineNo">378</span> restoreLocks(stack, restored);<a name="line.378"></a> |
| <span class="sourceLineNo">379</span> return;<a name="line.379"></a> |
| <span class="sourceLineNo">380</span> }<a name="line.380"></a> |
| <span class="sourceLineNo">381</span> if (!proc.hasParent()) {<a name="line.381"></a> |
| <span class="sourceLineNo">382</span> restoreLock(proc, restored);<a name="line.382"></a> |
| <span class="sourceLineNo">383</span> restoreLocks(stack, restored);<a name="line.383"></a> |
| <span class="sourceLineNo">384</span> return;<a name="line.384"></a> |
| <span class="sourceLineNo">385</span> }<a name="line.385"></a> |
| <span class="sourceLineNo">386</span> stack.push(proc);<a name="line.386"></a> |
| <span class="sourceLineNo">387</span> proc = procedures.get(proc.getParentProcId());<a name="line.387"></a> |
| <span class="sourceLineNo">388</span> }<a name="line.388"></a> |
| <span class="sourceLineNo">389</span> });<a name="line.389"></a> |
| <span class="sourceLineNo">390</span> }<a name="line.390"></a> |
| <span class="sourceLineNo">391</span><a name="line.391"></a> |
| <span class="sourceLineNo">392</span> private void loadProcedures(ProcedureIterator procIter, boolean abortOnCorruption)<a name="line.392"></a> |
| <span class="sourceLineNo">393</span> throws IOException {<a name="line.393"></a> |
| <span class="sourceLineNo">394</span> // 1. Build the rollback stack<a name="line.394"></a> |
| <span class="sourceLineNo">395</span> int runnableCount = 0;<a name="line.395"></a> |
| <span class="sourceLineNo">396</span> int failedCount = 0;<a name="line.396"></a> |
| <span class="sourceLineNo">397</span> int waitingCount = 0;<a name="line.397"></a> |
| <span class="sourceLineNo">398</span> int waitingTimeoutCount = 0;<a name="line.398"></a> |
| <span class="sourceLineNo">399</span> while (procIter.hasNext()) {<a name="line.399"></a> |
| <span class="sourceLineNo">400</span> boolean finished = procIter.isNextFinished();<a name="line.400"></a> |
| <span class="sourceLineNo">401</span> @SuppressWarnings("unchecked")<a name="line.401"></a> |
| <span class="sourceLineNo">402</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.402"></a> |
| <span class="sourceLineNo">403</span> NonceKey nonceKey = proc.getNonceKey();<a name="line.403"></a> |
| <span class="sourceLineNo">404</span> long procId = proc.getProcId();<a name="line.404"></a> |
| <span class="sourceLineNo">405</span><a name="line.405"></a> |
| <span class="sourceLineNo">406</span> if (finished) {<a name="line.406"></a> |
| <span class="sourceLineNo">407</span> completed.put(proc.getProcId(), new CompletedProcedureRetainer<>(proc));<a name="line.407"></a> |
| <span class="sourceLineNo">408</span> LOG.debug("Completed {}", proc);<a name="line.408"></a> |
| <span class="sourceLineNo">409</span> } else {<a name="line.409"></a> |
| <span class="sourceLineNo">410</span> if (!proc.hasParent()) {<a name="line.410"></a> |
| <span class="sourceLineNo">411</span> assert !proc.isFinished() : "unexpected finished procedure";<a name="line.411"></a> |
| <span class="sourceLineNo">412</span> rollbackStack.put(proc.getProcId(), new RootProcedureState<>());<a name="line.412"></a> |
| <span class="sourceLineNo">413</span> }<a name="line.413"></a> |
| <span class="sourceLineNo">414</span><a name="line.414"></a> |
| <span class="sourceLineNo">415</span> // add the procedure to the map<a name="line.415"></a> |
| <span class="sourceLineNo">416</span> proc.beforeReplay(getEnvironment());<a name="line.416"></a> |
| <span class="sourceLineNo">417</span> procedures.put(proc.getProcId(), proc);<a name="line.417"></a> |
| <span class="sourceLineNo">418</span> switch (proc.getState()) {<a name="line.418"></a> |
| <span class="sourceLineNo">419</span> case RUNNABLE:<a name="line.419"></a> |
| <span class="sourceLineNo">420</span> runnableCount++;<a name="line.420"></a> |
| <span class="sourceLineNo">421</span> break;<a name="line.421"></a> |
| <span class="sourceLineNo">422</span> case FAILED:<a name="line.422"></a> |
| <span class="sourceLineNo">423</span> failedCount++;<a name="line.423"></a> |
| <span class="sourceLineNo">424</span> break;<a name="line.424"></a> |
| <span class="sourceLineNo">425</span> case WAITING:<a name="line.425"></a> |
| <span class="sourceLineNo">426</span> waitingCount++;<a name="line.426"></a> |
| <span class="sourceLineNo">427</span> break;<a name="line.427"></a> |
| <span class="sourceLineNo">428</span> case WAITING_TIMEOUT:<a name="line.428"></a> |
| <span class="sourceLineNo">429</span> waitingTimeoutCount++;<a name="line.429"></a> |
| <span class="sourceLineNo">430</span> break;<a name="line.430"></a> |
| <span class="sourceLineNo">431</span> default:<a name="line.431"></a> |
| <span class="sourceLineNo">432</span> break;<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><a name="line.435"></a> |
| <span class="sourceLineNo">436</span> if (nonceKey != null) {<a name="line.436"></a> |
| <span class="sourceLineNo">437</span> nonceKeysToProcIdsMap.put(nonceKey, procId); // add the nonce to the map<a name="line.437"></a> |
| <span class="sourceLineNo">438</span> }<a name="line.438"></a> |
| <span class="sourceLineNo">439</span> }<a name="line.439"></a> |
| <span class="sourceLineNo">440</span><a name="line.440"></a> |
| <span class="sourceLineNo">441</span> // 2. Initialize the stacks: In the old implementation, for procedures in FAILED state, we will<a name="line.441"></a> |
| <span class="sourceLineNo">442</span> // push it into the ProcedureScheduler directly to execute the rollback. But this does not work<a name="line.442"></a> |
| <span class="sourceLineNo">443</span> // after we introduce the restore lock stage. For now, when we acquire a xlock, we will remove<a name="line.443"></a> |
| <span class="sourceLineNo">444</span> // the queue from runQueue in scheduler, and then when a procedure which has lock access, for<a name="line.444"></a> |
| <span class="sourceLineNo">445</span> // example, a sub procedure of the procedure which has the xlock, is pushed into the scheduler,<a name="line.445"></a> |
| <span class="sourceLineNo">446</span> // we will add the queue back to let the workers poll from it. The assumption here is that, the<a name="line.446"></a> |
| <span class="sourceLineNo">447</span> // procedure which has the xlock should have been polled out already, so when loading we can not<a name="line.447"></a> |
| <span class="sourceLineNo">448</span> // add the procedure to scheduler first and then call acquireLock, since the procedure is still<a name="line.448"></a> |
| <span class="sourceLineNo">449</span> // in the queue, and since we will remove the queue from runQueue, then no one can poll it out,<a name="line.449"></a> |
| <span class="sourceLineNo">450</span> // then there is a dead lock<a name="line.450"></a> |
| <span class="sourceLineNo">451</span> List<Procedure<TEnvironment>> runnableList = new ArrayList<>(runnableCount);<a name="line.451"></a> |
| <span class="sourceLineNo">452</span> List<Procedure<TEnvironment>> failedList = new ArrayList<>(failedCount);<a name="line.452"></a> |
| <span class="sourceLineNo">453</span> List<Procedure<TEnvironment>> waitingList = new ArrayList<>(waitingCount);<a name="line.453"></a> |
| <span class="sourceLineNo">454</span> List<Procedure<TEnvironment>> waitingTimeoutList = new ArrayList<>(waitingTimeoutCount);<a name="line.454"></a> |
| <span class="sourceLineNo">455</span> procIter.reset();<a name="line.455"></a> |
| <span class="sourceLineNo">456</span> while (procIter.hasNext()) {<a name="line.456"></a> |
| <span class="sourceLineNo">457</span> if (procIter.isNextFinished()) {<a name="line.457"></a> |
| <span class="sourceLineNo">458</span> procIter.skipNext();<a name="line.458"></a> |
| <span class="sourceLineNo">459</span> continue;<a name="line.459"></a> |
| <span class="sourceLineNo">460</span> }<a name="line.460"></a> |
| <span class="sourceLineNo">461</span><a name="line.461"></a> |
| <span class="sourceLineNo">462</span> @SuppressWarnings("unchecked")<a name="line.462"></a> |
| <span class="sourceLineNo">463</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.463"></a> |
| <span class="sourceLineNo">464</span> assert !(proc.isFinished() && !proc.hasParent()) : "unexpected completed proc=" + proc;<a name="line.464"></a> |
| <span class="sourceLineNo">465</span> LOG.debug("Loading {}", proc);<a name="line.465"></a> |
| <span class="sourceLineNo">466</span> Long rootProcId = getRootProcedureId(proc);<a name="line.466"></a> |
| <span class="sourceLineNo">467</span> // The orphan procedures will be passed to handleCorrupted, so add an assert here<a name="line.467"></a> |
| <span class="sourceLineNo">468</span> assert rootProcId != null;<a name="line.468"></a> |
| <span class="sourceLineNo">469</span><a name="line.469"></a> |
| <span class="sourceLineNo">470</span> if (proc.hasParent()) {<a name="line.470"></a> |
| <span class="sourceLineNo">471</span> Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());<a name="line.471"></a> |
| <span class="sourceLineNo">472</span> if (parent != null && !proc.isFinished()) {<a name="line.472"></a> |
| <span class="sourceLineNo">473</span> parent.incChildrenLatch();<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><a name="line.476"></a> |
| <span class="sourceLineNo">477</span> RootProcedureState<TEnvironment> procStack = rollbackStack.get(rootProcId);<a name="line.477"></a> |
| <span class="sourceLineNo">478</span> procStack.loadStack(proc);<a name="line.478"></a> |
| <span class="sourceLineNo">479</span><a name="line.479"></a> |
| <span class="sourceLineNo">480</span> proc.setRootProcId(rootProcId);<a name="line.480"></a> |
| <span class="sourceLineNo">481</span> switch (proc.getState()) {<a name="line.481"></a> |
| <span class="sourceLineNo">482</span> case RUNNABLE:<a name="line.482"></a> |
| <span class="sourceLineNo">483</span> runnableList.add(proc);<a name="line.483"></a> |
| <span class="sourceLineNo">484</span> break;<a name="line.484"></a> |
| <span class="sourceLineNo">485</span> case WAITING:<a name="line.485"></a> |
| <span class="sourceLineNo">486</span> waitingList.add(proc);<a name="line.486"></a> |
| <span class="sourceLineNo">487</span> break;<a name="line.487"></a> |
| <span class="sourceLineNo">488</span> case WAITING_TIMEOUT:<a name="line.488"></a> |
| <span class="sourceLineNo">489</span> waitingTimeoutList.add(proc);<a name="line.489"></a> |
| <span class="sourceLineNo">490</span> break;<a name="line.490"></a> |
| <span class="sourceLineNo">491</span> case FAILED:<a name="line.491"></a> |
| <span class="sourceLineNo">492</span> failedList.add(proc);<a name="line.492"></a> |
| <span class="sourceLineNo">493</span> break;<a name="line.493"></a> |
| <span class="sourceLineNo">494</span> case ROLLEDBACK:<a name="line.494"></a> |
| <span class="sourceLineNo">495</span> case INITIALIZING:<a name="line.495"></a> |
| <span class="sourceLineNo">496</span> String msg = "Unexpected " + proc.getState() + " state for " + proc;<a name="line.496"></a> |
| <span class="sourceLineNo">497</span> LOG.error(msg);<a name="line.497"></a> |
| <span class="sourceLineNo">498</span> throw new UnsupportedOperationException(msg);<a name="line.498"></a> |
| <span class="sourceLineNo">499</span> default:<a name="line.499"></a> |
| <span class="sourceLineNo">500</span> break;<a name="line.500"></a> |
| <span class="sourceLineNo">501</span> }<a name="line.501"></a> |
| <span class="sourceLineNo">502</span> }<a name="line.502"></a> |
| <span class="sourceLineNo">503</span><a name="line.503"></a> |
| <span class="sourceLineNo">504</span> // 3. Check the waiting procedures to see if some of them can be added to runnable.<a name="line.504"></a> |
| <span class="sourceLineNo">505</span> waitingList.forEach(proc -> {<a name="line.505"></a> |
| <span class="sourceLineNo">506</span> if (!proc.hasChildren()) {<a name="line.506"></a> |
| <span class="sourceLineNo">507</span> // Normally, WAITING procedures should be waken by its children. But, there is a case that,<a name="line.507"></a> |
| <span class="sourceLineNo">508</span> // all the children are successful and before they can wake up their parent procedure, the<a name="line.508"></a> |
| <span class="sourceLineNo">509</span> // master was killed. So, during recovering the procedures from ProcedureWal, its children<a name="line.509"></a> |
| <span class="sourceLineNo">510</span> // are not loaded because of their SUCCESS state. So we need to continue to run this WAITING<a name="line.510"></a> |
| <span class="sourceLineNo">511</span> // procedure. But before executing, we need to set its state to RUNNABLE, otherwise, a<a name="line.511"></a> |
| <span class="sourceLineNo">512</span> // exception will throw:<a name="line.512"></a> |
| <span class="sourceLineNo">513</span> // Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,<a name="line.513"></a> |
| <span class="sourceLineNo">514</span> // "NOT RUNNABLE! " + procedure.toString());<a name="line.514"></a> |
| <span class="sourceLineNo">515</span> proc.setState(ProcedureState.RUNNABLE);<a name="line.515"></a> |
| <span class="sourceLineNo">516</span> runnableList.add(proc);<a name="line.516"></a> |
| <span class="sourceLineNo">517</span> } else {<a name="line.517"></a> |
| <span class="sourceLineNo">518</span> proc.afterReplay(getEnvironment());<a name="line.518"></a> |
| <span class="sourceLineNo">519</span> }<a name="line.519"></a> |
| <span class="sourceLineNo">520</span> });<a name="line.520"></a> |
| <span class="sourceLineNo">521</span> // 4. restore locks<a name="line.521"></a> |
| <span class="sourceLineNo">522</span> restoreLocks();<a name="line.522"></a> |
| <span class="sourceLineNo">523</span><a name="line.523"></a> |
| <span class="sourceLineNo">524</span> // 5. Push the procedures to the timeout executor<a name="line.524"></a> |
| <span class="sourceLineNo">525</span> waitingTimeoutList.forEach(proc -> {<a name="line.525"></a> |
| <span class="sourceLineNo">526</span> proc.afterReplay(getEnvironment());<a name="line.526"></a> |
| <span class="sourceLineNo">527</span> timeoutExecutor.add(proc);<a name="line.527"></a> |
| <span class="sourceLineNo">528</span> });<a name="line.528"></a> |
| <span class="sourceLineNo">529</span><a name="line.529"></a> |
| <span class="sourceLineNo">530</span> // 6. Push the procedure to the scheduler<a name="line.530"></a> |
| <span class="sourceLineNo">531</span> failedList.forEach(scheduler::addBack);<a name="line.531"></a> |
| <span class="sourceLineNo">532</span> runnableList.forEach(p -> {<a name="line.532"></a> |
| <span class="sourceLineNo">533</span> p.afterReplay(getEnvironment());<a name="line.533"></a> |
| <span class="sourceLineNo">534</span> if (!p.hasParent()) {<a name="line.534"></a> |
| <span class="sourceLineNo">535</span> sendProcedureLoadedNotification(p.getProcId());<a name="line.535"></a> |
| <span class="sourceLineNo">536</span> }<a name="line.536"></a> |
| <span class="sourceLineNo">537</span> scheduler.addBack(p);<a name="line.537"></a> |
| <span class="sourceLineNo">538</span> });<a name="line.538"></a> |
| <span class="sourceLineNo">539</span> // After all procedures put into the queue, signal the worker threads.<a name="line.539"></a> |
| <span class="sourceLineNo">540</span> // Otherwise, there is a race condition. See HBASE-21364.<a name="line.540"></a> |
| <span class="sourceLineNo">541</span> scheduler.signalAll();<a name="line.541"></a> |
| <span class="sourceLineNo">542</span> }<a name="line.542"></a> |
| <span class="sourceLineNo">543</span><a name="line.543"></a> |
| <span class="sourceLineNo">544</span> /**<a name="line.544"></a> |
| <span class="sourceLineNo">545</span> * Initialize the procedure executor, but do not start workers. We will start them later.<a name="line.545"></a> |
| <span class="sourceLineNo">546</span> * <p/><a name="line.546"></a> |
| <span class="sourceLineNo">547</span> * It calls ProcedureStore.recoverLease() and ProcedureStore.load() to recover the lease, and<a name="line.547"></a> |
| <span class="sourceLineNo">548</span> * ensure a single executor, and start the procedure replay to resume and recover the previous<a name="line.548"></a> |
| <span class="sourceLineNo">549</span> * pending and in-progress procedures.<a name="line.549"></a> |
| <span class="sourceLineNo">550</span> * @param numThreads number of threads available for procedure execution.<a name="line.550"></a> |
| <span class="sourceLineNo">551</span> * @param abortOnCorruption true if you want to abort your service in case a corrupted procedure<a name="line.551"></a> |
| <span class="sourceLineNo">552</span> * is found on replay. otherwise false.<a name="line.552"></a> |
| <span class="sourceLineNo">553</span> */<a name="line.553"></a> |
| <span class="sourceLineNo">554</span> public void init(int numThreads, boolean abortOnCorruption) throws IOException {<a name="line.554"></a> |
| <span class="sourceLineNo">555</span> // We have numThreads executor + one timer thread used for timing out<a name="line.555"></a> |
| <span class="sourceLineNo">556</span> // procedures and triggering periodic procedures.<a name="line.556"></a> |
| <span class="sourceLineNo">557</span> this.corePoolSize = numThreads;<a name="line.557"></a> |
| <span class="sourceLineNo">558</span> this.maxPoolSize = 10 * numThreads;<a name="line.558"></a> |
| <span class="sourceLineNo">559</span> LOG.info("Starting {} core workers (bigger of cpus/4 or 16) with max (burst) worker count={}",<a name="line.559"></a> |
| <span class="sourceLineNo">560</span> corePoolSize, maxPoolSize);<a name="line.560"></a> |
| <span class="sourceLineNo">561</span><a name="line.561"></a> |
| <span class="sourceLineNo">562</span> this.threadGroup = new ThreadGroup("PEWorkerGroup");<a name="line.562"></a> |
| <span class="sourceLineNo">563</span> this.timeoutExecutor = new TimeoutExecutorThread<>(this, threadGroup);<a name="line.563"></a> |
| <span class="sourceLineNo">564</span><a name="line.564"></a> |
| <span class="sourceLineNo">565</span> // Create the workers<a name="line.565"></a> |
| <span class="sourceLineNo">566</span> workerId.set(0);<a name="line.566"></a> |
| <span class="sourceLineNo">567</span> workerThreads = new CopyOnWriteArrayList<>();<a name="line.567"></a> |
| <span class="sourceLineNo">568</span> for (int i = 0; i < corePoolSize; ++i) {<a name="line.568"></a> |
| <span class="sourceLineNo">569</span> workerThreads.add(new WorkerThread(threadGroup));<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> long st, et;<a name="line.572"></a> |
| <span class="sourceLineNo">573</span><a name="line.573"></a> |
| <span class="sourceLineNo">574</span> // Acquire the store lease.<a name="line.574"></a> |
| <span class="sourceLineNo">575</span> st = System.nanoTime();<a name="line.575"></a> |
| <span class="sourceLineNo">576</span> store.recoverLease();<a name="line.576"></a> |
| <span class="sourceLineNo">577</span> et = System.nanoTime();<a name="line.577"></a> |
| <span class="sourceLineNo">578</span> LOG.info("Recovered {} lease in {}", store.getClass().getSimpleName(),<a name="line.578"></a> |
| <span class="sourceLineNo">579</span> StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.579"></a> |
| <span class="sourceLineNo">580</span><a name="line.580"></a> |
| <span class="sourceLineNo">581</span> // start the procedure scheduler<a name="line.581"></a> |
| <span class="sourceLineNo">582</span> scheduler.start();<a name="line.582"></a> |
| <span class="sourceLineNo">583</span><a name="line.583"></a> |
| <span class="sourceLineNo">584</span> // TODO: Split in two steps.<a name="line.584"></a> |
| <span class="sourceLineNo">585</span> // TODO: Handle corrupted procedures (currently just a warn)<a name="line.585"></a> |
| <span class="sourceLineNo">586</span> // The first one will make sure that we have the latest id,<a name="line.586"></a> |
| <span class="sourceLineNo">587</span> // so we can start the threads and accept new procedures.<a name="line.587"></a> |
| <span class="sourceLineNo">588</span> // The second step will do the actual load of old procedures.<a name="line.588"></a> |
| <span class="sourceLineNo">589</span> st = System.nanoTime();<a name="line.589"></a> |
| <span class="sourceLineNo">590</span> load(abortOnCorruption);<a name="line.590"></a> |
| <span class="sourceLineNo">591</span> et = System.nanoTime();<a name="line.591"></a> |
| <span class="sourceLineNo">592</span> LOG.info("Loaded {} in {}", store.getClass().getSimpleName(),<a name="line.592"></a> |
| <span class="sourceLineNo">593</span> StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.593"></a> |
| <span class="sourceLineNo">594</span> }<a name="line.594"></a> |
| <span class="sourceLineNo">595</span><a name="line.595"></a> |
| <span class="sourceLineNo">596</span> /**<a name="line.596"></a> |
| <span class="sourceLineNo">597</span> * Start the workers.<a name="line.597"></a> |
| <span class="sourceLineNo">598</span> */<a name="line.598"></a> |
| <span class="sourceLineNo">599</span> public void startWorkers() throws IOException {<a name="line.599"></a> |
| <span class="sourceLineNo">600</span> if (!running.compareAndSet(false, true)) {<a name="line.600"></a> |
| <span class="sourceLineNo">601</span> LOG.warn("Already running");<a name="line.601"></a> |
| <span class="sourceLineNo">602</span> return;<a name="line.602"></a> |
| <span class="sourceLineNo">603</span> }<a name="line.603"></a> |
| <span class="sourceLineNo">604</span> // Start the executors. Here we must have the lastProcId set.<a name="line.604"></a> |
| <span class="sourceLineNo">605</span> LOG.trace("Start workers {}", workerThreads.size());<a name="line.605"></a> |
| <span class="sourceLineNo">606</span> timeoutExecutor.start();<a name="line.606"></a> |
| <span class="sourceLineNo">607</span> for (WorkerThread worker: workerThreads) {<a name="line.607"></a> |
| <span class="sourceLineNo">608</span> worker.start();<a name="line.608"></a> |
| <span class="sourceLineNo">609</span> }<a name="line.609"></a> |
| <span class="sourceLineNo">610</span><a name="line.610"></a> |
| <span class="sourceLineNo">611</span> // Internal chores<a name="line.611"></a> |
| <span class="sourceLineNo">612</span> timeoutExecutor.add(new WorkerMonitor());<a name="line.612"></a> |
| <span class="sourceLineNo">613</span><a name="line.613"></a> |
| <span class="sourceLineNo">614</span> // Add completed cleaner chore<a name="line.614"></a> |
| <span class="sourceLineNo">615</span> addChore(new CompletedProcedureCleaner<>(conf, store, procExecutionLock, completed,<a name="line.615"></a> |
| <span class="sourceLineNo">616</span> nonceKeysToProcIdsMap));<a name="line.616"></a> |
| <span class="sourceLineNo">617</span> }<a name="line.617"></a> |
| <span class="sourceLineNo">618</span><a name="line.618"></a> |
| <span class="sourceLineNo">619</span> public void stop() {<a name="line.619"></a> |
| <span class="sourceLineNo">620</span> if (!running.getAndSet(false)) {<a name="line.620"></a> |
| <span class="sourceLineNo">621</span> return;<a name="line.621"></a> |
| <span class="sourceLineNo">622</span> }<a name="line.622"></a> |
| <span class="sourceLineNo">623</span><a name="line.623"></a> |
| <span class="sourceLineNo">624</span> LOG.info("Stopping");<a name="line.624"></a> |
| <span class="sourceLineNo">625</span> scheduler.stop();<a name="line.625"></a> |
| <span class="sourceLineNo">626</span> timeoutExecutor.sendStopSignal();<a name="line.626"></a> |
| <span class="sourceLineNo">627</span> }<a name="line.627"></a> |
| <span class="sourceLineNo">628</span><a name="line.628"></a> |
| <span class="sourceLineNo">629</span> @VisibleForTesting<a name="line.629"></a> |
| <span class="sourceLineNo">630</span> public void join() {<a name="line.630"></a> |
| <span class="sourceLineNo">631</span> assert !isRunning() : "expected not running";<a name="line.631"></a> |
| <span class="sourceLineNo">632</span><a name="line.632"></a> |
| <span class="sourceLineNo">633</span> // stop the timeout executor<a name="line.633"></a> |
| <span class="sourceLineNo">634</span> timeoutExecutor.awaitTermination();<a name="line.634"></a> |
| <span class="sourceLineNo">635</span><a name="line.635"></a> |
| <span class="sourceLineNo">636</span> // stop the worker threads<a name="line.636"></a> |
| <span class="sourceLineNo">637</span> for (WorkerThread worker: workerThreads) {<a name="line.637"></a> |
| <span class="sourceLineNo">638</span> worker.awaitTermination();<a name="line.638"></a> |
| <span class="sourceLineNo">639</span> }<a name="line.639"></a> |
| <span class="sourceLineNo">640</span><a name="line.640"></a> |
| <span class="sourceLineNo">641</span> // Destroy the Thread Group for the executors<a name="line.641"></a> |
| <span class="sourceLineNo">642</span> // TODO: Fix. #join is not place to destroy resources.<a name="line.642"></a> |
| <span class="sourceLineNo">643</span> try {<a name="line.643"></a> |
| <span class="sourceLineNo">644</span> threadGroup.destroy();<a name="line.644"></a> |
| <span class="sourceLineNo">645</span> } catch (IllegalThreadStateException e) {<a name="line.645"></a> |
| <span class="sourceLineNo">646</span> LOG.error("ThreadGroup {} contains running threads; {}: See STDOUT",<a name="line.646"></a> |
| <span class="sourceLineNo">647</span> this.threadGroup, e.getMessage());<a name="line.647"></a> |
| <span class="sourceLineNo">648</span> // This dumps list of threads on STDOUT.<a name="line.648"></a> |
| <span class="sourceLineNo">649</span> this.threadGroup.list();<a name="line.649"></a> |
| <span class="sourceLineNo">650</span> }<a name="line.650"></a> |
| <span class="sourceLineNo">651</span><a name="line.651"></a> |
| <span class="sourceLineNo">652</span> // reset the in-memory state for testing<a name="line.652"></a> |
| <span class="sourceLineNo">653</span> completed.clear();<a name="line.653"></a> |
| <span class="sourceLineNo">654</span> rollbackStack.clear();<a name="line.654"></a> |
| <span class="sourceLineNo">655</span> procedures.clear();<a name="line.655"></a> |
| <span class="sourceLineNo">656</span> nonceKeysToProcIdsMap.clear();<a name="line.656"></a> |
| <span class="sourceLineNo">657</span> scheduler.clear();<a name="line.657"></a> |
| <span class="sourceLineNo">658</span> lastProcId.set(-1);<a name="line.658"></a> |
| <span class="sourceLineNo">659</span> }<a name="line.659"></a> |
| <span class="sourceLineNo">660</span><a name="line.660"></a> |
| <span class="sourceLineNo">661</span> public void refreshConfiguration(final Configuration conf) {<a name="line.661"></a> |
| <span class="sourceLineNo">662</span> this.conf = conf;<a name="line.662"></a> |
| <span class="sourceLineNo">663</span> setKeepAliveTime(conf.getLong(WORKER_KEEP_ALIVE_TIME_CONF_KEY,<a name="line.663"></a> |
| <span class="sourceLineNo">664</span> DEFAULT_WORKER_KEEP_ALIVE_TIME), TimeUnit.MILLISECONDS);<a name="line.664"></a> |
| <span class="sourceLineNo">665</span> }<a name="line.665"></a> |
| <span class="sourceLineNo">666</span><a name="line.666"></a> |
| <span class="sourceLineNo">667</span> // ==========================================================================<a name="line.667"></a> |
| <span class="sourceLineNo">668</span> // Accessors<a name="line.668"></a> |
| <span class="sourceLineNo">669</span> // ==========================================================================<a name="line.669"></a> |
| <span class="sourceLineNo">670</span> public boolean isRunning() {<a name="line.670"></a> |
| <span class="sourceLineNo">671</span> return running.get();<a name="line.671"></a> |
| <span class="sourceLineNo">672</span> }<a name="line.672"></a> |
| <span class="sourceLineNo">673</span><a name="line.673"></a> |
| <span class="sourceLineNo">674</span> /**<a name="line.674"></a> |
| <span class="sourceLineNo">675</span> * @return the current number of worker threads.<a name="line.675"></a> |
| <span class="sourceLineNo">676</span> */<a name="line.676"></a> |
| <span class="sourceLineNo">677</span> public int getWorkerThreadCount() {<a name="line.677"></a> |
| <span class="sourceLineNo">678</span> return workerThreads.size();<a name="line.678"></a> |
| <span class="sourceLineNo">679</span> }<a name="line.679"></a> |
| <span class="sourceLineNo">680</span><a name="line.680"></a> |
| <span class="sourceLineNo">681</span> /**<a name="line.681"></a> |
| <span class="sourceLineNo">682</span> * @return the core pool size settings.<a name="line.682"></a> |
| <span class="sourceLineNo">683</span> */<a name="line.683"></a> |
| <span class="sourceLineNo">684</span> public int getCorePoolSize() {<a name="line.684"></a> |
| <span class="sourceLineNo">685</span> return corePoolSize;<a name="line.685"></a> |
| <span class="sourceLineNo">686</span> }<a name="line.686"></a> |
| <span class="sourceLineNo">687</span><a name="line.687"></a> |
| <span class="sourceLineNo">688</span> public int getActiveExecutorCount() {<a name="line.688"></a> |
| <span class="sourceLineNo">689</span> return activeExecutorCount.get();<a name="line.689"></a> |
| <span class="sourceLineNo">690</span> }<a name="line.690"></a> |
| <span class="sourceLineNo">691</span><a name="line.691"></a> |
| <span class="sourceLineNo">692</span> public TEnvironment getEnvironment() {<a name="line.692"></a> |
| <span class="sourceLineNo">693</span> return this.environment;<a name="line.693"></a> |
| <span class="sourceLineNo">694</span> }<a name="line.694"></a> |
| <span class="sourceLineNo">695</span><a name="line.695"></a> |
| <span class="sourceLineNo">696</span> public ProcedureStore getStore() {<a name="line.696"></a> |
| <span class="sourceLineNo">697</span> return this.store;<a name="line.697"></a> |
| <span class="sourceLineNo">698</span> }<a name="line.698"></a> |
| <span class="sourceLineNo">699</span><a name="line.699"></a> |
| <span class="sourceLineNo">700</span> ProcedureScheduler getScheduler() {<a name="line.700"></a> |
| <span class="sourceLineNo">701</span> return scheduler;<a name="line.701"></a> |
| <span class="sourceLineNo">702</span> }<a name="line.702"></a> |
| <span class="sourceLineNo">703</span><a name="line.703"></a> |
| <span class="sourceLineNo">704</span> public void setKeepAliveTime(final long keepAliveTime, final TimeUnit timeUnit) {<a name="line.704"></a> |
| <span class="sourceLineNo">705</span> this.keepAliveTime = timeUnit.toMillis(keepAliveTime);<a name="line.705"></a> |
| <span class="sourceLineNo">706</span> this.scheduler.signalAll();<a name="line.706"></a> |
| <span class="sourceLineNo">707</span> }<a name="line.707"></a> |
| <span class="sourceLineNo">708</span><a name="line.708"></a> |
| <span class="sourceLineNo">709</span> public long getKeepAliveTime(final TimeUnit timeUnit) {<a name="line.709"></a> |
| <span class="sourceLineNo">710</span> return timeUnit.convert(keepAliveTime, TimeUnit.MILLISECONDS);<a name="line.710"></a> |
| <span class="sourceLineNo">711</span> }<a name="line.711"></a> |
| <span class="sourceLineNo">712</span><a name="line.712"></a> |
| <span class="sourceLineNo">713</span> // ==========================================================================<a name="line.713"></a> |
| <span class="sourceLineNo">714</span> // Submit/Remove Chores<a name="line.714"></a> |
| <span class="sourceLineNo">715</span> // ==========================================================================<a name="line.715"></a> |
| <span class="sourceLineNo">716</span><a name="line.716"></a> |
| <span class="sourceLineNo">717</span> /**<a name="line.717"></a> |
| <span class="sourceLineNo">718</span> * Add a chore procedure to the executor<a name="line.718"></a> |
| <span class="sourceLineNo">719</span> * @param chore the chore to add<a name="line.719"></a> |
| <span class="sourceLineNo">720</span> */<a name="line.720"></a> |
| <span class="sourceLineNo">721</span> public void addChore(ProcedureInMemoryChore<TEnvironment> chore) {<a name="line.721"></a> |
| <span class="sourceLineNo">722</span> chore.setState(ProcedureState.WAITING_TIMEOUT);<a name="line.722"></a> |
| <span class="sourceLineNo">723</span> timeoutExecutor.add(chore);<a name="line.723"></a> |
| <span class="sourceLineNo">724</span> }<a name="line.724"></a> |
| <span class="sourceLineNo">725</span><a name="line.725"></a> |
| <span class="sourceLineNo">726</span> /**<a name="line.726"></a> |
| <span class="sourceLineNo">727</span> * Remove a chore procedure from the executor<a name="line.727"></a> |
| <span class="sourceLineNo">728</span> * @param chore the chore to remove<a name="line.728"></a> |
| <span class="sourceLineNo">729</span> * @return whether the chore is removed, or it will be removed later<a name="line.729"></a> |
| <span class="sourceLineNo">730</span> */<a name="line.730"></a> |
| <span class="sourceLineNo">731</span> public boolean removeChore(ProcedureInMemoryChore<TEnvironment> chore) {<a name="line.731"></a> |
| <span class="sourceLineNo">732</span> chore.setState(ProcedureState.SUCCESS);<a name="line.732"></a> |
| <span class="sourceLineNo">733</span> return timeoutExecutor.remove(chore);<a name="line.733"></a> |
| <span class="sourceLineNo">734</span> }<a name="line.734"></a> |
| <span class="sourceLineNo">735</span><a name="line.735"></a> |
| <span class="sourceLineNo">736</span> // ==========================================================================<a name="line.736"></a> |
| <span class="sourceLineNo">737</span> // Nonce Procedure helpers<a name="line.737"></a> |
| <span class="sourceLineNo">738</span> // ==========================================================================<a name="line.738"></a> |
| <span class="sourceLineNo">739</span> /**<a name="line.739"></a> |
| <span class="sourceLineNo">740</span> * Create a NonceKey from the specified nonceGroup and nonce.<a name="line.740"></a> |
| <span class="sourceLineNo">741</span> * @param nonceGroup the group to use for the {@link NonceKey}<a name="line.741"></a> |
| <span class="sourceLineNo">742</span> * @param nonce the nonce to use in the {@link NonceKey}<a name="line.742"></a> |
| <span class="sourceLineNo">743</span> * @return the generated NonceKey<a name="line.743"></a> |
| <span class="sourceLineNo">744</span> */<a name="line.744"></a> |
| <span class="sourceLineNo">745</span> public NonceKey createNonceKey(final long nonceGroup, final long nonce) {<a name="line.745"></a> |
| <span class="sourceLineNo">746</span> return (nonce == HConstants.NO_NONCE) ? null : new NonceKey(nonceGroup, nonce);<a name="line.746"></a> |
| <span class="sourceLineNo">747</span> }<a name="line.747"></a> |
| <span class="sourceLineNo">748</span><a name="line.748"></a> |
| <span class="sourceLineNo">749</span> /**<a name="line.749"></a> |
| <span class="sourceLineNo">750</span> * Register a nonce for a procedure that is going to be submitted.<a name="line.750"></a> |
| <span class="sourceLineNo">751</span> * A procId will be reserved and on submitProcedure(),<a name="line.751"></a> |
| <span class="sourceLineNo">752</span> * the procedure with the specified nonce will take the reserved ProcId.<a name="line.752"></a> |
| <span class="sourceLineNo">753</span> * If someone already reserved the nonce, this method will return the procId reserved,<a name="line.753"></a> |
| <span class="sourceLineNo">754</span> * otherwise an invalid procId will be returned. and the caller should procede<a name="line.754"></a> |
| <span class="sourceLineNo">755</span> * and submit the procedure.<a name="line.755"></a> |
| <span class="sourceLineNo">756</span> *<a name="line.756"></a> |
| <span class="sourceLineNo">757</span> * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.757"></a> |
| <span class="sourceLineNo">758</span> * @return the procId associated with the nonce, if any otherwise an invalid procId.<a name="line.758"></a> |
| <span class="sourceLineNo">759</span> */<a name="line.759"></a> |
| <span class="sourceLineNo">760</span> public long registerNonce(final NonceKey nonceKey) {<a name="line.760"></a> |
| <span class="sourceLineNo">761</span> if (nonceKey == null) {<a name="line.761"></a> |
| <span class="sourceLineNo">762</span> return -1;<a name="line.762"></a> |
| <span class="sourceLineNo">763</span> }<a name="line.763"></a> |
| <span class="sourceLineNo">764</span><a name="line.764"></a> |
| <span class="sourceLineNo">765</span> // check if we have already a Reserved ID for the nonce<a name="line.765"></a> |
| <span class="sourceLineNo">766</span> Long oldProcId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.766"></a> |
| <span class="sourceLineNo">767</span> if (oldProcId == null) {<a name="line.767"></a> |
| <span class="sourceLineNo">768</span> // reserve a new Procedure ID, this will be associated with the nonce<a name="line.768"></a> |
| <span class="sourceLineNo">769</span> // and the procedure submitted with the specified nonce will use this ID.<a name="line.769"></a> |
| <span class="sourceLineNo">770</span> final long newProcId = nextProcId();<a name="line.770"></a> |
| <span class="sourceLineNo">771</span> oldProcId = nonceKeysToProcIdsMap.putIfAbsent(nonceKey, newProcId);<a name="line.771"></a> |
| <span class="sourceLineNo">772</span> if (oldProcId == null) {<a name="line.772"></a> |
| <span class="sourceLineNo">773</span> return -1;<a name="line.773"></a> |
| <span class="sourceLineNo">774</span> }<a name="line.774"></a> |
| <span class="sourceLineNo">775</span> }<a name="line.775"></a> |
| <span class="sourceLineNo">776</span><a name="line.776"></a> |
| <span class="sourceLineNo">777</span> // we found a registered nonce, but the procedure may not have been submitted yet.<a name="line.777"></a> |
| <span class="sourceLineNo">778</span> // since the client expect the procedure to be submitted, spin here until it is.<a name="line.778"></a> |
| <span class="sourceLineNo">779</span> final boolean traceEnabled = LOG.isTraceEnabled();<a name="line.779"></a> |
| <span class="sourceLineNo">780</span> while (isRunning() &&<a name="line.780"></a> |
| <span class="sourceLineNo">781</span> !(procedures.containsKey(oldProcId) || completed.containsKey(oldProcId)) &&<a name="line.781"></a> |
| <span class="sourceLineNo">782</span> nonceKeysToProcIdsMap.containsKey(nonceKey)) {<a name="line.782"></a> |
| <span class="sourceLineNo">783</span> if (traceEnabled) {<a name="line.783"></a> |
| <span class="sourceLineNo">784</span> LOG.trace("Waiting for pid=" + oldProcId.longValue() + " to be submitted");<a name="line.784"></a> |
| <span class="sourceLineNo">785</span> }<a name="line.785"></a> |
| <span class="sourceLineNo">786</span> Threads.sleep(100);<a name="line.786"></a> |
| <span class="sourceLineNo">787</span> }<a name="line.787"></a> |
| <span class="sourceLineNo">788</span> return oldProcId.longValue();<a name="line.788"></a> |
| <span class="sourceLineNo">789</span> }<a name="line.789"></a> |
| <span class="sourceLineNo">790</span><a name="line.790"></a> |
| <span class="sourceLineNo">791</span> /**<a name="line.791"></a> |
| <span class="sourceLineNo">792</span> * Remove the NonceKey if the procedure was not submitted to the executor.<a name="line.792"></a> |
| <span class="sourceLineNo">793</span> * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.793"></a> |
| <span class="sourceLineNo">794</span> */<a name="line.794"></a> |
| <span class="sourceLineNo">795</span> public void unregisterNonceIfProcedureWasNotSubmitted(final NonceKey nonceKey) {<a name="line.795"></a> |
| <span class="sourceLineNo">796</span> if (nonceKey == null) {<a name="line.796"></a> |
| <span class="sourceLineNo">797</span> return;<a name="line.797"></a> |
| <span class="sourceLineNo">798</span> }<a name="line.798"></a> |
| <span class="sourceLineNo">799</span><a name="line.799"></a> |
| <span class="sourceLineNo">800</span> final Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.800"></a> |
| <span class="sourceLineNo">801</span> if (procId == null) {<a name="line.801"></a> |
| <span class="sourceLineNo">802</span> return;<a name="line.802"></a> |
| <span class="sourceLineNo">803</span> }<a name="line.803"></a> |
| <span class="sourceLineNo">804</span><a name="line.804"></a> |
| <span class="sourceLineNo">805</span> // if the procedure was not submitted, remove the nonce<a name="line.805"></a> |
| <span class="sourceLineNo">806</span> if (!(procedures.containsKey(procId) || completed.containsKey(procId))) {<a name="line.806"></a> |
| <span class="sourceLineNo">807</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.807"></a> |
| <span class="sourceLineNo">808</span> }<a name="line.808"></a> |
| <span class="sourceLineNo">809</span> }<a name="line.809"></a> |
| <span class="sourceLineNo">810</span><a name="line.810"></a> |
| <span class="sourceLineNo">811</span> /**<a name="line.811"></a> |
| <span class="sourceLineNo">812</span> * If the failure failed before submitting it, we may want to give back the<a name="line.812"></a> |
| <span class="sourceLineNo">813</span> * same error to the requests with the same nonceKey.<a name="line.813"></a> |
| <span class="sourceLineNo">814</span> *<a name="line.814"></a> |
| <span class="sourceLineNo">815</span> * @param nonceKey A unique identifier for this operation from the client or process<a name="line.815"></a> |
| <span class="sourceLineNo">816</span> * @param procName name of the procedure, used to inform the user<a name="line.816"></a> |
| <span class="sourceLineNo">817</span> * @param procOwner name of the owner of the procedure, used to inform the user<a name="line.817"></a> |
| <span class="sourceLineNo">818</span> * @param exception the failure to report to the user<a name="line.818"></a> |
| <span class="sourceLineNo">819</span> */<a name="line.819"></a> |
| <span class="sourceLineNo">820</span> public void setFailureResultForNonce(NonceKey nonceKey, String procName, User procOwner,<a name="line.820"></a> |
| <span class="sourceLineNo">821</span> IOException exception) {<a name="line.821"></a> |
| <span class="sourceLineNo">822</span> if (nonceKey == null) {<a name="line.822"></a> |
| <span class="sourceLineNo">823</span> return;<a name="line.823"></a> |
| <span class="sourceLineNo">824</span> }<a name="line.824"></a> |
| <span class="sourceLineNo">825</span><a name="line.825"></a> |
| <span class="sourceLineNo">826</span> Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.826"></a> |
| <span class="sourceLineNo">827</span> if (procId == null || completed.containsKey(procId)) {<a name="line.827"></a> |
| <span class="sourceLineNo">828</span> return;<a name="line.828"></a> |
| <span class="sourceLineNo">829</span> }<a name="line.829"></a> |
| <span class="sourceLineNo">830</span><a name="line.830"></a> |
| <span class="sourceLineNo">831</span> Procedure<TEnvironment> proc =<a name="line.831"></a> |
| <span class="sourceLineNo">832</span> new FailedProcedure<>(procId.longValue(), procName, procOwner, nonceKey, exception);<a name="line.832"></a> |
| <span class="sourceLineNo">833</span><a name="line.833"></a> |
| <span class="sourceLineNo">834</span> completed.putIfAbsent(procId, new CompletedProcedureRetainer<>(proc));<a name="line.834"></a> |
| <span class="sourceLineNo">835</span> }<a name="line.835"></a> |
| <span class="sourceLineNo">836</span><a name="line.836"></a> |
| <span class="sourceLineNo">837</span> // ==========================================================================<a name="line.837"></a> |
| <span class="sourceLineNo">838</span> // Submit/Abort Procedure<a name="line.838"></a> |
| <span class="sourceLineNo">839</span> // ==========================================================================<a name="line.839"></a> |
| <span class="sourceLineNo">840</span> /**<a name="line.840"></a> |
| <span class="sourceLineNo">841</span> * Add a new root-procedure to the executor.<a name="line.841"></a> |
| <span class="sourceLineNo">842</span> * @param proc the new procedure to execute.<a name="line.842"></a> |
| <span class="sourceLineNo">843</span> * @return the procedure id, that can be used to monitor the operation<a name="line.843"></a> |
| <span class="sourceLineNo">844</span> */<a name="line.844"></a> |
| <span class="sourceLineNo">845</span> public long submitProcedure(Procedure<TEnvironment> proc) {<a name="line.845"></a> |
| <span class="sourceLineNo">846</span> return submitProcedure(proc, null);<a name="line.846"></a> |
| <span class="sourceLineNo">847</span> }<a name="line.847"></a> |
| <span class="sourceLineNo">848</span><a name="line.848"></a> |
| <span class="sourceLineNo">849</span> /**<a name="line.849"></a> |
| <span class="sourceLineNo">850</span> * Bypass a procedure. If the procedure is set to bypass, all the logic in<a name="line.850"></a> |
| <span class="sourceLineNo">851</span> * execute/rollback will be ignored and it will return success, whatever.<a name="line.851"></a> |
| <span class="sourceLineNo">852</span> * It is used to recover buggy stuck procedures, releasing the lock resources<a name="line.852"></a> |
| <span class="sourceLineNo">853</span> * and letting other procedures run. Bypassing one procedure (and its ancestors will<a name="line.853"></a> |
| <span class="sourceLineNo">854</span> * be bypassed automatically) may leave the cluster in a middle state, e.g. region<a name="line.854"></a> |
| <span class="sourceLineNo">855</span> * not assigned, or some hdfs files left behind. After getting rid of those stuck procedures,<a name="line.855"></a> |
| <span class="sourceLineNo">856</span> * the operators may have to do some clean up on hdfs or schedule some assign procedures<a name="line.856"></a> |
| <span class="sourceLineNo">857</span> * to let region online. DO AT YOUR OWN RISK.<a name="line.857"></a> |
| <span class="sourceLineNo">858</span> * <p><a name="line.858"></a> |
| <span class="sourceLineNo">859</span> * A procedure can be bypassed only if<a name="line.859"></a> |
| <span class="sourceLineNo">860</span> * 1. The procedure is in state of RUNNABLE, WAITING, WAITING_TIMEOUT<a name="line.860"></a> |
| <span class="sourceLineNo">861</span> * or it is a root procedure without any child.<a name="line.861"></a> |
| <span class="sourceLineNo">862</span> * 2. No other worker thread is executing it<a name="line.862"></a> |
| <span class="sourceLineNo">863</span> * 3. No child procedure has been submitted<a name="line.863"></a> |
| <span class="sourceLineNo">864</span> *<a name="line.864"></a> |
| <span class="sourceLineNo">865</span> * <p><a name="line.865"></a> |
| <span class="sourceLineNo">866</span> * If all the requirements are meet, the procedure and its ancestors will be<a name="line.866"></a> |
| <span class="sourceLineNo">867</span> * bypassed and persisted to WAL.<a name="line.867"></a> |
| <span class="sourceLineNo">868</span> *<a name="line.868"></a> |
| <span class="sourceLineNo">869</span> * <p><a name="line.869"></a> |
| <span class="sourceLineNo">870</span> * If the procedure is in WAITING state, will set it to RUNNABLE add it to run queue.<a name="line.870"></a> |
| <span class="sourceLineNo">871</span> * TODO: What about WAITING_TIMEOUT?<a name="line.871"></a> |
| <span class="sourceLineNo">872</span> * @param pids the procedure id<a name="line.872"></a> |
| <span class="sourceLineNo">873</span> * @param lockWait time to wait lock<a name="line.873"></a> |
| <span class="sourceLineNo">874</span> * @param force if force set to true, we will bypass the procedure even if it is executing.<a name="line.874"></a> |
| <span class="sourceLineNo">875</span> * This is for procedures which can't break out during executing(due to bug, mostly)<a name="line.875"></a> |
| <span class="sourceLineNo">876</span> * In this case, bypassing the procedure is not enough, since it is already stuck<a name="line.876"></a> |
| <span class="sourceLineNo">877</span> * there. We need to restart the master after bypassing, and letting the problematic<a name="line.877"></a> |
| <span class="sourceLineNo">878</span> * procedure to execute wth bypass=true, so in that condition, the procedure can be<a name="line.878"></a> |
| <span class="sourceLineNo">879</span> * successfully bypassed.<a name="line.879"></a> |
| <span class="sourceLineNo">880</span> * @param recursive We will do an expensive search for children of each pid. EXPENSIVE!<a name="line.880"></a> |
| <span class="sourceLineNo">881</span> * @return true if bypass success<a name="line.881"></a> |
| <span class="sourceLineNo">882</span> * @throws IOException IOException<a name="line.882"></a> |
| <span class="sourceLineNo">883</span> */<a name="line.883"></a> |
| <span class="sourceLineNo">884</span> public List<Boolean> bypassProcedure(List<Long> pids, long lockWait, boolean force,<a name="line.884"></a> |
| <span class="sourceLineNo">885</span> boolean recursive)<a name="line.885"></a> |
| <span class="sourceLineNo">886</span> throws IOException {<a name="line.886"></a> |
| <span class="sourceLineNo">887</span> List<Boolean> result = new ArrayList<Boolean>(pids.size());<a name="line.887"></a> |
| <span class="sourceLineNo">888</span> for(long pid: pids) {<a name="line.888"></a> |
| <span class="sourceLineNo">889</span> result.add(bypassProcedure(pid, lockWait, force, recursive));<a name="line.889"></a> |
| <span class="sourceLineNo">890</span> }<a name="line.890"></a> |
| <span class="sourceLineNo">891</span> return result;<a name="line.891"></a> |
| <span class="sourceLineNo">892</span> }<a name="line.892"></a> |
| <span class="sourceLineNo">893</span><a name="line.893"></a> |
| <span class="sourceLineNo">894</span> boolean bypassProcedure(long pid, long lockWait, boolean override, boolean recursive)<a name="line.894"></a> |
| <span class="sourceLineNo">895</span> throws IOException {<a name="line.895"></a> |
| <span class="sourceLineNo">896</span> Preconditions.checkArgument(lockWait > 0, "lockWait should be positive");<a name="line.896"></a> |
| <span class="sourceLineNo">897</span> final Procedure<TEnvironment> procedure = getProcedure(pid);<a name="line.897"></a> |
| <span class="sourceLineNo">898</span> if (procedure == null) {<a name="line.898"></a> |
| <span class="sourceLineNo">899</span> LOG.debug("Procedure pid={} does not exist, skipping bypass", pid);<a name="line.899"></a> |
| <span class="sourceLineNo">900</span> return false;<a name="line.900"></a> |
| <span class="sourceLineNo">901</span> }<a name="line.901"></a> |
| <span class="sourceLineNo">902</span><a name="line.902"></a> |
| <span class="sourceLineNo">903</span> LOG.debug("Begin bypass {} with lockWait={}, override={}, recursive={}",<a name="line.903"></a> |
| <span class="sourceLineNo">904</span> procedure, lockWait, override, recursive);<a name="line.904"></a> |
| <span class="sourceLineNo">905</span> IdLock.Entry lockEntry = procExecutionLock.tryLockEntry(procedure.getProcId(), lockWait);<a name="line.905"></a> |
| <span class="sourceLineNo">906</span> if (lockEntry == null && !override) {<a name="line.906"></a> |
| <span class="sourceLineNo">907</span> LOG.debug("Waited {} ms, but {} is still running, skipping bypass with force={}",<a name="line.907"></a> |
| <span class="sourceLineNo">908</span> lockWait, procedure, override);<a name="line.908"></a> |
| <span class="sourceLineNo">909</span> return false;<a name="line.909"></a> |
| <span class="sourceLineNo">910</span> } else if (lockEntry == null) {<a name="line.910"></a> |
| <span class="sourceLineNo">911</span> LOG.debug("Waited {} ms, but {} is still running, begin bypass with force={}",<a name="line.911"></a> |
| <span class="sourceLineNo">912</span> lockWait, procedure, override);<a name="line.912"></a> |
| <span class="sourceLineNo">913</span> }<a name="line.913"></a> |
| <span class="sourceLineNo">914</span> try {<a name="line.914"></a> |
| <span class="sourceLineNo">915</span> // check whether the procedure is already finished<a name="line.915"></a> |
| <span class="sourceLineNo">916</span> if (procedure.isFinished()) {<a name="line.916"></a> |
| <span class="sourceLineNo">917</span> LOG.debug("{} is already finished, skipping bypass", procedure);<a name="line.917"></a> |
| <span class="sourceLineNo">918</span> return false;<a name="line.918"></a> |
| <span class="sourceLineNo">919</span> }<a name="line.919"></a> |
| <span class="sourceLineNo">920</span><a name="line.920"></a> |
| <span class="sourceLineNo">921</span> if (procedure.hasChildren()) {<a name="line.921"></a> |
| <span class="sourceLineNo">922</span> if (recursive) {<a name="line.922"></a> |
| <span class="sourceLineNo">923</span> // EXPENSIVE. Checks each live procedure of which there could be many!!!<a name="line.923"></a> |
| <span class="sourceLineNo">924</span> // Is there another way to get children of a procedure?<a name="line.924"></a> |
| <span class="sourceLineNo">925</span> LOG.info("Recursive bypass on children of pid={}", procedure.getProcId());<a name="line.925"></a> |
| <span class="sourceLineNo">926</span> this.procedures.forEachValue(1 /*Single-threaded*/,<a name="line.926"></a> |
| <span class="sourceLineNo">927</span> // Transformer<a name="line.927"></a> |
| <span class="sourceLineNo">928</span> v -> v.getParentProcId() == procedure.getProcId()? v: null,<a name="line.928"></a> |
| <span class="sourceLineNo">929</span> // Consumer<a name="line.929"></a> |
| <span class="sourceLineNo">930</span> v -> {<a name="line.930"></a> |
| <span class="sourceLineNo">931</span> try {<a name="line.931"></a> |
| <span class="sourceLineNo">932</span> bypassProcedure(v.getProcId(), lockWait, override, recursive);<a name="line.932"></a> |
| <span class="sourceLineNo">933</span> } catch (IOException e) {<a name="line.933"></a> |
| <span class="sourceLineNo">934</span> LOG.warn("Recursive bypass of pid={}", v.getProcId(), e);<a name="line.934"></a> |
| <span class="sourceLineNo">935</span> }<a name="line.935"></a> |
| <span class="sourceLineNo">936</span> });<a name="line.936"></a> |
| <span class="sourceLineNo">937</span> } else {<a name="line.937"></a> |
| <span class="sourceLineNo">938</span> LOG.debug("{} has children, skipping bypass", procedure);<a name="line.938"></a> |
| <span class="sourceLineNo">939</span> return false;<a name="line.939"></a> |
| <span class="sourceLineNo">940</span> }<a name="line.940"></a> |
| <span class="sourceLineNo">941</span> }<a name="line.941"></a> |
| <span class="sourceLineNo">942</span><a name="line.942"></a> |
| <span class="sourceLineNo">943</span> // If the procedure has no parent or no child, we are safe to bypass it in whatever state<a name="line.943"></a> |
| <span class="sourceLineNo">944</span> if (procedure.hasParent() && procedure.getState() != ProcedureState.RUNNABLE<a name="line.944"></a> |
| <span class="sourceLineNo">945</span> && procedure.getState() != ProcedureState.WAITING<a name="line.945"></a> |
| <span class="sourceLineNo">946</span> && procedure.getState() != ProcedureState.WAITING_TIMEOUT) {<a name="line.946"></a> |
| <span class="sourceLineNo">947</span> LOG.debug("Bypassing procedures in RUNNABLE, WAITING and WAITING_TIMEOUT states "<a name="line.947"></a> |
| <span class="sourceLineNo">948</span> + "(with no parent), {}",<a name="line.948"></a> |
| <span class="sourceLineNo">949</span> procedure);<a name="line.949"></a> |
| <span class="sourceLineNo">950</span> // Question: how is the bypass done here?<a name="line.950"></a> |
| <span class="sourceLineNo">951</span> return false;<a name="line.951"></a> |
| <span class="sourceLineNo">952</span> }<a name="line.952"></a> |
| <span class="sourceLineNo">953</span><a name="line.953"></a> |
| <span class="sourceLineNo">954</span> // Now, the procedure is not finished, and no one can execute it since we take the lock now<a name="line.954"></a> |
| <span class="sourceLineNo">955</span> // And we can be sure that its ancestor is not running too, since their child has not<a name="line.955"></a> |
| <span class="sourceLineNo">956</span> // finished yet<a name="line.956"></a> |
| <span class="sourceLineNo">957</span> Procedure<TEnvironment> current = procedure;<a name="line.957"></a> |
| <span class="sourceLineNo">958</span> while (current != null) {<a name="line.958"></a> |
| <span class="sourceLineNo">959</span> LOG.debug("Bypassing {}", current);<a name="line.959"></a> |
| <span class="sourceLineNo">960</span> current.bypass(getEnvironment());<a name="line.960"></a> |
| <span class="sourceLineNo">961</span> store.update(procedure);<a name="line.961"></a> |
| <span class="sourceLineNo">962</span> long parentID = current.getParentProcId();<a name="line.962"></a> |
| <span class="sourceLineNo">963</span> current = getProcedure(parentID);<a name="line.963"></a> |
| <span class="sourceLineNo">964</span> }<a name="line.964"></a> |
| <span class="sourceLineNo">965</span><a name="line.965"></a> |
| <span class="sourceLineNo">966</span> //wake up waiting procedure, already checked there is no child<a name="line.966"></a> |
| <span class="sourceLineNo">967</span> if (procedure.getState() == ProcedureState.WAITING) {<a name="line.967"></a> |
| <span class="sourceLineNo">968</span> procedure.setState(ProcedureState.RUNNABLE);<a name="line.968"></a> |
| <span class="sourceLineNo">969</span> store.update(procedure);<a name="line.969"></a> |
| <span class="sourceLineNo">970</span> }<a name="line.970"></a> |
| <span class="sourceLineNo">971</span><a name="line.971"></a> |
| <span class="sourceLineNo">972</span> // If state of procedure is WAITING_TIMEOUT, we can directly submit it to the scheduler.<a name="line.972"></a> |
| <span class="sourceLineNo">973</span> // Instead we should remove it from timeout Executor queue and tranfer its state to RUNNABLE<a name="line.973"></a> |
| <span class="sourceLineNo">974</span> if (procedure.getState() == ProcedureState.WAITING_TIMEOUT) {<a name="line.974"></a> |
| <span class="sourceLineNo">975</span> LOG.debug("transform procedure {} from WAITING_TIMEOUT to RUNNABLE", procedure);<a name="line.975"></a> |
| <span class="sourceLineNo">976</span> if (timeoutExecutor.remove(procedure)) {<a name="line.976"></a> |
| <span class="sourceLineNo">977</span> LOG.debug("removed procedure {} from timeoutExecutor", procedure);<a name="line.977"></a> |
| <span class="sourceLineNo">978</span> timeoutExecutor.executeTimedoutProcedure(procedure);<a name="line.978"></a> |
| <span class="sourceLineNo">979</span> }<a name="line.979"></a> |
| <span class="sourceLineNo">980</span> } else if (lockEntry != null) {<a name="line.980"></a> |
| <span class="sourceLineNo">981</span> scheduler.addFront(procedure);<a name="line.981"></a> |
| <span class="sourceLineNo">982</span> LOG.debug("Bypassing {} and its ancestors successfully, adding to queue", procedure);<a name="line.982"></a> |
| <span class="sourceLineNo">983</span> } else {<a name="line.983"></a> |
| <span class="sourceLineNo">984</span> // If we don't have the lock, we can't re-submit the queue,<a name="line.984"></a> |
| <span class="sourceLineNo">985</span> // since it is already executing. To get rid of the stuck situation, we<a name="line.985"></a> |
| <span class="sourceLineNo">986</span> // need to restart the master. With the procedure set to bypass, the procedureExecutor<a name="line.986"></a> |
| <span class="sourceLineNo">987</span> // will bypass it and won't get stuck again.<a name="line.987"></a> |
| <span class="sourceLineNo">988</span> LOG.debug("Bypassing {} and its ancestors successfully, but since it is already running, "<a name="line.988"></a> |
| <span class="sourceLineNo">989</span> + "skipping add to queue",<a name="line.989"></a> |
| <span class="sourceLineNo">990</span> procedure);<a name="line.990"></a> |
| <span class="sourceLineNo">991</span> }<a name="line.991"></a> |
| <span class="sourceLineNo">992</span> return true;<a name="line.992"></a> |
| <span class="sourceLineNo">993</span><a name="line.993"></a> |
| <span class="sourceLineNo">994</span> } finally {<a name="line.994"></a> |
| <span class="sourceLineNo">995</span> if (lockEntry != null) {<a name="line.995"></a> |
| <span class="sourceLineNo">996</span> procExecutionLock.releaseLockEntry(lockEntry);<a name="line.996"></a> |
| <span class="sourceLineNo">997</span> }<a name="line.997"></a> |
| <span class="sourceLineNo">998</span> }<a name="line.998"></a> |
| <span class="sourceLineNo">999</span> }<a name="line.999"></a> |
| <span class="sourceLineNo">1000</span><a name="line.1000"></a> |
| <span class="sourceLineNo">1001</span> /**<a name="line.1001"></a> |
| <span class="sourceLineNo">1002</span> * Add a new root-procedure to the executor.<a name="line.1002"></a> |
| <span class="sourceLineNo">1003</span> * @param proc the new procedure to execute.<a name="line.1003"></a> |
| <span class="sourceLineNo">1004</span> * @param nonceKey the registered unique identifier for this operation from the client or process.<a name="line.1004"></a> |
| <span class="sourceLineNo">1005</span> * @return the procedure id, that can be used to monitor the operation<a name="line.1005"></a> |
| <span class="sourceLineNo">1006</span> */<a name="line.1006"></a> |
| <span class="sourceLineNo">1007</span> @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="NP_NULL_ON_SOME_PATH",<a name="line.1007"></a> |
| <span class="sourceLineNo">1008</span> justification = "FindBugs is blind to the check-for-null")<a name="line.1008"></a> |
| <span class="sourceLineNo">1009</span> public long submitProcedure(Procedure<TEnvironment> proc, NonceKey nonceKey) {<a name="line.1009"></a> |
| <span class="sourceLineNo">1010</span> Preconditions.checkArgument(lastProcId.get() >= 0);<a name="line.1010"></a> |
| <span class="sourceLineNo">1011</span><a name="line.1011"></a> |
| <span class="sourceLineNo">1012</span> prepareProcedure(proc);<a name="line.1012"></a> |
| <span class="sourceLineNo">1013</span><a name="line.1013"></a> |
| <span class="sourceLineNo">1014</span> final Long currentProcId;<a name="line.1014"></a> |
| <span class="sourceLineNo">1015</span> if (nonceKey != null) {<a name="line.1015"></a> |
| <span class="sourceLineNo">1016</span> currentProcId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.1016"></a> |
| <span class="sourceLineNo">1017</span> Preconditions.checkArgument(currentProcId != null,<a name="line.1017"></a> |
| <span class="sourceLineNo">1018</span> "Expected nonceKey=" + nonceKey + " to be reserved, use registerNonce(); proc=" + proc);<a name="line.1018"></a> |
| <span class="sourceLineNo">1019</span> } else {<a name="line.1019"></a> |
| <span class="sourceLineNo">1020</span> currentProcId = nextProcId();<a name="line.1020"></a> |
| <span class="sourceLineNo">1021</span> }<a name="line.1021"></a> |
| <span class="sourceLineNo">1022</span><a name="line.1022"></a> |
| <span class="sourceLineNo">1023</span> // Initialize the procedure<a name="line.1023"></a> |
| <span class="sourceLineNo">1024</span> proc.setNonceKey(nonceKey);<a name="line.1024"></a> |
| <span class="sourceLineNo">1025</span> proc.setProcId(currentProcId.longValue());<a name="line.1025"></a> |
| <span class="sourceLineNo">1026</span><a name="line.1026"></a> |
| <span class="sourceLineNo">1027</span> // Commit the transaction<a name="line.1027"></a> |
| <span class="sourceLineNo">1028</span> store.insert(proc, null);<a name="line.1028"></a> |
| <span class="sourceLineNo">1029</span> LOG.debug("Stored {}", proc);<a name="line.1029"></a> |
| <span class="sourceLineNo">1030</span><a name="line.1030"></a> |
| <span class="sourceLineNo">1031</span> // Add the procedure to the executor<a name="line.1031"></a> |
| <span class="sourceLineNo">1032</span> return pushProcedure(proc);<a name="line.1032"></a> |
| <span class="sourceLineNo">1033</span> }<a name="line.1033"></a> |
| <span class="sourceLineNo">1034</span><a name="line.1034"></a> |
| <span class="sourceLineNo">1035</span> /**<a name="line.1035"></a> |
| <span class="sourceLineNo">1036</span> * Add a set of new root-procedure to the executor.<a name="line.1036"></a> |
| <span class="sourceLineNo">1037</span> * @param procs the new procedures to execute.<a name="line.1037"></a> |
| <span class="sourceLineNo">1038</span> */<a name="line.1038"></a> |
| <span class="sourceLineNo">1039</span> // TODO: Do we need to take nonces here?<a name="line.1039"></a> |
| <span class="sourceLineNo">1040</span> public void submitProcedures(Procedure<TEnvironment>[] procs) {<a name="line.1040"></a> |
| <span class="sourceLineNo">1041</span> Preconditions.checkArgument(lastProcId.get() >= 0);<a name="line.1041"></a> |
| <span class="sourceLineNo">1042</span> if (procs == null || procs.length <= 0) {<a name="line.1042"></a> |
| <span class="sourceLineNo">1043</span> return;<a name="line.1043"></a> |
| <span class="sourceLineNo">1044</span> }<a name="line.1044"></a> |
| <span class="sourceLineNo">1045</span><a name="line.1045"></a> |
| <span class="sourceLineNo">1046</span> // Prepare procedure<a name="line.1046"></a> |
| <span class="sourceLineNo">1047</span> for (int i = 0; i < procs.length; ++i) {<a name="line.1047"></a> |
| <span class="sourceLineNo">1048</span> prepareProcedure(procs[i]).setProcId(nextProcId());<a name="line.1048"></a> |
| <span class="sourceLineNo">1049</span> }<a name="line.1049"></a> |
| <span class="sourceLineNo">1050</span><a name="line.1050"></a> |
| <span class="sourceLineNo">1051</span> // Commit the transaction<a name="line.1051"></a> |
| <span class="sourceLineNo">1052</span> store.insert(procs);<a name="line.1052"></a> |
| <span class="sourceLineNo">1053</span> if (LOG.isDebugEnabled()) {<a name="line.1053"></a> |
| <span class="sourceLineNo">1054</span> LOG.debug("Stored " + Arrays.toString(procs));<a name="line.1054"></a> |
| <span class="sourceLineNo">1055</span> }<a name="line.1055"></a> |
| <span class="sourceLineNo">1056</span><a name="line.1056"></a> |
| <span class="sourceLineNo">1057</span> // Add the procedure to the executor<a name="line.1057"></a> |
| <span class="sourceLineNo">1058</span> for (int i = 0; i < procs.length; ++i) {<a name="line.1058"></a> |
| <span class="sourceLineNo">1059</span> pushProcedure(procs[i]);<a name="line.1059"></a> |
| <span class="sourceLineNo">1060</span> }<a name="line.1060"></a> |
| <span class="sourceLineNo">1061</span> }<a name="line.1061"></a> |
| <span class="sourceLineNo">1062</span><a name="line.1062"></a> |
| <span class="sourceLineNo">1063</span> private Procedure<TEnvironment> prepareProcedure(Procedure<TEnvironment> proc) {<a name="line.1063"></a> |
| <span class="sourceLineNo">1064</span> Preconditions.checkArgument(proc.getState() == ProcedureState.INITIALIZING);<a name="line.1064"></a> |
| <span class="sourceLineNo">1065</span> Preconditions.checkArgument(!proc.hasParent(), "unexpected parent", proc);<a name="line.1065"></a> |
| <span class="sourceLineNo">1066</span> if (this.checkOwnerSet) {<a name="line.1066"></a> |
| <span class="sourceLineNo">1067</span> Preconditions.checkArgument(proc.hasOwner(), "missing owner");<a name="line.1067"></a> |
| <span class="sourceLineNo">1068</span> }<a name="line.1068"></a> |
| <span class="sourceLineNo">1069</span> return proc;<a name="line.1069"></a> |
| <span class="sourceLineNo">1070</span> }<a name="line.1070"></a> |
| <span class="sourceLineNo">1071</span><a name="line.1071"></a> |
| <span class="sourceLineNo">1072</span> private long pushProcedure(Procedure<TEnvironment> proc) {<a name="line.1072"></a> |
| <span class="sourceLineNo">1073</span> final long currentProcId = proc.getProcId();<a name="line.1073"></a> |
| <span class="sourceLineNo">1074</span><a name="line.1074"></a> |
| <span class="sourceLineNo">1075</span> // Update metrics on start of a procedure<a name="line.1075"></a> |
| <span class="sourceLineNo">1076</span> proc.updateMetricsOnSubmit(getEnvironment());<a name="line.1076"></a> |
| <span class="sourceLineNo">1077</span><a name="line.1077"></a> |
| <span class="sourceLineNo">1078</span> // Create the rollback stack for the procedure<a name="line.1078"></a> |
| <span class="sourceLineNo">1079</span> RootProcedureState<TEnvironment> stack = new RootProcedureState<>();<a name="line.1079"></a> |
| <span class="sourceLineNo">1080</span> rollbackStack.put(currentProcId, stack);<a name="line.1080"></a> |
| <span class="sourceLineNo">1081</span><a name="line.1081"></a> |
| <span class="sourceLineNo">1082</span> // Submit the new subprocedures<a name="line.1082"></a> |
| <span class="sourceLineNo">1083</span> assert !procedures.containsKey(currentProcId);<a name="line.1083"></a> |
| <span class="sourceLineNo">1084</span> procedures.put(currentProcId, proc);<a name="line.1084"></a> |
| <span class="sourceLineNo">1085</span> sendProcedureAddedNotification(currentProcId);<a name="line.1085"></a> |
| <span class="sourceLineNo">1086</span> scheduler.addBack(proc);<a name="line.1086"></a> |
| <span class="sourceLineNo">1087</span> return proc.getProcId();<a name="line.1087"></a> |
| <span class="sourceLineNo">1088</span> }<a name="line.1088"></a> |
| <span class="sourceLineNo">1089</span><a name="line.1089"></a> |
| <span class="sourceLineNo">1090</span> /**<a name="line.1090"></a> |
| <span class="sourceLineNo">1091</span> * Send an abort notification the specified procedure.<a name="line.1091"></a> |
| <span class="sourceLineNo">1092</span> * Depending on the procedure implementation the abort can be considered or ignored.<a name="line.1092"></a> |
| <span class="sourceLineNo">1093</span> * @param procId the procedure to abort<a name="line.1093"></a> |
| <span class="sourceLineNo">1094</span> * @return true if the procedure exists and has received the abort, otherwise false.<a name="line.1094"></a> |
| <span class="sourceLineNo">1095</span> */<a name="line.1095"></a> |
| <span class="sourceLineNo">1096</span> public boolean abort(long procId) {<a name="line.1096"></a> |
| <span class="sourceLineNo">1097</span> return abort(procId, true);<a name="line.1097"></a> |
| <span class="sourceLineNo">1098</span> }<a name="line.1098"></a> |
| <span class="sourceLineNo">1099</span><a name="line.1099"></a> |
| <span class="sourceLineNo">1100</span> /**<a name="line.1100"></a> |
| <span class="sourceLineNo">1101</span> * Send an abort notification to the specified procedure.<a name="line.1101"></a> |
| <span class="sourceLineNo">1102</span> * Depending on the procedure implementation, the abort can be considered or ignored.<a name="line.1102"></a> |
| <span class="sourceLineNo">1103</span> * @param procId the procedure to abort<a name="line.1103"></a> |
| <span class="sourceLineNo">1104</span> * @param mayInterruptIfRunning if the proc completed at least one step, should it be aborted?<a name="line.1104"></a> |
| <span class="sourceLineNo">1105</span> * @return true if the procedure exists and has received the abort, otherwise false.<a name="line.1105"></a> |
| <span class="sourceLineNo">1106</span> */<a name="line.1106"></a> |
| <span class="sourceLineNo">1107</span> public boolean abort(long procId, boolean mayInterruptIfRunning) {<a name="line.1107"></a> |
| <span class="sourceLineNo">1108</span> Procedure<TEnvironment> proc = procedures.get(procId);<a name="line.1108"></a> |
| <span class="sourceLineNo">1109</span> if (proc != null) {<a name="line.1109"></a> |
| <span class="sourceLineNo">1110</span> if (!mayInterruptIfRunning && proc.wasExecuted()) {<a name="line.1110"></a> |
| <span class="sourceLineNo">1111</span> return false;<a name="line.1111"></a> |
| <span class="sourceLineNo">1112</span> }<a name="line.1112"></a> |
| <span class="sourceLineNo">1113</span> return proc.abort(getEnvironment());<a name="line.1113"></a> |
| <span class="sourceLineNo">1114</span> }<a name="line.1114"></a> |
| <span class="sourceLineNo">1115</span> return false;<a name="line.1115"></a> |
| <span class="sourceLineNo">1116</span> }<a name="line.1116"></a> |
| <span class="sourceLineNo">1117</span><a name="line.1117"></a> |
| <span class="sourceLineNo">1118</span> // ==========================================================================<a name="line.1118"></a> |
| <span class="sourceLineNo">1119</span> // Executor query helpers<a name="line.1119"></a> |
| <span class="sourceLineNo">1120</span> // ==========================================================================<a name="line.1120"></a> |
| <span class="sourceLineNo">1121</span> public Procedure<TEnvironment> getProcedure(final long procId) {<a name="line.1121"></a> |
| <span class="sourceLineNo">1122</span> return procedures.get(procId);<a name="line.1122"></a> |
| <span class="sourceLineNo">1123</span> }<a name="line.1123"></a> |
| <span class="sourceLineNo">1124</span><a name="line.1124"></a> |
| <span class="sourceLineNo">1125</span> public <T extends Procedure<TEnvironment>> T getProcedure(Class<T> clazz, long procId) {<a name="line.1125"></a> |
| <span class="sourceLineNo">1126</span> Procedure<TEnvironment> proc = getProcedure(procId);<a name="line.1126"></a> |
| <span class="sourceLineNo">1127</span> if (clazz.isInstance(proc)) {<a name="line.1127"></a> |
| <span class="sourceLineNo">1128</span> return clazz.cast(proc);<a name="line.1128"></a> |
| <span class="sourceLineNo">1129</span> }<a name="line.1129"></a> |
| <span class="sourceLineNo">1130</span> return null;<a name="line.1130"></a> |
| <span class="sourceLineNo">1131</span> }<a name="line.1131"></a> |
| <span class="sourceLineNo">1132</span><a name="line.1132"></a> |
| <span class="sourceLineNo">1133</span> public Procedure<TEnvironment> getResult(long procId) {<a name="line.1133"></a> |
| <span class="sourceLineNo">1134</span> CompletedProcedureRetainer<TEnvironment> retainer = completed.get(procId);<a name="line.1134"></a> |
| <span class="sourceLineNo">1135</span> if (retainer == null) {<a name="line.1135"></a> |
| <span class="sourceLineNo">1136</span> return null;<a name="line.1136"></a> |
| <span class="sourceLineNo">1137</span> } else {<a name="line.1137"></a> |
| <span class="sourceLineNo">1138</span> return retainer.getProcedure();<a name="line.1138"></a> |
| <span class="sourceLineNo">1139</span> }<a name="line.1139"></a> |
| <span class="sourceLineNo">1140</span> }<a name="line.1140"></a> |
| <span class="sourceLineNo">1141</span><a name="line.1141"></a> |
| <span class="sourceLineNo">1142</span> /**<a name="line.1142"></a> |
| <span class="sourceLineNo">1143</span> * Return true if the procedure is finished.<a name="line.1143"></a> |
| <span class="sourceLineNo">1144</span> * The state may be "completed successfully" or "failed and rolledback".<a name="line.1144"></a> |
| <span class="sourceLineNo">1145</span> * Use getResult() to check the state or get the result data.<a name="line.1145"></a> |
| <span class="sourceLineNo">1146</span> * @param procId the ID of the procedure to check<a name="line.1146"></a> |
| <span class="sourceLineNo">1147</span> * @return true if the procedure execution is finished, otherwise false.<a name="line.1147"></a> |
| <span class="sourceLineNo">1148</span> */<a name="line.1148"></a> |
| <span class="sourceLineNo">1149</span> public boolean isFinished(final long procId) {<a name="line.1149"></a> |
| <span class="sourceLineNo">1150</span> return !procedures.containsKey(procId);<a name="line.1150"></a> |
| <span class="sourceLineNo">1151</span> }<a name="line.1151"></a> |
| <span class="sourceLineNo">1152</span><a name="line.1152"></a> |
| <span class="sourceLineNo">1153</span> /**<a name="line.1153"></a> |
| <span class="sourceLineNo">1154</span> * Return true if the procedure is started.<a name="line.1154"></a> |
| <span class="sourceLineNo">1155</span> * @param procId the ID of the procedure to check<a name="line.1155"></a> |
| <span class="sourceLineNo">1156</span> * @return true if the procedure execution is started, otherwise false.<a name="line.1156"></a> |
| <span class="sourceLineNo">1157</span> */<a name="line.1157"></a> |
| <span class="sourceLineNo">1158</span> public boolean isStarted(long procId) {<a name="line.1158"></a> |
| <span class="sourceLineNo">1159</span> Procedure<?> proc = procedures.get(procId);<a name="line.1159"></a> |
| <span class="sourceLineNo">1160</span> if (proc == null) {<a name="line.1160"></a> |
| <span class="sourceLineNo">1161</span> return completed.get(procId) != null;<a name="line.1161"></a> |
| <span class="sourceLineNo">1162</span> }<a name="line.1162"></a> |
| <span class="sourceLineNo">1163</span> return proc.wasExecuted();<a name="line.1163"></a> |
| <span class="sourceLineNo">1164</span> }<a name="line.1164"></a> |
| <span class="sourceLineNo">1165</span><a name="line.1165"></a> |
| <span class="sourceLineNo">1166</span> /**<a name="line.1166"></a> |
| <span class="sourceLineNo">1167</span> * Mark the specified completed procedure, as ready to remove.<a name="line.1167"></a> |
| <span class="sourceLineNo">1168</span> * @param procId the ID of the procedure to remove<a name="line.1168"></a> |
| <span class="sourceLineNo">1169</span> */<a name="line.1169"></a> |
| <span class="sourceLineNo">1170</span> public void removeResult(long procId) {<a name="line.1170"></a> |
| <span class="sourceLineNo">1171</span> CompletedProcedureRetainer<TEnvironment> retainer = completed.get(procId);<a name="line.1171"></a> |
| <span class="sourceLineNo">1172</span> if (retainer == null) {<a name="line.1172"></a> |
| <span class="sourceLineNo">1173</span> assert !procedures.containsKey(procId) : "pid=" + procId + " is still running";<a name="line.1173"></a> |
| <span class="sourceLineNo">1174</span> LOG.debug("pid={} already removed by the cleaner.", procId);<a name="line.1174"></a> |
| <span class="sourceLineNo">1175</span> return;<a name="line.1175"></a> |
| <span class="sourceLineNo">1176</span> }<a name="line.1176"></a> |
| <span class="sourceLineNo">1177</span><a name="line.1177"></a> |
| <span class="sourceLineNo">1178</span> // The CompletedProcedureCleaner will take care of deletion, once the TTL is expired.<a name="line.1178"></a> |
| <span class="sourceLineNo">1179</span> retainer.setClientAckTime(EnvironmentEdgeManager.currentTime());<a name="line.1179"></a> |
| <span class="sourceLineNo">1180</span> }<a name="line.1180"></a> |
| <span class="sourceLineNo">1181</span><a name="line.1181"></a> |
| <span class="sourceLineNo">1182</span> public Procedure<TEnvironment> getResultOrProcedure(long procId) {<a name="line.1182"></a> |
| <span class="sourceLineNo">1183</span> CompletedProcedureRetainer<TEnvironment> retainer = completed.get(procId);<a name="line.1183"></a> |
| <span class="sourceLineNo">1184</span> if (retainer == null) {<a name="line.1184"></a> |
| <span class="sourceLineNo">1185</span> return procedures.get(procId);<a name="line.1185"></a> |
| <span class="sourceLineNo">1186</span> } else {<a name="line.1186"></a> |
| <span class="sourceLineNo">1187</span> return retainer.getProcedure();<a name="line.1187"></a> |
| <span class="sourceLineNo">1188</span> }<a name="line.1188"></a> |
| <span class="sourceLineNo">1189</span> }<a name="line.1189"></a> |
| <span class="sourceLineNo">1190</span><a name="line.1190"></a> |
| <span class="sourceLineNo">1191</span> /**<a name="line.1191"></a> |
| <span class="sourceLineNo">1192</span> * Check if the user is this procedure's owner<a name="line.1192"></a> |
| <span class="sourceLineNo">1193</span> * @param procId the target procedure<a name="line.1193"></a> |
| <span class="sourceLineNo">1194</span> * @param user the user<a name="line.1194"></a> |
| <span class="sourceLineNo">1195</span> * @return true if the user is the owner of the procedure,<a name="line.1195"></a> |
| <span class="sourceLineNo">1196</span> * false otherwise or the owner is unknown.<a name="line.1196"></a> |
| <span class="sourceLineNo">1197</span> */<a name="line.1197"></a> |
| <span class="sourceLineNo">1198</span> public boolean isProcedureOwner(long procId, User user) {<a name="line.1198"></a> |
| <span class="sourceLineNo">1199</span> if (user == null) {<a name="line.1199"></a> |
| <span class="sourceLineNo">1200</span> return false;<a name="line.1200"></a> |
| <span class="sourceLineNo">1201</span> }<a name="line.1201"></a> |
| <span class="sourceLineNo">1202</span> final Procedure<TEnvironment> runningProc = procedures.get(procId);<a name="line.1202"></a> |
| <span class="sourceLineNo">1203</span> if (runningProc != null) {<a name="line.1203"></a> |
| <span class="sourceLineNo">1204</span> return runningProc.getOwner().equals(user.getShortName());<a name="line.1204"></a> |
| <span class="sourceLineNo">1205</span> }<a name="line.1205"></a> |
| <span class="sourceLineNo">1206</span><a name="line.1206"></a> |
| <span class="sourceLineNo">1207</span> final CompletedProcedureRetainer<TEnvironment> retainer = completed.get(procId);<a name="line.1207"></a> |
| <span class="sourceLineNo">1208</span> if (retainer != null) {<a name="line.1208"></a> |
| <span class="sourceLineNo">1209</span> return retainer.getProcedure().getOwner().equals(user.getShortName());<a name="line.1209"></a> |
| <span class="sourceLineNo">1210</span> }<a name="line.1210"></a> |
| <span class="sourceLineNo">1211</span><a name="line.1211"></a> |
| <span class="sourceLineNo">1212</span> // Procedure either does not exist or has already completed and got cleaned up.<a name="line.1212"></a> |
| <span class="sourceLineNo">1213</span> // At this time, we cannot check the owner of the procedure<a name="line.1213"></a> |
| <span class="sourceLineNo">1214</span> return false;<a name="line.1214"></a> |
| <span class="sourceLineNo">1215</span> }<a name="line.1215"></a> |
| <span class="sourceLineNo">1216</span><a name="line.1216"></a> |
| <span class="sourceLineNo">1217</span> /**<a name="line.1217"></a> |
| <span class="sourceLineNo">1218</span> * Should only be used when starting up, where the procedure workers have not been started.<a name="line.1218"></a> |
| <span class="sourceLineNo">1219</span> * <p/><a name="line.1219"></a> |
| <span class="sourceLineNo">1220</span> * If the procedure works has been started, the return values maybe changed when you are<a name="line.1220"></a> |
| <span class="sourceLineNo">1221</span> * processing it so usually this is not safe. Use {@link #getProcedures()} below for most cases as<a name="line.1221"></a> |
| <span class="sourceLineNo">1222</span> * it will do a copy, and also include the finished procedures.<a name="line.1222"></a> |
| <span class="sourceLineNo">1223</span> */<a name="line.1223"></a> |
| <span class="sourceLineNo">1224</span> public Collection<Procedure<TEnvironment>> getActiveProceduresNoCopy() {<a name="line.1224"></a> |
| <span class="sourceLineNo">1225</span> return procedures.values();<a name="line.1225"></a> |
| <span class="sourceLineNo">1226</span> }<a name="line.1226"></a> |
| <span class="sourceLineNo">1227</span><a name="line.1227"></a> |
| <span class="sourceLineNo">1228</span> /**<a name="line.1228"></a> |
| <span class="sourceLineNo">1229</span> * Get procedures.<a name="line.1229"></a> |
| <span class="sourceLineNo">1230</span> * @return the procedures in a list<a name="line.1230"></a> |
| <span class="sourceLineNo">1231</span> */<a name="line.1231"></a> |
| <span class="sourceLineNo">1232</span> public List<Procedure<TEnvironment>> getProcedures() {<a name="line.1232"></a> |
| <span class="sourceLineNo">1233</span> List<Procedure<TEnvironment>> procedureList =<a name="line.1233"></a> |
| <span class="sourceLineNo">1234</span> new ArrayList<>(procedures.size() + completed.size());<a name="line.1234"></a> |
| <span class="sourceLineNo">1235</span> procedureList.addAll(procedures.values());<a name="line.1235"></a> |
| <span class="sourceLineNo">1236</span> // Note: The procedure could show up twice in the list with different state, as<a name="line.1236"></a> |
| <span class="sourceLineNo">1237</span> // it could complete after we walk through procedures list and insert into<a name="line.1237"></a> |
| <span class="sourceLineNo">1238</span> // procedureList - it is ok, as we will use the information in the Procedure<a name="line.1238"></a> |
| <span class="sourceLineNo">1239</span> // to figure it out; to prevent this would increase the complexity of the logic.<a name="line.1239"></a> |
| <span class="sourceLineNo">1240</span> completed.values().stream().map(CompletedProcedureRetainer::getProcedure)<a name="line.1240"></a> |
| <span class="sourceLineNo">1241</span> .forEach(procedureList::add);<a name="line.1241"></a> |
| <span class="sourceLineNo">1242</span> return procedureList;<a name="line.1242"></a> |
| <span class="sourceLineNo">1243</span> }<a name="line.1243"></a> |
| <span class="sourceLineNo">1244</span><a name="line.1244"></a> |
| <span class="sourceLineNo">1245</span> // ==========================================================================<a name="line.1245"></a> |
| <span class="sourceLineNo">1246</span> // Listeners helpers<a name="line.1246"></a> |
| <span class="sourceLineNo">1247</span> // ==========================================================================<a name="line.1247"></a> |
| <span class="sourceLineNo">1248</span> public void registerListener(ProcedureExecutorListener listener) {<a name="line.1248"></a> |
| <span class="sourceLineNo">1249</span> this.listeners.add(listener);<a name="line.1249"></a> |
| <span class="sourceLineNo">1250</span> }<a name="line.1250"></a> |
| <span class="sourceLineNo">1251</span><a name="line.1251"></a> |
| <span class="sourceLineNo">1252</span> public boolean unregisterListener(ProcedureExecutorListener listener) {<a name="line.1252"></a> |
| <span class="sourceLineNo">1253</span> return this.listeners.remove(listener);<a name="line.1253"></a> |
| <span class="sourceLineNo">1254</span> }<a name="line.1254"></a> |
| <span class="sourceLineNo">1255</span><a name="line.1255"></a> |
| <span class="sourceLineNo">1256</span> private void sendProcedureLoadedNotification(final long procId) {<a name="line.1256"></a> |
| <span class="sourceLineNo">1257</span> if (!this.listeners.isEmpty()) {<a name="line.1257"></a> |
| <span class="sourceLineNo">1258</span> for (ProcedureExecutorListener listener: this.listeners) {<a name="line.1258"></a> |
| <span class="sourceLineNo">1259</span> try {<a name="line.1259"></a> |
| <span class="sourceLineNo">1260</span> listener.procedureLoaded(procId);<a name="line.1260"></a> |
| <span class="sourceLineNo">1261</span> } catch (Throwable e) {<a name="line.1261"></a> |
| <span class="sourceLineNo">1262</span> LOG.error("Listener " + listener + " had an error: " + e.getMessage(), e);<a name="line.1262"></a> |
| <span class="sourceLineNo">1263</span> }<a name="line.1263"></a> |
| <span class="sourceLineNo">1264</span> }<a name="line.1264"></a> |
| <span class="sourceLineNo">1265</span> }<a name="line.1265"></a> |
| <span class="sourceLineNo">1266</span> }<a name="line.1266"></a> |
| <span class="sourceLineNo">1267</span><a name="line.1267"></a> |
| <span class="sourceLineNo">1268</span> private void sendProcedureAddedNotification(final long procId) {<a name="line.1268"></a> |
| <span class="sourceLineNo">1269</span> if (!this.listeners.isEmpty()) {<a name="line.1269"></a> |
| <span class="sourceLineNo">1270</span> for (ProcedureExecutorListener listener: this.listeners) {<a name="line.1270"></a> |
| <span class="sourceLineNo">1271</span> try {<a name="line.1271"></a> |
| <span class="sourceLineNo">1272</span> listener.procedureAdded(procId);<a name="line.1272"></a> |
| <span class="sourceLineNo">1273</span> } catch (Throwable e) {<a name="line.1273"></a> |
| <span class="sourceLineNo">1274</span> LOG.error("Listener " + listener + " had an error: " + e.getMessage(), e);<a name="line.1274"></a> |
| <span class="sourceLineNo">1275</span> }<a name="line.1275"></a> |
| <span class="sourceLineNo">1276</span> }<a name="line.1276"></a> |
| <span class="sourceLineNo">1277</span> }<a name="line.1277"></a> |
| <span class="sourceLineNo">1278</span> }<a name="line.1278"></a> |
| <span class="sourceLineNo">1279</span><a name="line.1279"></a> |
| <span class="sourceLineNo">1280</span> private void sendProcedureFinishedNotification(final long procId) {<a name="line.1280"></a> |
| <span class="sourceLineNo">1281</span> if (!this.listeners.isEmpty()) {<a name="line.1281"></a> |
| <span class="sourceLineNo">1282</span> for (ProcedureExecutorListener listener: this.listeners) {<a name="line.1282"></a> |
| <span class="sourceLineNo">1283</span> try {<a name="line.1283"></a> |
| <span class="sourceLineNo">1284</span> listener.procedureFinished(procId);<a name="line.1284"></a> |
| <span class="sourceLineNo">1285</span> } catch (Throwable e) {<a name="line.1285"></a> |
| <span class="sourceLineNo">1286</span> LOG.error("Listener " + listener + " had an error: " + e.getMessage(), e);<a name="line.1286"></a> |
| <span class="sourceLineNo">1287</span> }<a name="line.1287"></a> |
| <span class="sourceLineNo">1288</span> }<a name="line.1288"></a> |
| <span class="sourceLineNo">1289</span> }<a name="line.1289"></a> |
| <span class="sourceLineNo">1290</span> }<a name="line.1290"></a> |
| <span class="sourceLineNo">1291</span><a name="line.1291"></a> |
| <span class="sourceLineNo">1292</span> // ==========================================================================<a name="line.1292"></a> |
| <span class="sourceLineNo">1293</span> // Procedure IDs helpers<a name="line.1293"></a> |
| <span class="sourceLineNo">1294</span> // ==========================================================================<a name="line.1294"></a> |
| <span class="sourceLineNo">1295</span> private long nextProcId() {<a name="line.1295"></a> |
| <span class="sourceLineNo">1296</span> long procId = lastProcId.incrementAndGet();<a name="line.1296"></a> |
| <span class="sourceLineNo">1297</span> if (procId < 0) {<a name="line.1297"></a> |
| <span class="sourceLineNo">1298</span> while (!lastProcId.compareAndSet(procId, 0)) {<a name="line.1298"></a> |
| <span class="sourceLineNo">1299</span> procId = lastProcId.get();<a name="line.1299"></a> |
| <span class="sourceLineNo">1300</span> if (procId >= 0) {<a name="line.1300"></a> |
| <span class="sourceLineNo">1301</span> break;<a name="line.1301"></a> |
| <span class="sourceLineNo">1302</span> }<a name="line.1302"></a> |
| <span class="sourceLineNo">1303</span> }<a name="line.1303"></a> |
| <span class="sourceLineNo">1304</span> while (procedures.containsKey(procId)) {<a name="line.1304"></a> |
| <span class="sourceLineNo">1305</span> procId = lastProcId.incrementAndGet();<a name="line.1305"></a> |
| <span class="sourceLineNo">1306</span> }<a name="line.1306"></a> |
| <span class="sourceLineNo">1307</span> }<a name="line.1307"></a> |
| <span class="sourceLineNo">1308</span> assert procId >= 0 : "Invalid procId " + procId;<a name="line.1308"></a> |
| <span class="sourceLineNo">1309</span> return procId;<a name="line.1309"></a> |
| <span class="sourceLineNo">1310</span> }<a name="line.1310"></a> |
| <span class="sourceLineNo">1311</span><a name="line.1311"></a> |
| <span class="sourceLineNo">1312</span> @VisibleForTesting<a name="line.1312"></a> |
| <span class="sourceLineNo">1313</span> protected long getLastProcId() {<a name="line.1313"></a> |
| <span class="sourceLineNo">1314</span> return lastProcId.get();<a name="line.1314"></a> |
| <span class="sourceLineNo">1315</span> }<a name="line.1315"></a> |
| <span class="sourceLineNo">1316</span><a name="line.1316"></a> |
| <span class="sourceLineNo">1317</span> @VisibleForTesting<a name="line.1317"></a> |
| <span class="sourceLineNo">1318</span> public Set<Long> getActiveProcIds() {<a name="line.1318"></a> |
| <span class="sourceLineNo">1319</span> return procedures.keySet();<a name="line.1319"></a> |
| <span class="sourceLineNo">1320</span> }<a name="line.1320"></a> |
| <span class="sourceLineNo">1321</span><a name="line.1321"></a> |
| <span class="sourceLineNo">1322</span> Long getRootProcedureId(Procedure<TEnvironment> proc) {<a name="line.1322"></a> |
| <span class="sourceLineNo">1323</span> return Procedure.getRootProcedureId(procedures, proc);<a name="line.1323"></a> |
| <span class="sourceLineNo">1324</span> }<a name="line.1324"></a> |
| <span class="sourceLineNo">1325</span><a name="line.1325"></a> |
| <span class="sourceLineNo">1326</span> // ==========================================================================<a name="line.1326"></a> |
| <span class="sourceLineNo">1327</span> // Executions<a name="line.1327"></a> |
| <span class="sourceLineNo">1328</span> // ==========================================================================<a name="line.1328"></a> |
| <span class="sourceLineNo">1329</span> private void executeProcedure(Procedure<TEnvironment> proc) {<a name="line.1329"></a> |
| <span class="sourceLineNo">1330</span> if (proc.isFinished()) {<a name="line.1330"></a> |
| <span class="sourceLineNo">1331</span> LOG.debug("{} is already finished, skipping execution", proc);<a name="line.1331"></a> |
| <span class="sourceLineNo">1332</span> return;<a name="line.1332"></a> |
| <span class="sourceLineNo">1333</span> }<a name="line.1333"></a> |
| <span class="sourceLineNo">1334</span> final Long rootProcId = getRootProcedureId(proc);<a name="line.1334"></a> |
| <span class="sourceLineNo">1335</span> if (rootProcId == null) {<a name="line.1335"></a> |
| <span class="sourceLineNo">1336</span> // The 'proc' was ready to run but the root procedure was rolledback<a name="line.1336"></a> |
| <span class="sourceLineNo">1337</span> LOG.warn("Rollback because parent is done/rolledback proc=" + proc);<a name="line.1337"></a> |
| <span class="sourceLineNo">1338</span> executeRollback(proc);<a name="line.1338"></a> |
| <span class="sourceLineNo">1339</span> return;<a name="line.1339"></a> |
| <span class="sourceLineNo">1340</span> }<a name="line.1340"></a> |
| <span class="sourceLineNo">1341</span><a name="line.1341"></a> |
| <span class="sourceLineNo">1342</span> RootProcedureState<TEnvironment> procStack = rollbackStack.get(rootProcId);<a name="line.1342"></a> |
| <span class="sourceLineNo">1343</span> if (procStack == null) {<a name="line.1343"></a> |
| <span class="sourceLineNo">1344</span> LOG.warn("RootProcedureState is null for " + proc.getProcId());<a name="line.1344"></a> |
| <span class="sourceLineNo">1345</span> return;<a name="line.1345"></a> |
| <span class="sourceLineNo">1346</span> }<a name="line.1346"></a> |
| <span class="sourceLineNo">1347</span> do {<a name="line.1347"></a> |
| <span class="sourceLineNo">1348</span> // Try to acquire the execution<a name="line.1348"></a> |
| <span class="sourceLineNo">1349</span> if (!procStack.acquire(proc)) {<a name="line.1349"></a> |
| <span class="sourceLineNo">1350</span> if (procStack.setRollback()) {<a name="line.1350"></a> |
| <span class="sourceLineNo">1351</span> // we have the 'rollback-lock' we can start rollingback<a name="line.1351"></a> |
| <span class="sourceLineNo">1352</span> switch (executeRollback(rootProcId, procStack)) {<a name="line.1352"></a> |
| <span class="sourceLineNo">1353</span> case LOCK_ACQUIRED:<a name="line.1353"></a> |
| <span class="sourceLineNo">1354</span> break;<a name="line.1354"></a> |
| <span class="sourceLineNo">1355</span> case LOCK_YIELD_WAIT:<a name="line.1355"></a> |
| <span class="sourceLineNo">1356</span> procStack.unsetRollback();<a name="line.1356"></a> |
| <span class="sourceLineNo">1357</span> scheduler.yield(proc);<a name="line.1357"></a> |
| <span class="sourceLineNo">1358</span> break;<a name="line.1358"></a> |
| <span class="sourceLineNo">1359</span> case LOCK_EVENT_WAIT:<a name="line.1359"></a> |
| <span class="sourceLineNo">1360</span> LOG.info("LOCK_EVENT_WAIT rollback..." + proc);<a name="line.1360"></a> |
| <span class="sourceLineNo">1361</span> procStack.unsetRollback();<a name="line.1361"></a> |
| <span class="sourceLineNo">1362</span> break;<a name="line.1362"></a> |
| <span class="sourceLineNo">1363</span> default:<a name="line.1363"></a> |
| <span class="sourceLineNo">1364</span> throw new UnsupportedOperationException();<a name="line.1364"></a> |
| <span class="sourceLineNo">1365</span> }<a name="line.1365"></a> |
| <span class="sourceLineNo">1366</span> } else {<a name="line.1366"></a> |
| <span class="sourceLineNo">1367</span> // if we can't rollback means that some child is still running.<a name="line.1367"></a> |
| <span class="sourceLineNo">1368</span> // the rollback will be executed after all the children are done.<a name="line.1368"></a> |
| <span class="sourceLineNo">1369</span> // If the procedure was never executed, remove and mark it as rolledback.<a name="line.1369"></a> |
| <span class="sourceLineNo">1370</span> if (!proc.wasExecuted()) {<a name="line.1370"></a> |
| <span class="sourceLineNo">1371</span> switch (executeRollback(proc)) {<a name="line.1371"></a> |
| <span class="sourceLineNo">1372</span> case LOCK_ACQUIRED:<a name="line.1372"></a> |
| <span class="sourceLineNo">1373</span> break;<a name="line.1373"></a> |
| <span class="sourceLineNo">1374</span> case LOCK_YIELD_WAIT:<a name="line.1374"></a> |
| <span class="sourceLineNo">1375</span> scheduler.yield(proc);<a name="line.1375"></a> |
| <span class="sourceLineNo">1376</span> break;<a name="line.1376"></a> |
| <span class="sourceLineNo">1377</span> case LOCK_EVENT_WAIT:<a name="line.1377"></a> |
| <span class="sourceLineNo">1378</span> LOG.info("LOCK_EVENT_WAIT can't rollback child running?..." + proc);<a name="line.1378"></a> |
| <span class="sourceLineNo">1379</span> break;<a name="line.1379"></a> |
| <span class="sourceLineNo">1380</span> default:<a name="line.1380"></a> |
| <span class="sourceLineNo">1381</span> throw new UnsupportedOperationException();<a name="line.1381"></a> |
| <span class="sourceLineNo">1382</span> }<a name="line.1382"></a> |
| <span class="sourceLineNo">1383</span> }<a name="line.1383"></a> |
| <span class="sourceLineNo">1384</span> }<a name="line.1384"></a> |
| <span class="sourceLineNo">1385</span> break;<a name="line.1385"></a> |
| <span class="sourceLineNo">1386</span> }<a name="line.1386"></a> |
| <span class="sourceLineNo">1387</span><a name="line.1387"></a> |
| <span class="sourceLineNo">1388</span> // Execute the procedure<a name="line.1388"></a> |
| <span class="sourceLineNo">1389</span> assert proc.getState() == ProcedureState.RUNNABLE : proc;<a name="line.1389"></a> |
| <span class="sourceLineNo">1390</span> // Note that lock is NOT about concurrency but rather about ensuring<a name="line.1390"></a> |
| <span class="sourceLineNo">1391</span> // ownership of a procedure of an entity such as a region or table<a name="line.1391"></a> |
| <span class="sourceLineNo">1392</span> LockState lockState = acquireLock(proc);<a name="line.1392"></a> |
| <span class="sourceLineNo">1393</span> switch (lockState) {<a name="line.1393"></a> |
| <span class="sourceLineNo">1394</span> case LOCK_ACQUIRED:<a name="line.1394"></a> |
| <span class="sourceLineNo">1395</span> execProcedure(procStack, proc);<a name="line.1395"></a> |
| <span class="sourceLineNo">1396</span> break;<a name="line.1396"></a> |
| <span class="sourceLineNo">1397</span> case LOCK_YIELD_WAIT:<a name="line.1397"></a> |
| <span class="sourceLineNo">1398</span> LOG.info(lockState + " " + proc);<a name="line.1398"></a> |
| <span class="sourceLineNo">1399</span> scheduler.yield(proc);<a name="line.1399"></a> |
| <span class="sourceLineNo">1400</span> break;<a name="line.1400"></a> |
| <span class="sourceLineNo">1401</span> case LOCK_EVENT_WAIT:<a name="line.1401"></a> |
| <span class="sourceLineNo">1402</span> // Someone will wake us up when the lock is available<a name="line.1402"></a> |
| <span class="sourceLineNo">1403</span> LOG.debug(lockState + " " + proc);<a name="line.1403"></a> |
| <span class="sourceLineNo">1404</span> break;<a name="line.1404"></a> |
| <span class="sourceLineNo">1405</span> default:<a name="line.1405"></a> |
| <span class="sourceLineNo">1406</span> throw new UnsupportedOperationException();<a name="line.1406"></a> |
| <span class="sourceLineNo">1407</span> }<a name="line.1407"></a> |
| <span class="sourceLineNo">1408</span> procStack.release(proc);<a name="line.1408"></a> |
| <span class="sourceLineNo">1409</span><a name="line.1409"></a> |
| <span class="sourceLineNo">1410</span> if (proc.isSuccess()) {<a name="line.1410"></a> |
| <span class="sourceLineNo">1411</span> // update metrics on finishing the procedure<a name="line.1411"></a> |
| <span class="sourceLineNo">1412</span> proc.updateMetricsOnFinish(getEnvironment(), proc.elapsedTime(), true);<a name="line.1412"></a> |
| <span class="sourceLineNo">1413</span> LOG.info("Finished " + proc + " in " + StringUtils.humanTimeDiff(proc.elapsedTime()));<a name="line.1413"></a> |
| <span class="sourceLineNo">1414</span> // Finalize the procedure state<a name="line.1414"></a> |
| <span class="sourceLineNo">1415</span> if (proc.getProcId() == rootProcId) {<a name="line.1415"></a> |
| <span class="sourceLineNo">1416</span> procedureFinished(proc);<a name="line.1416"></a> |
| <span class="sourceLineNo">1417</span> } else {<a name="line.1417"></a> |
| <span class="sourceLineNo">1418</span> execCompletionCleanup(proc);<a name="line.1418"></a> |
| <span class="sourceLineNo">1419</span> }<a name="line.1419"></a> |
| <span class="sourceLineNo">1420</span> break;<a name="line.1420"></a> |
| <span class="sourceLineNo">1421</span> }<a name="line.1421"></a> |
| <span class="sourceLineNo">1422</span> } while (procStack.isFailed());<a name="line.1422"></a> |
| <span class="sourceLineNo">1423</span> }<a name="line.1423"></a> |
| <span class="sourceLineNo">1424</span><a name="line.1424"></a> |
| <span class="sourceLineNo">1425</span> private LockState acquireLock(Procedure<TEnvironment> proc) {<a name="line.1425"></a> |
| <span class="sourceLineNo">1426</span> TEnvironment env = getEnvironment();<a name="line.1426"></a> |
| <span class="sourceLineNo">1427</span> // if holdLock is true, then maybe we already have the lock, so just return LOCK_ACQUIRED if<a name="line.1427"></a> |
| <span class="sourceLineNo">1428</span> // hasLock is true.<a name="line.1428"></a> |
| <span class="sourceLineNo">1429</span> if (proc.hasLock()) {<a name="line.1429"></a> |
| <span class="sourceLineNo">1430</span> return LockState.LOCK_ACQUIRED;<a name="line.1430"></a> |
| <span class="sourceLineNo">1431</span> }<a name="line.1431"></a> |
| <span class="sourceLineNo">1432</span> return proc.doAcquireLock(env, store);<a name="line.1432"></a> |
| <span class="sourceLineNo">1433</span> }<a name="line.1433"></a> |
| <span class="sourceLineNo">1434</span><a name="line.1434"></a> |
| <span class="sourceLineNo">1435</span> private void releaseLock(Procedure<TEnvironment> proc, boolean force) {<a name="line.1435"></a> |
| <span class="sourceLineNo">1436</span> TEnvironment env = getEnvironment();<a name="line.1436"></a> |
| <span class="sourceLineNo">1437</span> // For how the framework works, we know that we will always have the lock<a name="line.1437"></a> |
| <span class="sourceLineNo">1438</span> // when we call releaseLock(), so we can avoid calling proc.hasLock()<a name="line.1438"></a> |
| <span class="sourceLineNo">1439</span> if (force || !proc.holdLock(env) || proc.isFinished()) {<a name="line.1439"></a> |
| <span class="sourceLineNo">1440</span> proc.doReleaseLock(env, store);<a name="line.1440"></a> |
| <span class="sourceLineNo">1441</span> }<a name="line.1441"></a> |
| <span class="sourceLineNo">1442</span> }<a name="line.1442"></a> |
| <span class="sourceLineNo">1443</span><a name="line.1443"></a> |
| <span class="sourceLineNo">1444</span> /**<a name="line.1444"></a> |
| <span class="sourceLineNo">1445</span> * Execute the rollback of the full procedure stack. Once the procedure is rolledback, the<a name="line.1445"></a> |
| <span class="sourceLineNo">1446</span> * root-procedure will be visible as finished to user, and the result will be the fatal exception.<a name="line.1446"></a> |
| <span class="sourceLineNo">1447</span> */<a name="line.1447"></a> |
| <span class="sourceLineNo">1448</span> private LockState executeRollback(long rootProcId, RootProcedureState<TEnvironment> procStack) {<a name="line.1448"></a> |
| <span class="sourceLineNo">1449</span> Procedure<TEnvironment> rootProc = procedures.get(rootProcId);<a name="line.1449"></a> |
| <span class="sourceLineNo">1450</span> RemoteProcedureException exception = rootProc.getException();<a name="line.1450"></a> |
| <span class="sourceLineNo">1451</span> // TODO: This needs doc. The root proc doesn't have an exception. Maybe we are<a name="line.1451"></a> |
| <span class="sourceLineNo">1452</span> // rolling back because the subprocedure does. Clarify.<a name="line.1452"></a> |
| <span class="sourceLineNo">1453</span> if (exception == null) {<a name="line.1453"></a> |
| <span class="sourceLineNo">1454</span> exception = procStack.getException();<a name="line.1454"></a> |
| <span class="sourceLineNo">1455</span> rootProc.setFailure(exception);<a name="line.1455"></a> |
| <span class="sourceLineNo">1456</span> store.update(rootProc);<a name="line.1456"></a> |
| <span class="sourceLineNo">1457</span> }<a name="line.1457"></a> |
| <span class="sourceLineNo">1458</span><a name="line.1458"></a> |
| <span class="sourceLineNo">1459</span> List<Procedure<TEnvironment>> subprocStack = procStack.getSubproceduresStack();<a name="line.1459"></a> |
| <span class="sourceLineNo">1460</span> assert subprocStack != null : "Called rollback with no steps executed rootProc=" + rootProc;<a name="line.1460"></a> |
| <span class="sourceLineNo">1461</span><a name="line.1461"></a> |
| <span class="sourceLineNo">1462</span> int stackTail = subprocStack.size();<a name="line.1462"></a> |
| <span class="sourceLineNo">1463</span> while (stackTail-- > 0) {<a name="line.1463"></a> |
| <span class="sourceLineNo">1464</span> Procedure<TEnvironment> proc = subprocStack.get(stackTail);<a name="line.1464"></a> |
| <span class="sourceLineNo">1465</span> IdLock.Entry lockEntry = null;<a name="line.1465"></a> |
| <span class="sourceLineNo">1466</span> // Hold the execution lock if it is not held by us. The IdLock is not reentrant so we need<a name="line.1466"></a> |
| <span class="sourceLineNo">1467</span> // this check, as the worker will hold the lock before executing a procedure. This is the only<a name="line.1467"></a> |
| <span class="sourceLineNo">1468</span> // place where we may hold two procedure execution locks, and there is a fence in the<a name="line.1468"></a> |
| <span class="sourceLineNo">1469</span> // RootProcedureState where we can make sure that only one worker can execute the rollback of<a name="line.1469"></a> |
| <span class="sourceLineNo">1470</span> // a RootProcedureState, so there is no dead lock problem. And the lock here is necessary to<a name="line.1470"></a> |
| <span class="sourceLineNo">1471</span> // prevent race between us and the force update thread.<a name="line.1471"></a> |
| <span class="sourceLineNo">1472</span> if (!procExecutionLock.isHeldByCurrentThread(proc.getProcId())) {<a name="line.1472"></a> |
| <span class="sourceLineNo">1473</span> try {<a name="line.1473"></a> |
| <span class="sourceLineNo">1474</span> lockEntry = procExecutionLock.getLockEntry(proc.getProcId());<a name="line.1474"></a> |
| <span class="sourceLineNo">1475</span> } catch (IOException e) {<a name="line.1475"></a> |
| <span class="sourceLineNo">1476</span> // can only happen if interrupted, so not a big deal to propagate it<a name="line.1476"></a> |
| <span class="sourceLineNo">1477</span> throw new UncheckedIOException(e);<a name="line.1477"></a> |
| <span class="sourceLineNo">1478</span> }<a name="line.1478"></a> |
| <span class="sourceLineNo">1479</span> }<a name="line.1479"></a> |
| <span class="sourceLineNo">1480</span> try {<a name="line.1480"></a> |
| <span class="sourceLineNo">1481</span> // For the sub procedures which are successfully finished, we do not rollback them.<a name="line.1481"></a> |
| <span class="sourceLineNo">1482</span> // Typically, if we want to rollback a procedure, we first need to rollback it, and then<a name="line.1482"></a> |
| <span class="sourceLineNo">1483</span> // recursively rollback its ancestors. The state changes which are done by sub procedures<a name="line.1483"></a> |
| <span class="sourceLineNo">1484</span> // should be handled by parent procedures when rolling back. For example, when rolling back<a name="line.1484"></a> |
| <span class="sourceLineNo">1485</span> // a MergeTableProcedure, we will schedule new procedures to bring the offline regions<a name="line.1485"></a> |
| <span class="sourceLineNo">1486</span> // online, instead of rolling back the original procedures which offlined the regions(in<a name="line.1486"></a> |
| <span class="sourceLineNo">1487</span> // fact these procedures can not be rolled back...).<a name="line.1487"></a> |
| <span class="sourceLineNo">1488</span> if (proc.isSuccess()) {<a name="line.1488"></a> |
| <span class="sourceLineNo">1489</span> // Just do the cleanup work, without actually executing the rollback<a name="line.1489"></a> |
| <span class="sourceLineNo">1490</span> subprocStack.remove(stackTail);<a name="line.1490"></a> |
| <span class="sourceLineNo">1491</span> cleanupAfterRollbackOneStep(proc);<a name="line.1491"></a> |
| <span class="sourceLineNo">1492</span> continue;<a name="line.1492"></a> |
| <span class="sourceLineNo">1493</span> }<a name="line.1493"></a> |
| <span class="sourceLineNo">1494</span> LockState lockState = acquireLock(proc);<a name="line.1494"></a> |
| <span class="sourceLineNo">1495</span> if (lockState != LockState.LOCK_ACQUIRED) {<a name="line.1495"></a> |
| <span class="sourceLineNo">1496</span> // can't take a lock on the procedure, add the root-proc back on the<a name="line.1496"></a> |
| <span class="sourceLineNo">1497</span> // queue waiting for the lock availability<a name="line.1497"></a> |
| <span class="sourceLineNo">1498</span> return lockState;<a name="line.1498"></a> |
| <span class="sourceLineNo">1499</span> }<a name="line.1499"></a> |
| <span class="sourceLineNo">1500</span><a name="line.1500"></a> |
| <span class="sourceLineNo">1501</span> lockState = executeRollback(proc);<a name="line.1501"></a> |
| <span class="sourceLineNo">1502</span> releaseLock(proc, false);<a name="line.1502"></a> |
| <span class="sourceLineNo">1503</span> boolean abortRollback = lockState != LockState.LOCK_ACQUIRED;<a name="line.1503"></a> |
| <span class="sourceLineNo">1504</span> abortRollback |= !isRunning() || !store.isRunning();<a name="line.1504"></a> |
| <span class="sourceLineNo">1505</span><a name="line.1505"></a> |
| <span class="sourceLineNo">1506</span> // allows to kill the executor before something is stored to the wal.<a name="line.1506"></a> |
| <span class="sourceLineNo">1507</span> // useful to test the procedure recovery.<a name="line.1507"></a> |
| <span class="sourceLineNo">1508</span> if (abortRollback) {<a name="line.1508"></a> |
| <span class="sourceLineNo">1509</span> return lockState;<a name="line.1509"></a> |
| <span class="sourceLineNo">1510</span> }<a name="line.1510"></a> |
| <span class="sourceLineNo">1511</span><a name="line.1511"></a> |
| <span class="sourceLineNo">1512</span> subprocStack.remove(stackTail);<a name="line.1512"></a> |
| <span class="sourceLineNo">1513</span><a name="line.1513"></a> |
| <span class="sourceLineNo">1514</span> // if the procedure is kind enough to pass the slot to someone else, yield<a name="line.1514"></a> |
| <span class="sourceLineNo">1515</span> // if the proc is already finished, do not yield<a name="line.1515"></a> |
| <span class="sourceLineNo">1516</span> if (!proc.isFinished() && proc.isYieldAfterExecutionStep(getEnvironment())) {<a name="line.1516"></a> |
| <span class="sourceLineNo">1517</span> return LockState.LOCK_YIELD_WAIT;<a name="line.1517"></a> |
| <span class="sourceLineNo">1518</span> }<a name="line.1518"></a> |
| <span class="sourceLineNo">1519</span><a name="line.1519"></a> |
| <span class="sourceLineNo">1520</span> if (proc != rootProc) {<a name="line.1520"></a> |
| <span class="sourceLineNo">1521</span> execCompletionCleanup(proc);<a name="line.1521"></a> |
| <span class="sourceLineNo">1522</span> }<a name="line.1522"></a> |
| <span class="sourceLineNo">1523</span> } finally {<a name="line.1523"></a> |
| <span class="sourceLineNo">1524</span> if (lockEntry != null) {<a name="line.1524"></a> |
| <span class="sourceLineNo">1525</span> procExecutionLock.releaseLockEntry(lockEntry);<a name="line.1525"></a> |
| <span class="sourceLineNo">1526</span> }<a name="line.1526"></a> |
| <span class="sourceLineNo">1527</span> }<a name="line.1527"></a> |
| <span class="sourceLineNo">1528</span> }<a name="line.1528"></a> |
| <span class="sourceLineNo">1529</span><a name="line.1529"></a> |
| <span class="sourceLineNo">1530</span> // Finalize the procedure state<a name="line.1530"></a> |
| <span class="sourceLineNo">1531</span> LOG.info("Rolled back {} exec-time={}", rootProc,<a name="line.1531"></a> |
| <span class="sourceLineNo">1532</span> StringUtils.humanTimeDiff(rootProc.elapsedTime()));<a name="line.1532"></a> |
| <span class="sourceLineNo">1533</span> procedureFinished(rootProc);<a name="line.1533"></a> |
| <span class="sourceLineNo">1534</span> return LockState.LOCK_ACQUIRED;<a name="line.1534"></a> |
| <span class="sourceLineNo">1535</span> }<a name="line.1535"></a> |
| <span class="sourceLineNo">1536</span><a name="line.1536"></a> |
| <span class="sourceLineNo">1537</span> private void cleanupAfterRollbackOneStep(Procedure<TEnvironment> proc) {<a name="line.1537"></a> |
| <span class="sourceLineNo">1538</span> if (proc.removeStackIndex()) {<a name="line.1538"></a> |
| <span class="sourceLineNo">1539</span> if (!proc.isSuccess()) {<a name="line.1539"></a> |
| <span class="sourceLineNo">1540</span> proc.setState(ProcedureState.ROLLEDBACK);<a name="line.1540"></a> |
| <span class="sourceLineNo">1541</span> }<a name="line.1541"></a> |
| <span class="sourceLineNo">1542</span><a name="line.1542"></a> |
| <span class="sourceLineNo">1543</span> // update metrics on finishing the procedure (fail)<a name="line.1543"></a> |
| <span class="sourceLineNo">1544</span> proc.updateMetricsOnFinish(getEnvironment(), proc.elapsedTime(), false);<a name="line.1544"></a> |
| <span class="sourceLineNo">1545</span><a name="line.1545"></a> |
| <span class="sourceLineNo">1546</span> if (proc.hasParent()) {<a name="line.1546"></a> |
| <span class="sourceLineNo">1547</span> store.delete(proc.getProcId());<a name="line.1547"></a> |
| <span class="sourceLineNo">1548</span> procedures.remove(proc.getProcId());<a name="line.1548"></a> |
| <span class="sourceLineNo">1549</span> } else {<a name="line.1549"></a> |
| <span class="sourceLineNo">1550</span> final long[] childProcIds = rollbackStack.get(proc.getProcId()).getSubprocedureIds();<a name="line.1550"></a> |
| <span class="sourceLineNo">1551</span> if (childProcIds != null) {<a name="line.1551"></a> |
| <span class="sourceLineNo">1552</span> store.delete(proc, childProcIds);<a name="line.1552"></a> |
| <span class="sourceLineNo">1553</span> } else {<a name="line.1553"></a> |
| <span class="sourceLineNo">1554</span> store.update(proc);<a name="line.1554"></a> |
| <span class="sourceLineNo">1555</span> }<a name="line.1555"></a> |
| <span class="sourceLineNo">1556</span> }<a name="line.1556"></a> |
| <span class="sourceLineNo">1557</span> } else {<a name="line.1557"></a> |
| <span class="sourceLineNo">1558</span> store.update(proc);<a name="line.1558"></a> |
| <span class="sourceLineNo">1559</span> }<a name="line.1559"></a> |
| <span class="sourceLineNo">1560</span> }<a name="line.1560"></a> |
| <span class="sourceLineNo">1561</span><a name="line.1561"></a> |
| <span class="sourceLineNo">1562</span> /**<a name="line.1562"></a> |
| <span class="sourceLineNo">1563</span> * Execute the rollback of the procedure step.<a name="line.1563"></a> |
| <span class="sourceLineNo">1564</span> * It updates the store with the new state (stack index)<a name="line.1564"></a> |
| <span class="sourceLineNo">1565</span> * or will remove completly the procedure in case it is a child.<a name="line.1565"></a> |
| <span class="sourceLineNo">1566</span> */<a name="line.1566"></a> |
| <span class="sourceLineNo">1567</span> private LockState executeRollback(Procedure<TEnvironment> proc) {<a name="line.1567"></a> |
| <span class="sourceLineNo">1568</span> try {<a name="line.1568"></a> |
| <span class="sourceLineNo">1569</span> proc.doRollback(getEnvironment());<a name="line.1569"></a> |
| <span class="sourceLineNo">1570</span> } catch (IOException e) {<a name="line.1570"></a> |
| <span class="sourceLineNo">1571</span> LOG.debug("Roll back attempt failed for {}", proc, e);<a name="line.1571"></a> |
| <span class="sourceLineNo">1572</span> return LockState.LOCK_YIELD_WAIT;<a name="line.1572"></a> |
| <span class="sourceLineNo">1573</span> } catch (InterruptedException e) {<a name="line.1573"></a> |
| <span class="sourceLineNo">1574</span> handleInterruptedException(proc, e);<a name="line.1574"></a> |
| <span class="sourceLineNo">1575</span> return LockState.LOCK_YIELD_WAIT;<a name="line.1575"></a> |
| <span class="sourceLineNo">1576</span> } catch (Throwable e) {<a name="line.1576"></a> |
| <span class="sourceLineNo">1577</span> // Catch NullPointerExceptions or similar errors...<a name="line.1577"></a> |
| <span class="sourceLineNo">1578</span> LOG.error(HBaseMarkers.FATAL, "CODE-BUG: Uncaught runtime exception for " + proc, e);<a name="line.1578"></a> |
| <span class="sourceLineNo">1579</span> }<a name="line.1579"></a> |
| <span class="sourceLineNo">1580</span><a name="line.1580"></a> |
| <span class="sourceLineNo">1581</span> // allows to kill the executor before something is stored to the wal.<a name="line.1581"></a> |
| <span class="sourceLineNo">1582</span> // useful to test the procedure recovery.<a name="line.1582"></a> |
| <span class="sourceLineNo">1583</span> if (testing != null && testing.shouldKillBeforeStoreUpdate()) {<a name="line.1583"></a> |
| <span class="sourceLineNo">1584</span> String msg = "TESTING: Kill before store update";<a name="line.1584"></a> |
| <span class="sourceLineNo">1585</span> LOG.debug(msg);<a name="line.1585"></a> |
| <span class="sourceLineNo">1586</span> stop();<a name="line.1586"></a> |
| <span class="sourceLineNo">1587</span> throw new RuntimeException(msg);<a name="line.1587"></a> |
| <span class="sourceLineNo">1588</span> }<a name="line.1588"></a> |
| <span class="sourceLineNo">1589</span><a name="line.1589"></a> |
| <span class="sourceLineNo">1590</span> cleanupAfterRollbackOneStep(proc);<a name="line.1590"></a> |
| <span class="sourceLineNo">1591</span><a name="line.1591"></a> |
| <span class="sourceLineNo">1592</span> return LockState.LOCK_ACQUIRED;<a name="line.1592"></a> |
| <span class="sourceLineNo">1593</span> }<a name="line.1593"></a> |
| <span class="sourceLineNo">1594</span><a name="line.1594"></a> |
| <span class="sourceLineNo">1595</span> private void yieldProcedure(Procedure<TEnvironment> proc) {<a name="line.1595"></a> |
| <span class="sourceLineNo">1596</span> releaseLock(proc, false);<a name="line.1596"></a> |
| <span class="sourceLineNo">1597</span> scheduler.yield(proc);<a name="line.1597"></a> |
| <span class="sourceLineNo">1598</span> }<a name="line.1598"></a> |
| <span class="sourceLineNo">1599</span><a name="line.1599"></a> |
| <span class="sourceLineNo">1600</span> /**<a name="line.1600"></a> |
| <span class="sourceLineNo">1601</span> * Executes <code>procedure</code><a name="line.1601"></a> |
| <span class="sourceLineNo">1602</span> * <ul><a name="line.1602"></a> |
| <span class="sourceLineNo">1603</span> * <li>Calls the doExecute() of the procedure<a name="line.1603"></a> |
| <span class="sourceLineNo">1604</span> * <li>If the procedure execution didn't fail (i.e. valid user input)<a name="line.1604"></a> |
| <span class="sourceLineNo">1605</span> * <ul><a name="line.1605"></a> |
| <span class="sourceLineNo">1606</span> * <li>...and returned subprocedures<a name="line.1606"></a> |
| <span class="sourceLineNo">1607</span> * <ul><li>The subprocedures are initialized.<a name="line.1607"></a> |
| <span class="sourceLineNo">1608</span> * <li>The subprocedures are added to the store<a name="line.1608"></a> |
| <span class="sourceLineNo">1609</span> * <li>The subprocedures are added to the runnable queue<a name="line.1609"></a> |
| <span class="sourceLineNo">1610</span> * <li>The procedure is now in a WAITING state, waiting for the subprocedures to complete<a name="line.1610"></a> |
| <span class="sourceLineNo">1611</span> * </ul><a name="line.1611"></a> |
| <span class="sourceLineNo">1612</span> * </li><a name="line.1612"></a> |
| <span class="sourceLineNo">1613</span> * <li>...if there are no subprocedure<a name="line.1613"></a> |
| <span class="sourceLineNo">1614</span> * <ul><li>the procedure completed successfully<a name="line.1614"></a> |
| <span class="sourceLineNo">1615</span> * <li>if there is a parent (WAITING)<a name="line.1615"></a> |
| <span class="sourceLineNo">1616</span> * <li>the parent state will be set to RUNNABLE<a name="line.1616"></a> |
| <span class="sourceLineNo">1617</span> * </ul><a name="line.1617"></a> |
| <span class="sourceLineNo">1618</span> * </li><a name="line.1618"></a> |
| <span class="sourceLineNo">1619</span> * </ul><a name="line.1619"></a> |
| <span class="sourceLineNo">1620</span> * </li><a name="line.1620"></a> |
| <span class="sourceLineNo">1621</span> * <li>In case of failure<a name="line.1621"></a> |
| <span class="sourceLineNo">1622</span> * <ul><a name="line.1622"></a> |
| <span class="sourceLineNo">1623</span> * <li>The store is updated with the new state</li><a name="line.1623"></a> |
| <span class="sourceLineNo">1624</span> * <li>The executor (caller of this method) will start the rollback of the procedure</li><a name="line.1624"></a> |
| <span class="sourceLineNo">1625</span> * </ul><a name="line.1625"></a> |
| <span class="sourceLineNo">1626</span> * </li><a name="line.1626"></a> |
| <span class="sourceLineNo">1627</span> * </ul><a name="line.1627"></a> |
| <span class="sourceLineNo">1628</span> */<a name="line.1628"></a> |
| <span class="sourceLineNo">1629</span> private void execProcedure(RootProcedureState<TEnvironment> procStack,<a name="line.1629"></a> |
| <span class="sourceLineNo">1630</span> Procedure<TEnvironment> procedure) {<a name="line.1630"></a> |
| <span class="sourceLineNo">1631</span> Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,<a name="line.1631"></a> |
| <span class="sourceLineNo">1632</span> "NOT RUNNABLE! " + procedure.toString());<a name="line.1632"></a> |
| <span class="sourceLineNo">1633</span><a name="line.1633"></a> |
| <span class="sourceLineNo">1634</span> // Procedures can suspend themselves. They skip out by throwing a ProcedureSuspendedException.<a name="line.1634"></a> |
| <span class="sourceLineNo">1635</span> // The exception is caught below and then we hurry to the exit without disturbing state. The<a name="line.1635"></a> |
| <span class="sourceLineNo">1636</span> // idea is that the processing of this procedure will be unsuspended later by an external event<a name="line.1636"></a> |
| <span class="sourceLineNo">1637</span> // such the report of a region open.<a name="line.1637"></a> |
| <span class="sourceLineNo">1638</span> boolean suspended = false;<a name="line.1638"></a> |
| <span class="sourceLineNo">1639</span><a name="line.1639"></a> |
| <span class="sourceLineNo">1640</span> // Whether to 're-' -execute; run through the loop again.<a name="line.1640"></a> |
| <span class="sourceLineNo">1641</span> boolean reExecute = false;<a name="line.1641"></a> |
| <span class="sourceLineNo">1642</span><a name="line.1642"></a> |
| <span class="sourceLineNo">1643</span> Procedure<TEnvironment>[] subprocs = null;<a name="line.1643"></a> |
| <span class="sourceLineNo">1644</span> do {<a name="line.1644"></a> |
| <span class="sourceLineNo">1645</span> reExecute = false;<a name="line.1645"></a> |
| <span class="sourceLineNo">1646</span> procedure.resetPersistence();<a name="line.1646"></a> |
| <span class="sourceLineNo">1647</span> try {<a name="line.1647"></a> |
| <span class="sourceLineNo">1648</span> subprocs = procedure.doExecute(getEnvironment());<a name="line.1648"></a> |
| <span class="sourceLineNo">1649</span> if (subprocs != null && subprocs.length == 0) {<a name="line.1649"></a> |
| <span class="sourceLineNo">1650</span> subprocs = null;<a name="line.1650"></a> |
| <span class="sourceLineNo">1651</span> }<a name="line.1651"></a> |
| <span class="sourceLineNo">1652</span> } catch (ProcedureSuspendedException e) {<a name="line.1652"></a> |
| <span class="sourceLineNo">1653</span> LOG.trace("Suspend {}", procedure);<a name="line.1653"></a> |
| <span class="sourceLineNo">1654</span> suspended = true;<a name="line.1654"></a> |
| <span class="sourceLineNo">1655</span> } catch (ProcedureYieldException e) {<a name="line.1655"></a> |
| <span class="sourceLineNo">1656</span> LOG.trace("Yield {}", procedure, e);<a name="line.1656"></a> |
| <span class="sourceLineNo">1657</span> yieldProcedure(procedure);<a name="line.1657"></a> |
| <span class="sourceLineNo">1658</span> return;<a name="line.1658"></a> |
| <span class="sourceLineNo">1659</span> } catch (InterruptedException e) {<a name="line.1659"></a> |
| <span class="sourceLineNo">1660</span> LOG.trace("Yield interrupt {}", procedure, e);<a name="line.1660"></a> |
| <span class="sourceLineNo">1661</span> handleInterruptedException(procedure, e);<a name="line.1661"></a> |
| <span class="sourceLineNo">1662</span> yieldProcedure(procedure);<a name="line.1662"></a> |
| <span class="sourceLineNo">1663</span> return;<a name="line.1663"></a> |
| <span class="sourceLineNo">1664</span> } catch (Throwable e) {<a name="line.1664"></a> |
| <span class="sourceLineNo">1665</span> // Catch NullPointerExceptions or similar errors...<a name="line.1665"></a> |
| <span class="sourceLineNo">1666</span> String msg = "CODE-BUG: Uncaught runtime exception: " + procedure;<a name="line.1666"></a> |
| <span class="sourceLineNo">1667</span> LOG.error(msg, e);<a name="line.1667"></a> |
| <span class="sourceLineNo">1668</span> procedure.setFailure(new RemoteProcedureException(msg, e));<a name="line.1668"></a> |
| <span class="sourceLineNo">1669</span> }<a name="line.1669"></a> |
| <span class="sourceLineNo">1670</span><a name="line.1670"></a> |
| <span class="sourceLineNo">1671</span> if (!procedure.isFailed()) {<a name="line.1671"></a> |
| <span class="sourceLineNo">1672</span> if (subprocs != null) {<a name="line.1672"></a> |
| <span class="sourceLineNo">1673</span> if (subprocs.length == 1 && subprocs[0] == procedure) {<a name="line.1673"></a> |
| <span class="sourceLineNo">1674</span> // Procedure returned itself. Quick-shortcut for a state machine-like procedure;<a name="line.1674"></a> |
| <span class="sourceLineNo">1675</span> // i.e. we go around this loop again rather than go back out on the scheduler queue.<a name="line.1675"></a> |
| <span class="sourceLineNo">1676</span> subprocs = null;<a name="line.1676"></a> |
| <span class="sourceLineNo">1677</span> reExecute = true;<a name="line.1677"></a> |
| <span class="sourceLineNo">1678</span> LOG.trace("Short-circuit to next step on pid={}", procedure.getProcId());<a name="line.1678"></a> |
| <span class="sourceLineNo">1679</span> } else {<a name="line.1679"></a> |
| <span class="sourceLineNo">1680</span> // Yield the current procedure, and make the subprocedure runnable<a name="line.1680"></a> |
| <span class="sourceLineNo">1681</span> // subprocs may come back 'null'.<a name="line.1681"></a> |
| <span class="sourceLineNo">1682</span> subprocs = initializeChildren(procStack, procedure, subprocs);<a name="line.1682"></a> |
| <span class="sourceLineNo">1683</span> LOG.info("Initialized subprocedures=" +<a name="line.1683"></a> |
| <span class="sourceLineNo">1684</span> (subprocs == null? null:<a name="line.1684"></a> |
| <span class="sourceLineNo">1685</span> Stream.of(subprocs).map(e -> "{" + e.toString() + "}").<a name="line.1685"></a> |
| <span class="sourceLineNo">1686</span> collect(Collectors.toList()).toString()));<a name="line.1686"></a> |
| <span class="sourceLineNo">1687</span> }<a name="line.1687"></a> |
| <span class="sourceLineNo">1688</span> } else if (procedure.getState() == ProcedureState.WAITING_TIMEOUT) {<a name="line.1688"></a> |
| <span class="sourceLineNo">1689</span> LOG.trace("Added to timeoutExecutor {}", procedure);<a name="line.1689"></a> |
| <span class="sourceLineNo">1690</span> timeoutExecutor.add(procedure);<a name="line.1690"></a> |
| <span class="sourceLineNo">1691</span> } else if (!suspended) {<a name="line.1691"></a> |
| <span class="sourceLineNo">1692</span> // No subtask, so we are done<a name="line.1692"></a> |
| <span class="sourceLineNo">1693</span> procedure.setState(ProcedureState.SUCCESS);<a name="line.1693"></a> |
| <span class="sourceLineNo">1694</span> }<a name="line.1694"></a> |
| <span class="sourceLineNo">1695</span> }<a name="line.1695"></a> |
| <span class="sourceLineNo">1696</span><a name="line.1696"></a> |
| <span class="sourceLineNo">1697</span> // Add the procedure to the stack<a name="line.1697"></a> |
| <span class="sourceLineNo">1698</span> procStack.addRollbackStep(procedure);<a name="line.1698"></a> |
| <span class="sourceLineNo">1699</span><a name="line.1699"></a> |
| <span class="sourceLineNo">1700</span> // allows to kill the executor before something is stored to the wal.<a name="line.1700"></a> |
| <span class="sourceLineNo">1701</span> // useful to test the procedure recovery.<a name="line.1701"></a> |
| <span class="sourceLineNo">1702</span> if (testing != null &&<a name="line.1702"></a> |
| <span class="sourceLineNo">1703</span> testing.shouldKillBeforeStoreUpdate(suspended, procedure.hasParent())) {<a name="line.1703"></a> |
| <span class="sourceLineNo">1704</span> kill("TESTING: Kill BEFORE store update: " + procedure);<a name="line.1704"></a> |
| <span class="sourceLineNo">1705</span> }<a name="line.1705"></a> |
| <span class="sourceLineNo">1706</span><a name="line.1706"></a> |
| <span class="sourceLineNo">1707</span> // TODO: The code here doesn't check if store is running before persisting to the store as<a name="line.1707"></a> |
| <span class="sourceLineNo">1708</span> // it relies on the method call below to throw RuntimeException to wind up the stack and<a name="line.1708"></a> |
| <span class="sourceLineNo">1709</span> // executor thread to stop. The statement following the method call below seems to check if<a name="line.1709"></a> |
| <span class="sourceLineNo">1710</span> // store is not running, to prevent scheduling children procedures, re-execution or yield<a name="line.1710"></a> |
| <span class="sourceLineNo">1711</span> // of this procedure. This may need more scrutiny and subsequent cleanup in future<a name="line.1711"></a> |
| <span class="sourceLineNo">1712</span> //<a name="line.1712"></a> |
| <span class="sourceLineNo">1713</span> // Commit the transaction even if a suspend (state may have changed). Note this append<a name="line.1713"></a> |
| <span class="sourceLineNo">1714</span> // can take a bunch of time to complete.<a name="line.1714"></a> |
| <span class="sourceLineNo">1715</span> if (procedure.needPersistence()) {<a name="line.1715"></a> |
| <span class="sourceLineNo">1716</span> updateStoreOnExec(procStack, procedure, subprocs);<a name="line.1716"></a> |
| <span class="sourceLineNo">1717</span> }<a name="line.1717"></a> |
| <span class="sourceLineNo">1718</span><a name="line.1718"></a> |
| <span class="sourceLineNo">1719</span> // if the store is not running we are aborting<a name="line.1719"></a> |
| <span class="sourceLineNo">1720</span> if (!store.isRunning()) {<a name="line.1720"></a> |
| <span class="sourceLineNo">1721</span> return;<a name="line.1721"></a> |
| <span class="sourceLineNo">1722</span> }<a name="line.1722"></a> |
| <span class="sourceLineNo">1723</span> // if the procedure is kind enough to pass the slot to someone else, yield<a name="line.1723"></a> |
| <span class="sourceLineNo">1724</span> if (procedure.isRunnable() && !suspended &&<a name="line.1724"></a> |
| <span class="sourceLineNo">1725</span> procedure.isYieldAfterExecutionStep(getEnvironment())) {<a name="line.1725"></a> |
| <span class="sourceLineNo">1726</span> yieldProcedure(procedure);<a name="line.1726"></a> |
| <span class="sourceLineNo">1727</span> return;<a name="line.1727"></a> |
| <span class="sourceLineNo">1728</span> }<a name="line.1728"></a> |
| <span class="sourceLineNo">1729</span><a name="line.1729"></a> |
| <span class="sourceLineNo">1730</span> assert (reExecute && subprocs == null) || !reExecute;<a name="line.1730"></a> |
| <span class="sourceLineNo">1731</span> } while (reExecute);<a name="line.1731"></a> |
| <span class="sourceLineNo">1732</span><a name="line.1732"></a> |
| <span class="sourceLineNo">1733</span> // Allows to kill the executor after something is stored to the WAL but before the below<a name="line.1733"></a> |
| <span class="sourceLineNo">1734</span> // state settings are done -- in particular the one on the end where we make parent<a name="line.1734"></a> |
| <span class="sourceLineNo">1735</span> // RUNNABLE again when its children are done; see countDownChildren.<a name="line.1735"></a> |
| <span class="sourceLineNo">1736</span> if (testing != null && testing.shouldKillAfterStoreUpdate(suspended)) {<a name="line.1736"></a> |
| <span class="sourceLineNo">1737</span> kill("TESTING: Kill AFTER store update: " + procedure);<a name="line.1737"></a> |
| <span class="sourceLineNo">1738</span> }<a name="line.1738"></a> |
| <span class="sourceLineNo">1739</span><a name="line.1739"></a> |
| <span class="sourceLineNo">1740</span> // Submit the new subprocedures<a name="line.1740"></a> |
| <span class="sourceLineNo">1741</span> if (subprocs != null && !procedure.isFailed()) {<a name="line.1741"></a> |
| <span class="sourceLineNo">1742</span> submitChildrenProcedures(subprocs);<a name="line.1742"></a> |
| <span class="sourceLineNo">1743</span> }<a name="line.1743"></a> |
| <span class="sourceLineNo">1744</span><a name="line.1744"></a> |
| <span class="sourceLineNo">1745</span> // we need to log the release lock operation before waking up the parent procedure, as there<a name="line.1745"></a> |
| <span class="sourceLineNo">1746</span> // could be race that the parent procedure may call updateStoreOnExec ahead of us and remove all<a name="line.1746"></a> |
| <span class="sourceLineNo">1747</span> // the sub procedures from store and cause problems...<a name="line.1747"></a> |
| <span class="sourceLineNo">1748</span> releaseLock(procedure, false);<a name="line.1748"></a> |
| <span class="sourceLineNo">1749</span><a name="line.1749"></a> |
| <span class="sourceLineNo">1750</span> // if the procedure is complete and has a parent, count down the children latch.<a name="line.1750"></a> |
| <span class="sourceLineNo">1751</span> // If 'suspended', do nothing to change state -- let other threads handle unsuspend event.<a name="line.1751"></a> |
| <span class="sourceLineNo">1752</span> if (!suspended && procedure.isFinished() && procedure.hasParent()) {<a name="line.1752"></a> |
| <span class="sourceLineNo">1753</span> countDownChildren(procStack, procedure);<a name="line.1753"></a> |
| <span class="sourceLineNo">1754</span> }<a name="line.1754"></a> |
| <span class="sourceLineNo">1755</span> }<a name="line.1755"></a> |
| <span class="sourceLineNo">1756</span><a name="line.1756"></a> |
| <span class="sourceLineNo">1757</span> private void kill(String msg) {<a name="line.1757"></a> |
| <span class="sourceLineNo">1758</span> LOG.debug(msg);<a name="line.1758"></a> |
| <span class="sourceLineNo">1759</span> stop();<a name="line.1759"></a> |
| <span class="sourceLineNo">1760</span> throw new RuntimeException(msg);<a name="line.1760"></a> |
| <span class="sourceLineNo">1761</span> }<a name="line.1761"></a> |
| <span class="sourceLineNo">1762</span><a name="line.1762"></a> |
| <span class="sourceLineNo">1763</span> private Procedure<TEnvironment>[] initializeChildren(RootProcedureState<TEnvironment> procStack,<a name="line.1763"></a> |
| <span class="sourceLineNo">1764</span> Procedure<TEnvironment> procedure, Procedure<TEnvironment>[] subprocs) {<a name="line.1764"></a> |
| <span class="sourceLineNo">1765</span> assert subprocs != null : "expected subprocedures";<a name="line.1765"></a> |
| <span class="sourceLineNo">1766</span> final long rootProcId = getRootProcedureId(procedure);<a name="line.1766"></a> |
| <span class="sourceLineNo">1767</span> for (int i = 0; i < subprocs.length; ++i) {<a name="line.1767"></a> |
| <span class="sourceLineNo">1768</span> Procedure<TEnvironment> subproc = subprocs[i];<a name="line.1768"></a> |
| <span class="sourceLineNo">1769</span> if (subproc == null) {<a name="line.1769"></a> |
| <span class="sourceLineNo">1770</span> String msg = "subproc[" + i + "] is null, aborting the procedure";<a name="line.1770"></a> |
| <span class="sourceLineNo">1771</span> procedure.setFailure(new RemoteProcedureException(msg,<a name="line.1771"></a> |
| <span class="sourceLineNo">1772</span> new IllegalArgumentIOException(msg)));<a name="line.1772"></a> |
| <span class="sourceLineNo">1773</span> return null;<a name="line.1773"></a> |
| <span class="sourceLineNo">1774</span> }<a name="line.1774"></a> |
| <span class="sourceLineNo">1775</span><a name="line.1775"></a> |
| <span class="sourceLineNo">1776</span> assert subproc.getState() == ProcedureState.INITIALIZING : subproc;<a name="line.1776"></a> |
| <span class="sourceLineNo">1777</span> subproc.setParentProcId(procedure.getProcId());<a name="line.1777"></a> |
| <span class="sourceLineNo">1778</span> subproc.setRootProcId(rootProcId);<a name="line.1778"></a> |
| <span class="sourceLineNo">1779</span> subproc.setProcId(nextProcId());<a name="line.1779"></a> |
| <span class="sourceLineNo">1780</span> procStack.addSubProcedure(subproc);<a name="line.1780"></a> |
| <span class="sourceLineNo">1781</span> }<a name="line.1781"></a> |
| <span class="sourceLineNo">1782</span><a name="line.1782"></a> |
| <span class="sourceLineNo">1783</span> if (!procedure.isFailed()) {<a name="line.1783"></a> |
| <span class="sourceLineNo">1784</span> procedure.setChildrenLatch(subprocs.length);<a name="line.1784"></a> |
| <span class="sourceLineNo">1785</span> switch (procedure.getState()) {<a name="line.1785"></a> |
| <span class="sourceLineNo">1786</span> case RUNNABLE:<a name="line.1786"></a> |
| <span class="sourceLineNo">1787</span> procedure.setState(ProcedureState.WAITING);<a name="line.1787"></a> |
| <span class="sourceLineNo">1788</span> break;<a name="line.1788"></a> |
| <span class="sourceLineNo">1789</span> case WAITING_TIMEOUT:<a name="line.1789"></a> |
| <span class="sourceLineNo">1790</span> timeoutExecutor.add(procedure);<a name="line.1790"></a> |
| <span class="sourceLineNo">1791</span> break;<a name="line.1791"></a> |
| <span class="sourceLineNo">1792</span> default:<a name="line.1792"></a> |
| <span class="sourceLineNo">1793</span> break;<a name="line.1793"></a> |
| <span class="sourceLineNo">1794</span> }<a name="line.1794"></a> |
| <span class="sourceLineNo">1795</span> }<a name="line.1795"></a> |
| <span class="sourceLineNo">1796</span> return subprocs;<a name="line.1796"></a> |
| <span class="sourceLineNo">1797</span> }<a name="line.1797"></a> |
| <span class="sourceLineNo">1798</span><a name="line.1798"></a> |
| <span class="sourceLineNo">1799</span> private void submitChildrenProcedures(Procedure<TEnvironment>[] subprocs) {<a name="line.1799"></a> |
| <span class="sourceLineNo">1800</span> for (int i = 0; i < subprocs.length; ++i) {<a name="line.1800"></a> |
| <span class="sourceLineNo">1801</span> Procedure<TEnvironment> subproc = subprocs[i];<a name="line.1801"></a> |
| <span class="sourceLineNo">1802</span> subproc.updateMetricsOnSubmit(getEnvironment());<a name="line.1802"></a> |
| <span class="sourceLineNo">1803</span> assert !procedures.containsKey(subproc.getProcId());<a name="line.1803"></a> |
| <span class="sourceLineNo">1804</span> procedures.put(subproc.getProcId(), subproc);<a name="line.1804"></a> |
| <span class="sourceLineNo">1805</span> scheduler.addFront(subproc);<a name="line.1805"></a> |
| <span class="sourceLineNo">1806</span> }<a name="line.1806"></a> |
| <span class="sourceLineNo">1807</span> }<a name="line.1807"></a> |
| <span class="sourceLineNo">1808</span><a name="line.1808"></a> |
| <span class="sourceLineNo">1809</span> private void countDownChildren(RootProcedureState<TEnvironment> procStack,<a name="line.1809"></a> |
| <span class="sourceLineNo">1810</span> Procedure<TEnvironment> procedure) {<a name="line.1810"></a> |
| <span class="sourceLineNo">1811</span> Procedure<TEnvironment> parent = procedures.get(procedure.getParentProcId());<a name="line.1811"></a> |
| <span class="sourceLineNo">1812</span> if (parent == null) {<a name="line.1812"></a> |
| <span class="sourceLineNo">1813</span> assert procStack.isRollingback();<a name="line.1813"></a> |
| <span class="sourceLineNo">1814</span> return;<a name="line.1814"></a> |
| <span class="sourceLineNo">1815</span> }<a name="line.1815"></a> |
| <span class="sourceLineNo">1816</span><a name="line.1816"></a> |
| <span class="sourceLineNo">1817</span> // If this procedure is the last child awake the parent procedure<a name="line.1817"></a> |
| <span class="sourceLineNo">1818</span> if (parent.tryRunnable()) {<a name="line.1818"></a> |
| <span class="sourceLineNo">1819</span> // If we succeeded in making the parent runnable -- i.e. all of its<a name="line.1819"></a> |
| <span class="sourceLineNo">1820</span> // children have completed, move parent to front of the queue.<a name="line.1820"></a> |
| <span class="sourceLineNo">1821</span> store.update(parent);<a name="line.1821"></a> |
| <span class="sourceLineNo">1822</span> scheduler.addFront(parent);<a name="line.1822"></a> |
| <span class="sourceLineNo">1823</span> LOG.info("Finished subprocedure(s) of " + parent + "; resume parent processing.");<a name="line.1823"></a> |
| <span class="sourceLineNo">1824</span> return;<a name="line.1824"></a> |
| <span class="sourceLineNo">1825</span> }<a name="line.1825"></a> |
| <span class="sourceLineNo">1826</span> }<a name="line.1826"></a> |
| <span class="sourceLineNo">1827</span><a name="line.1827"></a> |
| <span class="sourceLineNo">1828</span> private void updateStoreOnExec(RootProcedureState<TEnvironment> procStack,<a name="line.1828"></a> |
| <span class="sourceLineNo">1829</span> Procedure<TEnvironment> procedure, Procedure<TEnvironment>[] subprocs) {<a name="line.1829"></a> |
| <span class="sourceLineNo">1830</span> if (subprocs != null && !procedure.isFailed()) {<a name="line.1830"></a> |
| <span class="sourceLineNo">1831</span> if (LOG.isTraceEnabled()) {<a name="line.1831"></a> |
| <span class="sourceLineNo">1832</span> LOG.trace("Stored " + procedure + ", children " + Arrays.toString(subprocs));<a name="line.1832"></a> |
| <span class="sourceLineNo">1833</span> }<a name="line.1833"></a> |
| <span class="sourceLineNo">1834</span> store.insert(procedure, subprocs);<a name="line.1834"></a> |
| <span class="sourceLineNo">1835</span> } else {<a name="line.1835"></a> |
| <span class="sourceLineNo">1836</span> LOG.trace("Store update {}", procedure);<a name="line.1836"></a> |
| <span class="sourceLineNo">1837</span> if (procedure.isFinished() && !procedure.hasParent()) {<a name="line.1837"></a> |
| <span class="sourceLineNo">1838</span> // remove child procedures<a name="line.1838"></a> |
| <span class="sourceLineNo">1839</span> final long[] childProcIds = procStack.getSubprocedureIds();<a name="line.1839"></a> |
| <span class="sourceLineNo">1840</span> if (childProcIds != null) {<a name="line.1840"></a> |
| <span class="sourceLineNo">1841</span> store.delete(procedure, childProcIds);<a name="line.1841"></a> |
| <span class="sourceLineNo">1842</span> for (int i = 0; i < childProcIds.length; ++i) {<a name="line.1842"></a> |
| <span class="sourceLineNo">1843</span> procedures.remove(childProcIds[i]);<a name="line.1843"></a> |
| <span class="sourceLineNo">1844</span> }<a name="line.1844"></a> |
| <span class="sourceLineNo">1845</span> } else {<a name="line.1845"></a> |
| <span class="sourceLineNo">1846</span> store.update(procedure);<a name="line.1846"></a> |
| <span class="sourceLineNo">1847</span> }<a name="line.1847"></a> |
| <span class="sourceLineNo">1848</span> } else {<a name="line.1848"></a> |
| <span class="sourceLineNo">1849</span> store.update(procedure);<a name="line.1849"></a> |
| <span class="sourceLineNo">1850</span> }<a name="line.1850"></a> |
| <span class="sourceLineNo">1851</span> }<a name="line.1851"></a> |
| <span class="sourceLineNo">1852</span> }<a name="line.1852"></a> |
| <span class="sourceLineNo">1853</span><a name="line.1853"></a> |
| <span class="sourceLineNo">1854</span> private void handleInterruptedException(Procedure<TEnvironment> proc, InterruptedException e) {<a name="line.1854"></a> |
| <span class="sourceLineNo">1855</span> LOG.trace("Interrupt during {}. suspend and retry it later.", proc, e);<a name="line.1855"></a> |
| <span class="sourceLineNo">1856</span> // NOTE: We don't call Thread.currentThread().interrupt()<a name="line.1856"></a> |
| <span class="sourceLineNo">1857</span> // because otherwise all the subsequent calls e.g. Thread.sleep() will throw<a name="line.1857"></a> |
| <span class="sourceLineNo">1858</span> // the InterruptedException. If the master is going down, we will be notified<a name="line.1858"></a> |
| <span class="sourceLineNo">1859</span> // and the executor/store will be stopped.<a name="line.1859"></a> |
| <span class="sourceLineNo">1860</span> // (The interrupted procedure will be retried on the next run)<a name="line.1860"></a> |
| <span class="sourceLineNo">1861</span> }<a name="line.1861"></a> |
| <span class="sourceLineNo">1862</span><a name="line.1862"></a> |
| <span class="sourceLineNo">1863</span> private void execCompletionCleanup(Procedure<TEnvironment> proc) {<a name="line.1863"></a> |
| <span class="sourceLineNo">1864</span> final TEnvironment env = getEnvironment();<a name="line.1864"></a> |
| <span class="sourceLineNo">1865</span> if (proc.hasLock()) {<a name="line.1865"></a> |
| <span class="sourceLineNo">1866</span> LOG.warn("Usually this should not happen, we will release the lock before if the procedure" +<a name="line.1866"></a> |
| <span class="sourceLineNo">1867</span> " is finished, even if the holdLock is true, arrive here means we have some holes where" +<a name="line.1867"></a> |
| <span class="sourceLineNo">1868</span> " we do not release the lock. And the releaseLock below may fail since the procedure may" +<a name="line.1868"></a> |
| <span class="sourceLineNo">1869</span> " have already been deleted from the procedure store.");<a name="line.1869"></a> |
| <span class="sourceLineNo">1870</span> releaseLock(proc, true);<a name="line.1870"></a> |
| <span class="sourceLineNo">1871</span> }<a name="line.1871"></a> |
| <span class="sourceLineNo">1872</span> try {<a name="line.1872"></a> |
| <span class="sourceLineNo">1873</span> proc.completionCleanup(env);<a name="line.1873"></a> |
| <span class="sourceLineNo">1874</span> } catch (Throwable e) {<a name="line.1874"></a> |
| <span class="sourceLineNo">1875</span> // Catch NullPointerExceptions or similar errors...<a name="line.1875"></a> |
| <span class="sourceLineNo">1876</span> LOG.error("CODE-BUG: uncatched runtime exception for procedure: " + proc, e);<a name="line.1876"></a> |
| <span class="sourceLineNo">1877</span> }<a name="line.1877"></a> |
| <span class="sourceLineNo">1878</span> }<a name="line.1878"></a> |
| <span class="sourceLineNo">1879</span><a name="line.1879"></a> |
| <span class="sourceLineNo">1880</span> private void procedureFinished(Procedure<TEnvironment> proc) {<a name="line.1880"></a> |
| <span class="sourceLineNo">1881</span> // call the procedure completion cleanup handler<a name="line.1881"></a> |
| <span class="sourceLineNo">1882</span> execCompletionCleanup(proc);<a name="line.1882"></a> |
| <span class="sourceLineNo">1883</span><a name="line.1883"></a> |
| <span class="sourceLineNo">1884</span> CompletedProcedureRetainer<TEnvironment> retainer = new CompletedProcedureRetainer<>(proc);<a name="line.1884"></a> |
| <span class="sourceLineNo">1885</span><a name="line.1885"></a> |
| <span class="sourceLineNo">1886</span> // update the executor internal state maps<a name="line.1886"></a> |
| <span class="sourceLineNo">1887</span> if (!proc.shouldWaitClientAck(getEnvironment())) {<a name="line.1887"></a> |
| <span class="sourceLineNo">1888</span> retainer.setClientAckTime(0);<a name="line.1888"></a> |
| <span class="sourceLineNo">1889</span> }<a name="line.1889"></a> |
| <span class="sourceLineNo">1890</span><a name="line.1890"></a> |
| <span class="sourceLineNo">1891</span> completed.put(proc.getProcId(), retainer);<a name="line.1891"></a> |
| <span class="sourceLineNo">1892</span> rollbackStack.remove(proc.getProcId());<a name="line.1892"></a> |
| <span class="sourceLineNo">1893</span> procedures.remove(proc.getProcId());<a name="line.1893"></a> |
| <span class="sourceLineNo">1894</span><a name="line.1894"></a> |
| <span class="sourceLineNo">1895</span> // call the runnableSet completion cleanup handler<a name="line.1895"></a> |
| <span class="sourceLineNo">1896</span> try {<a name="line.1896"></a> |
| <span class="sourceLineNo">1897</span> scheduler.completionCleanup(proc);<a name="line.1897"></a> |
| <span class="sourceLineNo">1898</span> } catch (Throwable e) {<a name="line.1898"></a> |
| <span class="sourceLineNo">1899</span> // Catch NullPointerExceptions or similar errors...<a name="line.1899"></a> |
| <span class="sourceLineNo">1900</span> LOG.error("CODE-BUG: uncatched runtime exception for completion cleanup: {}", proc, e);<a name="line.1900"></a> |
| <span class="sourceLineNo">1901</span> }<a name="line.1901"></a> |
| <span class="sourceLineNo">1902</span><a name="line.1902"></a> |
| <span class="sourceLineNo">1903</span> // Notify the listeners<a name="line.1903"></a> |
| <span class="sourceLineNo">1904</span> sendProcedureFinishedNotification(proc.getProcId());<a name="line.1904"></a> |
| <span class="sourceLineNo">1905</span> }<a name="line.1905"></a> |
| <span class="sourceLineNo">1906</span><a name="line.1906"></a> |
| <span class="sourceLineNo">1907</span> RootProcedureState<TEnvironment> getProcStack(long rootProcId) {<a name="line.1907"></a> |
| <span class="sourceLineNo">1908</span> return rollbackStack.get(rootProcId);<a name="line.1908"></a> |
| <span class="sourceLineNo">1909</span> }<a name="line.1909"></a> |
| <span class="sourceLineNo">1910</span><a name="line.1910"></a> |
| <span class="sourceLineNo">1911</span> @VisibleForTesting<a name="line.1911"></a> |
| <span class="sourceLineNo">1912</span> ProcedureScheduler getProcedureScheduler() {<a name="line.1912"></a> |
| <span class="sourceLineNo">1913</span> return scheduler;<a name="line.1913"></a> |
| <span class="sourceLineNo">1914</span> }<a name="line.1914"></a> |
| <span class="sourceLineNo">1915</span><a name="line.1915"></a> |
| <span class="sourceLineNo">1916</span> @VisibleForTesting<a name="line.1916"></a> |
| <span class="sourceLineNo">1917</span> int getCompletedSize() {<a name="line.1917"></a> |
| <span class="sourceLineNo">1918</span> return completed.size();<a name="line.1918"></a> |
| <span class="sourceLineNo">1919</span> }<a name="line.1919"></a> |
| <span class="sourceLineNo">1920</span><a name="line.1920"></a> |
| <span class="sourceLineNo">1921</span> @VisibleForTesting<a name="line.1921"></a> |
| <span class="sourceLineNo">1922</span> public IdLock getProcExecutionLock() {<a name="line.1922"></a> |
| <span class="sourceLineNo">1923</span> return procExecutionLock;<a name="line.1923"></a> |
| <span class="sourceLineNo">1924</span> }<a name="line.1924"></a> |
| <span class="sourceLineNo">1925</span><a name="line.1925"></a> |
| <span class="sourceLineNo">1926</span> // ==========================================================================<a name="line.1926"></a> |
| <span class="sourceLineNo">1927</span> // Worker Thread<a name="line.1927"></a> |
| <span class="sourceLineNo">1928</span> // ==========================================================================<a name="line.1928"></a> |
| <span class="sourceLineNo">1929</span> private class WorkerThread extends StoppableThread {<a name="line.1929"></a> |
| <span class="sourceLineNo">1930</span> private final AtomicLong executionStartTime = new AtomicLong(Long.MAX_VALUE);<a name="line.1930"></a> |
| <span class="sourceLineNo">1931</span> private volatile Procedure<TEnvironment> activeProcedure;<a name="line.1931"></a> |
| <span class="sourceLineNo">1932</span><a name="line.1932"></a> |
| <span class="sourceLineNo">1933</span> public WorkerThread(ThreadGroup group) {<a name="line.1933"></a> |
| <span class="sourceLineNo">1934</span> this(group, "PEWorker-");<a name="line.1934"></a> |
| <span class="sourceLineNo">1935</span> }<a name="line.1935"></a> |
| <span class="sourceLineNo">1936</span><a name="line.1936"></a> |
| <span class="sourceLineNo">1937</span> protected WorkerThread(ThreadGroup group, String prefix) {<a name="line.1937"></a> |
| <span class="sourceLineNo">1938</span> super(group, prefix + workerId.incrementAndGet());<a name="line.1938"></a> |
| <span class="sourceLineNo">1939</span> setDaemon(true);<a name="line.1939"></a> |
| <span class="sourceLineNo">1940</span> }<a name="line.1940"></a> |
| <span class="sourceLineNo">1941</span><a name="line.1941"></a> |
| <span class="sourceLineNo">1942</span> @Override<a name="line.1942"></a> |
| <span class="sourceLineNo">1943</span> public void sendStopSignal() {<a name="line.1943"></a> |
| <span class="sourceLineNo">1944</span> scheduler.signalAll();<a name="line.1944"></a> |
| <span class="sourceLineNo">1945</span> }<a name="line.1945"></a> |
| <span class="sourceLineNo">1946</span> @Override<a name="line.1946"></a> |
| <span class="sourceLineNo">1947</span> public void run() {<a name="line.1947"></a> |
| <span class="sourceLineNo">1948</span> long lastUpdate = EnvironmentEdgeManager.currentTime();<a name="line.1948"></a> |
| <span class="sourceLineNo">1949</span> try {<a name="line.1949"></a> |
| <span class="sourceLineNo">1950</span> while (isRunning() && keepAlive(lastUpdate)) {<a name="line.1950"></a> |
| <span class="sourceLineNo">1951</span> @SuppressWarnings("unchecked")<a name="line.1951"></a> |
| <span class="sourceLineNo">1952</span> Procedure<TEnvironment> proc = scheduler.poll(keepAliveTime, TimeUnit.MILLISECONDS);<a name="line.1952"></a> |
| <span class="sourceLineNo">1953</span> if (proc == null) {<a name="line.1953"></a> |
| <span class="sourceLineNo">1954</span> continue;<a name="line.1954"></a> |
| <span class="sourceLineNo">1955</span> }<a name="line.1955"></a> |
| <span class="sourceLineNo">1956</span> this.activeProcedure = proc;<a name="line.1956"></a> |
| <span class="sourceLineNo">1957</span> int activeCount = activeExecutorCount.incrementAndGet();<a name="line.1957"></a> |
| <span class="sourceLineNo">1958</span> int runningCount = store.setRunningProcedureCount(activeCount);<a name="line.1958"></a> |
| <span class="sourceLineNo">1959</span> LOG.trace("Execute pid={} runningCount={}, activeCount={}", proc.getProcId(),<a name="line.1959"></a> |
| <span class="sourceLineNo">1960</span> runningCount, activeCount);<a name="line.1960"></a> |
| <span class="sourceLineNo">1961</span> executionStartTime.set(EnvironmentEdgeManager.currentTime());<a name="line.1961"></a> |
| <span class="sourceLineNo">1962</span> IdLock.Entry lockEntry = procExecutionLock.getLockEntry(proc.getProcId());<a name="line.1962"></a> |
| <span class="sourceLineNo">1963</span> try {<a name="line.1963"></a> |
| <span class="sourceLineNo">1964</span> executeProcedure(proc);<a name="line.1964"></a> |
| <span class="sourceLineNo">1965</span> } catch (AssertionError e) {<a name="line.1965"></a> |
| <span class="sourceLineNo">1966</span> LOG.info("ASSERT pid=" + proc.getProcId(), e);<a name="line.1966"></a> |
| <span class="sourceLineNo">1967</span> throw e;<a name="line.1967"></a> |
| <span class="sourceLineNo">1968</span> } finally {<a name="line.1968"></a> |
| <span class="sourceLineNo">1969</span> procExecutionLock.releaseLockEntry(lockEntry);<a name="line.1969"></a> |
| <span class="sourceLineNo">1970</span> activeCount = activeExecutorCount.decrementAndGet();<a name="line.1970"></a> |
| <span class="sourceLineNo">1971</span> runningCount = store.setRunningProcedureCount(activeCount);<a name="line.1971"></a> |
| <span class="sourceLineNo">1972</span> LOG.trace("Halt pid={} runningCount={}, activeCount={}", proc.getProcId(),<a name="line.1972"></a> |
| <span class="sourceLineNo">1973</span> runningCount, activeCount);<a name="line.1973"></a> |
| <span class="sourceLineNo">1974</span> this.activeProcedure = null;<a name="line.1974"></a> |
| <span class="sourceLineNo">1975</span> lastUpdate = EnvironmentEdgeManager.currentTime();<a name="line.1975"></a> |
| <span class="sourceLineNo">1976</span> executionStartTime.set(Long.MAX_VALUE);<a name="line.1976"></a> |
| <span class="sourceLineNo">1977</span> }<a name="line.1977"></a> |
| <span class="sourceLineNo">1978</span> }<a name="line.1978"></a> |
| <span class="sourceLineNo">1979</span> } catch (Throwable t) {<a name="line.1979"></a> |
| <span class="sourceLineNo">1980</span> LOG.warn("Worker terminating UNNATURALLY {}", this.activeProcedure, t);<a name="line.1980"></a> |
| <span class="sourceLineNo">1981</span> } finally {<a name="line.1981"></a> |
| <span class="sourceLineNo">1982</span> LOG.trace("Worker terminated.");<a name="line.1982"></a> |
| <span class="sourceLineNo">1983</span> }<a name="line.1983"></a> |
| <span class="sourceLineNo">1984</span> workerThreads.remove(this);<a name="line.1984"></a> |
| <span class="sourceLineNo">1985</span> }<a name="line.1985"></a> |
| <span class="sourceLineNo">1986</span><a name="line.1986"></a> |
| <span class="sourceLineNo">1987</span> @Override<a name="line.1987"></a> |
| <span class="sourceLineNo">1988</span> public String toString() {<a name="line.1988"></a> |
| <span class="sourceLineNo">1989</span> Procedure<?> p = this.activeProcedure;<a name="line.1989"></a> |
| <span class="sourceLineNo">1990</span> return getName() + "(pid=" + (p == null? Procedure.NO_PROC_ID: p.getProcId() + ")");<a name="line.1990"></a> |
| <span class="sourceLineNo">1991</span> }<a name="line.1991"></a> |
| <span class="sourceLineNo">1992</span><a name="line.1992"></a> |
| <span class="sourceLineNo">1993</span> /**<a name="line.1993"></a> |
| <span class="sourceLineNo">1994</span> * @return the time since the current procedure is running<a name="line.1994"></a> |
| <span class="sourceLineNo">1995</span> */<a name="line.1995"></a> |
| <span class="sourceLineNo">1996</span> public long getCurrentRunTime() {<a name="line.1996"></a> |
| <span class="sourceLineNo">1997</span> return EnvironmentEdgeManager.currentTime() - executionStartTime.get();<a name="line.1997"></a> |
| <span class="sourceLineNo">1998</span> }<a name="line.1998"></a> |
| <span class="sourceLineNo">1999</span><a name="line.1999"></a> |
| <span class="sourceLineNo">2000</span> // core worker never timeout<a name="line.2000"></a> |
| <span class="sourceLineNo">2001</span> protected boolean keepAlive(long lastUpdate) {<a name="line.2001"></a> |
| <span class="sourceLineNo">2002</span> return true;<a name="line.2002"></a> |
| <span class="sourceLineNo">2003</span> }<a name="line.2003"></a> |
| <span class="sourceLineNo">2004</span> }<a name="line.2004"></a> |
| <span class="sourceLineNo">2005</span><a name="line.2005"></a> |
| <span class="sourceLineNo">2006</span> // A worker thread which can be added when core workers are stuck. Will timeout after<a name="line.2006"></a> |
| <span class="sourceLineNo">2007</span> // keepAliveTime if there is no procedure to run.<a name="line.2007"></a> |
| <span class="sourceLineNo">2008</span> private final class KeepAliveWorkerThread extends WorkerThread {<a name="line.2008"></a> |
| <span class="sourceLineNo">2009</span> public KeepAliveWorkerThread(ThreadGroup group) {<a name="line.2009"></a> |
| <span class="sourceLineNo">2010</span> super(group, "KeepAlivePEWorker-");<a name="line.2010"></a> |
| <span class="sourceLineNo">2011</span> }<a name="line.2011"></a> |
| <span class="sourceLineNo">2012</span><a name="line.2012"></a> |
| <span class="sourceLineNo">2013</span> @Override<a name="line.2013"></a> |
| <span class="sourceLineNo">2014</span> protected boolean keepAlive(long lastUpdate) {<a name="line.2014"></a> |
| <span class="sourceLineNo">2015</span> return EnvironmentEdgeManager.currentTime() - lastUpdate < keepAliveTime;<a name="line.2015"></a> |
| <span class="sourceLineNo">2016</span> }<a name="line.2016"></a> |
| <span class="sourceLineNo">2017</span> }<a name="line.2017"></a> |
| <span class="sourceLineNo">2018</span><a name="line.2018"></a> |
| <span class="sourceLineNo">2019</span> // ----------------------------------------------------------------------------<a name="line.2019"></a> |
| <span class="sourceLineNo">2020</span> // TODO-MAYBE: Should we provide a InlineChore to notify the store with the<a name="line.2020"></a> |
| <span class="sourceLineNo">2021</span> // full set of procedures pending and completed to write a compacted<a name="line.2021"></a> |
| <span class="sourceLineNo">2022</span> // version of the log (in case is a log)?<a name="line.2022"></a> |
| <span class="sourceLineNo">2023</span> // In theory no, procedures are have a short life, so at some point the store<a name="line.2023"></a> |
| <span class="sourceLineNo">2024</span> // will have the tracker saying everything is in the last log.<a name="line.2024"></a> |
| <span class="sourceLineNo">2025</span> // ----------------------------------------------------------------------------<a name="line.2025"></a> |
| <span class="sourceLineNo">2026</span><a name="line.2026"></a> |
| <span class="sourceLineNo">2027</span> private final class WorkerMonitor extends InlineChore {<a name="line.2027"></a> |
| <span class="sourceLineNo">2028</span> public static final String WORKER_MONITOR_INTERVAL_CONF_KEY =<a name="line.2028"></a> |
| <span class="sourceLineNo">2029</span> "hbase.procedure.worker.monitor.interval.msec";<a name="line.2029"></a> |
| <span class="sourceLineNo">2030</span> private static final int DEFAULT_WORKER_MONITOR_INTERVAL = 5000; // 5sec<a name="line.2030"></a> |
| <span class="sourceLineNo">2031</span><a name="line.2031"></a> |
| <span class="sourceLineNo">2032</span> public static final String WORKER_STUCK_THRESHOLD_CONF_KEY =<a name="line.2032"></a> |
| <span class="sourceLineNo">2033</span> "hbase.procedure.worker.stuck.threshold.msec";<a name="line.2033"></a> |
| <span class="sourceLineNo">2034</span> private static final int DEFAULT_WORKER_STUCK_THRESHOLD = 10000; // 10sec<a name="line.2034"></a> |
| <span class="sourceLineNo">2035</span><a name="line.2035"></a> |
| <span class="sourceLineNo">2036</span> public static final String WORKER_ADD_STUCK_PERCENTAGE_CONF_KEY =<a name="line.2036"></a> |
| <span class="sourceLineNo">2037</span> "hbase.procedure.worker.add.stuck.percentage";<a name="line.2037"></a> |
| <span class="sourceLineNo">2038</span> private static final float DEFAULT_WORKER_ADD_STUCK_PERCENTAGE = 0.5f; // 50% stuck<a name="line.2038"></a> |
| <span class="sourceLineNo">2039</span><a name="line.2039"></a> |
| <span class="sourceLineNo">2040</span> private float addWorkerStuckPercentage = DEFAULT_WORKER_ADD_STUCK_PERCENTAGE;<a name="line.2040"></a> |
| <span class="sourceLineNo">2041</span> private int timeoutInterval = DEFAULT_WORKER_MONITOR_INTERVAL;<a name="line.2041"></a> |
| <span class="sourceLineNo">2042</span> private int stuckThreshold = DEFAULT_WORKER_STUCK_THRESHOLD;<a name="line.2042"></a> |
| <span class="sourceLineNo">2043</span><a name="line.2043"></a> |
| <span class="sourceLineNo">2044</span> public WorkerMonitor() {<a name="line.2044"></a> |
| <span class="sourceLineNo">2045</span> refreshConfig();<a name="line.2045"></a> |
| <span class="sourceLineNo">2046</span> }<a name="line.2046"></a> |
| <span class="sourceLineNo">2047</span><a name="line.2047"></a> |
| <span class="sourceLineNo">2048</span> @Override<a name="line.2048"></a> |
| <span class="sourceLineNo">2049</span> public void run() {<a name="line.2049"></a> |
| <span class="sourceLineNo">2050</span> final int stuckCount = checkForStuckWorkers();<a name="line.2050"></a> |
| <span class="sourceLineNo">2051</span> checkThreadCount(stuckCount);<a name="line.2051"></a> |
| <span class="sourceLineNo">2052</span><a name="line.2052"></a> |
| <span class="sourceLineNo">2053</span> // refresh interval (poor man dynamic conf update)<a name="line.2053"></a> |
| <span class="sourceLineNo">2054</span> refreshConfig();<a name="line.2054"></a> |
| <span class="sourceLineNo">2055</span> }<a name="line.2055"></a> |
| <span class="sourceLineNo">2056</span><a name="line.2056"></a> |
| <span class="sourceLineNo">2057</span> private int checkForStuckWorkers() {<a name="line.2057"></a> |
| <span class="sourceLineNo">2058</span> // check if any of the worker is stuck<a name="line.2058"></a> |
| <span class="sourceLineNo">2059</span> int stuckCount = 0;<a name="line.2059"></a> |
| <span class="sourceLineNo">2060</span> for (WorkerThread worker : workerThreads) {<a name="line.2060"></a> |
| <span class="sourceLineNo">2061</span> if (worker.getCurrentRunTime() < stuckThreshold) {<a name="line.2061"></a> |
| <span class="sourceLineNo">2062</span> continue;<a name="line.2062"></a> |
| <span class="sourceLineNo">2063</span> }<a name="line.2063"></a> |
| <span class="sourceLineNo">2064</span><a name="line.2064"></a> |
| <span class="sourceLineNo">2065</span> // WARN the worker is stuck<a name="line.2065"></a> |
| <span class="sourceLineNo">2066</span> stuckCount++;<a name="line.2066"></a> |
| <span class="sourceLineNo">2067</span> LOG.warn("Worker stuck {}, run time {}", worker,<a name="line.2067"></a> |
| <span class="sourceLineNo">2068</span> StringUtils.humanTimeDiff(worker.getCurrentRunTime()));<a name="line.2068"></a> |
| <span class="sourceLineNo">2069</span> }<a name="line.2069"></a> |
| <span class="sourceLineNo">2070</span> return stuckCount;<a name="line.2070"></a> |
| <span class="sourceLineNo">2071</span> }<a name="line.2071"></a> |
| <span class="sourceLineNo">2072</span><a name="line.2072"></a> |
| <span class="sourceLineNo">2073</span> private void checkThreadCount(final int stuckCount) {<a name="line.2073"></a> |
| <span class="sourceLineNo">2074</span> // nothing to do if there are no runnable tasks<a name="line.2074"></a> |
| <span class="sourceLineNo">2075</span> if (stuckCount < 1 || !scheduler.hasRunnables()) {<a name="line.2075"></a> |
| <span class="sourceLineNo">2076</span> return;<a name="line.2076"></a> |
| <span class="sourceLineNo">2077</span> }<a name="line.2077"></a> |
| <span class="sourceLineNo">2078</span><a name="line.2078"></a> |
| <span class="sourceLineNo">2079</span> // add a new thread if the worker stuck percentage exceed the threshold limit<a name="line.2079"></a> |
| <span class="sourceLineNo">2080</span> // and every handler is active.<a name="line.2080"></a> |
| <span class="sourceLineNo">2081</span> final float stuckPerc = ((float) stuckCount) / workerThreads.size();<a name="line.2081"></a> |
| <span class="sourceLineNo">2082</span> // let's add new worker thread more aggressively, as they will timeout finally if there is no<a name="line.2082"></a> |
| <span class="sourceLineNo">2083</span> // work to do.<a name="line.2083"></a> |
| <span class="sourceLineNo">2084</span> if (stuckPerc >= addWorkerStuckPercentage && workerThreads.size() < maxPoolSize) {<a name="line.2084"></a> |
| <span class="sourceLineNo">2085</span> final KeepAliveWorkerThread worker = new KeepAliveWorkerThread(threadGroup);<a name="line.2085"></a> |
| <span class="sourceLineNo">2086</span> workerThreads.add(worker);<a name="line.2086"></a> |
| <span class="sourceLineNo">2087</span> worker.start();<a name="line.2087"></a> |
| <span class="sourceLineNo">2088</span> LOG.debug("Added new worker thread {}", worker);<a name="line.2088"></a> |
| <span class="sourceLineNo">2089</span> }<a name="line.2089"></a> |
| <span class="sourceLineNo">2090</span> }<a name="line.2090"></a> |
| <span class="sourceLineNo">2091</span><a name="line.2091"></a> |
| <span class="sourceLineNo">2092</span> private void refreshConfig() {<a name="line.2092"></a> |
| <span class="sourceLineNo">2093</span> addWorkerStuckPercentage = conf.getFloat(WORKER_ADD_STUCK_PERCENTAGE_CONF_KEY,<a name="line.2093"></a> |
| <span class="sourceLineNo">2094</span> DEFAULT_WORKER_ADD_STUCK_PERCENTAGE);<a name="line.2094"></a> |
| <span class="sourceLineNo">2095</span> timeoutInterval = conf.getInt(WORKER_MONITOR_INTERVAL_CONF_KEY,<a name="line.2095"></a> |
| <span class="sourceLineNo">2096</span> DEFAULT_WORKER_MONITOR_INTERVAL);<a name="line.2096"></a> |
| <span class="sourceLineNo">2097</span> stuckThreshold = conf.getInt(WORKER_STUCK_THRESHOLD_CONF_KEY,<a name="line.2097"></a> |
| <span class="sourceLineNo">2098</span> DEFAULT_WORKER_STUCK_THRESHOLD);<a name="line.2098"></a> |
| <span class="sourceLineNo">2099</span> }<a name="line.2099"></a> |
| <span class="sourceLineNo">2100</span><a name="line.2100"></a> |
| <span class="sourceLineNo">2101</span> @Override<a name="line.2101"></a> |
| <span class="sourceLineNo">2102</span> public int getTimeoutInterval() {<a name="line.2102"></a> |
| <span class="sourceLineNo">2103</span> return timeoutInterval;<a name="line.2103"></a> |
| <span class="sourceLineNo">2104</span> }<a name="line.2104"></a> |
| <span class="sourceLineNo">2105</span> }<a name="line.2105"></a> |
| <span class="sourceLineNo">2106</span>}<a name="line.2106"></a> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |