http://git-wip-us.apache.org/repos/asf/hbase-site/blob/fa1bebf8/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.WorkerThread.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.WorkerThread.html b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.WorkerThread.html index 2c14c50..43c66a8 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.WorkerThread.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.WorkerThread.html @@ -46,2104 +46,2113 @@ <span class="sourceLineNo">038</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.38"></a> <span class="sourceLineNo">039</span>import java.util.stream.Collectors;<a name="line.39"></a> <span class="sourceLineNo">040</span>import java.util.stream.Stream;<a name="line.40"></a> -<span class="sourceLineNo">041</span>import org.apache.hadoop.conf.Configuration;<a name="line.41"></a> -<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.HConstants;<a name="line.42"></a> -<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.exceptions.IllegalArgumentIOException;<a name="line.43"></a> -<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.log.HBaseMarkers;<a name="line.44"></a> -<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.procedure2.Procedure.LockState;<a name="line.45"></a> -<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStore;<a name="line.46"></a> -<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;<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><a name="line.60"></a> -<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureState;<a name="line.61"></a> -<span class="sourceLineNo">062</span><a name="line.62"></a> -<span class="sourceLineNo">063</span>/**<a name="line.63"></a> -<span class="sourceLineNo">064</span> * Thread Pool that executes the submitted procedures.<a name="line.64"></a> -<span class="sourceLineNo">065</span> * The executor has a ProcedureStore associated.<a name="line.65"></a> -<span class="sourceLineNo">066</span> * Each operation is logged and on restart the pending procedures are resumed.<a name="line.66"></a> -<span class="sourceLineNo">067</span> *<a name="line.67"></a> -<span class="sourceLineNo">068</span> * Unless the Procedure code throws an error (e.g. invalid user input)<a name="line.68"></a> -<span class="sourceLineNo">069</span> * the procedure will complete (at some point in time), On restart the pending<a name="line.69"></a> -<span class="sourceLineNo">070</span> * procedures are resumed and the once failed will be rolledback.<a name="line.70"></a> -<span class="sourceLineNo">071</span> *<a name="line.71"></a> -<span class="sourceLineNo">072</span> * The user can add procedures to the executor via submitProcedure(proc)<a name="line.72"></a> -<span class="sourceLineNo">073</span> * check for the finished state via isFinished(procId)<a name="line.73"></a> -<span class="sourceLineNo">074</span> * and get the result via getResult(procId)<a name="line.74"></a> -<span class="sourceLineNo">075</span> */<a name="line.75"></a> -<span class="sourceLineNo">076</span>@InterfaceAudience.Private<a name="line.76"></a> -<span class="sourceLineNo">077</span>public class ProcedureExecutor<TEnvironment> {<a name="line.77"></a> -<span class="sourceLineNo">078</span> private static final Logger LOG = LoggerFactory.getLogger(ProcedureExecutor.class);<a name="line.78"></a> -<span class="sourceLineNo">079</span><a name="line.79"></a> -<span class="sourceLineNo">080</span> public static final String CHECK_OWNER_SET_CONF_KEY = "hbase.procedure.check.owner.set";<a name="line.80"></a> -<span class="sourceLineNo">081</span> private static final boolean DEFAULT_CHECK_OWNER_SET = false;<a name="line.81"></a> -<span class="sourceLineNo">082</span><a name="line.82"></a> -<span class="sourceLineNo">083</span> public static final String WORKER_KEEP_ALIVE_TIME_CONF_KEY =<a name="line.83"></a> -<span class="sourceLineNo">084</span> "hbase.procedure.worker.keep.alive.time.msec";<a name="line.84"></a> -<span class="sourceLineNo">085</span> private static final long DEFAULT_WORKER_KEEP_ALIVE_TIME = TimeUnit.MINUTES.toMillis(1);<a name="line.85"></a> -<span class="sourceLineNo">086</span><a name="line.86"></a> -<span class="sourceLineNo">087</span> /**<a name="line.87"></a> -<span class="sourceLineNo">088</span> * {@link #testing} is non-null when ProcedureExecutor is being tested. Tests will try to<a name="line.88"></a> -<span class="sourceLineNo">089</span> * break PE having it fail at various junctures. When non-null, testing is set to an instance of<a name="line.89"></a> -<span class="sourceLineNo">090</span> * the below internal {@link Testing} class with flags set for the particular test.<a name="line.90"></a> -<span class="sourceLineNo">091</span> */<a name="line.91"></a> -<span class="sourceLineNo">092</span> Testing testing = null;<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> * Class with parameters describing how to fail/die when in testing-context.<a name="line.95"></a> -<span class="sourceLineNo">096</span> */<a name="line.96"></a> -<span class="sourceLineNo">097</span> public static class Testing {<a name="line.97"></a> -<span class="sourceLineNo">098</span> protected boolean killIfHasParent = true;<a name="line.98"></a> -<span class="sourceLineNo">099</span> protected boolean killIfSuspended = false;<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> * Kill the PE BEFORE we store state to the WAL. Good for figuring out if a Procedure is<a name="line.102"></a> -<span class="sourceLineNo">103</span> * persisting all the state it needs to recover after a crash.<a name="line.103"></a> -<span class="sourceLineNo">104</span> */<a name="line.104"></a> -<span class="sourceLineNo">105</span> protected boolean killBeforeStoreUpdate = false;<a name="line.105"></a> -<span class="sourceLineNo">106</span> protected boolean toggleKillBeforeStoreUpdate = 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> * Set when we want to fail AFTER state has been stored into the WAL. Rarely used. HBASE-20978<a name="line.109"></a> -<span class="sourceLineNo">110</span> * is about a case where memory-state was being set after store to WAL where a crash could<a name="line.110"></a> -<span class="sourceLineNo">111</span> * cause us to get stuck. This flag allows killing at what was a vulnerable time.<a name="line.111"></a> -<span class="sourceLineNo">112</span> */<a name="line.112"></a> -<span class="sourceLineNo">113</span> protected boolean killAfterStoreUpdate = false;<a name="line.113"></a> -<span class="sourceLineNo">114</span> protected boolean toggleKillAfterStoreUpdate = false;<a name="line.114"></a> -<span class="sourceLineNo">115</span><a name="line.115"></a> -<span class="sourceLineNo">116</span> protected boolean shouldKillBeforeStoreUpdate() {<a name="line.116"></a> -<span class="sourceLineNo">117</span> final boolean kill = this.killBeforeStoreUpdate;<a name="line.117"></a> -<span class="sourceLineNo">118</span> if (this.toggleKillBeforeStoreUpdate) {<a name="line.118"></a> -<span class="sourceLineNo">119</span> this.killBeforeStoreUpdate = !kill;<a name="line.119"></a> -<span class="sourceLineNo">120</span> LOG.warn("Toggle KILL before store update to: " + this.killBeforeStoreUpdate);<a name="line.120"></a> -<span class="sourceLineNo">121</span> }<a name="line.121"></a> -<span class="sourceLineNo">122</span> return kill;<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> protected boolean shouldKillBeforeStoreUpdate(boolean isSuspended, boolean hasParent) {<a name="line.125"></a> -<span class="sourceLineNo">126</span> if (isSuspended && !killIfSuspended) {<a name="line.126"></a> -<span class="sourceLineNo">127</span> return false;<a name="line.127"></a> -<span class="sourceLineNo">128</span> }<a name="line.128"></a> -<span class="sourceLineNo">129</span> if (hasParent && !killIfHasParent) {<a name="line.129"></a> -<span class="sourceLineNo">130</span> return false;<a name="line.130"></a> -<span class="sourceLineNo">131</span> }<a name="line.131"></a> -<span class="sourceLineNo">132</span> return shouldKillBeforeStoreUpdate();<a name="line.132"></a> -<span class="sourceLineNo">133</span> }<a name="line.133"></a> -<span class="sourceLineNo">134</span><a name="line.134"></a> -<span class="sourceLineNo">135</span> protected boolean shouldKillAfterStoreUpdate() {<a name="line.135"></a> -<span class="sourceLineNo">136</span> final boolean kill = this.killAfterStoreUpdate;<a name="line.136"></a> -<span class="sourceLineNo">137</span> if (this.toggleKillAfterStoreUpdate) {<a name="line.137"></a> -<span class="sourceLineNo">138</span> this.killAfterStoreUpdate = !kill;<a name="line.138"></a> -<span class="sourceLineNo">139</span> LOG.warn("Toggle KILL after store update to: " + this.killAfterStoreUpdate);<a name="line.139"></a> -<span class="sourceLineNo">140</span> }<a name="line.140"></a> -<span class="sourceLineNo">141</span> return kill;<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> protected boolean shouldKillAfterStoreUpdate(final boolean isSuspended) {<a name="line.144"></a> -<span class="sourceLineNo">145</span> return (isSuspended && !killIfSuspended) ? false : shouldKillAfterStoreUpdate();<a name="line.145"></a> -<span class="sourceLineNo">146</span> }<a name="line.146"></a> -<span class="sourceLineNo">147</span> }<a name="line.147"></a> -<span class="sourceLineNo">148</span><a name="line.148"></a> -<span class="sourceLineNo">149</span> public interface ProcedureExecutorListener {<a name="line.149"></a> -<span class="sourceLineNo">150</span> void procedureLoaded(long procId);<a name="line.150"></a> -<span class="sourceLineNo">151</span> void procedureAdded(long procId);<a name="line.151"></a> -<span class="sourceLineNo">152</span> void procedureFinished(long procId);<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> private static final class CompletedProcedureRetainer<TEnvironment> {<a name="line.155"></a> -<span class="sourceLineNo">156</span> private final Procedure<TEnvironment> procedure;<a name="line.156"></a> -<span class="sourceLineNo">157</span> private long clientAckTime;<a name="line.157"></a> -<span class="sourceLineNo">158</span><a name="line.158"></a> -<span class="sourceLineNo">159</span> public CompletedProcedureRetainer(Procedure<TEnvironment> procedure) {<a name="line.159"></a> -<span class="sourceLineNo">160</span> this.procedure = procedure;<a name="line.160"></a> -<span class="sourceLineNo">161</span> 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> public Procedure<TEnvironment> getProcedure() {<a name="line.164"></a> -<span class="sourceLineNo">165</span> return procedure;<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 boolean hasClientAckTime() {<a name="line.168"></a> -<span class="sourceLineNo">169</span> return clientAckTime != -1;<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 long getClientAckTime() {<a name="line.172"></a> -<span class="sourceLineNo">173</span> return clientAckTime;<a name="line.173"></a> -<span class="sourceLineNo">174</span> }<a name="line.174"></a> -<span class="sourceLineNo">175</span><a name="line.175"></a> -<span class="sourceLineNo">176</span> public void setClientAckTime(long clientAckTime) {<a name="line.176"></a> -<span class="sourceLineNo">177</span> this.clientAckTime = clientAckTime;<a name="line.177"></a> -<span class="sourceLineNo">178</span> }<a name="line.178"></a> -<span class="sourceLineNo">179</span><a name="line.179"></a> -<span class="sourceLineNo">180</span> public boolean isExpired(long now, long evictTtl, long evictAckTtl) {<a name="line.180"></a> -<span class="sourceLineNo">181</span> return (hasClientAckTime() && (now - getClientAckTime()) >= evictAckTtl) ||<a name="line.181"></a> -<span class="sourceLineNo">182</span> (now - procedure.getLastUpdate()) >= evictTtl;<a name="line.182"></a> -<span class="sourceLineNo">183</span> }<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> /**<a name="line.186"></a> -<span class="sourceLineNo">187</span> * Internal cleaner that removes the completed procedure results after a TTL.<a name="line.187"></a> -<span class="sourceLineNo">188</span> * NOTE: This is a special case handled in timeoutLoop().<a name="line.188"></a> -<span class="sourceLineNo">189</span> *<a name="line.189"></a> -<span class="sourceLineNo">190</span> * <p>Since the client code looks more or less like:<a name="line.190"></a> -<span class="sourceLineNo">191</span> * <pre><a name="line.191"></a> -<span class="sourceLineNo">192</span> * procId = master.doOperation()<a name="line.192"></a> -<span class="sourceLineNo">193</span> * while (master.getProcResult(procId) == ProcInProgress);<a name="line.193"></a> -<span class="sourceLineNo">194</span> * </pre><a name="line.194"></a> -<span class="sourceLineNo">195</span> * The master should not throw away the proc result as soon as the procedure is done<a name="line.195"></a> -<span class="sourceLineNo">196</span> * but should wait a result request from the client (see executor.removeResult(procId))<a name="line.196"></a> -<span class="sourceLineNo">197</span> * The client will call something like master.isProcDone() or master.getProcResult()<a name="line.197"></a> -<span class="sourceLineNo">198</span> * which will return the result/state to the client, and it will mark the completed<a name="line.198"></a> -<span class="sourceLineNo">199</span> * proc as ready to delete. note that the client may not receive the response from<a name="line.199"></a> -<span class="sourceLineNo">200</span> * the master (e.g. master failover) so, if we delay a bit the real deletion of<a name="line.200"></a> -<span class="sourceLineNo">201</span> * the proc result the client will be able to get the result the next try.<a name="line.201"></a> -<span class="sourceLineNo">202</span> */<a name="line.202"></a> -<span class="sourceLineNo">203</span> private static class CompletedProcedureCleaner<TEnvironment><a name="line.203"></a> -<span class="sourceLineNo">204</span> extends ProcedureInMemoryChore<TEnvironment> {<a name="line.204"></a> -<span class="sourceLineNo">205</span> private static final Logger LOG = LoggerFactory.getLogger(CompletedProcedureCleaner.class);<a name="line.205"></a> -<span class="sourceLineNo">206</span><a name="line.206"></a> -<span class="sourceLineNo">207</span> private static final String CLEANER_INTERVAL_CONF_KEY = "hbase.procedure.cleaner.interval";<a name="line.207"></a> -<span class="sourceLineNo">208</span> private static final int DEFAULT_CLEANER_INTERVAL = 30 * 1000; // 30sec<a name="line.208"></a> -<span class="sourceLineNo">209</span><a name="line.209"></a> -<span class="sourceLineNo">210</span> private static final String EVICT_TTL_CONF_KEY = "hbase.procedure.cleaner.evict.ttl";<a name="line.210"></a> -<span class="sourceLineNo">211</span> private static final int DEFAULT_EVICT_TTL = 15 * 60000; // 15min<a name="line.211"></a> -<span class="sourceLineNo">212</span><a name="line.212"></a> -<span class="sourceLineNo">213</span> private static final String EVICT_ACKED_TTL_CONF_KEY ="hbase.procedure.cleaner.acked.evict.ttl";<a name="line.213"></a> -<span class="sourceLineNo">214</span> private static final int DEFAULT_ACKED_EVICT_TTL = 5 * 60000; // 5min<a name="line.214"></a> -<span class="sourceLineNo">215</span><a name="line.215"></a> -<span class="sourceLineNo">216</span> private static final String BATCH_SIZE_CONF_KEY = "hbase.procedure.cleaner.evict.batch.size";<a name="line.216"></a> -<span class="sourceLineNo">217</span> private static final int DEFAULT_BATCH_SIZE = 32;<a name="line.217"></a> -<span class="sourceLineNo">218</span><a name="line.218"></a> -<span class="sourceLineNo">219</span> private final Map<Long, CompletedProcedureRetainer<TEnvironment>> completed;<a name="line.219"></a> -<span class="sourceLineNo">220</span> private final Map<NonceKey, Long> nonceKeysToProcIdsMap;<a name="line.220"></a> -<span class="sourceLineNo">221</span> private final ProcedureStore store;<a name="line.221"></a> -<span class="sourceLineNo">222</span> private Configuration conf;<a name="line.222"></a> -<span class="sourceLineNo">223</span><a name="line.223"></a> -<span class="sourceLineNo">224</span> public CompletedProcedureCleaner(Configuration conf, final ProcedureStore store,<a name="line.224"></a> -<span class="sourceLineNo">225</span> final Map<Long, CompletedProcedureRetainer<TEnvironment>> completedMap,<a name="line.225"></a> -<span class="sourceLineNo">226</span> final Map<NonceKey, Long> nonceKeysToProcIdsMap) {<a name="line.226"></a> -<span class="sourceLineNo">227</span> // set the timeout interval that triggers the periodic-procedure<a name="line.227"></a> -<span class="sourceLineNo">228</span> super(conf.getInt(CLEANER_INTERVAL_CONF_KEY, DEFAULT_CLEANER_INTERVAL));<a name="line.228"></a> -<span class="sourceLineNo">229</span> this.completed = completedMap;<a name="line.229"></a> -<span class="sourceLineNo">230</span> this.nonceKeysToProcIdsMap = nonceKeysToProcIdsMap;<a name="line.230"></a> -<span class="sourceLineNo">231</span> this.store = store;<a name="line.231"></a> -<span class="sourceLineNo">232</span> this.conf = conf;<a name="line.232"></a> -<span class="sourceLineNo">233</span> }<a name="line.233"></a> -<span class="sourceLineNo">234</span><a name="line.234"></a> -<span class="sourceLineNo">235</span> @Override<a name="line.235"></a> -<span class="sourceLineNo">236</span> protected void periodicExecute(final TEnvironment env) {<a name="line.236"></a> -<span class="sourceLineNo">237</span> if (completed.isEmpty()) {<a name="line.237"></a> -<span class="sourceLineNo">238</span> if (LOG.isTraceEnabled()) {<a name="line.238"></a> -<span class="sourceLineNo">239</span> LOG.trace("No completed procedures to cleanup.");<a name="line.239"></a> -<span class="sourceLineNo">240</span> }<a name="line.240"></a> -<span class="sourceLineNo">241</span> return;<a name="line.241"></a> -<span class="sourceLineNo">242</span> }<a name="line.242"></a> -<span class="sourceLineNo">243</span><a name="line.243"></a> -<span class="sourceLineNo">244</span> final long evictTtl = conf.getInt(EVICT_TTL_CONF_KEY, DEFAULT_EVICT_TTL);<a name="line.244"></a> -<span class="sourceLineNo">245</span> final long evictAckTtl = conf.getInt(EVICT_ACKED_TTL_CONF_KEY, DEFAULT_ACKED_EVICT_TTL);<a name="line.245"></a> -<span class="sourceLineNo">246</span> final int batchSize = conf.getInt(BATCH_SIZE_CONF_KEY, DEFAULT_BATCH_SIZE);<a name="line.246"></a> -<span class="sourceLineNo">247</span><a name="line.247"></a> -<span class="sourceLineNo">248</span> final long[] batchIds = new long[batchSize];<a name="line.248"></a> -<span class="sourceLineNo">249</span> int batchCount = 0;<a name="line.249"></a> -<span class="sourceLineNo">250</span><a name="line.250"></a> -<span class="sourceLineNo">251</span> final long now = EnvironmentEdgeManager.currentTime();<a name="line.251"></a> -<span class="sourceLineNo">252</span> final Iterator<Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>>> it =<a name="line.252"></a> -<span class="sourceLineNo">253</span> completed.entrySet().iterator();<a name="line.253"></a> -<span class="sourceLineNo">254</span> while (it.hasNext() && store.isRunning()) {<a name="line.254"></a> -<span class="sourceLineNo">255</span> final Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>> entry = it.next();<a name="line.255"></a> -<span class="sourceLineNo">256</span> final CompletedProcedureRetainer<TEnvironment> retainer = entry.getValue();<a name="line.256"></a> -<span class="sourceLineNo">257</span> final Procedure<?> proc = retainer.getProcedure();<a name="line.257"></a> -<span class="sourceLineNo">258</span><a name="line.258"></a> -<span class="sourceLineNo">259</span> // TODO: Select TTL based on Procedure type<a name="line.259"></a> -<span class="sourceLineNo">260</span> if (retainer.isExpired(now, evictTtl, evictAckTtl)) {<a name="line.260"></a> -<span class="sourceLineNo">261</span> // Failed procedures aren't persisted in WAL.<a name="line.261"></a> -<span class="sourceLineNo">262</span> if (!(proc instanceof FailedProcedure)) {<a name="line.262"></a> -<span class="sourceLineNo">263</span> batchIds[batchCount++] = entry.getKey();<a name="line.263"></a> -<span class="sourceLineNo">264</span> if (batchCount == batchIds.length) {<a name="line.264"></a> -<span class="sourceLineNo">265</span> store.delete(batchIds, 0, batchCount);<a name="line.265"></a> -<span class="sourceLineNo">266</span> batchCount = 0;<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> final NonceKey nonceKey = proc.getNonceKey();<a name="line.269"></a> -<span class="sourceLineNo">270</span> if (nonceKey != null) {<a name="line.270"></a> -<span class="sourceLineNo">271</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.271"></a> -<span class="sourceLineNo">272</span> }<a name="line.272"></a> -<span class="sourceLineNo">273</span> it.remove();<a name="line.273"></a> -<span class="sourceLineNo">274</span> LOG.trace("Evict completed {}", proc);<a name="line.274"></a> -<span class="sourceLineNo">275</span> }<a name="line.275"></a> -<span class="sourceLineNo">276</span> }<a name="line.276"></a> -<span class="sourceLineNo">277</span> if (batchCount > 0) {<a name="line.277"></a> -<span class="sourceLineNo">278</span> store.delete(batchIds, 0, batchCount);<a name="line.278"></a> -<span class="sourceLineNo">279</span> }<a name="line.279"></a> -<span class="sourceLineNo">280</span> }<a name="line.280"></a> -<span class="sourceLineNo">281</span> }<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 Procedure.<a name="line.284"></a> -<span class="sourceLineNo">285</span> * Once a Root-Procedure completes (success or failure), the result will be added to this map.<a name="line.285"></a> -<span class="sourceLineNo">286</span> * The user of ProcedureExecutor should call getResult(procId) to get the result.<a name="line.286"></a> -<span class="sourceLineNo">287</span> */<a name="line.287"></a> -<span class="sourceLineNo">288</span> private final ConcurrentHashMap<Long, CompletedProcedureRetainer<TEnvironment>> completed =<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> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the RootProcedureState.<a name="line.292"></a> -<span class="sourceLineNo">293</span> * The RootProcedureState contains the execution stack of the Root-Procedure,<a name="line.293"></a> -<span class="sourceLineNo">294</span> * It is added to the map by submitProcedure() and removed on procedure completion.<a name="line.294"></a> -<span class="sourceLineNo">295</span> */<a name="line.295"></a> -<span class="sourceLineNo">296</span> private final ConcurrentHashMap<Long, RootProcedureState<TEnvironment>> rollbackStack =<a name="line.296"></a> -<span class="sourceLineNo">297</span> new ConcurrentHashMap<>();<a name="line.297"></a> -<span class="sourceLineNo">298</span><a name="line.298"></a> -<span class="sourceLineNo">299</span> /**<a name="line.299"></a> -<span class="sourceLineNo">300</span> * Helper map to lookup the live procedures by ID.<a name="line.300"></a> -<span class="sourceLineNo">301</span> * This map contains every procedure. root-procedures and subprocedures.<a name="line.301"></a> -<span class="sourceLineNo">302</span> */<a name="line.302"></a> -<span class="sourceLineNo">303</span> private final ConcurrentHashMap<Long, Procedure<TEnvironment>> procedures =<a name="line.303"></a> -<span class="sourceLineNo">304</span> new ConcurrentHashMap<>();<a name="line.304"></a> -<span class="sourceLineNo">305</span><a name="line.305"></a> -<span class="sourceLineNo">306</span> /**<a name="line.306"></a> -<span class="sourceLineNo">307</span> * Helper map to lookup whether the procedure already issued from the same client. This map<a name="line.307"></a> -<span class="sourceLineNo">308</span> * contains every root procedure.<a name="line.308"></a> -<span class="sourceLineNo">309</span> */<a name="line.309"></a> -<span class="sourceLineNo">310</span> private final ConcurrentHashMap<NonceKey, Long> nonceKeysToProcIdsMap = new ConcurrentHashMap<>();<a name="line.310"></a> -<span class="sourceLineNo">311</span><a name="line.311"></a> -<span class="sourceLineNo">312</span> private final CopyOnWriteArrayList<ProcedureExecutorListener> listeners =<a name="line.312"></a> -<span class="sourceLineNo">313</span> new CopyOnWriteArrayList<>();<a name="line.313"></a> -<span class="sourceLineNo">314</span><a name="line.314"></a> -<span class="sourceLineNo">315</span> private Configuration conf;<a name="line.315"></a> -<span class="sourceLineNo">316</span><a name="line.316"></a> -<span class="sourceLineNo">317</span> /**<a name="line.317"></a> -<span class="sourceLineNo">318</span> * Created in the {@link #start(int, boolean)} method. Destroyed 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 ThreadGroup threadGroup;<a name="line.323"></a> -<span class="sourceLineNo">324</span><a name="line.324"></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 CopyOnWriteArrayList<WorkerThread> workerThreads;<a name="line.331"></a> -<span class="sourceLineNo">332</span><a name="line.332"></a> -<span class="sourceLineNo">333</span> /**<a name="line.333"></a> -<span class="sourceLineNo">334</span> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.334"></a> -<span class="sourceLineNo">335</span> * resource handling rather than observing in a #join is unexpected).<a name="line.335"></a> -<span class="sourceLineNo">336</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.336"></a> -<span class="sourceLineNo">337</span> * (Should be ok).<a name="line.337"></a> -<span class="sourceLineNo">338</span> */<a name="line.338"></a> -<span class="sourceLineNo">339</span> private TimeoutExecutorThread<TEnvironment> timeoutExecutor;<a name="line.339"></a> -<span class="sourceLineNo">340</span><a name="line.340"></a> -<span class="sourceLineNo">341</span> private int corePoolSize;<a name="line.341"></a> -<span class="sourceLineNo">342</span> private int maxPoolSize;<a name="line.342"></a> -<span class="sourceLineNo">343</span><a name="line.343"></a> -<span class="sourceLineNo">344</span> private volatile long keepAliveTime;<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> * Scheduler/Queue that contains runnable procedures.<a name="line.347"></a> -<span class="sourceLineNo">348</span> */<a name="line.348"></a> -<span class="sourceLineNo">349</span> private final ProcedureScheduler scheduler;<a name="line.349"></a> -<span class="sourceLineNo">350</span><a name="line.350"></a> -<span class="sourceLineNo">351</span> private final AtomicLong lastProcId = new AtomicLong(-1);<a name="line.351"></a> -<span class="sourceLineNo">352</span> private final AtomicLong workerId = new AtomicLong(0);<a name="line.352"></a> -<span class="sourceLineNo">353</span> private final AtomicInteger activeExecutorCount = new AtomicInteger(0);<a name="line.353"></a> -<span class="sourceLineNo">354</span> private final AtomicBoolean running = new AtomicBoolean(false);<a name="line.354"></a> -<span class="sourceLineNo">355</span> private final TEnvironment environment;<a name="line.355"></a> -<span class="sourceLineNo">356</span> private final ProcedureStore store;<a name="line.356"></a> -<span class="sourceLineNo">357</span><a name="line.357"></a> -<span class="sourceLineNo">358</span> private final boolean checkOwnerSet;<a name="line.358"></a> -<span class="sourceLineNo">359</span><a name="line.359"></a> -<span class="sourceLineNo">360</span> // To prevent concurrent execution of the same procedure.<a name="line.360"></a> -<span class="sourceLineNo">361</span> // For some rare cases, especially if the procedure uses ProcedureEvent, it is possible that the<a name="line.361"></a> -<span class="sourceLineNo">362</span> // procedure is woken up before we finish the suspend which causes the same procedures to be<a name="line.362"></a> -<span class="sourceLineNo">363</span> // executed in parallel. This does lead to some problems, see HBASE-20939&HBASE-20949, and is also<a name="line.363"></a> -<span class="sourceLineNo">364</span> // a bit confusing to the developers. So here we introduce this lock to prevent the concurrent<a name="line.364"></a> -<span class="sourceLineNo">365</span> // execution of the same procedure.<a name="line.365"></a> -<span class="sourceLineNo">366</span> private final IdLock procExecutionLock = new IdLock();<a name="line.366"></a> -<span class="sourceLineNo">367</span><a name="line.367"></a> -<span class="sourceLineNo">368</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.368"></a> -<span class="sourceLineNo">369</span> final ProcedureStore store) {<a name="line.369"></a> -<span class="sourceLineNo">370</span> this(conf, environment, store, new SimpleProcedureScheduler());<a name="line.370"></a> -<span class="sourceLineNo">371</span> }<a name="line.371"></a> -<span class="sourceLineNo">372</span><a name="line.372"></a> -<span class="sourceLineNo">373</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.373"></a> -<span class="sourceLineNo">374</span> final ProcedureStore store, final ProcedureScheduler scheduler) {<a name="line.374"></a> -<span class="sourceLineNo">375</span> this.environment = environment;<a name="line.375"></a> -<span class="sourceLineNo">376</span> this.scheduler = scheduler;<a name="line.376"></a> -<span class="sourceLineNo">377</span> this.store = store;<a name="line.377"></a> -<span class="sourceLineNo">378</span> this.conf = conf;<a name="line.378"></a> -<span class="sourceLineNo">379</span> this.checkOwnerSet = conf.getBoolean(CHECK_OWNER_SET_CONF_KEY, DEFAULT_CHECK_OWNER_SET);<a name="line.379"></a> -<span class="sourceLineNo">380</span> refreshConfiguration(conf);<a name="line.380"></a> -<span class="sourceLineNo">381</span><a name="line.381"></a> -<span class="sourceLineNo">382</span> }<a name="line.382"></a> -<span class="sourceLineNo">383</span><a name="line.383"></a> -<span class="sourceLineNo">384</span> private void load(final boolean abortOnCorruption) throws IOException {<a name="line.384"></a> -<span class="sourceLineNo">385</span> Preconditions.checkArgument(completed.isEmpty(), "completed not empty");<a name="line.385"></a> -<span class="sourceLineNo">386</span> Preconditions.checkArgument(rollbackStack.isEmpty(), "rollback state not empty");<a name="line.386"></a> -<span class="sourceLineNo">387</span> Preconditions.checkArgument(procedures.isEmpty(), "procedure map not empty");<a name="line.387"></a> -<span class="sourceLineNo">388</span> Preconditions.checkArgument(scheduler.size() == 0, "run queue not empty");<a name="line.388"></a> -<span class="sourceLineNo">389</span><a name="line.389"></a> -<span class="sourceLineNo">390</span> store.load(new ProcedureStore.ProcedureLoader() {<a name="line.390"></a> -<span class="sourceLineNo">391</span> @Override<a name="line.391"></a> -<span class="sourceLineNo">392</span> public void setMaxProcId(long maxProcId) {<a name="line.392"></a> -<span class="sourceLineNo">393</span> assert lastProcId.get() < 0 : "expected only one call to setMaxProcId()";<a name="line.393"></a> -<span class="sourceLineNo">394</span> lastProcId.set(maxProcId);<a name="line.394"></a> -<span class="sourceLineNo">395</span> }<a name="line.395"></a> -<span class="sourceLineNo">396</span><a name="line.396"></a> -<span class="sourceLineNo">397</span> @Override<a name="line.397"></a> -<span class="sourceLineNo">398</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.398"></a> -<span class="sourceLineNo">399</span> loadProcedures(procIter, abortOnCorruption);<a name="line.399"></a> -<span class="sourceLineNo">400</span> }<a name="line.400"></a> -<span class="sourceLineNo">401</span><a name="line.401"></a> -<span class="sourceLineNo">402</span> @Override<a name="line.402"></a> -<span class="sourceLineNo">403</span> public void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.403"></a> -<span class="sourceLineNo">404</span> int corruptedCount = 0;<a name="line.404"></a> -<span class="sourceLineNo">405</span> while (procIter.hasNext()) {<a name="line.405"></a> -<span class="sourceLineNo">406</span> Procedure<?> proc = procIter.next();<a name="line.406"></a> -<span class="sourceLineNo">407</span> LOG.error("Corrupt " + proc);<a name="line.407"></a> -<span class="sourceLineNo">408</span> corruptedCount++;<a name="line.408"></a> -<span class="sourceLineNo">409</span> }<a name="line.409"></a> -<span class="sourceLineNo">410</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.410"></a> -<span class="sourceLineNo">411</span> throw new IOException("found " + corruptedCount + " corrupted procedure(s) on replay");<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> });<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> private void restoreLock(Procedure<TEnvironment> proc, Set<Long> restored) {<a name="line.417"></a> -<span class="sourceLineNo">418</span> proc.restoreLock(getEnvironment());<a name="line.418"></a> -<span class="sourceLineNo">419</span> restored.add(proc.getProcId());<a name="line.419"></a> -<span class="sourceLineNo">420</span> }<a name="line.420"></a> -<span class="sourceLineNo">421</span><a name="line.421"></a> -<span class="sourceLineNo">422</span> private void restoreLocks(Deque<Procedure<TEnvironment>> stack, Set<Long> restored) {<a name="line.422"></a> -<span class="sourceLineNo">423</span> while (!stack.isEmpty()) {<a name="line.423"></a> -<span class="sourceLineNo">424</span> restoreLock(stack.pop(), restored);<a name="line.424"></a> -<span class="sourceLineNo">425</span> }<a name="line.425"></a> -<span class="sourceLineNo">426</span> }<a name="line.426"></a> -<span class="sourceLineNo">427</span><a name="line.427"></a> -<span class="sourceLineNo">428</span> // Restore the locks for all the procedures.<a name="line.428"></a> -<span class="sourceLineNo">429</span> // Notice that we need to restore the locks starting from the root proc, otherwise there will be<a name="line.429"></a> -<span class="sourceLineNo">430</span> // problem that a sub procedure may hold the exclusive lock first and then we are stuck when<a name="line.430"></a> -<span class="sourceLineNo">431</span> // calling the acquireLock method for the parent procedure.<a name="line.431"></a> -<span class="sourceLineNo">432</span> // The algorithm is straight-forward:<a name="line.432"></a> -<span class="sourceLineNo">433</span> // 1. Use a set to record the procedures which locks have already been restored.<a name="line.433"></a> -<span class="sourceLineNo">434</span> // 2. Use a stack to store the hierarchy of the procedures<a name="line.434"></a> -<span class="sourceLineNo">435</span> // 3. For all the procedure, we will first try to find its parent and push it into the stack,<a name="line.435"></a> -<span class="sourceLineNo">436</span> // unless<a name="line.436"></a> -<span class="sourceLineNo">437</span> // a. We have no parent, i.e, we are the root procedure<a name="line.437"></a> -<span class="sourceLineNo">438</span> // b. The lock has already been restored(by checking the set introduced in #1)<a name="line.438"></a> -<span class="sourceLineNo">439</span> // then we start to pop the stack and call acquireLock for each procedure.<a name="line.439"></a> -<span class="sourceLineNo">440</span> // Notice that this should be done for all procedures, not only the ones in runnableList.<a name="line.440"></a> -<span class="sourceLineNo">441</span> private void restoreLocks() {<a name="line.441"></a> -<span class="sourceLineNo">442</span> Set<Long> restored = new HashSet<>();<a name="line.442"></a> -<span class="sourceLineNo">443</span> Deque<Procedure<TEnvironment>> stack = new ArrayDeque<>();<a name="line.443"></a> -<span class="sourceLineNo">444</span> procedures.values().forEach(proc -> {<a name="line.444"></a> -<span class="sourceLineNo">445</span> for (;;) {<a name="line.445"></a> -<span class="sourceLineNo">446</span> if (restored.contains(proc.getProcId())) {<a name="line.446"></a> -<span class="sourceLineNo">447</span> restoreLocks(stack, restored);<a name="line.447"></a> -<span class="sourceLineNo">448</span> return;<a name="line.448"></a> -<span class="sourceLineNo">449</span> }<a name="line.449"></a> -<span class="sourceLineNo">450</span> if (!proc.hasParent()) {<a name="line.450"></a> -<span class="sourceLineNo">451</span> restoreLock(proc, restored);<a name="line.451"></a> -<span class="sourceLineNo">452</span> restoreLocks(stack, restored);<a name="line.452"></a> -<span class="sourceLineNo">453</span> return;<a name="line.453"></a> -<span class="sourceLineNo">454</span> }<a name="line.454"></a> -<span class="sourceLineNo">455</span> stack.push(proc);<a name="line.455"></a> -<span class="sourceLineNo">456</span> proc = procedures.get(proc.getParentProcId());<a name="line.456"></a> -<span class="sourceLineNo">457</span> }<a name="line.457"></a> -<span class="sourceLineNo">458</span> });<a name="line.458"></a> -<span class="sourceLineNo">459</span> }<a name="line.459"></a> -<span class="sourceLineNo">460</span><a name="line.460"></a> -<span class="sourceLineNo">461</span> private void loadProcedures(ProcedureIterator procIter, boolean abortOnCorruption)<a name="line.461"></a> -<span class="sourceLineNo">462</span> throws IOException {<a name="line.462"></a> -<span class="sourceLineNo">463</span> // 1. Build the rollback stack<a name="line.463"></a> -<span class="sourceLineNo">464</span> int runnablesCount = 0;<a name="line.464"></a> -<span class="sourceLineNo">465</span> int failedCount = 0;<a name="line.465"></a> -<span class="sourceLineNo">466</span> while (procIter.hasNext()) {<a name="line.466"></a> -<span class="sourceLineNo">467</span> boolean finished = procIter.isNextFinished();<a name="line.467"></a> -<span class="sourceLineNo">468</span> @SuppressWarnings("unchecked")<a name="line.468"></a> -<span class="sourceLineNo">469</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.469"></a> -<span class="sourceLineNo">470</span> NonceKey nonceKey = proc.getNonceKey();<a name="line.470"></a> -<span class="sourceLineNo">471</span> long procId = proc.getProcId();<a name="line.471"></a> -<span class="sourceLineNo">472</span><a name="line.472"></a> -<span class="sourceLineNo">473</span> if (finished) {<a name="line.473"></a> -<span class="sourceLineNo">474</span> completed.put(proc.getProcId(), new CompletedProcedureRetainer<>(proc));<a name="line.474"></a> -<span class="sourceLineNo">475</span> LOG.debug("Completed {}", proc);<a name="line.475"></a> -<span class="sourceLineNo">476</span> } else {<a name="line.476"></a> -<span class="sourceLineNo">477</span> if (!proc.hasParent()) {<a name="line.477"></a> -<span class="sourceLineNo">478</span> assert !proc.isFinished() : "unexpected finished procedure";<a name="line.478"></a> -<span class="sourceLineNo">479</span> rollbackStack.put(proc.getProcId(), new RootProcedureState<>());<a name="line.479"></a> -<span class="sourceLineNo">480</span> }<a name="line.480"></a> -<span class="sourceLineNo">481</span><a name="line.481"></a> -<span class="sourceLineNo">482</span> // add the procedure to the map<a name="line.482"></a> -<span class="sourceLineNo">483</span> proc.beforeReplay(getEnvironment());<a name="line.483"></a> -<span class="sourceLineNo">484</span> procedures.put(proc.getProcId(), proc);<a name="line.484"></a> -<span class="sourceLineNo">485</span><a name="line.485"></a> -<span class="sourceLineNo">486</span> if (proc.getState() == ProcedureState.RUNNABLE) {<a name="line.486"></a> -<span class="sourceLineNo">487</span> runnablesCount++;<a name="line.487"></a> -<span class="sourceLineNo">488</span> } else if (proc.getState() == ProcedureState.FAILED) {<a name="line.488"></a> -<span class="sourceLineNo">489</span> failedCount++;<a name="line.489"></a> -<span class="sourceLineNo">490</span> }<a name="line.490"></a> -<span class="sourceLineNo">491</span> }<a name="line.491"></a> -<span class="sourceLineNo">492</span><a name="line.492"></a> -<span class="sourceLineNo">493</span> // add the nonce to the map<a name="line.493"></a> -<span class="sourceLineNo">494</span> if (nonceKey != null) {<a name="line.494"></a> -<span class="sourceLineNo">495</span> nonceKeysToProcIdsMap.put(nonceKey, procId);<a name="line.495"></a> -<span class="sourceLineNo">496</span> }<a name="line.496"></a> -<span class="sourceLineNo">497</span> }<a name="line.497"></a> -<span class="sourceLineNo">498</span><a name="line.498"></a> -<span class="sourceLineNo">499</span> // 2. Initialize the stacks<a name="line.499"></a> -<span class="sourceLineNo">500</span> // In the old implementation, for procedures in FAILED state, we will push it into the<a name="line.500"></a> -<span class="sourceLineNo">501</span> // ProcedureScheduler directly to execute the rollback. But this does not work after we<a name="line.501"></a> -<span class="sourceLineNo">502</span> // introduce the restore lock stage.<a name="line.502"></a> -<span class="sourceLineNo">503</span> // For now, when we acquire a xlock, we will remove the queue from runQueue in scheduler, and<a name="line.503"></a> -<span class="sourceLineNo">504</span> // then when a procedure which has lock access, for example, a sub procedure of the procedure<a name="line.504"></a> -<span class="sourceLineNo">505</span> // which has the xlock, is pushed into the scheduler, we will add the queue back to let the<a name="line.505"></a> -<span class="sourceLineNo">506</span> // workers poll from it. The assumption here is that, the procedure which has the xlock should<a name="line.506"></a> -<span class="sourceLineNo">507</span> // have been polled out already, so when loading we can not add the procedure to scheduler first<a name="line.507"></a> -<span class="sourceLineNo">508</span> // and then call acquireLock, since the procedure is still in the queue, and since we will<a name="line.508"></a> -<span class="sourceLineNo">509</span> // remove the queue from runQueue, then no one can poll it out, then there is a dead lock<a name="line.509"></a> -<span class="sourceLineNo">510</span> List<Procedure<TEnvironment>> runnableList = new ArrayList<>(runnablesCount);<a name="line.510"></a> -<span class="sourceLineNo">511</span> List<Procedure<TEnvironment>> failedList = new ArrayList<>(failedCount);<a name="line.511"></a> -<span class="sourceLineNo">512</span> Set<Procedure<TEnvironment>> waitingSet = null;<a name="line.512"></a> -<span class="sourceLineNo">513</span> procIter.reset();<a name="line.513"></a> -<span class="sourceLineNo">514</span> while (procIter.hasNext()) {<a name="line.514"></a> -<span class="sourceLineNo">515</span> if (procIter.isNextFinished()) {<a name="line.515"></a> -<span class="sourceLineNo">516</span> procIter.skipNext();<a name="line.516"></a> -<span class="sourceLineNo">517</span> continue;<a name="line.517"></a> -<span class="sourceLineNo">518</span> }<a name="line.518"></a> -<span class="sourceLineNo">519</span><a name="line.519"></a> -<span class="sourceLineNo">520</span> @SuppressWarnings("unchecked")<a name="line.520"></a> -<span class="sourceLineNo">521</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.521"></a> -<span class="sourceLineNo">522</span> assert !(proc.isFinished() && !proc.hasParent()) : "unexpected completed proc=" + proc;<a name="line.522"></a> -<span class="sourceLineNo">523</span><a name="line.523"></a> -<span class="sourceLineNo">524</span> LOG.debug("Loading {}", proc);<a name="line.524"></a> -<span class="sourceLineNo">525</span><a name="line.525"></a> -<span class="sourceLineNo">526</span> Long rootProcId = getRootProcedureId(proc);<a name="line.526"></a> -<span class="sourceLineNo">527</span> if (rootProcId == null) {<a name="line.527"></a> -<span class="sourceLineNo">528</span> // The 'proc' was ready to run but the root procedure was rolledback?<a name="line.528"></a> -<span class="sourceLineNo">529</span> scheduler.addBack(proc);<a name="line.529"></a> -<span class="sourceLineNo">530</span> continue;<a name="line.530"></a> -<span class="sourceLineNo">531</span> }<a name="line.531"></a> -<span class="sourceLineNo">532</span><a name="line.532"></a> -<span class="sourceLineNo">533</span> if (proc.hasParent()) {<a name="line.533"></a> -<span class="sourceLineNo">534</span> Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());<a name="line.534"></a> -<span class="sourceLineNo">535</span> // corrupted procedures are handled later at step 3<a name="line.535"></a> -<span class="sourceLineNo">536</span> if (parent != null && !proc.isFinished()) {<a name="line.536"></a> -<span class="sourceLineNo">537</span> parent.incChildrenLatch();<a name="line.537"></a> -<span class="sourceLineNo">538</span> }<a name="line.538"></a> -<span class="sourceLineNo">539</span> }<a name="line.539"></a> -<span class="sourceLineNo">540</span><a name="line.540"></a> -<span class="sourceLineNo">541</span> RootProcedureState<TEnvironment> procStack = rollbackStack.get(rootProcId);<a name="line.541"></a> -<span class="sourceLineNo">542</span> procStack.loadStack(proc);<a name="line.542"></a> -<span class="sourceLineNo">543</span><a name="line.543"></a> -<span class="sourceLineNo">544</span> proc.setRootProcId(rootProcId);<a name="line.544"></a> -<span class="sourceLineNo">545</span> switch (proc.getState()) {<a name="line.545"></a> -<span class="sourceLineNo">546</span> case RUNNABLE:<a name="line.546"></a> -<span class="sourceLineNo">547</span> runnableList.add(proc);<a name="line.547"></a> -<span class="sourceLineNo">548</span> break;<a name="line.548"></a> -<span class="sourceLineNo">549</span> case WAITING:<a name="line.549"></a> -<span class="sourceLineNo">550</span> if (!proc.hasChildren()) {<a name="line.550"></a> -<span class="sourceLineNo">551</span> // Normally, WAITING procedures should be waken by its children.<a name="line.551"></a> -<span class="sourceLineNo">552</span> // But, there is a case that, all the children are successful and before<a name="line.552"></a> -<span class="sourceLineNo">553</span> // they can wake up their parent procedure, the master was killed.<a name="line.553"></a> -<span class="sourceLineNo">554</span> // So, during recovering the procedures from ProcedureWal, its children<a name="line.554"></a> -<span class="sourceLineNo">555</span> // are not loaded because of their SUCCESS state.<a name="line.555"></a> -<span class="sourceLineNo">556</span> // So we need to continue to run this WAITING procedure. But before<a name="line.556"></a> -<span class="sourceLineNo">557</span> // executing, we need to set its state to RUNNABLE, otherwise, a exception<a name="line.557"></a> -<span class="sourceLineNo">558</span> // will throw:<a name="line.558"></a> -<span class="sourceLineNo">559</span> // Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,<a name="line.559"></a> -<span class="sourceLineNo">560</span> // "NOT RUNNABLE! " + procedure.toString());<a name="line.560"></a> -<span class="sourceLineNo">561</span> proc.setState(ProcedureState.RUNNABLE);<a name="line.561"></a> -<span class="sourceLineNo">562</span> runnableList.add(proc);<a name="line.562"></a> -<span class="sourceLineNo">563</span> }<a name="line.563"></a> -<span class="sourceLineNo">564</span> break;<a name="line.564"></a> -<span class="sourceLineNo">565</span> case WAITING_TIMEOUT:<a name="line.565"></a> -<span class="sourceLineNo">566</span> if (waitingSet == null) {<a name="line.566"></a> -<span class="sourceLineNo">567</span> waitingSet = new HashSet<>();<a name="line.567"></a> -<span class="sourceLineNo">568</span> }<a name="line.568"></a> -<span class="sourceLineNo">569</span> waitingSet.add(proc);<a name="line.569"></a> -<span class="sourceLineNo">570</span> break;<a name="line.570"></a> -<span class="sourceLineNo">571</span> case FAILED:<a name="line.571"></a> -<span class="sourceLineNo">572</span> failedList.add(proc);<a name="line.572"></a> -<span class="sourceLineNo">573</span> break;<a name="line.573"></a> -<span class="sourceLineNo">574</span> case ROLLEDBACK:<a name="line.574"></a> -<span class="sourceLineNo">575</span> case INITIALIZING:<a name="line.575"></a> -<span class="sourceLineNo">576</span> String msg = "Unexpected " + proc.getState() + " state for " + proc;<a name="line.576"></a> -<span class="sourceLineNo">577</span> LOG.error(msg);<a name="line.577"></a> -<span class="sourceLineNo">578</span> throw new UnsupportedOperationException(msg);<a name="line.578"></a> -<span class="sourceLineNo">579</span> default:<a name="line.579"></a> -<span class="sourceLineNo">580</span> break;<a name="line.580"></a> -<span class="sourceLineNo">581</span> }<a name="line.581"></a> -<span class="sourceLineNo">582</span> }<a name="line.582"></a> -<span class="sourceLineNo">583</span><a name="line.583"></a> -<span class="sourceLineNo">584</span> // 3. Validate the stacks<a name="line.584"></a> -<span class="sourceLineNo">585</span> int corruptedCount = 0;<a name="line.585"></a> -<span class="sourceLineNo">586</span> Iterator<Map.Entry<Long, RootProcedureState<TEnvironment>>> itStack =<a name="line.586"></a> -<span class="sourceLineNo">587</span> rollbackStack.entrySet().iterator();<a name="line.587"></a> -<span class="sourceLineNo">588</span> while (itStack.hasNext()) {<a name="line.588"></a> -<span class="sourceLineNo">589</span> Map.Entry<Long, RootProcedureState<TEnvironment>> entry = itStack.next();<a name="line.589"></a> -<span class="sourceLineNo">590</span> RootProcedureState<TEnvironment> procStack = entry.getValue();<a name="line.590"></a> -<span class="sourceLineNo">591</span> if (procStack.isValid()) continue;<a name="line.591"></a> -<span class="sourceLineNo">592</span><a name="line.592"></a> -<span class="sourceLineNo">593</span> for (Procedure<TEnvironment> proc : procStack.getSubproceduresStack()) {<a name="line.593"></a> -<span class="sourceLineNo">594</span> LOG.error("Corrupted " + proc);<a name="line.594"></a> -<span class="sourceLineNo">595</span> procedures.remove(proc.getProcId());<a name="line.595"></a> -<span class="sourceLineNo">596</span> runnableList.remove(proc);<a name="line.596"></a> -<span class="sourceLineNo">597</span> if (waitingSet != null) waitingSet.remove(proc);<a name="line.597"></a> -<span class="sourceLineNo">598</span> corruptedCount++;<a name="line.598"></a> -<span class="sourceLineNo">599</span> }<a name="line.599"></a> -<span class="sourceLineNo">600</span> itStack.remove();<a name="line.600"></a> -<span class="sourceLineNo">601</span> }<a name="line.601"></a> -<span class="sourceLineNo">602</span><a name="line.602"></a> -<span class="sourceLineNo">603</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.603"></a> -<span class="sourceLineNo">604</span> throw new IOException("found " + corruptedCount + " procedures on replay");<a name="line.604"></a> -<span class="sourceLineNo">605</span> }<a name="line.605"></a> -<span class="sourceLineNo">606</span><a name="line.606"></a> -<span class="sourceLineNo">607</span> // 4. Push the procedures to the timeout executor<a name="line.607"></a> -<span class="sourceLineNo">608</span> if (waitingSet != null && !waitingSet.isEmpty()) {<a name="line.608"></a> -<span class="sourceLineNo">609</span> for (Procedure<TEnvironment> proc: waitingSet) {<a name="line.609"></a> -<span class="sourceLineNo">610</span> proc.afterReplay(getEnvironment());<a name="line.610"></a> -<span class="sourceLineNo">611</span> timeoutExecutor.add(proc);<a name="line.611"></a> -<span class="sourceLineNo">612</span> }<a name="line.612"></a> -<span class="sourceLineNo">613</span> }<a name="line.613"></a> -<span class="sourceLineNo">614</span> // 5. restore locks<a name="line.614"></a> -<span class="sourceLineNo">615</span> restoreLocks();<a name="line.615"></a> -<span class="sourceLineNo">616</span> // 6. Push the procedure to the scheduler<a name="line.616"></a> -<span class="sourceLineNo">617</span> failedList.forEach(scheduler::addBack);<a name="line.617"></a> -<span class="sourceLineNo">618</span> runnableList.forEach(p -> {<a name="line.618"></a> -<span class="sourceLineNo">619</span> p.afterReplay(getEnvironment());<a name="line.619"></a> -<span class="sourceLineNo">620</span> if (!p.hasParent()) {<a name="line.620"></a> -<span class="sourceLineNo">621</span> sendProcedureLoadedNotification(p.getProcId());<a name="line.621"></a> -<span class="sourceLineNo">622</span> }<a name="line.622"></a> -<span class="sourceLineNo">623</span> scheduler.addBack(p);<a name="line.623"></a> -<span class="sourceLineNo">624</span> });<a name="line.624"></a> -<span class="sourceLineNo">625</span> }<a name="line.625"></a> -<span class="sourceLineNo">626</span><a name="line.626"></a> -<span class="sourceLineNo">627</span> /**<a name="line.627"></a> -<span class="sourceLineNo">628</span> * Initialize the procedure executor, but do not start workers. We will start them later.<a name="line.628"></a> -<span class="sourceLineNo">629</span> * <p/><a name="line.629"></a> -<span class="sourceLineNo">630</span> * It calls ProcedureStore.recoverLease() and ProcedureStore.load() to recover the lease, and<a name="line.630"></a> -<span class="sourceLineNo">631</span> * ensure a single executor, and start the procedure replay to resume and recover the previous<a name="line.631"></a> -<span class="sourceLineNo">632</span> * pending and in-progress procedures.<a name="line.632"></a> -<span class="sourceLineNo">633</span> * @param numThreads number of threads available for procedure execution.<a name="line.633"></a> -<span class="sourceLineNo">634</span> * @param abortOnCorruption true if you want to abort your service in case a corrupted procedure<a name="line.634"></a> -<span class="sourceLineNo">635</span> * is found on replay. otherwise false.<a name="line.635"></a> -<span class="sourceLineNo">636</span> */<a name="line.636"></a> -<span class="sourceLineNo">637</span> public void init(int numThreads, boolean abortOnCorruption) throws IOException {<a name="line.637"></a> -<span class="sourceLineNo">638</span> // We have numThreads executor + one timer thread used for timing out<a name="line.638"></a> -<span class="sourceLineNo">639</span> // procedures and triggering periodic procedures.<a name="line.639"></a> -<span class="sourceLineNo">640</span> this.corePoolSize = numThreads;<a name="line.640"></a> -<span class="sourceLineNo">641</span> this.maxPoolSize = 10 * numThreads;<a name="line.641"></a> -<span class="sourceLineNo">642</span> LOG.info("Starting {} core workers (bigger of cpus/4 or 16) with max (burst) worker count={}",<a name="line.642"></a> -<span class="sourceLineNo">643</span> corePoolSize, maxPoolSize);<a name="line.643"></a> -<span class="sourceLineNo">644</span><a name="line.644"></a> -<span class="sourceLineNo">645</span> this.threadGroup = new ThreadGroup("PEWorkerGroup");<a name="line.645"></a> -<span class="sourceLineNo">646</span> this.timeoutExecutor = new TimeoutExecutorThread<>(this, threadGroup);<a name="line.646"></a> -<span class="sourceLineNo">647</span><a name="line.647"></a> -<span class="sourceLineNo">648</span> // Create the workers<a name="line.648"></a> -<span class="sourceLineNo">649</span> workerId.set(0);<a name="line.649"></a> -<span class="sourceLineNo">650</span> workerThreads = new CopyOnWriteArrayList<>();<a name="line.650"></a> -<span class="sourceLineNo">651</span> for (int i = 0; i < corePoolSize; ++i) {<a name="line.651"></a> -<span class="sourceLineNo">652</span> workerThreads.add(new WorkerThread(threadGroup));<a name="line.652"></a> -<span class="sourceLineNo">653</span> }<a name="line.653"></a> -<span class="sourceLineNo">654</span><a name="line.654"></a> -<span class="sourceLineNo">655</span> long st, et;<a name="line.655"></a> -<span class="sourceLineNo">656</span><a name="line.656"></a> -<span class="sourceLineNo">657</span> // Acquire the store lease.<a name="line.657"></a> -<span class="sourceLineNo">658</span> st = System.nanoTime();<a name="line.658"></a> -<span class="sourceLineNo">659</span> store.recoverLease();<a name="line.659"></a> -<span class="sourceLineNo">660</span> et = System.nanoTime();<a name="line.660"></a> -<span class="sourceLineNo">661</span> LOG.info("Recovered {} lease in {}", store.getClass().getSimpleName(),<a name="line.661"></a> -<span class="sourceLineNo">662</span> StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.662"></a> -<span class="sourceLineNo">663</span><a name="line.663"></a> -<span class="sourceLineNo">664</span> // start the procedure scheduler<a name="line.664"></a> -<span class="sourceLineNo">665</span> scheduler.start();<a name="line.665"></a> -<span class="sourceLineNo">666</span><a name="line.666"></a> -<span class="sourceLineNo">667</span> // TODO: Split in two steps.<a name="line.667"></a> -<span class="sourceLineNo">668</span> // TODO: Handle corrupted procedures (currently just a warn)<a name="line.668"></a> -<span class="sourceLineNo">669</span> // The first one will make sure that we have the latest id,<a name="line.669"></a> -<span class="sourceLineNo">670</span> // so we can start the threads and accept new procedures.<a name="line.670"></a> -<span class="sourceLineNo">671</span> // The second step will do the actual load of old procedures.<a name="line.671"></a> -<span class="sourceLineNo">672</span> st = System.nanoTime();<a name="line.672"></a> -<span class="sourceLineNo">673</span> load(abortOnCorruption);<a name="line.673"></a> -<span class="sourceLineNo">674</span> et = System.nanoTime();<a name="line.674"></a> -<span class="sourceLineNo">675</span> LOG.info("Loaded {} in {}", store.getClass().getSimpleName(),<a name="line.675"></a> -<span class="sourceLineNo">676</span> StringUtils.humanTimeDiff(TimeUnit.NANOSECONDS.toMillis(et - st)));<a name="line.676"></a> -<span class="sourceLineNo">677</span> }<a name="line.677"></a> -<span class="sourceLineNo">678</span><a name="line.678"></a> -<span class="sourceLineNo">679</span> /**<a name="line.679"></a> -<span class="sourceLineNo">680</span> * Start the workers.<a name="line.680"></a> -<span class="sourceLineNo">681</span> */<a name="line.681"></a> -<span class="sourceLineNo">682</span> public void startWorkers() throws IOException {<a name="line.682"></a> -<span class="sourceLineNo">683</span> if (!running.compareAndSet(false, true)) {<a name="line.683"></a> -<span class="sourceLineNo">684</span> LOG.warn("Already running");<a name="line.684"></a> -<span class="sourceLineNo">685</span> return;<a name="line.685"></a> -<span class="sourceLineNo">686</span> }<a name="line.686"></a> -<span class="sourceLineNo">687</span> // Start the executors. Here we must have the lastProcId set.<a name="line.687"></a> -<span class="sourceLineNo">688</span> LOG.trace("Start workers {}", workerThreads.size());<a name="line.688"></a> -<span class="sourceLineNo">689</span> timeoutExecutor.start();<a name="line.689"></a> -<span class="sourceLineNo">690</span> for (WorkerThread worker: workerThreads) {<a name="line.690"></a> -<span class="sourceLineNo">691</span> worker.start();<a name="line.691"></a> -<span class="sourceLineNo">692</span> }<a name="line.692"></a> -<span class="sourceLineNo">693</span><a name="line.693"></a> -<span class="sourceLineNo">694</span> // Internal chores<a name="line.694"></a> -<span class="sourceLineNo">695</span> timeoutExecutor.add(new WorkerMonitor());<a name="line.695"></a> -<span class="sourceLineNo">696</span><a name="line.696"></a> -<span class="sourceLineNo">697</span> // Add completed cleaner chore<a name="line.697"></a> -<span class="sourceLineNo">698</span> addChore(new CompletedProcedureCleaner<>(conf, store, completed, nonceKeysToProcIdsMap));<a name="line.698"></a> -<span class="sourceLineNo">699</span> }<a name="line.699"></a> -<span class="sourceLineNo">700</span><a name="line.700"></a> -<span class="sourceLineNo">701</span> public void stop() {<a name="line.701"></a> -<span class="sourceLineNo">702</span> if (!running.getAndSet(false)) {<a name="line.702"></a> -<span class="sourceLineNo">703</span> return;<a name="line.703"></a> -<span class="sourceLineNo">704</span> }<a name="line.704"></a> -<span class="sourceLineNo">705</span><a name="line.705"></a> -<span class="sourceLineNo">706</span> LOG.info("Stopping");<a name="line.706"></a> -<span class="sourceLineNo">707</span> scheduler.stop();<a name="line.707"></a> -<span class="sourceLineNo">708</span> timeoutExecutor.sendStopSignal();<a name="line.708"></a> -<span class="sourceLineNo">709</span> }<a name="line.709"></a> -<span class="sourceLineNo">710</span><a name="line.710"></a> -<span class="sourceLineNo">711</span> @VisibleForTesting<a name="line.711"></a> -<span class="sourceLineNo">712</span> public void join() {<a name="line.712"></a> -<span class="sourceLineNo">713</span> assert !isRunning() : "expected not running";<a name="line.713"></a> -<span class="sourceLineNo">714</span><a name="line.714"></a> -<span class="sourceLineNo">715</span> // stop the timeout executor<a name="line.715"></a> -<span class="sourceLineNo">716</span> timeoutExecutor.awaitTermination();<a name="line.716"></a> -<span class="sourceLineNo">717</span><a name="line.717"></a> -<span class="sourceLineNo">718</span> // stop the worker threads<a name="line.718"></a> -<span class="sourceLineNo">719</span> for (WorkerThread worker: workerThreads) {<a name="line.719"></a> -<span class="sourceLineNo">720</span> worker.awaitTermination();<a name="line.720"></a> -<span class="sourceLineNo">721</span> }<a name="line.721"></a> -<span class="sourceLineNo">722</span><a name="line.722"></a> -<span class="sourceLineNo">723</span> // Destroy the Thread Group for the executors<a name="line.723"></a> -<span class="sourceLineNo">724</span> // TODO: Fix. #join is not place to destroy resources.<a name="line.724"></a> -<span class="sourceLineNo">725</span> try {<a name="line.725"></a> -<span class="sourceLineNo">726</span> threadGroup.destroy();<a name="line.726"></a> -<span class="sourceLineNo">727</span> } catch (IllegalThreadStateException e) {<a name="line.727"></a> -<span class="sourceLineNo">728</span> LOG.error("ThreadGroup {} contains running threads; {}: See STDOUT",<a name="line.728"></a> -<span class="sourceLineNo">729</span> this.threadGroup, e.getMessage());<a name="line.729"></a> -<span class="sourceLineNo">730</span> // This dumps list of threads on STDOUT.<a name="line.730"></a> -<span class="sourceLineNo">731</span> this.threadGroup.list();<a name="line.731"></a> -<span class="sourceLineNo">732</span> }<a name="line.732"></a> -<span class="sourceLineNo">733</span><a name="line.733"></a> -<span class="sourceLineNo">734</span> // reset the in-memory state for testing<a name="line.734"></a> -<span class="sourceLineNo">735</span> completed.clear();<a name="line.735"></a> -<span class="sourceLineNo">736</span> rollbackStack.clear();<a name="line.736"></a> -<span class="sourceLineNo">737</span> procedures.clear();<a name="line.737"></a> -<span class="sourceLineNo">738</span> nonceKeysToProcIdsMap.clear();<a name="line.738"></a> -<span class="sourceLineNo">739</span> scheduler.clear();<a name="line.739"></a> -<span class="sourceLineNo">740</span> lastProcId.set(-1);<a name="line.740"></a> -<span class="sourceLineNo">741</span> }<a name="line.741"></a> -<span class="sourceLineNo">742</span><a name="line.742"></a> -<span class="sourceLineNo">743</span> public void refreshConfiguration(final Configuration conf) {<a name="line.743"></a> -<span class="sourceLineNo">744</span> this.conf = conf;<a name="line.744"></a> -<span class="sourceLineNo">745</span> setKeepAliveTime(conf.getLong(WORKER_KEEP_ALIVE_TIME_CONF_KEY,<a name="line.745"></a> -<span class="sourceLineNo">746</span> DEFAULT_WORKER_KEEP_ALIVE_TIME), TimeUnit.MILLISECONDS);<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> // Accessors<a name="line.750"></a> -<span class="sourceLineNo">751</span> // ==========================================================================<a name="line.751"></a> -<span class="sourceLineNo">752</span> public boolean isRunning() {<a name="line.752"></a> -<span class="sourceLineNo">753</span> return running.get();<a name="line.753"></a> -<span class="sourceLineNo">754</span> }<a name="line.754"></a> -<span class="sourceLineNo">755</span><a name="line.755"></a> -<span class="sourceLineNo">756</span> /**<a name="line.756"></a> -<span class="sourceLineNo">757</span> * @return the current number of worker threads.<a name="line.757"></a> -<span class="sourceLineNo">758</span> */<a name="line.758"></a> -<span class="sourceLineNo">759</span> public int getWorkerThreadCount() {<a name="line.759"></a> -<span class="sourceLineNo">760</span> return workerThreads.size();<a name="line.760"></a> -<span class="sourceLineNo">761</span> }<a name="line.761"></a> -<span class="sourceLineNo">762</span><a name="line.762"></a> -<span class="sourceLineNo">763</span> /**<a name="line.763"></a> -<span class="sourceLineNo">764</span> * @return the core pool size settings.<a name="line.764"></a> -<span class="sourceLineNo">765</span> */<a name="line.765"></a> -<span class="sourceLineNo">766</span> public int getCorePoolSize() {<a name="line.766"></a> -<span class="sourceLineNo">767</span> return corePoolSize;<a name="line.767"></a> -<span class="sourceLineNo">768</span> }<a name="line.768"></a> -<span class="sourceLineNo">769</span><a name="line.769"></a> -<span class="sourceLineNo">770</span> public int getActiveExecutorCount() {<a name="line.770"></a> -<span class="sourceLineNo">771</span> return activeExecutorCount.get();<a name="line.771"></a> -<span class="sourceLineNo">772</span> }<a name="line.772"></a> -<span class="sourceLineNo">773</span><a name="line.773"></a> -<span class="sourceLineNo">774</span> public TEnvironment getEnvironment() {<a name="line.774"></a> -<span class="sourceLineNo">775</span> return this.environment;<a name="line.775"></a> -<span class="sourceLineNo">776</span> }<a name="line.776"></a> -<span class="sourceLineNo">777</span><a name="line.777"></a> -<span class="sourceLineNo">778</span> public ProcedureStore getStore() {<a name="line.778"></a> -<span class="sourceLineNo">779</span> return this.store;<a name="line.779"></a> -<span class="sourceLineNo">780</span> }<a name="line.780"></a> -<span class="sourceLineNo">781</span><a name="line.781"></a> -<span class="sourceLineNo">782</span> ProcedureScheduler getScheduler() {<a name="line.782"></a> -<span class="sourceLineNo">783</span> return scheduler;<a name="line.783"></a> -<span class="sourceLineNo">784</span> }<a name="line.784"></a> -<span class="sourceLineNo">785</span><a name="line.785"></a> -<span class="sourceLineNo">786</span> public void setKeepAliveTime(final long keepAliveTime, final TimeUnit timeUnit) {<a name="line.786"></a> -<span class="sourceLineNo">787</span> this.keepAliveTime = timeUnit.toMillis(keepAliveTime);<a name="line.787"></a> -<span class="sourceLineNo">788</span> this.scheduler.signalAll();<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> public long getKeepAliveTime(final TimeUnit timeUnit) {<a name="line.791"></a> -<span class="sourceLineNo">792</span> return timeUnit.convert(keepAliveTime, TimeUnit.MILLISECONDS);<a name="line.792"></a> -<span class="sourceLineNo">793</span> }<a name="line.793"></a> -<span class="sourceLineNo">794</span><a name="line.794"></a> -<span class="sourceLineNo">795</span> // ==========================================================================<a name="line.795"></a> -<span class="sourceLineNo">796</span> // Submit/Remove Chores<a name="line.796"></a> -<span class="sourceLineNo">797</span> // ==========================================================================<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> * Add a chore procedure to the executor<a name="line.800"></a> -<span class="sourceLineNo">801</span> * @param chore the chore to add<a name="line.801"></a> -<span class="sourceLineNo">802</span> */<a name="line.802"></a> -<span class="sourceLineNo">803</span> public void addChore(ProcedureInMemoryChore<TEnvironment> chore) {<a name="line.803"></a> -<span class="sourceLineNo">804</span> chore.setState(ProcedureState.WAITING_TIMEOUT);<a name="line.804"></a> -<span class="sourceLineNo">805</span> timeoutExecutor.add(chore);<a name="line.805"></a> -<span class="sourceLineNo">806</span> }<a name="line.806"></a> -<span class="sourceLineNo">807</span><a name="line.807"></a> -<span class="sourceLineNo">808</span> /**<a name="line.808"></a> -<span class="sourceLineNo">809</span> * Remove a chore procedure from the executor<a name="line.809"></a> -<span class="sourceLineNo">810</span> * @param chore the chore to remove<a name="line.810"></a> -<span class="sourceLineNo">811</span> * @return whether the chore is removed, or it will be removed later<a name="line.811"></a> -<span class="sourceLineNo">812</span> */<a name="line.812"></a> -<span class="sourceLineNo">813</span> public boolean removeChore(ProcedureInMemoryChore<TEnvironment> chore) {<a name="line.813"></a> -<span class="sourceLineNo">814</span> chore.setState(ProcedureState.SUCCESS);<a name="line.814"></a> -<span class="sourceLineNo">815</span> return timeoutExecutor.remove(chore);<a name="line.815"></a> -<span class="sourceLineNo">816</span> }<a name="line.816"></a> -<span class="sourceLineNo">817</span><a name="line.817"></a> -<span class="sourceLineNo">818</span> // ==========================================================================<a name="line.818"></a> -<span class="sourceLineNo">819</span> // Nonce Procedure helpers<a name="line.819"></a> -<span class="sourceLineNo">820</span> // ==========================================================================<a name="line.820"></a> -<span class="sourceLineNo">821</span> /**<a name="line.821"></a> -<span class="sourceLineNo">822</span> * Create a NoneKey from the specified nonceGroup and nonce.<a name="line.822"></a> -<span class="sourceLineNo">823</span> * @param nonceGroup<a name="line.823"></a> -<span class="sourceLineNo">824</span> * @param nonce<a name="line.824"></a> -<span class="sourceLineNo">825</span> * @return the generated NonceKey<a name="line.825"></a> -<span class="sourceLineNo">826</span> */<a name="line.826"></a> -<span class="sourceLineNo">827</span> public NonceKey createNonceKey(final long nonceGroup, final long nonce) {<a name="line.827"></a> -<span class="sourceLineNo">828</span> return (nonce == HConstants.NO_NONCE) ? null : new NonceKey(nonceGroup, nonce);<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> /**<a name="line.831"></a> -<span class="sourceLineNo">832</span> * Register a nonce for a procedure that is going to be submitted.<a name="line.832"></a> -<span class="sourceLineNo">833</span> * A procId will be reserved and on submitProcedure(),<a name="line.833"></a> -<span class="sourceLineNo">834</span> * the procedure with the specified nonce will take the reserved ProcId.<a name="line.834"></a> -<span class="sourceLineNo">835</span> * If someone already reserved the nonce, this method will return the procId reserved,<a name="line.835"></a> -<span class="sourceLineNo">836</span> * otherwise an invalid procId will be returned. and the caller should procede<a name="line.836"></a> -<span class="sourceLineNo">837</span> * and submit the procedure.<a name="line.837"></a> -<span class="sourceLineNo">838</span> *<a name="line.838"></a> -<span class="sourceLineNo">839</span> * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.839"></a> -<span class="sourceLineNo">840</span> * @return the procId associated with the nonce, if any otherwise an invalid procId.<a name="line.840"></a> -<span class="sourceLineNo">841</span> */<a name="line.841"></a> -<span class="sourceLineNo">842</span> public long registerNonce(final NonceKey nonceKey) {<a name="line.842"></a> -<span class="sourceLineNo">843</span> if (nonceKey == null) return -1;<a name="line.843"></a> -<span class="sourceLineNo">844</span><a name="line.844"></a> -<span class="sourceLineNo">845</span> // check if we have already a Reserved ID for the nonce<a name="line.845"></a> -<span class="sourceLineNo">846</span> Long oldProcId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.846"></a> -<span class="sourceLineNo">847</span> if (oldProcId == null) {<a name="line.847"></a> -<span class="sourceLineNo">848</span> // reserve a new Procedure ID, this will be associated with the nonce<a name="line.848"></a> -<span class="sourceLineNo">849</span> // and the procedure submitted with the specified nonce will use this ID.<a name="line.849"></a> -<span class="sourceLineNo">850</span> final long newProcId = nextProcId();<a name="line.850"></a> -<span class="sourceLineNo">851</span> oldProcId = nonceKeysToProcIdsMap.putIfAbsent(nonceKey, newProcId);<a name="line.851"></a> -<span class="sourceLineNo">852</span> if (oldProcId == null) return -1;<a name="line.852"></a> -<span class="sourceLineNo">853</span> }<a name="line.853"></a> -<span class="sourceLineNo">854</span><a name="line.854"></a> -<span class="sourceLineNo">855</span> // we found a registered nonce, but the procedure may not have been submitted yet.<a name="line.855"></a> -<span class="sourceLineNo">856</span> // since the client expect the procedure to be submitted, spin here until it is.<a name="line.856"></a> -<span class="sourceLineNo">857</span> final boolean traceEnabled = LOG.isTraceEnabled();<a name="line.857"></a> -<span class="sourceLineNo">858</span> while (isRunning() &&<a name="line.858"></a> -<span class="sourceLineNo">859</span> !(procedures.containsKey(oldProcId) || completed.containsKey(oldProcId)) &&<a name="line.859"></a> -<span class="sourceLineNo">860</span> nonceKeysToProcIdsMap.containsKey(nonceKey)) {<a name="line.860"></a> -<span class="sourceLineNo">861</span> if (traceEnabled) {<a name="line.861"></a> -<span class="sourceLineNo">862</span> LOG.trace("Waiting for pid=" + oldProcId.longValue() + " to be submitted");<a name="line.862"></a> -<span class="sourceLineNo">863</span> }<a name="line.863"></a> -<span class="sourceLineNo">864</span> Threads.sleep(100);<a name="line.864"></a> -<span class="sourceLineNo">865</span> }<a name="line.865"></a> -<span class="sourceLineNo">866</span> return oldProcId.longValue();<a name="line.866"></a> -<span class="sourceLineNo">867</span> }<a name="line.867"></a> -<span class="sourceLineNo">868</span><a name="line.868"></a> -<span class="sourceLineNo">869</span> /**<a name="line.869"></a> -<span class="sourceLineNo">870</span> * Remove the NonceKey if the procedure was not submitted to the executor.<a name="line.870"></a> -<span class="sourceLineNo">871</span> * @param nonceKey A unique identifier for this operation from the client or process.<a name="line.871"></a> -<span class="sourceLineNo">872</span> */<a name="line.872"></a> -<span class="sourceLineNo">873</span> public void unregisterNonceIfProcedureWasNotSubmitted(final NonceKey nonceKey) {<a name="line.873"></a> -<span class="sourceLineNo">874</span> if (nonceKey == null) return;<a name="line.874"></a> -<span class="sourceLineNo">875</span><a name="line.875"></a> -<span class="sourceLineNo">876</span> final Long procId = nonceKeysToProcIdsMap.get(nonceKey);<a name="line.876"></a> -<span class="sourceLineNo">877</span> if (procId == null) return;<a name="line.877"></a> -<span class="sourceLineNo">878</span><a name="line.878"></a> -<span class="sourceLineNo">879</span> // if the procedure was not submitted, remove the nonce<a name="line.879"></a> -<span class="sourceLineNo">880</span> if (!(procedures.containsKey(procId) || completed.containsKey(procId))) {<a name="line.880"></a> -<span class="sourceLineNo">881</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.881"></a> -<span class="sourceLineNo">882</span> }<a name="line.882"></a> -<span class="sourceLineNo">883</span> }<a name="line.883"></a> -<span class="sourceLineNo">884</span><a name="line.884"></a> -<span class="sourceLineNo">885</span> public static class FailedProcedure<TEnvironment> extends Procedure<TEnvironment> {<a name="line.885"></a> -<span class="sourceLineNo">886</span> private String procName;<a name="line.886"></a> -<span class="sourceLineNo">887</span><a name="line.887"></a> -<span class="sourceLineNo">888</span> public FailedProcedure() {<a name="line.888"></a> -<span class="sourceLineNo">889</span> }<a name="line.889"></a> -<span class="sourceLineNo">890</span><a name="line.890"></a> -<span class="sourceLineNo">891</span> public FailedProcedure(long procId, String procName, User owner,<a name="line.891"></a> -<span class="sourceLineNo">892</span> NonceKey nonceKey, IOException exception) {<a name="line.892"></a> -<span class="sourceLineNo">893</span> this.procName = procName;<a name="line.893"></a> -<span class="sourceLineNo">894</span> setProcId(procId);<a name="line.894"></a> -<span class="sourceLineNo">895</span> setState(ProcedureState.ROLLEDBACK);<a name="line.895"></a> -<span class="sourceLineNo">896</span> setOwner(owner);<a name="line.896"></a> -<span class="sourceLineNo">897</span> setNonceKey(nonceKey);<a name="line.897"></a> -<span class="sourceLineNo">898</span> long currentTime = EnvironmentEdgeManager.currentTime();<a name="line.898"></a> -<span class="sourceLineNo">899</span> setSubmittedTime(currentTime);<a name="line.899"></a> -<span class="sourceLineNo">900</span> setLastUpdate(currentTime);<a name="line.900"></a> -<span class="sourceLineNo">901</span> setFailure(Objects.toString(exception.getMessage(), ""), exception);<a name="line.901"></a> -<span class="sourceLineNo">902</span> }<a name="line.902"></a> -<span class="sourceLineNo">903</span><a name="line.903"></a> -<span class="sourceLineNo">904</span> @Override<a name="line.904"></a> -<span class="sourceLineNo">905</span> public String getProcName() {<a name="line.905"></a> -<span class="sourceLineNo">906</span> return procName;<a name="line.906"></a> -<span class="sourceLineNo">907</span> }<a name="line.907"></a> -<span class="sourceLineNo">908</span><a name="line.908"></a> -<span class="sourceLineNo">909</span> @Override<a name="line.909"></a> -<span class="sourceLineNo">910</span> protected Procedure<TEnvironment>[] execute(TEnvironment env)<a name="line.910"></a> -<span class="sourceLineNo">911</span> throws ProcedureYieldException, ProcedureSuspendedException,<a name="line.911"></a> -<span class="sourceLineNo">912</span> InterruptedException {<a name="line.912"></a> -<span class="sourceLineNo">913</span> throw new UnsupportedOperationException();<a name="line.913"></a> -<span class="sourceLineNo">914</span> }<a name="line.914"></a> -<span class="sourceLineNo">915</span><a name="line.915"></a> -<span class="sourceLineNo">916</span> @Override<a name="line.916"></a> -<span class="sourceLineNo">917</span> protected void rollback(TEnvironment env)<a name="line.917"></a> -<span class="sourceLineNo">918</span> throws IOException, InterruptedException {<a name="line.918"></a> -<span class="sourceLineNo">919</span> throw new UnsupportedOperationException();<a name="line.919"></a> -<span class="sourceLineNo">920</span> }<a name="line.920"></a> -<span class="sourceLineNo">921</span><a name="line.921"></a> -<span class="sourceLineNo">922</span> @Override<a name="line.922"></a> -<span class="sourceLineNo">923</span> protected boolean abort(TEnvironment env) {<a name="line.923"></a> -<span class="sourceLineNo">924</span> throw new UnsupportedOperationException();<a name="line.924"></a> -<span class="sourceLineNo">925</span> }<a name="line.925"></a> -<span class="sourceLineNo">926</span><a name="line.926"></a> -<span class="sourceLineNo">927</span> @Override<a name="line.927"></a> -<span class="sourceLineNo">928</span> protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.928"></a> -<span class="sourceLineNo">929</span> th
<TRUNCATED>