You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hbase.apache.org by gi...@apache.org on 2018/08/15 14:48:54 UTC
[08/20] hbase-site git commit: Published site at
1114a1a65e78a00f643ef15b9a13bc710cb80845.
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/3b0a06ec/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.html b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.html
index 0a656db..e8799af 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.html
@@ -91,1861 +91,1918 @@
<span class="sourceLineNo">083</span> "hbase.procedure.worker.keep.alive.time.msec";<a name="line.83"></a>
<span class="sourceLineNo">084</span> private static final long DEFAULT_WORKER_KEEP_ALIVE_TIME = TimeUnit.MINUTES.toMillis(1);<a name="line.84"></a>
<span class="sourceLineNo">085</span><a name="line.85"></a>
-<span class="sourceLineNo">086</span> Testing testing = null;<a name="line.86"></a>
-<span class="sourceLineNo">087</span> public static class Testing {<a name="line.87"></a>
-<span class="sourceLineNo">088</span> protected boolean killIfSuspended = false;<a name="line.88"></a>
-<span class="sourceLineNo">089</span> protected boolean killBeforeStoreUpdate = false;<a name="line.89"></a>
-<span class="sourceLineNo">090</span> protected boolean toggleKillBeforeStoreUpdate = false;<a name="line.90"></a>
-<span class="sourceLineNo">091</span><a name="line.91"></a>
-<span class="sourceLineNo">092</span> protected boolean shouldKillBeforeStoreUpdate() {<a name="line.92"></a>
-<span class="sourceLineNo">093</span> final boolean kill = this.killBeforeStoreUpdate;<a name="line.93"></a>
-<span class="sourceLineNo">094</span> if (this.toggleKillBeforeStoreUpdate) {<a name="line.94"></a>
-<span class="sourceLineNo">095</span> this.killBeforeStoreUpdate = !kill;<a name="line.95"></a>
-<span class="sourceLineNo">096</span> LOG.warn("Toggle KILL before store update to: " + this.killBeforeStoreUpdate);<a name="line.96"></a>
-<span class="sourceLineNo">097</span> }<a name="line.97"></a>
-<span class="sourceLineNo">098</span> return kill;<a name="line.98"></a>
-<span class="sourceLineNo">099</span> }<a name="line.99"></a>
-<span class="sourceLineNo">100</span><a name="line.100"></a>
-<span class="sourceLineNo">101</span> protected boolean shouldKillBeforeStoreUpdate(final boolean isSuspended) {<a name="line.101"></a>
-<span class="sourceLineNo">102</span> return (isSuspended && !killIfSuspended) ? false : shouldKillBeforeStoreUpdate();<a name="line.102"></a>
-<span class="sourceLineNo">103</span> }<a name="line.103"></a>
-<span class="sourceLineNo">104</span> }<a name="line.104"></a>
+<span class="sourceLineNo">086</span> /**<a name="line.86"></a>
+<span class="sourceLineNo">087</span> * {@link #testing} is non-null when ProcedureExecutor is being tested. Tests will try to<a name="line.87"></a>
+<span class="sourceLineNo">088</span> * break PE having it fail at various junctures. When non-null, testing is set to an instance of<a name="line.88"></a>
+<span class="sourceLineNo">089</span> * the below internal {@link Testing} class with flags set for the particular test.<a name="line.89"></a>
+<span class="sourceLineNo">090</span> */<a name="line.90"></a>
+<span class="sourceLineNo">091</span> Testing testing = null;<a name="line.91"></a>
+<span class="sourceLineNo">092</span><a name="line.92"></a>
+<span class="sourceLineNo">093</span> /**<a name="line.93"></a>
+<span class="sourceLineNo">094</span> * Class with parameters describing how to fail/die when in testing-context.<a name="line.94"></a>
+<span class="sourceLineNo">095</span> */<a name="line.95"></a>
+<span class="sourceLineNo">096</span> public static class Testing {<a name="line.96"></a>
+<span class="sourceLineNo">097</span> protected boolean killIfSuspended = false;<a name="line.97"></a>
+<span class="sourceLineNo">098</span><a name="line.98"></a>
+<span class="sourceLineNo">099</span> /**<a name="line.99"></a>
+<span class="sourceLineNo">100</span> * Kill the PE BEFORE we store state to the WAL. Good for figuring out if a Procedure is<a name="line.100"></a>
+<span class="sourceLineNo">101</span> * persisting all the state it needs to recover after a crash.<a name="line.101"></a>
+<span class="sourceLineNo">102</span> */<a name="line.102"></a>
+<span class="sourceLineNo">103</span> protected boolean killBeforeStoreUpdate = false;<a name="line.103"></a>
+<span class="sourceLineNo">104</span> protected boolean toggleKillBeforeStoreUpdate = false;<a name="line.104"></a>
<span class="sourceLineNo">105</span><a name="line.105"></a>
-<span class="sourceLineNo">106</span> public interface ProcedureExecutorListener {<a name="line.106"></a>
-<span class="sourceLineNo">107</span> void procedureLoaded(long procId);<a name="line.107"></a>
-<span class="sourceLineNo">108</span> void procedureAdded(long procId);<a name="line.108"></a>
-<span class="sourceLineNo">109</span> void procedureFinished(long procId);<a name="line.109"></a>
-<span class="sourceLineNo">110</span> }<a name="line.110"></a>
-<span class="sourceLineNo">111</span><a name="line.111"></a>
-<span class="sourceLineNo">112</span> private static final class CompletedProcedureRetainer<TEnvironment> {<a name="line.112"></a>
-<span class="sourceLineNo">113</span> private final Procedure<TEnvironment> procedure;<a name="line.113"></a>
-<span class="sourceLineNo">114</span> private long clientAckTime;<a name="line.114"></a>
-<span class="sourceLineNo">115</span><a name="line.115"></a>
-<span class="sourceLineNo">116</span> public CompletedProcedureRetainer(Procedure<TEnvironment> procedure) {<a name="line.116"></a>
-<span class="sourceLineNo">117</span> this.procedure = procedure;<a name="line.117"></a>
-<span class="sourceLineNo">118</span> clientAckTime = -1;<a name="line.118"></a>
-<span class="sourceLineNo">119</span> }<a name="line.119"></a>
-<span class="sourceLineNo">120</span><a name="line.120"></a>
-<span class="sourceLineNo">121</span> public Procedure<TEnvironment> getProcedure() {<a name="line.121"></a>
-<span class="sourceLineNo">122</span> return procedure;<a name="line.122"></a>
-<span class="sourceLineNo">123</span> }<a name="line.123"></a>
-<span class="sourceLineNo">124</span><a name="line.124"></a>
-<span class="sourceLineNo">125</span> public boolean hasClientAckTime() {<a name="line.125"></a>
-<span class="sourceLineNo">126</span> return clientAckTime != -1;<a name="line.126"></a>
-<span class="sourceLineNo">127</span> }<a name="line.127"></a>
-<span class="sourceLineNo">128</span><a name="line.128"></a>
-<span class="sourceLineNo">129</span> public long getClientAckTime() {<a name="line.129"></a>
-<span class="sourceLineNo">130</span> return clientAckTime;<a name="line.130"></a>
-<span class="sourceLineNo">131</span> }<a name="line.131"></a>
-<span class="sourceLineNo">132</span><a name="line.132"></a>
-<span class="sourceLineNo">133</span> public void setClientAckTime(long clientAckTime) {<a name="line.133"></a>
-<span class="sourceLineNo">134</span> this.clientAckTime = clientAckTime;<a name="line.134"></a>
-<span class="sourceLineNo">135</span> }<a name="line.135"></a>
-<span class="sourceLineNo">136</span><a name="line.136"></a>
-<span class="sourceLineNo">137</span> public boolean isExpired(long now, long evictTtl, long evictAckTtl) {<a name="line.137"></a>
-<span class="sourceLineNo">138</span> return (hasClientAckTime() && (now - getClientAckTime()) >= evictAckTtl) ||<a name="line.138"></a>
-<span class="sourceLineNo">139</span> (now - procedure.getLastUpdate()) >= evictTtl;<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><a name="line.142"></a>
-<span class="sourceLineNo">143</span> /**<a name="line.143"></a>
-<span class="sourceLineNo">144</span> * Internal cleaner that removes the completed procedure results after a TTL.<a name="line.144"></a>
-<span class="sourceLineNo">145</span> * NOTE: This is a special case handled in timeoutLoop().<a name="line.145"></a>
-<span class="sourceLineNo">146</span> *<a name="line.146"></a>
-<span class="sourceLineNo">147</span> * <p>Since the client code looks more or less like:<a name="line.147"></a>
-<span class="sourceLineNo">148</span> * <pre><a name="line.148"></a>
-<span class="sourceLineNo">149</span> * procId = master.doOperation()<a name="line.149"></a>
-<span class="sourceLineNo">150</span> * while (master.getProcResult(procId) == ProcInProgress);<a name="line.150"></a>
-<span class="sourceLineNo">151</span> * </pre><a name="line.151"></a>
-<span class="sourceLineNo">152</span> * The master should not throw away the proc result as soon as the procedure is done<a name="line.152"></a>
-<span class="sourceLineNo">153</span> * but should wait a result request from the client (see executor.removeResult(procId))<a name="line.153"></a>
-<span class="sourceLineNo">154</span> * The client will call something like master.isProcDone() or master.getProcResult()<a name="line.154"></a>
-<span class="sourceLineNo">155</span> * which will return the result/state to the client, and it will mark the completed<a name="line.155"></a>
-<span class="sourceLineNo">156</span> * proc as ready to delete. note that the client may not receive the response from<a name="line.156"></a>
-<span class="sourceLineNo">157</span> * the master (e.g. master failover) so, if we delay a bit the real deletion of<a name="line.157"></a>
-<span class="sourceLineNo">158</span> * the proc result the client will be able to get the result the next try.<a name="line.158"></a>
-<span class="sourceLineNo">159</span> */<a name="line.159"></a>
-<span class="sourceLineNo">160</span> private static class CompletedProcedureCleaner<TEnvironment><a name="line.160"></a>
-<span class="sourceLineNo">161</span> extends ProcedureInMemoryChore<TEnvironment> {<a name="line.161"></a>
-<span class="sourceLineNo">162</span> private static final Logger LOG = LoggerFactory.getLogger(CompletedProcedureCleaner.class);<a name="line.162"></a>
+<span class="sourceLineNo">106</span> /**<a name="line.106"></a>
+<span class="sourceLineNo">107</span> * Set when we want to fail AFTER state has been stored into the WAL. Rarely used. HBASE-20978<a name="line.107"></a>
+<span class="sourceLineNo">108</span> * is about a case where memory-state was being set after store to WAL where a crash could<a name="line.108"></a>
+<span class="sourceLineNo">109</span> * cause us to get stuck. This flag allows killing at what was a vulnerable time.<a name="line.109"></a>
+<span class="sourceLineNo">110</span> */<a name="line.110"></a>
+<span class="sourceLineNo">111</span> protected boolean killAfterStoreUpdate = false;<a name="line.111"></a>
+<span class="sourceLineNo">112</span> protected boolean toggleKillAfterStoreUpdate = false;<a name="line.112"></a>
+<span class="sourceLineNo">113</span><a name="line.113"></a>
+<span class="sourceLineNo">114</span> protected boolean shouldKillBeforeStoreUpdate() {<a name="line.114"></a>
+<span class="sourceLineNo">115</span> final boolean kill = this.killBeforeStoreUpdate;<a name="line.115"></a>
+<span class="sourceLineNo">116</span> if (this.toggleKillBeforeStoreUpdate) {<a name="line.116"></a>
+<span class="sourceLineNo">117</span> this.killBeforeStoreUpdate = !kill;<a name="line.117"></a>
+<span class="sourceLineNo">118</span> LOG.warn("Toggle KILL before store update to: " + this.killBeforeStoreUpdate);<a name="line.118"></a>
+<span class="sourceLineNo">119</span> }<a name="line.119"></a>
+<span class="sourceLineNo">120</span> return kill;<a name="line.120"></a>
+<span class="sourceLineNo">121</span> }<a name="line.121"></a>
+<span class="sourceLineNo">122</span><a name="line.122"></a>
+<span class="sourceLineNo">123</span> protected boolean shouldKillBeforeStoreUpdate(final boolean isSuspended) {<a name="line.123"></a>
+<span class="sourceLineNo">124</span> return (isSuspended && !killIfSuspended) ? false : shouldKillBeforeStoreUpdate();<a name="line.124"></a>
+<span class="sourceLineNo">125</span> }<a name="line.125"></a>
+<span class="sourceLineNo">126</span><a name="line.126"></a>
+<span class="sourceLineNo">127</span> protected boolean shouldKillAfterStoreUpdate() {<a name="line.127"></a>
+<span class="sourceLineNo">128</span> final boolean kill = this.killAfterStoreUpdate;<a name="line.128"></a>
+<span class="sourceLineNo">129</span> if (this.toggleKillAfterStoreUpdate) {<a name="line.129"></a>
+<span class="sourceLineNo">130</span> this.killAfterStoreUpdate = !kill;<a name="line.130"></a>
+<span class="sourceLineNo">131</span> LOG.warn("Toggle KILL after store update to: " + this.killAfterStoreUpdate);<a name="line.131"></a>
+<span class="sourceLineNo">132</span> }<a name="line.132"></a>
+<span class="sourceLineNo">133</span> return kill;<a name="line.133"></a>
+<span class="sourceLineNo">134</span> }<a name="line.134"></a>
+<span class="sourceLineNo">135</span><a name="line.135"></a>
+<span class="sourceLineNo">136</span> protected boolean shouldKillAfterStoreUpdate(final boolean isSuspended) {<a name="line.136"></a>
+<span class="sourceLineNo">137</span> return (isSuspended && !killIfSuspended) ? false : shouldKillAfterStoreUpdate();<a name="line.137"></a>
+<span class="sourceLineNo">138</span> }<a name="line.138"></a>
+<span class="sourceLineNo">139</span> }<a name="line.139"></a>
+<span class="sourceLineNo">140</span><a name="line.140"></a>
+<span class="sourceLineNo">141</span> public interface ProcedureExecutorListener {<a name="line.141"></a>
+<span class="sourceLineNo">142</span> void procedureLoaded(long procId);<a name="line.142"></a>
+<span class="sourceLineNo">143</span> void procedureAdded(long procId);<a name="line.143"></a>
+<span class="sourceLineNo">144</span> void procedureFinished(long procId);<a name="line.144"></a>
+<span class="sourceLineNo">145</span> }<a name="line.145"></a>
+<span class="sourceLineNo">146</span><a name="line.146"></a>
+<span class="sourceLineNo">147</span> private static final class CompletedProcedureRetainer<TEnvironment> {<a name="line.147"></a>
+<span class="sourceLineNo">148</span> private final Procedure<TEnvironment> procedure;<a name="line.148"></a>
+<span class="sourceLineNo">149</span> private long clientAckTime;<a name="line.149"></a>
+<span class="sourceLineNo">150</span><a name="line.150"></a>
+<span class="sourceLineNo">151</span> public CompletedProcedureRetainer(Procedure<TEnvironment> procedure) {<a name="line.151"></a>
+<span class="sourceLineNo">152</span> this.procedure = procedure;<a name="line.152"></a>
+<span class="sourceLineNo">153</span> clientAckTime = -1;<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 Procedure<TEnvironment> getProcedure() {<a name="line.156"></a>
+<span class="sourceLineNo">157</span> return procedure;<a name="line.157"></a>
+<span class="sourceLineNo">158</span> }<a name="line.158"></a>
+<span class="sourceLineNo">159</span><a name="line.159"></a>
+<span class="sourceLineNo">160</span> public boolean hasClientAckTime() {<a name="line.160"></a>
+<span class="sourceLineNo">161</span> return clientAckTime != -1;<a name="line.161"></a>
+<span class="sourceLineNo">162</span> }<a name="line.162"></a>
<span class="sourceLineNo">163</span><a name="line.163"></a>
-<span class="sourceLineNo">164</span> private static final String CLEANER_INTERVAL_CONF_KEY = "hbase.procedure.cleaner.interval";<a name="line.164"></a>
-<span class="sourceLineNo">165</span> private static final int DEFAULT_CLEANER_INTERVAL = 30 * 1000; // 30sec<a name="line.165"></a>
-<span class="sourceLineNo">166</span><a name="line.166"></a>
-<span class="sourceLineNo">167</span> private static final String EVICT_TTL_CONF_KEY = "hbase.procedure.cleaner.evict.ttl";<a name="line.167"></a>
-<span class="sourceLineNo">168</span> private static final int DEFAULT_EVICT_TTL = 15 * 60000; // 15min<a name="line.168"></a>
-<span class="sourceLineNo">169</span><a name="line.169"></a>
-<span class="sourceLineNo">170</span> private static final String EVICT_ACKED_TTL_CONF_KEY ="hbase.procedure.cleaner.acked.evict.ttl";<a name="line.170"></a>
-<span class="sourceLineNo">171</span> private static final int DEFAULT_ACKED_EVICT_TTL = 5 * 60000; // 5min<a name="line.171"></a>
-<span class="sourceLineNo">172</span><a name="line.172"></a>
-<span class="sourceLineNo">173</span> private static final String BATCH_SIZE_CONF_KEY = "hbase.procedure.cleaner.evict.batch.size";<a name="line.173"></a>
-<span class="sourceLineNo">174</span> private static final int DEFAULT_BATCH_SIZE = 32;<a name="line.174"></a>
-<span class="sourceLineNo">175</span><a name="line.175"></a>
-<span class="sourceLineNo">176</span> private final Map<Long, CompletedProcedureRetainer<TEnvironment>> completed;<a name="line.176"></a>
-<span class="sourceLineNo">177</span> private final Map<NonceKey, Long> nonceKeysToProcIdsMap;<a name="line.177"></a>
-<span class="sourceLineNo">178</span> private final ProcedureStore store;<a name="line.178"></a>
-<span class="sourceLineNo">179</span> private Configuration conf;<a name="line.179"></a>
-<span class="sourceLineNo">180</span><a name="line.180"></a>
-<span class="sourceLineNo">181</span> public CompletedProcedureCleaner(Configuration conf, final ProcedureStore store,<a name="line.181"></a>
-<span class="sourceLineNo">182</span> final Map<Long, CompletedProcedureRetainer<TEnvironment>> completedMap,<a name="line.182"></a>
-<span class="sourceLineNo">183</span> final Map<NonceKey, Long> nonceKeysToProcIdsMap) {<a name="line.183"></a>
-<span class="sourceLineNo">184</span> // set the timeout interval that triggers the periodic-procedure<a name="line.184"></a>
-<span class="sourceLineNo">185</span> super(conf.getInt(CLEANER_INTERVAL_CONF_KEY, DEFAULT_CLEANER_INTERVAL));<a name="line.185"></a>
-<span class="sourceLineNo">186</span> this.completed = completedMap;<a name="line.186"></a>
-<span class="sourceLineNo">187</span> this.nonceKeysToProcIdsMap = nonceKeysToProcIdsMap;<a name="line.187"></a>
-<span class="sourceLineNo">188</span> this.store = store;<a name="line.188"></a>
-<span class="sourceLineNo">189</span> this.conf = conf;<a name="line.189"></a>
-<span class="sourceLineNo">190</span> }<a name="line.190"></a>
-<span class="sourceLineNo">191</span><a name="line.191"></a>
-<span class="sourceLineNo">192</span> @Override<a name="line.192"></a>
-<span class="sourceLineNo">193</span> protected void periodicExecute(final TEnvironment env) {<a name="line.193"></a>
-<span class="sourceLineNo">194</span> if (completed.isEmpty()) {<a name="line.194"></a>
-<span class="sourceLineNo">195</span> if (LOG.isTraceEnabled()) {<a name="line.195"></a>
-<span class="sourceLineNo">196</span> LOG.trace("No completed procedures to cleanup.");<a name="line.196"></a>
-<span class="sourceLineNo">197</span> }<a name="line.197"></a>
-<span class="sourceLineNo">198</span> return;<a name="line.198"></a>
-<span class="sourceLineNo">199</span> }<a name="line.199"></a>
-<span class="sourceLineNo">200</span><a name="line.200"></a>
-<span class="sourceLineNo">201</span> final long evictTtl = conf.getInt(EVICT_TTL_CONF_KEY, DEFAULT_EVICT_TTL);<a name="line.201"></a>
-<span class="sourceLineNo">202</span> final long evictAckTtl = conf.getInt(EVICT_ACKED_TTL_CONF_KEY, DEFAULT_ACKED_EVICT_TTL);<a name="line.202"></a>
-<span class="sourceLineNo">203</span> final int batchSize = conf.getInt(BATCH_SIZE_CONF_KEY, DEFAULT_BATCH_SIZE);<a name="line.203"></a>
+<span class="sourceLineNo">164</span> public long getClientAckTime() {<a name="line.164"></a>
+<span class="sourceLineNo">165</span> return clientAckTime;<a name="line.165"></a>
+<span class="sourceLineNo">166</span> }<a name="line.166"></a>
+<span class="sourceLineNo">167</span><a name="line.167"></a>
+<span class="sourceLineNo">168</span> public void setClientAckTime(long clientAckTime) {<a name="line.168"></a>
+<span class="sourceLineNo">169</span> this.clientAckTime = clientAckTime;<a name="line.169"></a>
+<span class="sourceLineNo">170</span> }<a name="line.170"></a>
+<span class="sourceLineNo">171</span><a name="line.171"></a>
+<span class="sourceLineNo">172</span> public boolean isExpired(long now, long evictTtl, long evictAckTtl) {<a name="line.172"></a>
+<span class="sourceLineNo">173</span> return (hasClientAckTime() && (now - getClientAckTime()) >= evictAckTtl) ||<a name="line.173"></a>
+<span class="sourceLineNo">174</span> (now - procedure.getLastUpdate()) >= evictTtl;<a name="line.174"></a>
+<span class="sourceLineNo">175</span> }<a name="line.175"></a>
+<span class="sourceLineNo">176</span> }<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> * Internal cleaner that removes the completed procedure results after a TTL.<a name="line.179"></a>
+<span class="sourceLineNo">180</span> * NOTE: This is a special case handled in timeoutLoop().<a name="line.180"></a>
+<span class="sourceLineNo">181</span> *<a name="line.181"></a>
+<span class="sourceLineNo">182</span> * <p>Since the client code looks more or less like:<a name="line.182"></a>
+<span class="sourceLineNo">183</span> * <pre><a name="line.183"></a>
+<span class="sourceLineNo">184</span> * procId = master.doOperation()<a name="line.184"></a>
+<span class="sourceLineNo">185</span> * while (master.getProcResult(procId) == ProcInProgress);<a name="line.185"></a>
+<span class="sourceLineNo">186</span> * </pre><a name="line.186"></a>
+<span class="sourceLineNo">187</span> * The master should not throw away the proc result as soon as the procedure is done<a name="line.187"></a>
+<span class="sourceLineNo">188</span> * but should wait a result request from the client (see executor.removeResult(procId))<a name="line.188"></a>
+<span class="sourceLineNo">189</span> * The client will call something like master.isProcDone() or master.getProcResult()<a name="line.189"></a>
+<span class="sourceLineNo">190</span> * which will return the result/state to the client, and it will mark the completed<a name="line.190"></a>
+<span class="sourceLineNo">191</span> * proc as ready to delete. note that the client may not receive the response from<a name="line.191"></a>
+<span class="sourceLineNo">192</span> * the master (e.g. master failover) so, if we delay a bit the real deletion of<a name="line.192"></a>
+<span class="sourceLineNo">193</span> * the proc result the client will be able to get the result the next try.<a name="line.193"></a>
+<span class="sourceLineNo">194</span> */<a name="line.194"></a>
+<span class="sourceLineNo">195</span> private static class CompletedProcedureCleaner<TEnvironment><a name="line.195"></a>
+<span class="sourceLineNo">196</span> extends ProcedureInMemoryChore<TEnvironment> {<a name="line.196"></a>
+<span class="sourceLineNo">197</span> private static final Logger LOG = LoggerFactory.getLogger(CompletedProcedureCleaner.class);<a name="line.197"></a>
+<span class="sourceLineNo">198</span><a name="line.198"></a>
+<span class="sourceLineNo">199</span> private static final String CLEANER_INTERVAL_CONF_KEY = "hbase.procedure.cleaner.interval";<a name="line.199"></a>
+<span class="sourceLineNo">200</span> private static final int DEFAULT_CLEANER_INTERVAL = 30 * 1000; // 30sec<a name="line.200"></a>
+<span class="sourceLineNo">201</span><a name="line.201"></a>
+<span class="sourceLineNo">202</span> private static final String EVICT_TTL_CONF_KEY = "hbase.procedure.cleaner.evict.ttl";<a name="line.202"></a>
+<span class="sourceLineNo">203</span> private static final int DEFAULT_EVICT_TTL = 15 * 60000; // 15min<a name="line.203"></a>
<span class="sourceLineNo">204</span><a name="line.204"></a>
-<span class="sourceLineNo">205</span> final long[] batchIds = new long[batchSize];<a name="line.205"></a>
-<span class="sourceLineNo">206</span> int batchCount = 0;<a name="line.206"></a>
+<span class="sourceLineNo">205</span> private static final String EVICT_ACKED_TTL_CONF_KEY ="hbase.procedure.cleaner.acked.evict.ttl";<a name="line.205"></a>
+<span class="sourceLineNo">206</span> private static final int DEFAULT_ACKED_EVICT_TTL = 5 * 60000; // 5min<a name="line.206"></a>
<span class="sourceLineNo">207</span><a name="line.207"></a>
-<span class="sourceLineNo">208</span> final long now = EnvironmentEdgeManager.currentTime();<a name="line.208"></a>
-<span class="sourceLineNo">209</span> final Iterator<Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>>> it =<a name="line.209"></a>
-<span class="sourceLineNo">210</span> completed.entrySet().iterator();<a name="line.210"></a>
-<span class="sourceLineNo">211</span> while (it.hasNext() && store.isRunning()) {<a name="line.211"></a>
-<span class="sourceLineNo">212</span> final Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>> entry = it.next();<a name="line.212"></a>
-<span class="sourceLineNo">213</span> final CompletedProcedureRetainer<TEnvironment> retainer = entry.getValue();<a name="line.213"></a>
-<span class="sourceLineNo">214</span> final Procedure<?> proc = retainer.getProcedure();<a name="line.214"></a>
+<span class="sourceLineNo">208</span> private static final String BATCH_SIZE_CONF_KEY = "hbase.procedure.cleaner.evict.batch.size";<a name="line.208"></a>
+<span class="sourceLineNo">209</span> private static final int DEFAULT_BATCH_SIZE = 32;<a name="line.209"></a>
+<span class="sourceLineNo">210</span><a name="line.210"></a>
+<span class="sourceLineNo">211</span> private final Map<Long, CompletedProcedureRetainer<TEnvironment>> completed;<a name="line.211"></a>
+<span class="sourceLineNo">212</span> private final Map<NonceKey, Long> nonceKeysToProcIdsMap;<a name="line.212"></a>
+<span class="sourceLineNo">213</span> private final ProcedureStore store;<a name="line.213"></a>
+<span class="sourceLineNo">214</span> private Configuration conf;<a name="line.214"></a>
<span class="sourceLineNo">215</span><a name="line.215"></a>
-<span class="sourceLineNo">216</span> // TODO: Select TTL based on Procedure type<a name="line.216"></a>
-<span class="sourceLineNo">217</span> if (retainer.isExpired(now, evictTtl, evictAckTtl)) {<a name="line.217"></a>
-<span class="sourceLineNo">218</span> // Failed procedures aren't persisted in WAL.<a name="line.218"></a>
-<span class="sourceLineNo">219</span> if (!(proc instanceof FailedProcedure)) {<a name="line.219"></a>
-<span class="sourceLineNo">220</span> batchIds[batchCount++] = entry.getKey();<a name="line.220"></a>
-<span class="sourceLineNo">221</span> if (batchCount == batchIds.length) {<a name="line.221"></a>
-<span class="sourceLineNo">222</span> store.delete(batchIds, 0, batchCount);<a name="line.222"></a>
-<span class="sourceLineNo">223</span> batchCount = 0;<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> final NonceKey nonceKey = proc.getNonceKey();<a name="line.226"></a>
-<span class="sourceLineNo">227</span> if (nonceKey != null) {<a name="line.227"></a>
-<span class="sourceLineNo">228</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.228"></a>
-<span class="sourceLineNo">229</span> }<a name="line.229"></a>
-<span class="sourceLineNo">230</span> it.remove();<a name="line.230"></a>
-<span class="sourceLineNo">231</span> LOG.trace("Evict completed {}", proc);<a name="line.231"></a>
+<span class="sourceLineNo">216</span> public CompletedProcedureCleaner(Configuration conf, final ProcedureStore store,<a name="line.216"></a>
+<span class="sourceLineNo">217</span> final Map<Long, CompletedProcedureRetainer<TEnvironment>> completedMap,<a name="line.217"></a>
+<span class="sourceLineNo">218</span> final Map<NonceKey, Long> nonceKeysToProcIdsMap) {<a name="line.218"></a>
+<span class="sourceLineNo">219</span> // set the timeout interval that triggers the periodic-procedure<a name="line.219"></a>
+<span class="sourceLineNo">220</span> super(conf.getInt(CLEANER_INTERVAL_CONF_KEY, DEFAULT_CLEANER_INTERVAL));<a name="line.220"></a>
+<span class="sourceLineNo">221</span> this.completed = completedMap;<a name="line.221"></a>
+<span class="sourceLineNo">222</span> this.nonceKeysToProcIdsMap = nonceKeysToProcIdsMap;<a name="line.222"></a>
+<span class="sourceLineNo">223</span> this.store = store;<a name="line.223"></a>
+<span class="sourceLineNo">224</span> this.conf = conf;<a name="line.224"></a>
+<span class="sourceLineNo">225</span> }<a name="line.225"></a>
+<span class="sourceLineNo">226</span><a name="line.226"></a>
+<span class="sourceLineNo">227</span> @Override<a name="line.227"></a>
+<span class="sourceLineNo">228</span> protected void periodicExecute(final TEnvironment env) {<a name="line.228"></a>
+<span class="sourceLineNo">229</span> if (completed.isEmpty()) {<a name="line.229"></a>
+<span class="sourceLineNo">230</span> if (LOG.isTraceEnabled()) {<a name="line.230"></a>
+<span class="sourceLineNo">231</span> LOG.trace("No completed procedures to cleanup.");<a name="line.231"></a>
<span class="sourceLineNo">232</span> }<a name="line.232"></a>
-<span class="sourceLineNo">233</span> }<a name="line.233"></a>
-<span class="sourceLineNo">234</span> if (batchCount > 0) {<a name="line.234"></a>
-<span class="sourceLineNo">235</span> store.delete(batchIds, 0, batchCount);<a name="line.235"></a>
-<span class="sourceLineNo">236</span> }<a name="line.236"></a>
-<span class="sourceLineNo">237</span> }<a name="line.237"></a>
-<span class="sourceLineNo">238</span> }<a name="line.238"></a>
+<span class="sourceLineNo">233</span> return;<a name="line.233"></a>
+<span class="sourceLineNo">234</span> }<a name="line.234"></a>
+<span class="sourceLineNo">235</span><a name="line.235"></a>
+<span class="sourceLineNo">236</span> final long evictTtl = conf.getInt(EVICT_TTL_CONF_KEY, DEFAULT_EVICT_TTL);<a name="line.236"></a>
+<span class="sourceLineNo">237</span> final long evictAckTtl = conf.getInt(EVICT_ACKED_TTL_CONF_KEY, DEFAULT_ACKED_EVICT_TTL);<a name="line.237"></a>
+<span class="sourceLineNo">238</span> final int batchSize = conf.getInt(BATCH_SIZE_CONF_KEY, DEFAULT_BATCH_SIZE);<a name="line.238"></a>
<span class="sourceLineNo">239</span><a name="line.239"></a>
-<span class="sourceLineNo">240</span> /**<a name="line.240"></a>
-<span class="sourceLineNo">241</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the Procedure.<a name="line.241"></a>
-<span class="sourceLineNo">242</span> * Once a Root-Procedure completes (success or failure), the result will be added to this map.<a name="line.242"></a>
-<span class="sourceLineNo">243</span> * The user of ProcedureExecutor should call getResult(procId) to get the result.<a name="line.243"></a>
-<span class="sourceLineNo">244</span> */<a name="line.244"></a>
-<span class="sourceLineNo">245</span> private final ConcurrentHashMap<Long, CompletedProcedureRetainer<TEnvironment>> completed =<a name="line.245"></a>
-<span class="sourceLineNo">246</span> new ConcurrentHashMap<>();<a name="line.246"></a>
-<span class="sourceLineNo">247</span><a name="line.247"></a>
-<span class="sourceLineNo">248</span> /**<a name="line.248"></a>
-<span class="sourceLineNo">249</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the RootProcedureState.<a name="line.249"></a>
-<span class="sourceLineNo">250</span> * The RootProcedureState contains the execution stack of the Root-Procedure,<a name="line.250"></a>
-<span class="sourceLineNo">251</span> * It is added to the map by submitProcedure() and removed on procedure completion.<a name="line.251"></a>
-<span class="sourceLineNo">252</span> */<a name="line.252"></a>
-<span class="sourceLineNo">253</span> private final ConcurrentHashMap<Long, RootProcedureState<TEnvironment>> rollbackStack =<a name="line.253"></a>
-<span class="sourceLineNo">254</span> new ConcurrentHashMap<>();<a name="line.254"></a>
-<span class="sourceLineNo">255</span><a name="line.255"></a>
-<span class="sourceLineNo">256</span> /**<a name="line.256"></a>
-<span class="sourceLineNo">257</span> * Helper map to lookup the live procedures by ID.<a name="line.257"></a>
-<span class="sourceLineNo">258</span> * This map contains every procedure. root-procedures and subprocedures.<a name="line.258"></a>
-<span class="sourceLineNo">259</span> */<a name="line.259"></a>
-<span class="sourceLineNo">260</span> private final ConcurrentHashMap<Long, Procedure<TEnvironment>> procedures =<a name="line.260"></a>
-<span class="sourceLineNo">261</span> new ConcurrentHashMap<>();<a name="line.261"></a>
-<span class="sourceLineNo">262</span><a name="line.262"></a>
-<span class="sourceLineNo">263</span> /**<a name="line.263"></a>
-<span class="sourceLineNo">264</span> * Helper map to lookup whether the procedure already issued from the same client. This map<a name="line.264"></a>
-<span class="sourceLineNo">265</span> * contains every root procedure.<a name="line.265"></a>
-<span class="sourceLineNo">266</span> */<a name="line.266"></a>
-<span class="sourceLineNo">267</span> private final ConcurrentHashMap<NonceKey, Long> nonceKeysToProcIdsMap = new ConcurrentHashMap<>();<a name="line.267"></a>
-<span class="sourceLineNo">268</span><a name="line.268"></a>
-<span class="sourceLineNo">269</span> private final CopyOnWriteArrayList<ProcedureExecutorListener> listeners =<a name="line.269"></a>
-<span class="sourceLineNo">270</span> new CopyOnWriteArrayList<>();<a name="line.270"></a>
-<span class="sourceLineNo">271</span><a name="line.271"></a>
-<span class="sourceLineNo">272</span> private Configuration conf;<a name="line.272"></a>
-<span class="sourceLineNo">273</span><a name="line.273"></a>
-<span class="sourceLineNo">274</span> /**<a name="line.274"></a>
-<span class="sourceLineNo">275</span> * Created in the {@link #start(int, boolean)} method. Destroyed in {@link #join()} (FIX! Doing<a name="line.275"></a>
-<span class="sourceLineNo">276</span> * resource handling rather than observing in a #join is unexpected).<a name="line.276"></a>
-<span class="sourceLineNo">277</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.277"></a>
-<span class="sourceLineNo">278</span> * (Should be ok).<a name="line.278"></a>
+<span class="sourceLineNo">240</span> final long[] batchIds = new long[batchSize];<a name="line.240"></a>
+<span class="sourceLineNo">241</span> int batchCount = 0;<a name="line.241"></a>
+<span class="sourceLineNo">242</span><a name="line.242"></a>
+<span class="sourceLineNo">243</span> final long now = EnvironmentEdgeManager.currentTime();<a name="line.243"></a>
+<span class="sourceLineNo">244</span> final Iterator<Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>>> it =<a name="line.244"></a>
+<span class="sourceLineNo">245</span> completed.entrySet().iterator();<a name="line.245"></a>
+<span class="sourceLineNo">246</span> while (it.hasNext() && store.isRunning()) {<a name="line.246"></a>
+<span class="sourceLineNo">247</span> final Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>> entry = it.next();<a name="line.247"></a>
+<span class="sourceLineNo">248</span> final CompletedProcedureRetainer<TEnvironment> retainer = entry.getValue();<a name="line.248"></a>
+<span class="sourceLineNo">249</span> final Procedure<?> proc = retainer.getProcedure();<a name="line.249"></a>
+<span class="sourceLineNo">250</span><a name="line.250"></a>
+<span class="sourceLineNo">251</span> // TODO: Select TTL based on Procedure type<a name="line.251"></a>
+<span class="sourceLineNo">252</span> if (retainer.isExpired(now, evictTtl, evictAckTtl)) {<a name="line.252"></a>
+<span class="sourceLineNo">253</span> // Failed procedures aren't persisted in WAL.<a name="line.253"></a>
+<span class="sourceLineNo">254</span> if (!(proc instanceof FailedProcedure)) {<a name="line.254"></a>
+<span class="sourceLineNo">255</span> batchIds[batchCount++] = entry.getKey();<a name="line.255"></a>
+<span class="sourceLineNo">256</span> if (batchCount == batchIds.length) {<a name="line.256"></a>
+<span class="sourceLineNo">257</span> store.delete(batchIds, 0, batchCount);<a name="line.257"></a>
+<span class="sourceLineNo">258</span> batchCount = 0;<a name="line.258"></a>
+<span class="sourceLineNo">259</span> }<a name="line.259"></a>
+<span class="sourceLineNo">260</span> }<a name="line.260"></a>
+<span class="sourceLineNo">261</span> final NonceKey nonceKey = proc.getNonceKey();<a name="line.261"></a>
+<span class="sourceLineNo">262</span> if (nonceKey != null) {<a name="line.262"></a>
+<span class="sourceLineNo">263</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.263"></a>
+<span class="sourceLineNo">264</span> }<a name="line.264"></a>
+<span class="sourceLineNo">265</span> it.remove();<a name="line.265"></a>
+<span class="sourceLineNo">266</span> LOG.trace("Evict completed {}", proc);<a name="line.266"></a>
+<span class="sourceLineNo">267</span> }<a name="line.267"></a>
+<span class="sourceLineNo">268</span> }<a name="line.268"></a>
+<span class="sourceLineNo">269</span> if (batchCount > 0) {<a name="line.269"></a>
+<span class="sourceLineNo">270</span> store.delete(batchIds, 0, batchCount);<a name="line.270"></a>
+<span class="sourceLineNo">271</span> }<a name="line.271"></a>
+<span class="sourceLineNo">272</span> }<a name="line.272"></a>
+<span class="sourceLineNo">273</span> }<a name="line.273"></a>
+<span class="sourceLineNo">274</span><a name="line.274"></a>
+<span class="sourceLineNo">275</span> /**<a name="line.275"></a>
+<span class="sourceLineNo">276</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the Procedure.<a name="line.276"></a>
+<span class="sourceLineNo">277</span> * Once a Root-Procedure completes (success or failure), the result will be added to this map.<a name="line.277"></a>
+<span class="sourceLineNo">278</span> * The user of ProcedureExecutor should call getResult(procId) to get the result.<a name="line.278"></a>
<span class="sourceLineNo">279</span> */<a name="line.279"></a>
-<span class="sourceLineNo">280</span> private ThreadGroup threadGroup;<a name="line.280"></a>
-<span class="sourceLineNo">281</span><a name="line.281"></a>
-<span class="sourceLineNo">282</span> /**<a name="line.282"></a>
-<span class="sourceLineNo">283</span> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.283"></a>
-<span class="sourceLineNo">284</span> * resource handling rather than observing in a #join is unexpected).<a name="line.284"></a>
-<span class="sourceLineNo">285</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.285"></a>
-<span class="sourceLineNo">286</span> * (Should be ok).<a name="line.286"></a>
+<span class="sourceLineNo">280</span> private final ConcurrentHashMap<Long, CompletedProcedureRetainer<TEnvironment>> completed =<a name="line.280"></a>
+<span class="sourceLineNo">281</span> new ConcurrentHashMap<>();<a name="line.281"></a>
+<span class="sourceLineNo">282</span><a name="line.282"></a>
+<span class="sourceLineNo">283</span> /**<a name="line.283"></a>
+<span class="sourceLineNo">284</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the RootProcedureState.<a name="line.284"></a>
+<span class="sourceLineNo">285</span> * The RootProcedureState contains the execution stack of the Root-Procedure,<a name="line.285"></a>
+<span class="sourceLineNo">286</span> * It is added to the map by submitProcedure() and removed on procedure completion.<a name="line.286"></a>
<span class="sourceLineNo">287</span> */<a name="line.287"></a>
-<span class="sourceLineNo">288</span> private CopyOnWriteArrayList<WorkerThread> workerThreads;<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> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.291"></a>
-<span class="sourceLineNo">292</span> * resource handling rather than observing in a #join is unexpected).<a name="line.292"></a>
-<span class="sourceLineNo">293</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.293"></a>
-<span class="sourceLineNo">294</span> * (Should be ok).<a name="line.294"></a>
-<span class="sourceLineNo">295</span> */<a name="line.295"></a>
-<span class="sourceLineNo">296</span> private TimeoutExecutorThread<TEnvironment> timeoutExecutor;<a name="line.296"></a>
+<span class="sourceLineNo">288</span> private final ConcurrentHashMap<Long, RootProcedureState<TEnvironment>> rollbackStack =<a name="line.288"></a>
+<span class="sourceLineNo">289</span> new ConcurrentHashMap<>();<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> * Helper map to lookup the live procedures by ID.<a name="line.292"></a>
+<span class="sourceLineNo">293</span> * This map contains every procedure. root-procedures and subprocedures.<a name="line.293"></a>
+<span class="sourceLineNo">294</span> */<a name="line.294"></a>
+<span class="sourceLineNo">295</span> private final ConcurrentHashMap<Long, Procedure<TEnvironment>> procedures =<a name="line.295"></a>
+<span class="sourceLineNo">296</span> new ConcurrentHashMap<>();<a name="line.296"></a>
<span class="sourceLineNo">297</span><a name="line.297"></a>
-<span class="sourceLineNo">298</span> private int corePoolSize;<a name="line.298"></a>
-<span class="sourceLineNo">299</span> private int maxPoolSize;<a name="line.299"></a>
-<span class="sourceLineNo">300</span><a name="line.300"></a>
-<span class="sourceLineNo">301</span> private volatile long keepAliveTime;<a name="line.301"></a>
-<span class="sourceLineNo">302</span><a name="line.302"></a>
-<span class="sourceLineNo">303</span> /**<a name="line.303"></a>
-<span class="sourceLineNo">304</span> * Scheduler/Queue that contains runnable procedures.<a name="line.304"></a>
-<span class="sourceLineNo">305</span> */<a name="line.305"></a>
-<span class="sourceLineNo">306</span> private final ProcedureScheduler scheduler;<a name="line.306"></a>
-<span class="sourceLineNo">307</span><a name="line.307"></a>
-<span class="sourceLineNo">308</span> private final AtomicLong lastProcId = new AtomicLong(-1);<a name="line.308"></a>
-<span class="sourceLineNo">309</span> private final AtomicLong workerId = new AtomicLong(0);<a name="line.309"></a>
-<span class="sourceLineNo">310</span> private final AtomicInteger activeExecutorCount = new AtomicInteger(0);<a name="line.310"></a>
-<span class="sourceLineNo">311</span> private final AtomicBoolean running = new AtomicBoolean(false);<a name="line.311"></a>
-<span class="sourceLineNo">312</span> private final TEnvironment environment;<a name="line.312"></a>
-<span class="sourceLineNo">313</span> private final ProcedureStore store;<a name="line.313"></a>
-<span class="sourceLineNo">314</span><a name="line.314"></a>
-<span class="sourceLineNo">315</span> private final boolean checkOwnerSet;<a name="line.315"></a>
+<span class="sourceLineNo">298</span> /**<a name="line.298"></a>
+<span class="sourceLineNo">299</span> * Helper map to lookup whether the procedure already issued from the same client. This map<a name="line.299"></a>
+<span class="sourceLineNo">300</span> * contains every root procedure.<a name="line.300"></a>
+<span class="sourceLineNo">301</span> */<a name="line.301"></a>
+<span class="sourceLineNo">302</span> private final ConcurrentHashMap<NonceKey, Long> nonceKeysToProcIdsMap = new ConcurrentHashMap<>();<a name="line.302"></a>
+<span class="sourceLineNo">303</span><a name="line.303"></a>
+<span class="sourceLineNo">304</span> private final CopyOnWriteArrayList<ProcedureExecutorListener> listeners =<a name="line.304"></a>
+<span class="sourceLineNo">305</span> new CopyOnWriteArrayList<>();<a name="line.305"></a>
+<span class="sourceLineNo">306</span><a name="line.306"></a>
+<span class="sourceLineNo">307</span> private Configuration conf;<a name="line.307"></a>
+<span class="sourceLineNo">308</span><a name="line.308"></a>
+<span class="sourceLineNo">309</span> /**<a name="line.309"></a>
+<span class="sourceLineNo">310</span> * Created in the {@link #start(int, boolean)} method. Destroyed in {@link #join()} (FIX! Doing<a name="line.310"></a>
+<span class="sourceLineNo">311</span> * resource handling rather than observing in a #join is unexpected).<a name="line.311"></a>
+<span class="sourceLineNo">312</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.312"></a>
+<span class="sourceLineNo">313</span> * (Should be ok).<a name="line.313"></a>
+<span class="sourceLineNo">314</span> */<a name="line.314"></a>
+<span class="sourceLineNo">315</span> private ThreadGroup threadGroup;<a name="line.315"></a>
<span class="sourceLineNo">316</span><a name="line.316"></a>
-<span class="sourceLineNo">317</span> // To prevent concurrent execution of the same procedure.<a name="line.317"></a>
-<span class="sourceLineNo">318</span> // For some rare cases, especially if the procedure uses ProcedureEvent, it is possible that the<a name="line.318"></a>
-<span class="sourceLineNo">319</span> // procedure is woken up before we finish the suspend which causes the same procedures to be<a name="line.319"></a>
-<span class="sourceLineNo">320</span> // executed in parallel. This does lead to some problems, see HBASE-20939&HBASE-20949, and is also<a name="line.320"></a>
-<span class="sourceLineNo">321</span> // a bit confusing to the developers. So here we introduce this lock to prevent the concurrent<a name="line.321"></a>
-<span class="sourceLineNo">322</span> // execution of the same procedure.<a name="line.322"></a>
-<span class="sourceLineNo">323</span> private final IdLock procExecutionLock = new IdLock();<a name="line.323"></a>
+<span class="sourceLineNo">317</span> /**<a name="line.317"></a>
+<span class="sourceLineNo">318</span> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.318"></a>
+<span class="sourceLineNo">319</span> * resource handling rather than observing in a #join is unexpected).<a name="line.319"></a>
+<span class="sourceLineNo">320</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.320"></a>
+<span class="sourceLineNo">321</span> * (Should be ok).<a name="line.321"></a>
+<span class="sourceLineNo">322</span> */<a name="line.322"></a>
+<span class="sourceLineNo">323</span> private CopyOnWriteArrayList<WorkerThread> workerThreads;<a name="line.323"></a>
<span class="sourceLineNo">324</span><a name="line.324"></a>
-<span class="sourceLineNo">325</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.325"></a>
-<span class="sourceLineNo">326</span> final ProcedureStore store) {<a name="line.326"></a>
-<span class="sourceLineNo">327</span> this(conf, environment, store, new SimpleProcedureScheduler());<a name="line.327"></a>
-<span class="sourceLineNo">328</span> }<a name="line.328"></a>
-<span class="sourceLineNo">329</span><a name="line.329"></a>
-<span class="sourceLineNo">330</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.330"></a>
-<span class="sourceLineNo">331</span> final ProcedureStore store, final ProcedureScheduler scheduler) {<a name="line.331"></a>
-<span class="sourceLineNo">332</span> this.environment = environment;<a name="line.332"></a>
-<span class="sourceLineNo">333</span> this.scheduler = scheduler;<a name="line.333"></a>
-<span class="sourceLineNo">334</span> this.store = store;<a name="line.334"></a>
-<span class="sourceLineNo">335</span> this.conf = conf;<a name="line.335"></a>
-<span class="sourceLineNo">336</span> this.checkOwnerSet = conf.getBoolean(CHECK_OWNER_SET_CONF_KEY, DEFAULT_CHECK_OWNER_SET);<a name="line.336"></a>
-<span class="sourceLineNo">337</span> refreshConfiguration(conf);<a name="line.337"></a>
-<span class="sourceLineNo">338</span><a name="line.338"></a>
-<span class="sourceLineNo">339</span> }<a name="line.339"></a>
-<span class="sourceLineNo">340</span><a name="line.340"></a>
-<span class="sourceLineNo">341</span> private void load(final boolean abortOnCorruption) throws IOException {<a name="line.341"></a>
-<span class="sourceLineNo">342</span> Preconditions.checkArgument(completed.isEmpty(), "completed not empty");<a name="line.342"></a>
-<span class="sourceLineNo">343</span> Preconditions.checkArgument(rollbackStack.isEmpty(), "rollback state not empty");<a name="line.343"></a>
-<span class="sourceLineNo">344</span> Preconditions.checkArgument(procedures.isEmpty(), "procedure map not empty");<a name="line.344"></a>
-<span class="sourceLineNo">345</span> Preconditions.checkArgument(scheduler.size() == 0, "run queue not empty");<a name="line.345"></a>
-<span class="sourceLineNo">346</span><a name="line.346"></a>
-<span class="sourceLineNo">347</span> store.load(new ProcedureStore.ProcedureLoader() {<a name="line.347"></a>
-<span class="sourceLineNo">348</span> @Override<a name="line.348"></a>
-<span class="sourceLineNo">349</span> public void setMaxProcId(long maxProcId) {<a name="line.349"></a>
-<span class="sourceLineNo">350</span> assert lastProcId.get() < 0 : "expected only one call to setMaxProcId()";<a name="line.350"></a>
-<span class="sourceLineNo">351</span> lastProcId.set(maxProcId);<a name="line.351"></a>
-<span class="sourceLineNo">352</span> }<a name="line.352"></a>
-<span class="sourceLineNo">353</span><a name="line.353"></a>
-<span class="sourceLineNo">354</span> @Override<a name="line.354"></a>
-<span class="sourceLineNo">355</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.355"></a>
-<span class="sourceLineNo">356</span> loadProcedures(procIter, abortOnCorruption);<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> @Override<a name="line.359"></a>
-<span class="sourceLineNo">360</span> public void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.360"></a>
-<span class="sourceLineNo">361</span> int corruptedCount = 0;<a name="line.361"></a>
-<span class="sourceLineNo">362</span> while (procIter.hasNext()) {<a name="line.362"></a>
-<span class="sourceLineNo">363</span> Procedure<?> proc = procIter.next();<a name="line.363"></a>
-<span class="sourceLineNo">364</span> LOG.error("Corrupt " + proc);<a name="line.364"></a>
-<span class="sourceLineNo">365</span> corruptedCount++;<a name="line.365"></a>
-<span class="sourceLineNo">366</span> }<a name="line.366"></a>
-<span class="sourceLineNo">367</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.367"></a>
-<span class="sourceLineNo">368</span> throw new IOException("found " + corruptedCount + " corrupted procedure(s) on replay");<a name="line.368"></a>
-<span class="sourceLineNo">369</span> }<a name="line.369"></a>
-<span class="sourceLineNo">370</span> }<a name="line.370"></a>
-<span class="sourceLineNo">371</span> });<a name="line.371"></a>
-<span class="sourceLineNo">372</span> }<a name="line.372"></a>
+<span class="sourceLineNo">325</span> /**<a name="line.325"></a>
+<span class="sourceLineNo">326</span> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.326"></a>
+<span class="sourceLineNo">327</span> * resource handling rather than observing in a #join is unexpected).<a name="line.327"></a>
+<span class="sourceLineNo">328</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.328"></a>
+<span class="sourceLineNo">329</span> * (Should be ok).<a name="line.329"></a>
+<span class="sourceLineNo">330</span> */<a name="line.330"></a>
+<span class="sourceLineNo">331</span> private TimeoutExecutorThread<TEnvironment> timeoutExecutor;<a name="line.331"></a>
+<span class="sourceLineNo">332</span><a name="line.332"></a>
+<span class="sourceLineNo">333</span> private int corePoolSize;<a name="line.333"></a>
+<span class="sourceLineNo">334</span> private int maxPoolSize;<a name="line.334"></a>
+<span class="sourceLineNo">335</span><a name="line.335"></a>
+<span class="sourceLineNo">336</span> private volatile long keepAliveTime;<a name="line.336"></a>
+<span class="sourceLineNo">337</span><a name="line.337"></a>
+<span class="sourceLineNo">338</span> /**<a name="line.338"></a>
+<span class="sourceLineNo">339</span> * Scheduler/Queue that contains runnable procedures.<a name="line.339"></a>
+<span class="sourceLineNo">340</span> */<a name="line.340"></a>
+<span class="sourceLineNo">341</span> private final ProcedureScheduler scheduler;<a name="line.341"></a>
+<span class="sourceLineNo">342</span><a name="line.342"></a>
+<span class="sourceLineNo">343</span> private final AtomicLong lastProcId = new AtomicLong(-1);<a name="line.343"></a>
+<span class="sourceLineNo">344</span> private final AtomicLong workerId = new AtomicLong(0);<a name="line.344"></a>
+<span class="sourceLineNo">345</span> private final AtomicInteger activeExecutorCount = new AtomicInteger(0);<a name="line.345"></a>
+<span class="sourceLineNo">346</span> private final AtomicBoolean running = new AtomicBoolean(false);<a name="line.346"></a>
+<span class="sourceLineNo">347</span> private final TEnvironment environment;<a name="line.347"></a>
+<span class="sourceLineNo">348</span> private final ProcedureStore store;<a name="line.348"></a>
+<span class="sourceLineNo">349</span><a name="line.349"></a>
+<span class="sourceLineNo">350</span> private final boolean checkOwnerSet;<a name="line.350"></a>
+<span class="sourceLineNo">351</span><a name="line.351"></a>
+<span class="sourceLineNo">352</span> // To prevent concurrent execution of the same procedure.<a name="line.352"></a>
+<span class="sourceLineNo">353</span> // For some rare cases, especially if the procedure uses ProcedureEvent, it is possible that the<a name="line.353"></a>
+<span class="sourceLineNo">354</span> // procedure is woken up before we finish the suspend which causes the same procedures to be<a name="line.354"></a>
+<span class="sourceLineNo">355</span> // executed in parallel. This does lead to some problems, see HBASE-20939&HBASE-20949, and is also<a name="line.355"></a>
+<span class="sourceLineNo">356</span> // a bit confusing to the developers. So here we introduce this lock to prevent the concurrent<a name="line.356"></a>
+<span class="sourceLineNo">357</span> // execution of the same procedure.<a name="line.357"></a>
+<span class="sourceLineNo">358</span> private final IdLock procExecutionLock = new IdLock();<a name="line.358"></a>
+<span class="sourceLineNo">359</span><a name="line.359"></a>
+<span class="sourceLineNo">360</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.360"></a>
+<span class="sourceLineNo">361</span> final ProcedureStore store) {<a name="line.361"></a>
+<span class="sourceLineNo">362</span> this(conf, environment, store, new SimpleProcedureScheduler());<a name="line.362"></a>
+<span class="sourceLineNo">363</span> }<a name="line.363"></a>
+<span class="sourceLineNo">364</span><a name="line.364"></a>
+<span class="sourceLineNo">365</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.365"></a>
+<span class="sourceLineNo">366</span> final ProcedureStore store, final ProcedureScheduler scheduler) {<a name="line.366"></a>
+<span class="sourceLineNo">367</span> this.environment = environment;<a name="line.367"></a>
+<span class="sourceLineNo">368</span> this.scheduler = scheduler;<a name="line.368"></a>
+<span class="sourceLineNo">369</span> this.store = store;<a name="line.369"></a>
+<span class="sourceLineNo">370</span> this.conf = conf;<a name="line.370"></a>
+<span class="sourceLineNo">371</span> this.checkOwnerSet = conf.getBoolean(CHECK_OWNER_SET_CONF_KEY, DEFAULT_CHECK_OWNER_SET);<a name="line.371"></a>
+<span class="sourceLineNo">372</span> refreshConfiguration(conf);<a name="line.372"></a>
<span class="sourceLineNo">373</span><a name="line.373"></a>
-<span class="sourceLineNo">374</span> private void restoreLock(Procedure<TEnvironment> proc, Set<Long> restored) {<a name="line.374"></a>
-<span class="sourceLineNo">375</span> proc.restoreLock(getEnvironment());<a name="line.375"></a>
-<span class="sourceLineNo">376</span> restored.add(proc.getProcId());<a name="line.376"></a>
-<span class="sourceLineNo">377</span> }<a name="line.377"></a>
-<span class="sourceLineNo">378</span><a name="line.378"></a>
-<span class="sourceLineNo">379</span> private void restoreLocks(Deque<Procedure<TEnvironment>> stack, Set<Long> restored) {<a name="line.379"></a>
-<span class="sourceLineNo">380</span> while (!stack.isEmpty()) {<a name="line.380"></a>
-<span class="sourceLineNo">381</span> restoreLock(stack.pop(), restored);<a name="line.381"></a>
-<span class="sourceLineNo">382</span> }<a name="line.382"></a>
-<span class="sourceLineNo">383</span> }<a name="line.383"></a>
-<span class="sourceLineNo">384</span><a name="line.384"></a>
-<span class="sourceLineNo">385</span> // Restore the locks for all the procedures.<a name="line.385"></a>
-<span class="sourceLineNo">386</span> // Notice that we need to restore the locks starting from the root proc, otherwise there will be<a name="line.386"></a>
-<span class="sourceLineNo">387</span> // problem that a sub procedure may hold the exclusive lock first and then we are stuck when<a name="line.387"></a>
-<span class="sourceLineNo">388</span> // calling the acquireLock method for the parent procedure.<a name="line.388"></a>
-<span class="sourceLineNo">389</span> // The algorithm is straight-forward:<a name="line.389"></a>
-<span class="sourceLineNo">390</span> // 1. Use a set to record the procedures which locks have already been restored.<a name="line.390"></a>
-<span class="sourceLineNo">391</span> // 2. Use a stack to store the hierarchy of the procedures<a name="line.391"></a>
-<span class="sourceLineNo">392</span> // 3. For all the procedure, we will first try to find its parent and push it into the stack,<a name="line.392"></a>
-<span class="sourceLineNo">393</span> // unless<a name="line.393"></a>
-<span class="sourceLineNo">394</span> // a. We have no parent, i.e, we are the root procedure<a name="line.394"></a>
-<span class="sourceLineNo">395</span> // b. The lock has already been restored(by checking the set introduced in #1)<a name="line.395"></a>
-<span class="sourceLineNo">396</span> // then we start to pop the stack and call acquireLock for each procedure.<a name="line.396"></a>
-<span class="sourceLineNo">397</span> // Notice that this should be done for all procedures, not only the ones in runnableList.<a name="line.397"></a>
-<span class="sourceLineNo">398</span> private void restoreLocks() {<a name="line.398"></a>
-<span class="sourceLineNo">399</span> Set<Long> restored = new HashSet<>();<a name="line.399"></a>
-<span class="sourceLineNo">400</span> Deque<Procedure<TEnvironment>> stack = new ArrayDeque<>();<a name="line.400"></a>
-<span class="sourceLineNo">401</span> procedures.values().forEach(proc -> {<a name="line.401"></a>
-<span class="sourceLineNo">402</span> for (;;) {<a name="line.402"></a>
-<span class="sourceLineNo">403</span> if (restored.contains(proc.getProcId())) {<a name="line.403"></a>
-<span class="sourceLineNo">404</span> restoreLocks(stack, restored);<a name="line.404"></a>
-<span class="sourceLineNo">405</span> return;<a name="line.405"></a>
-<span class="sourceLineNo">406</span> }<a name="line.406"></a>
-<span class="sourceLineNo">407</span> if (!proc.hasParent()) {<a name="line.407"></a>
-<span class="sourceLineNo">408</span> restoreLock(proc, restored);<a name="line.408"></a>
-<span class="sourceLineNo">409</span> restoreLocks(stack, restored);<a name="line.409"></a>
-<span class="sourceLineNo">410</span> return;<a name="line.410"></a>
-<span class="sourceLineNo">411</span> }<a name="line.411"></a>
-<span class="sourceLineNo">412</span> stack.push(proc);<a name="line.412"></a>
-<span class="sourceLineNo">413</span> proc = procedures.get(proc.getParentProcId());<a name="line.413"></a>
-<span class="sourceLineNo">414</span> }<a name="line.414"></a>
-<span class="sourceLineNo">415</span> });<a name="line.415"></a>
-<span class="sourceLineNo">416</span> }<a name="line.416"></a>
-<span class="sourceLineNo">417</span><a name="line.417"></a>
-<span class="sourceLineNo">418</span> private void loadProcedures(ProcedureIterator procIter, boolean abortOnCorruption)<a name="line.418"></a>
-<span class="sourceLineNo">419</span> throws IOException {<a name="line.419"></a>
-<span class="sourceLineNo">420</span> // 1. Build the rollback stack<a name="line.420"></a>
-<span class="sourceLineNo">421</span> int runnablesCount = 0;<a name="line.421"></a>
-<span class="sourceLineNo">422</span> int failedCount = 0;<a name="line.422"></a>
-<span class="sourceLineNo">423</span> while (procIter.hasNext()) {<a name="line.423"></a>
-<span class="sourceLineNo">424</span> boolean finished = procIter.isNextFinished();<a name="line.424"></a>
-<span class="sourceLineNo">425</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.425"></a>
-<span class="sourceLineNo">426</span> NonceKey nonceKey = proc.getNonceKey();<a name="line.426"></a>
-<span class="sourceLineNo">427</span> long procId = proc.getProcId();<a name="line.427"></a>
-<span class="sourceLineNo">428</span><a name="line.428"></a>
-<span class="sourceLineNo">429</span> if (finished) {<a name="line.429"></a>
-<span class="sourceLineNo">430</span> completed.put(proc.getProcId(), new CompletedProcedureRetainer<>(proc));<a name="line.430"></a>
-<span class="sourceLineNo">431</span> LOG.debug("Completed {}", proc);<a name="line.431"></a>
-<span class="sourceLineNo">432</span> } else {<a name="line.432"></a>
-<span class="sourceLineNo">433</span> if (!proc.hasParent()) {<a name="line.433"></a>
-<span class="sourceLineNo">434</span> assert !proc.isFinished() : "unexpected finished procedure";<a name="line.434"></a>
-<span class="sourceLineNo">435</span> rollbackStack.put(proc.getProcId(), new RootProcedureState<>());<a name="line.435"></a>
-<span class="sourceLineNo">436</span> }<a name="line.436"></a>
-<span class="sourceLineNo">437</span><a name="line.437"></a>
-<span class="sourceLineNo">438</span> // add the procedure to the map<a name="line.438"></a>
-<span class="sourceLineNo">439</span> proc.beforeReplay(getEnvironment());<a name="line.439"></a>
-<span class="sourceLineNo">440</span> procedures.put(proc.getProcId(), proc);<a name="line.440"></a>
-<span class="sourceLineNo">441</span><a name="line.441"></a>
-<span class="sourceLineNo">442</span> if (proc.getState() == ProcedureState.RUNNABLE) {<a name="line.442"></a>
-<span class="sourceLineNo">443</span> runnablesCount++;<a name="line.443"></a>
-<span class="sourceLineNo">444</span> } else if (proc.getState() == ProcedureState.FAILED) {<a name="line.444"></a>
-<span class="sourceLineNo">445</span> failedCount++;<a name="line.445"></a>
+<span class="sourceLineNo">374</span> }<a name="line.374"></a>
+<span class="sourceLineNo">375</span><a name="line.375"></a>
+<span class="sourceLineNo">376</span> private void load(final boolean abortOnCorruption) throws IOException {<a name="line.376"></a>
+<span class="sourceLineNo">377</span> Preconditions.checkArgument(completed.isEmpty(), "completed not empty");<a name="line.377"></a>
+<span class="sourceLineNo">378</span> Preconditions.checkArgument(rollbackStack.isEmpty(), "rollback state not empty");<a name="line.378"></a>
+<span class="sourceLineNo">379</span> Preconditions.checkArgument(procedures.isEmpty(), "procedure map not empty");<a name="line.379"></a>
+<span class="sourceLineNo">380</span> Preconditions.checkArgument(scheduler.size() == 0, "run queue not empty");<a name="line.380"></a>
+<span class="sourceLineNo">381</span><a name="line.381"></a>
+<span class="sourceLineNo">382</span> store.load(new ProcedureStore.ProcedureLoader() {<a name="line.382"></a>
+<span class="sourceLineNo">383</span> @Override<a name="line.383"></a>
+<span class="sourceLineNo">384</span> public void setMaxProcId(long maxProcId) {<a name="line.384"></a>
+<span class="sourceLineNo">385</span> assert lastProcId.get() < 0 : "expected only one call to setMaxProcId()";<a name="line.385"></a>
+<span class="sourceLineNo">386</span> lastProcId.set(maxProcId);<a name="line.386"></a>
+<span class="sourceLineNo">387</span> }<a name="line.387"></a>
+<span class="sourceLineNo">388</span><a name="line.388"></a>
+<span class="sourceLineNo">389</span> @Override<a name="line.389"></a>
+<span class="sourceLineNo">390</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.390"></a>
+<span class="sourceLineNo">391</span> loadProcedures(procIter, abortOnCorruption);<a name="line.391"></a>
+<span class="sourceLineNo">392</span> }<a name="line.392"></a>
+<span class="sourceLineNo">393</span><a name="line.393"></a>
+<span class="sourceLineNo">394</span> @Override<a name="line.394"></a>
+<span class="sourceLineNo">395</span> public void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.395"></a>
+<span class="sourceLineNo">396</span> int corruptedCount = 0;<a name="line.396"></a>
+<span class="sourceLineNo">397</span> while (procIter.hasNext()) {<a name="line.397"></a>
+<span class="sourceLineNo">398</span> Procedure<?> proc = procIter.next();<a name="line.398"></a>
+<span class="sourceLineNo">399</span> LOG.error("Corrupt " + proc);<a name="line.399"></a>
+<span class="sourceLineNo">400</span> corruptedCount++;<a name="line.400"></a>
+<span class="sourceLineNo">401</span> }<a name="line.401"></a>
+<span class="sourceLineNo">402</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.402"></a>
+<span class="sourceLineNo">403</span> throw new IOException("found " + corruptedCount + " corrupted procedure(s) on replay");<a name="line.403"></a>
+<span class="sourceLineNo">404</span> }<a name="line.404"></a>
+<span class="sourceLineNo">405</span> }<a name="line.405"></a>
+<span class="sourceLineNo">406</span> });<a name="line.406"></a>
+<span class="sourceLineNo">407</span> }<a name="line.407"></a>
+<span class="sourceLineNo">408</span><a name="line.408"></a>
+<span class="sourceLineNo">409</span> private void restoreLock(Procedure<TEnvironment> proc, Set<Long> restored) {<a name="line.409"></a>
+<span class="sourceLineNo">410</span> proc.restoreLock(getEnvironment());<a name="line.410"></a>
+<span class="sourceLineNo">411</span> restored.add(proc.getProcId());<a name="line.411"></a>
+<span class="sourceLineNo">412</span> }<a name="line.412"></a>
+<span class="sourceLineNo">413</span><a name="line.413"></a>
+<span class="sourceLineNo">414</span> private void restoreLocks(Deque<Procedure<TEnvironment>> stack, Set<Long> restored) {<a name="line.414"></a>
+<span class="sourceLineNo">415</span> while (!stack.isEmpty()) {<a name="line.415"></a>
+<span class="sourceLineNo">416</span> restoreLock(stack.pop(), restored);<a name="line.416"></a>
+<span class="sourceLineNo">417</span> }<a name="line.417"></a>
+<span class="sourceLineNo">418</span> }<a name="line.418"></a>
+<span class="sourceLineNo">419</span><a name="line.419"></a>
+<span class="sourceLineNo">420</span> // Restore the locks for all the procedures.<a name="line.420"></a>
+<span class="sourceLineNo">421</span> // Notice that we need to restore the locks starting from the root proc, otherwise there will be<a name="line.421"></a>
+<span class="sourceLineNo">422</span> // problem that a sub procedure may hold the exclusive lock first and then we are stuck when<a name="line.422"></a>
+<span class="sourceLineNo">423</span> // calling the acquireLock method for the parent procedure.<a name="line.423"></a>
+<span class="sourceLineNo">424</span> // The algorithm is straight-forward:<a name="line.424"></a>
+<span class="sourceLineNo">425</span> // 1. Use a set to record the procedures which locks have already been restored.<a name="line.425"></a>
+<span class="sourceLineNo">426</span> // 2. Use a stack to store the hierarchy of the procedures<a name="line.426"></a>
+<span class="sourceLineNo">427</span> // 3. For all the procedure, we will first try to find its parent and push it into the stack,<a name="line.427"></a>
+<span class="sourceLineNo">428</span> // unless<a name="line.428"></a>
+<span class="sourceLineNo">429</span> // a. We have no parent, i.e, we are the root procedure<a name="line.429"></a>
+<span class="sourceLineNo">430</span> // b. The lock has already been restored(by checking the set introduced in #1)<a name="line.430"></a>
+<span class="sourceLineNo">431</span> // then we start to pop the stack and call acquireLock for each procedure.<a name="line.431"></a>
+<span class="sourceLineNo">432</span> // Notice that this should be done for all procedures, not only the ones in runnableList.<a name="line.432"></a>
+<span class="sourceLineNo">433</span> private void restoreLocks() {<a name="line.433"></a>
+<span class="sourceLineNo">434</span> Set<Long> restored = new HashSet<>();<a name="line.434"></a>
+<span class="sourceLineNo">435</span> Deque<Procedure<TEnvironment>> stack = new ArrayDeque<>();<a name="line.435"></a>
+<span class="sourceLineNo">436</span> procedures.values().forEach(proc -> {<a name="line.436"></a>
+<span class="sourceLineNo">437</span> for (;;) {<a name="line.437"></a>
+<span class="sourceLineNo">438</span> if (restored.contains(proc.getProcId())) {<a name="line.438"></a>
+<span class="sourceLineNo">439</span> restoreLocks(stack, restored);<a name="line.439"></a>
+<span class="sourceLineNo">440</span> return;<a name="line.440"></a>
+<span class="sourceLineNo">441</span> }<a name="line.441"></a>
+<span class="sourceLineNo">442</span> if (!proc.hasParent()) {<a name="line.442"></a>
+<span class="sourceLineNo">443</span> restoreLock(proc, restored);<a name="line.443"></a>
+<span class="sourceLineNo">444</span> restoreLocks(stack, restored);<a name="line.444"></a>
+<span class="sourceLineNo">445</span> return;<a name="line.445"></a>
<span class="sourceLineNo">446</span> }<a name="line.446"></a>
-<span class="sourceLineNo">447</span> }<a name="line.447"></a>
-<span class="sourceLineNo">448</span><a name="line.448"></a>
-<span class="sourceLineNo">449</span> // add the nonce to the map<a name="line.449"></a>
-<span class="sourceLineNo">450</span> if (nonceKey != null) {<a name="line.450"></a>
-<span class="sourceLineNo">451</span> nonceKeysToProcIdsMap.put(nonceKey, procId);<a name="line.451"></a>
-<span class="sourceLineNo">452</span> }<a name="line.452"></a>
-<span class="sourceLineNo">453</span> }<a name="line.453"></a>
-<span class="sourceLineNo">454</span><a name="line.454"></a>
-<span class="sourceLineNo">455</span> // 2. Initialize the stacks<a name="line.455"></a>
-<span class="sourceLineNo">456</span> // In the old implementation, for procedures in FAILED state, we will push it into the<a name="line.456"></a>
-<span class="sourceLineNo">457</span> // ProcedureScheduler directly to execute the rollback. But this does not work after we<a name="line.457"></a>
-<span class="sourceLineNo">458</span> // introduce the restore lock stage.<a name="line.458"></a>
-<span class="sourceLineNo">459</span> // For now, when we acquire a xlock, we will remove the queue from runQueue in scheduler, and<a name="line.459"></a>
-<span class="sourceLineNo">460</span> // then when a procedure which has lock access, for example, a sub procedure of the procedure<a name="line.460"></a>
-<span class="sourceLineNo">461</span> // which has the xlock, is pushed into the scheduler, we will add the queue back to let the<a name="line.461"></a>
-<span class="sourceLineNo">462</span> // workers poll from it. The assumption here is that, the procedure which has the xlock should<a name="line.462"></a>
-<span class="sourceLineNo">463</span> // have been polled out already, so when loading we can not add the procedure to scheduler first<a name="line.463"></a>
-<span class="sourceLineNo">464</span> // and then call acquireLock, since the procedure is still in the queue, and since we will<a name="line.464"></a>
-<span class="sourceLineNo">465</span> // remove the queue from runQueue, then no one can poll it out, then there is a dead lock<a name="line.465"></a>
-<span class="sourceLineNo">466</span> List<Procedure<TEnvironment>> runnableList = new ArrayList<>(runnablesCount);<a name="line.466"></a>
-<span class="sourceLineNo">467</span> List<Procedure<TEnvironment>> failedList = new ArrayList<>(failedCount);<a name="line.467"></a>
-<span class="sourceLineNo">468</span> Set<Procedure<TEnvironment>> waitingSet = null;<a name="line.468"></a>
-<span class="sourceLineNo">469</span> procIter.reset();<a name="line.469"></a>
-<span class="sourceLineNo">470</span> while (procIter.hasNext()) {<a name="line.470"></a>
-<span class="sourceLineNo">471</span> if (procIter.isNextFinished()) {<a name="line.471"></a>
-<span class="sourceLineNo">472</span> procIter.skipNext();<a name="line.472"></a>
-<span class="sourceLineNo">473</span> continue;<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> Procedure<TEnvironment> proc = procIter.next();<a name="line.476"></a>
-<span class="sourceLineNo">477</span> assert !(proc.isFinished() && !proc.hasParent()) : "unexpected completed proc=" + proc;<a name="line.477"></a>
-<span class="sourceLineNo">478</span><a name="line.478"></a>
-<span class="sourceLineNo">479</span> LOG.debug("Loading {}", proc);<a name="line.479"></a>
-<span class="sourceLineNo">480</span><a name="line.480"></a>
-<span class="sourceLineNo">481</span> Long rootProcId = getRootProcedureId(proc);<a name="line.481"></a>
-<span class="sourceLineNo">482</span> if (rootProcId == null) {<a name="line.482"></a>
-<span class="sourceLineNo">483</span> // The 'proc' was ready to run but the root procedure was rolledback?<a name="line.483"></a>
-<span class="sourceLineNo">484</span> scheduler.addBack(proc);<a name="line.484"></a>
-<span class="sourceLineNo">485</span> continue;<a name="line.485"></a>
-<span class="sourceLineNo">486</span> }<a name="line.486"></a>
-<span class="sourceLineNo">487</span><a name="line.487"></a>
-<span class="sourceLineNo">488</span> if (proc.hasParent()) {<a name="line.488"></a>
-<span class="sourceLineNo">489</span> Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());<a name="line.489"></a>
-<span class="sourceLineNo">490</span> // corrupted procedures are handled later at step 3<a name="line.490"></a>
-<span class="sourceLineNo">491</span> if (parent != null && !proc.isFinished()) {<a name="line.491"></a>
-<span class="sourceLineNo">492</span> parent.incChildrenLatch();<a name="line.492"></a>
-<span class="sourceLineNo">493</span> }<a name="line.493"></a>
-<span class="sourceLineNo">494</span> }<a name="line.494"></a>
-<span class="sourceLineNo">495</span><a name="line.495"></a>
-<span class="sourceLineNo">496</span> RootProcedureState<TEnvironment> procStack = rollbackStack.get(rootProcId);<a name="line.496"></a>
-<span class="sourceLineNo">497</span> procStack.loadStack(proc);<a name="line.497"></a>
-<span class="sourceLineNo">498</span><a name="line.498"></a>
-<span class="sourceLineNo">499</span> proc.setRootProcId(rootProcId);<a name="line.499"></a>
-<span class="sourceLineNo">500</span> switch (proc.getState()) {<a name="line.500"></a>
-<span class="sourceLineNo">501</span> case RUNNABLE:<a name="line.501"></a>
-<span class="sourceLineNo">502</span> runnableList.add(proc);<a name="line.502"></a>
-<span class="sourceLineNo">503</span> break;<a name="line.503"></a>
-<span class="sourceLineNo">504</span> case WAITING:<a name="line.504"></a>
-<span class="sourceLineNo">505</span> if (!proc.hasChildren()) {<a name="line.505"></a>
-<span class="sourceLineNo">506</span> runnableList.add(proc);<a name="line.506"></a>
-<span class="sourceLineNo">507</span> }<a name="line.507"></a>
-<span class="sourceLineNo">508</span> break;<a name="line.508"></a>
-<span class="sourceLineNo">509</span> case WAITING_TIMEOUT:<a name="line.509"></a>
-<span class="sourceLineNo">510</span> if (waitingSet == null) {<a name="line.510"></a>
-<span class="sourceLineNo">511</span> waitingSet = new HashSet<>();<a name="line.511"></a>
-<span class="sourceLineNo">512</span> }<a name="line.512"></a>
-<span class="sourceLineNo">513</span> waitingSet.add(proc);<a name="line.513"></a>
-<span class="sourceLineNo">514</span> break;<a name="line.514"></a>
-<span class="sourceLineNo">515</span> case FAILED:<a name="line.515"></a>
-<span class="sourceLineNo">516</span> failedList.add(proc);<a name="line.516"></a>
-<span class="sourceLineNo">517</span> break;<a name="line.517"></a>
-<span class="sourceLineNo">518</span> case ROLLEDBACK:<a name="line.518"></a>
-<span class="sourceLineNo">519</span> case INITIALIZING:<a name="line.519"></a>
-<span class="sourceLineNo">520</span> String msg = "Unexpected " + proc.getState() + " state for " + proc;<a name="line.520"></a>
-<span class="sourceLineNo">521</span> LOG.error(msg);<a name="line.521"></a>
-<span class="sourceLineNo">522</span> throw new UnsupportedOperationException(msg);<a name="line.522"></a>
-<span class="sourceLineNo">523</span> default:<a name="line.523"></a>
-<span class="sourceLineNo">524</span> break;<a name="line.524"></a>
-<span class="sourceLineNo">525</span> }<a name="line.525"></a>
-<span class="sourceLineNo">526</span> }<a name="line.526"></a>
-<span class="sourceLineNo">527</span><a name="line.527"></a>
-<span class="sourceLineNo">528</span> // 3. Validate the stacks<a name="line.528"></a>
-<span class="sourceLineNo">529</span> int corruptedCount = 0;<a name="line.529"></a>
-<span class="sourceLineNo">530</span> Iterator<Map.Entry<Long, RootProcedureState<TEnvironment>>> itStack =<a name="line.530"></a>
-<span class="sourceLineNo">531</span> rollbackStack.entrySet().iterator();<a name="line.531"></a>
-<span class="sourceLineNo">532</span> while (itStack.hasNext()) {<a name="line.532"></a>
-<span class="sourceLineNo">533</span> Map.Entry<Long, RootProcedureState<TEnvironment>> entry = itStack.next();<a name="line.533"></a>
-<span class="sourceLineNo">534</span> RootProcedureState<TEnvironment> procStack = entry.getValue();<a name="line.534"></a>
-<span class="sourceLineNo">535</span> if (procStack.isValid()) continue;<a name="line.535"></a>
-<span class="sourceLineNo">536</span><a name="line.536"></a>
-<span class="sourceLineNo">537</span> for (Procedure<TEnvironment> proc : procStack.getSubproceduresStack()) {<a name="line.537"></a>
-<span class="sourceLineNo">538</span> LOG.error("Corrupted " + proc);<a name="line.538"></a>
-<span class="sourceLineNo">539</span> procedures.remove(proc.getProcId());<a name="line.539"></a>
-<span class="sourceLineNo">540</span> runnableList.remove(proc);<a name="line.540"></a>
-<span class="sourceLineNo">541</span> if (waitingSet != null) waitingSet.remove(proc);<a name="line.541"></a>
-<span class="sourceLineNo">542</span> corruptedCount++;<a name="line.542"></a>
-<span class="sourceLineNo">543</span> }<a name="line.543"></a>
-<span class="sourceLineNo">544</span> itStack.remove();<a name="line.544"></a>
-<span class="sourceLineNo">545</span> }<a name="line.545"></a>
-<span class="sourceLineNo">546</span><a name="line.546"></a>
-<span class="sourceLineNo">547</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.547"></a>
-<span class="sourceLineNo">548</span> throw new IOException("found " + corruptedCount + " procedures on replay");<a name="line.548"></a>
-<span class="sourceLineNo">549</span> }<a name="line.549"></a>
-<span class="sourceLineNo">550</span><a name="line.550"></a>
-<span class="sourceLineNo">551</span> // 4. Push the procedures to the timeout executor<a name="line.551"></a>
-<span class="sourceLineNo">552</span> if (waitingSet != null && !waitingSet.isEmpty()) {<a name="line.552"></a>
-<span class="sourceLineNo">553</span> for (Procedure<TEnvironment> proc: waitingSet) {<a name="line.553"></a>
-<span class="sourceLineNo">554</span> proc.afterReplay(getEnvironment());<a name="line.554"></a>
-<span class="sourceLineNo">555</span> timeoutExecutor.add(proc);<a name="line.555"></a>
-<span class="sourceLineNo">556</span> }<a name="line.556"></a>
-<span class="sourceLineNo">557</span> }<a name="line.557"></a>
-<span class="sourceLineNo">558</span> // 5. restore locks<a name="line.558"></a>
-<span class="sourceLineNo">559</span> restoreLocks();<a name="line.559"></a>
-<span class="sourceLineNo">560</span> // 6. Push the procedure to the scheduler<a name="line.560"></a>
-<span class="sourceLineNo">561</span> failedList.forEach(scheduler::addBack);<a name="line.561"></a>
-<span class="sourceLineNo">562</span> runnableList.forEach(p -> {<a name="line.562"></a>
-<span class="sourceLineNo">563</span> p.afterReplay(getEnvironment());<a name="line.563"></a>
-<span class="sourceLineNo">564</span> if (!p.hasParent()) {<a name="line.564"></a>
-<span class="sourceLineNo">565</span> sendProcedureLoadedNotification(p.getProcId());<a name="line.565"></a>
-<span class="sourceLineNo">566</span> }<a name="line.566"></a>
-<span class="sourceLineNo">567</span> scheduler.addBack(p);<a name="line.567"></a>
-<span class="sourceLineNo">568</span> });<a name="line.568"></a>
-<span class="sourceLineNo">569</span> }<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> * Initialize the procedure executor, but do not start workers. We will start them later.<a name="line.572"></a>
-<span class="sourceLineNo">573</span> * <p/><a name="line.573"></a>
-<span class="sourceLineNo">574</span> * It calls ProcedureStore.recoverLease() and ProcedureStore.load() to recover the lease, and<a name="line.574"></a>
-<span class="sourceLineNo">575</span> * ensure a single executor, and start the procedure replay to resume and recover the previous<a name="line.575"></a>
-<span class="sourceLineNo">576</span> * pending and in-progress procedures.<a name="line.576"></a>
-<span class="sourceLineNo">577</span> * @param numThreads number of threads available for procedure execution.<a name="line.577"></a>
-<span class="sourceLineNo">578</span> * @param abortOnCorruption true if you want to abort your service in case a corrupted procedure<a name="line.578"></a>
-<span class="sourceLineNo">579</span> * is found on replay. otherwise false.<a name="line.579"></a>
-<span class="sourceLineNo">580</span> */<a name="line.580"></a>
-<span class="sourceLineNo">581</span> public void init(int numThreads, boolean abortOnCorruption) throws IOException {<a name="line.581"></a>
-<span class="sourceLineNo">582</span> // We have numThreads executor + one timer thread used for timing out<a name="line.582"></a>
-<span class="sourceLineNo">583</span> // procedures and triggering periodic procedures.<a name="line.583"></a>
-<span class="sourceLineNo">584</span> this.corePoolSize = numThreads;<a name="line.584"></a>
-<span class="sourceLineNo">585</span> this.maxPoolSize = 10 * numThreads;<a name="line.585"></a>
-<span class="sourceLineNo">586</span> LOG.info("Starting {} core workers (bigger of cpus/4 or 16) with max (burst) worker count={}",<a name="line.586"></a>
-<span class="sourceLineNo">587</span> corePoolSize, maxPoolSize);<a name="line.587"></a>
-<span class="sourceLineNo">588</span><a name="line.588"></a>
-<span class="sourceLineNo">589</span> this.threadGroup = new ThreadGroup("PEWorkerGroup");<a name="line.589"></a>
-<span class="sourceLineNo">590</span> this.timeoutExecutor = new TimeoutExecutorThread<>(this, threadGroup);<a name="line.590"></a>
-<span class="sourceLineNo">591</span><a name="line.591"></a>
-<span class="sourceLineNo">592</span> // Create the workers<a name="line.592"></a>
-<span class="sourceLineNo">593</span> workerId.set(0);<a name="line.593"></a>
-<span class="sourceLineNo">594</span> workerThreads = new CopyOnWriteArrayList<>();<a name="line.594"></a>
-<span class="sourceLineNo">595</span> for (int i = 0; i < corePoolSize; ++i) {<a name="line.595"></a>
-<span class="sourceLineNo">596</span> workerThreads.add(new WorkerThread(threadGroup));<a name="line.596"></a>
-<span class="sourceLineNo">597</span> }<a name="line.597"></a>
-<span class="sourceLineNo">598</span><a name="line.598"></a>
-<span class="sourceLineNo">599</span> long st, et;<a name="line.599"></a>
-<span class="sourceLineNo">600</span><a name="line.600"></a>
-<span class="sourceLineNo">601</span> // Acquire the store lease.<a name
<TRUNCATED>