http://git-wip-us.apache.org/repos/asf/hbase-site/blob/9cc58aff/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.FailedProcedure.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.FailedProcedure.html b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.FailedProcedure.html index 1aed387..22f68b3 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.FailedProcedure.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/procedure2/ProcedureExecutor.FailedProcedure.html @@ -55,1901 +55,1905 @@ <span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.procedure2.util.StringUtils;<a name="line.47"></a> <span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.security.User;<a name="line.48"></a> <span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.util.NonceKey;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.slf4j.Logger;<a name="line.53"></a> -<span class="sourceLineNo">054</span>import org.slf4j.LoggerFactory;<a name="line.54"></a> -<span class="sourceLineNo">055</span><a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;<a name="line.57"></a> -<span class="sourceLineNo">058</span><a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureState;<a name="line.59"></a> -<span class="sourceLineNo">060</span><a name="line.60"></a> -<span class="sourceLineNo">061</span>/**<a name="line.61"></a> -<span class="sourceLineNo">062</span> * Thread Pool that executes the submitted procedures.<a name="line.62"></a> -<span class="sourceLineNo">063</span> * The executor has a ProcedureStore associated.<a name="line.63"></a> -<span class="sourceLineNo">064</span> * Each operation is logged and on restart the pending procedures are resumed.<a name="line.64"></a> -<span class="sourceLineNo">065</span> *<a name="line.65"></a> -<span class="sourceLineNo">066</span> * Unless the Procedure code throws an error (e.g. invalid user input)<a name="line.66"></a> -<span class="sourceLineNo">067</span> * the procedure will complete (at some point in time), On restart the pending<a name="line.67"></a> -<span class="sourceLineNo">068</span> * procedures are resumed and the once failed will be rolledback.<a name="line.68"></a> -<span class="sourceLineNo">069</span> *<a name="line.69"></a> -<span class="sourceLineNo">070</span> * The user can add procedures to the executor via submitProcedure(proc)<a name="line.70"></a> -<span class="sourceLineNo">071</span> * check for the finished state via isFinished(procId)<a name="line.71"></a> -<span class="sourceLineNo">072</span> * and get the result via getResult(procId)<a name="line.72"></a> -<span class="sourceLineNo">073</span> */<a name="line.73"></a> -<span class="sourceLineNo">074</span>@InterfaceAudience.Private<a name="line.74"></a> -<span class="sourceLineNo">075</span>public class ProcedureExecutor<TEnvironment> {<a name="line.75"></a> -<span class="sourceLineNo">076</span> private static final Logger LOG = LoggerFactory.getLogger(ProcedureExecutor.class);<a name="line.76"></a> -<span class="sourceLineNo">077</span><a name="line.77"></a> -<span class="sourceLineNo">078</span> public static final String CHECK_OWNER_SET_CONF_KEY = "hbase.procedure.check.owner.set";<a name="line.78"></a> -<span class="sourceLineNo">079</span> private static final boolean DEFAULT_CHECK_OWNER_SET = false;<a name="line.79"></a> -<span class="sourceLineNo">080</span><a name="line.80"></a> -<span class="sourceLineNo">081</span> public static final String WORKER_KEEP_ALIVE_TIME_CONF_KEY =<a name="line.81"></a> -<span class="sourceLineNo">082</span> "hbase.procedure.worker.keep.alive.time.msec";<a name="line.82"></a> -<span class="sourceLineNo">083</span> private static final long DEFAULT_WORKER_KEEP_ALIVE_TIME = TimeUnit.MINUTES.toMillis(1);<a name="line.83"></a> -<span class="sourceLineNo">084</span><a name="line.84"></a> -<span class="sourceLineNo">085</span> Testing testing = null;<a name="line.85"></a> -<span class="sourceLineNo">086</span> public static class Testing {<a name="line.86"></a> -<span class="sourceLineNo">087</span> protected boolean killIfSuspended = false;<a name="line.87"></a> -<span class="sourceLineNo">088</span> protected boolean killBeforeStoreUpdate = false;<a name="line.88"></a> -<span class="sourceLineNo">089</span> protected boolean toggleKillBeforeStoreUpdate = false;<a name="line.89"></a> -<span class="sourceLineNo">090</span><a name="line.90"></a> -<span class="sourceLineNo">091</span> protected boolean shouldKillBeforeStoreUpdate() {<a name="line.91"></a> -<span class="sourceLineNo">092</span> final boolean kill = this.killBeforeStoreUpdate;<a name="line.92"></a> -<span class="sourceLineNo">093</span> if (this.toggleKillBeforeStoreUpdate) {<a name="line.93"></a> -<span class="sourceLineNo">094</span> this.killBeforeStoreUpdate = !kill;<a name="line.94"></a> -<span class="sourceLineNo">095</span> LOG.warn("Toggle KILL before store update to: " + this.killBeforeStoreUpdate);<a name="line.95"></a> -<span class="sourceLineNo">096</span> }<a name="line.96"></a> -<span class="sourceLineNo">097</span> return kill;<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> protected boolean shouldKillBeforeStoreUpdate(final boolean isSuspended) {<a name="line.100"></a> -<span class="sourceLineNo">101</span> return (isSuspended && !killIfSuspended) ? false : shouldKillBeforeStoreUpdate();<a name="line.101"></a> -<span class="sourceLineNo">102</span> }<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">105</span> public interface ProcedureExecutorListener {<a name="line.105"></a> -<span class="sourceLineNo">106</span> void procedureLoaded(long procId);<a name="line.106"></a> -<span class="sourceLineNo">107</span> void procedureAdded(long procId);<a name="line.107"></a> -<span class="sourceLineNo">108</span> void procedureFinished(long procId);<a name="line.108"></a> -<span class="sourceLineNo">109</span> }<a name="line.109"></a> -<span class="sourceLineNo">110</span><a name="line.110"></a> -<span class="sourceLineNo">111</span> private static final class CompletedProcedureRetainer<TEnvironment> {<a name="line.111"></a> -<span class="sourceLineNo">112</span> private final Procedure<TEnvironment> procedure;<a name="line.112"></a> -<span class="sourceLineNo">113</span> private long clientAckTime;<a name="line.113"></a> -<span class="sourceLineNo">114</span><a name="line.114"></a> -<span class="sourceLineNo">115</span> public CompletedProcedureRetainer(Procedure<TEnvironment> procedure) {<a name="line.115"></a> -<span class="sourceLineNo">116</span> this.procedure = procedure;<a name="line.116"></a> -<span class="sourceLineNo">117</span> clientAckTime = -1;<a name="line.117"></a> -<span class="sourceLineNo">118</span> }<a name="line.118"></a> -<span class="sourceLineNo">119</span><a name="line.119"></a> -<span class="sourceLineNo">120</span> public Procedure<TEnvironment> getProcedure() {<a name="line.120"></a> -<span class="sourceLineNo">121</span> return procedure;<a name="line.121"></a> -<span class="sourceLineNo">122</span> }<a name="line.122"></a> -<span class="sourceLineNo">123</span><a name="line.123"></a> -<span class="sourceLineNo">124</span> public boolean hasClientAckTime() {<a name="line.124"></a> -<span class="sourceLineNo">125</span> return clientAckTime != -1;<a name="line.125"></a> -<span class="sourceLineNo">126</span> }<a name="line.126"></a> -<span class="sourceLineNo">127</span><a name="line.127"></a> -<span class="sourceLineNo">128</span> public long getClientAckTime() {<a name="line.128"></a> -<span class="sourceLineNo">129</span> return clientAckTime;<a name="line.129"></a> -<span class="sourceLineNo">130</span> }<a name="line.130"></a> -<span class="sourceLineNo">131</span><a name="line.131"></a> -<span class="sourceLineNo">132</span> public void setClientAckTime(long clientAckTime) {<a name="line.132"></a> -<span class="sourceLineNo">133</span> this.clientAckTime = clientAckTime;<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> public boolean isExpired(long now, long evictTtl, long evictAckTtl) {<a name="line.136"></a> -<span class="sourceLineNo">137</span> return (hasClientAckTime() && (now - getClientAckTime()) >= evictAckTtl) ||<a name="line.137"></a> -<span class="sourceLineNo">138</span> (now - procedure.getLastUpdate()) >= evictTtl;<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><a name="line.141"></a> -<span class="sourceLineNo">142</span> /**<a name="line.142"></a> -<span class="sourceLineNo">143</span> * Internal cleaner that removes the completed procedure results after a TTL.<a name="line.143"></a> -<span class="sourceLineNo">144</span> * NOTE: This is a special case handled in timeoutLoop().<a name="line.144"></a> -<span class="sourceLineNo">145</span> *<a name="line.145"></a> -<span class="sourceLineNo">146</span> * <p>Since the client code looks more or less like:<a name="line.146"></a> -<span class="sourceLineNo">147</span> * <pre><a name="line.147"></a> -<span class="sourceLineNo">148</span> * procId = master.doOperation()<a name="line.148"></a> -<span class="sourceLineNo">149</span> * while (master.getProcResult(procId) == ProcInProgress);<a name="line.149"></a> -<span class="sourceLineNo">150</span> * </pre><a name="line.150"></a> -<span class="sourceLineNo">151</span> * The master should not throw away the proc result as soon as the procedure is done<a name="line.151"></a> -<span class="sourceLineNo">152</span> * but should wait a result request from the client (see executor.removeResult(procId))<a name="line.152"></a> -<span class="sourceLineNo">153</span> * The client will call something like master.isProcDone() or master.getProcResult()<a name="line.153"></a> -<span class="sourceLineNo">154</span> * which will return the result/state to the client, and it will mark the completed<a name="line.154"></a> -<span class="sourceLineNo">155</span> * proc as ready to delete. note that the client may not receive the response from<a name="line.155"></a> -<span class="sourceLineNo">156</span> * the master (e.g. master failover) so, if we delay a bit the real deletion of<a name="line.156"></a> -<span class="sourceLineNo">157</span> * the proc result the client will be able to get the result the next try.<a name="line.157"></a> -<span class="sourceLineNo">158</span> */<a name="line.158"></a> -<span class="sourceLineNo">159</span> private static class CompletedProcedureCleaner<TEnvironment><a name="line.159"></a> -<span class="sourceLineNo">160</span> extends ProcedureInMemoryChore<TEnvironment> {<a name="line.160"></a> -<span class="sourceLineNo">161</span> private static final Logger LOG = LoggerFactory.getLogger(CompletedProcedureCleaner.class);<a name="line.161"></a> -<span class="sourceLineNo">162</span><a name="line.162"></a> -<span class="sourceLineNo">163</span> private static final String CLEANER_INTERVAL_CONF_KEY = "hbase.procedure.cleaner.interval";<a name="line.163"></a> -<span class="sourceLineNo">164</span> private static final int DEFAULT_CLEANER_INTERVAL = 30 * 1000; // 30sec<a name="line.164"></a> -<span class="sourceLineNo">165</span><a name="line.165"></a> -<span class="sourceLineNo">166</span> private static final String EVICT_TTL_CONF_KEY = "hbase.procedure.cleaner.evict.ttl";<a name="line.166"></a> -<span class="sourceLineNo">167</span> private static final int DEFAULT_EVICT_TTL = 15 * 60000; // 15min<a name="line.167"></a> -<span class="sourceLineNo">168</span><a name="line.168"></a> -<span class="sourceLineNo">169</span> private static final String EVICT_ACKED_TTL_CONF_KEY ="hbase.procedure.cleaner.acked.evict.ttl";<a name="line.169"></a> -<span class="sourceLineNo">170</span> private static final int DEFAULT_ACKED_EVICT_TTL = 5 * 60000; // 5min<a name="line.170"></a> -<span class="sourceLineNo">171</span><a name="line.171"></a> -<span class="sourceLineNo">172</span> private static final String BATCH_SIZE_CONF_KEY = "hbase.procedure.cleaner.evict.batch.size";<a name="line.172"></a> -<span class="sourceLineNo">173</span> private static final int DEFAULT_BATCH_SIZE = 32;<a name="line.173"></a> -<span class="sourceLineNo">174</span><a name="line.174"></a> -<span class="sourceLineNo">175</span> private final Map<Long, CompletedProcedureRetainer<TEnvironment>> completed;<a name="line.175"></a> -<span class="sourceLineNo">176</span> private final Map<NonceKey, Long> nonceKeysToProcIdsMap;<a name="line.176"></a> -<span class="sourceLineNo">177</span> private final ProcedureStore store;<a name="line.177"></a> -<span class="sourceLineNo">178</span> private Configuration conf;<a name="line.178"></a> -<span class="sourceLineNo">179</span><a name="line.179"></a> -<span class="sourceLineNo">180</span> public CompletedProcedureCleaner(Configuration conf, final ProcedureStore store,<a name="line.180"></a> -<span class="sourceLineNo">181</span> final Map<Long, CompletedProcedureRetainer<TEnvironment>> completedMap,<a name="line.181"></a> -<span class="sourceLineNo">182</span> final Map<NonceKey, Long> nonceKeysToProcIdsMap) {<a name="line.182"></a> -<span class="sourceLineNo">183</span> // set the timeout interval that triggers the periodic-procedure<a name="line.183"></a> -<span class="sourceLineNo">184</span> super(conf.getInt(CLEANER_INTERVAL_CONF_KEY, DEFAULT_CLEANER_INTERVAL));<a name="line.184"></a> -<span class="sourceLineNo">185</span> this.completed = completedMap;<a name="line.185"></a> -<span class="sourceLineNo">186</span> this.nonceKeysToProcIdsMap = nonceKeysToProcIdsMap;<a name="line.186"></a> -<span class="sourceLineNo">187</span> this.store = store;<a name="line.187"></a> -<span class="sourceLineNo">188</span> this.conf = conf;<a name="line.188"></a> -<span class="sourceLineNo">189</span> }<a name="line.189"></a> -<span class="sourceLineNo">190</span><a name="line.190"></a> -<span class="sourceLineNo">191</span> @Override<a name="line.191"></a> -<span class="sourceLineNo">192</span> protected void periodicExecute(final TEnvironment env) {<a name="line.192"></a> -<span class="sourceLineNo">193</span> if (completed.isEmpty()) {<a name="line.193"></a> -<span class="sourceLineNo">194</span> if (LOG.isTraceEnabled()) {<a name="line.194"></a> -<span class="sourceLineNo">195</span> LOG.trace("No completed procedures to cleanup.");<a name="line.195"></a> -<span class="sourceLineNo">196</span> }<a name="line.196"></a> -<span class="sourceLineNo">197</span> return;<a name="line.197"></a> -<span class="sourceLineNo">198</span> }<a name="line.198"></a> -<span class="sourceLineNo">199</span><a name="line.199"></a> -<span class="sourceLineNo">200</span> final long evictTtl = conf.getInt(EVICT_TTL_CONF_KEY, DEFAULT_EVICT_TTL);<a name="line.200"></a> -<span class="sourceLineNo">201</span> final long evictAckTtl = conf.getInt(EVICT_ACKED_TTL_CONF_KEY, DEFAULT_ACKED_EVICT_TTL);<a name="line.201"></a> -<span class="sourceLineNo">202</span> final int batchSize = conf.getInt(BATCH_SIZE_CONF_KEY, DEFAULT_BATCH_SIZE);<a name="line.202"></a> -<span class="sourceLineNo">203</span><a name="line.203"></a> -<span class="sourceLineNo">204</span> final long[] batchIds = new long[batchSize];<a name="line.204"></a> -<span class="sourceLineNo">205</span> int batchCount = 0;<a name="line.205"></a> -<span class="sourceLineNo">206</span><a name="line.206"></a> -<span class="sourceLineNo">207</span> final long now = EnvironmentEdgeManager.currentTime();<a name="line.207"></a> -<span class="sourceLineNo">208</span> final Iterator<Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>>> it =<a name="line.208"></a> -<span class="sourceLineNo">209</span> completed.entrySet().iterator();<a name="line.209"></a> -<span class="sourceLineNo">210</span> while (it.hasNext() && store.isRunning()) {<a name="line.210"></a> -<span class="sourceLineNo">211</span> final Map.Entry<Long, CompletedProcedureRetainer<TEnvironment>> entry = it.next();<a name="line.211"></a> -<span class="sourceLineNo">212</span> final CompletedProcedureRetainer<TEnvironment> retainer = entry.getValue();<a name="line.212"></a> -<span class="sourceLineNo">213</span> final Procedure<?> proc = retainer.getProcedure();<a name="line.213"></a> -<span class="sourceLineNo">214</span><a name="line.214"></a> -<span class="sourceLineNo">215</span> // TODO: Select TTL based on Procedure type<a name="line.215"></a> -<span class="sourceLineNo">216</span> if (retainer.isExpired(now, evictTtl, evictAckTtl)) {<a name="line.216"></a> -<span class="sourceLineNo">217</span> // Failed procedures aren't persisted in WAL.<a name="line.217"></a> -<span class="sourceLineNo">218</span> if (!(proc instanceof FailedProcedure)) {<a name="line.218"></a> -<span class="sourceLineNo">219</span> batchIds[batchCount++] = entry.getKey();<a name="line.219"></a> -<span class="sourceLineNo">220</span> if (batchCount == batchIds.length) {<a name="line.220"></a> -<span class="sourceLineNo">221</span> store.delete(batchIds, 0, batchCount);<a name="line.221"></a> -<span class="sourceLineNo">222</span> batchCount = 0;<a name="line.222"></a> -<span class="sourceLineNo">223</span> }<a name="line.223"></a> -<span class="sourceLineNo">224</span> }<a name="line.224"></a> -<span class="sourceLineNo">225</span> final NonceKey nonceKey = proc.getNonceKey();<a name="line.225"></a> -<span class="sourceLineNo">226</span> if (nonceKey != null) {<a name="line.226"></a> -<span class="sourceLineNo">227</span> nonceKeysToProcIdsMap.remove(nonceKey);<a name="line.227"></a> -<span class="sourceLineNo">228</span> }<a name="line.228"></a> -<span class="sourceLineNo">229</span> it.remove();<a name="line.229"></a> -<span class="sourceLineNo">230</span> LOG.trace("Evict completed {}", proc);<a name="line.230"></a> -<span class="sourceLineNo">231</span> }<a name="line.231"></a> -<span class="sourceLineNo">232</span> }<a name="line.232"></a> -<span class="sourceLineNo">233</span> if (batchCount > 0) {<a name="line.233"></a> -<span class="sourceLineNo">234</span> store.delete(batchIds, 0, batchCount);<a name="line.234"></a> -<span class="sourceLineNo">235</span> }<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">239</span> /**<a name="line.239"></a> -<span class="sourceLineNo">240</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the Procedure.<a name="line.240"></a> -<span class="sourceLineNo">241</span> * Once a Root-Procedure completes (success or failure), the result will be added to this map.<a name="line.241"></a> -<span class="sourceLineNo">242</span> * The user of ProcedureExecutor should call getResult(procId) to get the result.<a name="line.242"></a> -<span class="sourceLineNo">243</span> */<a name="line.243"></a> -<span class="sourceLineNo">244</span> private final ConcurrentHashMap<Long, CompletedProcedureRetainer<TEnvironment>> completed =<a name="line.244"></a> -<span class="sourceLineNo">245</span> new ConcurrentHashMap<>();<a name="line.245"></a> -<span class="sourceLineNo">246</span><a name="line.246"></a> -<span class="sourceLineNo">247</span> /**<a name="line.247"></a> -<span class="sourceLineNo">248</span> * Map the the procId returned by submitProcedure(), the Root-ProcID, to the RootProcedureState.<a name="line.248"></a> -<span class="sourceLineNo">249</span> * The RootProcedureState contains the execution stack of the Root-Procedure,<a name="line.249"></a> -<span class="sourceLineNo">250</span> * It is added to the map by submitProcedure() and removed on procedure completion.<a name="line.250"></a> -<span class="sourceLineNo">251</span> */<a name="line.251"></a> -<span class="sourceLineNo">252</span> private final ConcurrentHashMap<Long, RootProcedureState<TEnvironment>> rollbackStack =<a name="line.252"></a> -<span class="sourceLineNo">253</span> new ConcurrentHashMap<>();<a name="line.253"></a> -<span class="sourceLineNo">254</span><a name="line.254"></a> -<span class="sourceLineNo">255</span> /**<a name="line.255"></a> -<span class="sourceLineNo">256</span> * Helper map to lookup the live procedures by ID.<a name="line.256"></a> -<span class="sourceLineNo">257</span> * This map contains every procedure. root-procedures and subprocedures.<a name="line.257"></a> -<span class="sourceLineNo">258</span> */<a name="line.258"></a> -<span class="sourceLineNo">259</span> private final ConcurrentHashMap<Long, Procedure<TEnvironment>> procedures =<a name="line.259"></a> -<span class="sourceLineNo">260</span> new ConcurrentHashMap<>();<a name="line.260"></a> -<span class="sourceLineNo">261</span><a name="line.261"></a> -<span class="sourceLineNo">262</span> /**<a name="line.262"></a> -<span class="sourceLineNo">263</span> * Helper map to lookup whether the procedure already issued from the same client. This map<a name="line.263"></a> -<span class="sourceLineNo">264</span> * contains every root procedure.<a name="line.264"></a> -<span class="sourceLineNo">265</span> */<a name="line.265"></a> -<span class="sourceLineNo">266</span> private final ConcurrentHashMap<NonceKey, Long> nonceKeysToProcIdsMap = new ConcurrentHashMap<>();<a name="line.266"></a> -<span class="sourceLineNo">267</span><a name="line.267"></a> -<span class="sourceLineNo">268</span> private final CopyOnWriteArrayList<ProcedureExecutorListener> listeners =<a name="line.268"></a> -<span class="sourceLineNo">269</span> new CopyOnWriteArrayList<>();<a name="line.269"></a> -<span class="sourceLineNo">270</span><a name="line.270"></a> -<span class="sourceLineNo">271</span> private Configuration conf;<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> * Created in the {@link #start(int, boolean)} method. Destroyed in {@link #join()} (FIX! Doing<a name="line.274"></a> -<span class="sourceLineNo">275</span> * resource handling rather than observing in a #join is unexpected).<a name="line.275"></a> -<span class="sourceLineNo">276</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.276"></a> -<span class="sourceLineNo">277</span> * (Should be ok).<a name="line.277"></a> -<span class="sourceLineNo">278</span> */<a name="line.278"></a> -<span class="sourceLineNo">279</span> private ThreadGroup threadGroup;<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> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.282"></a> -<span class="sourceLineNo">283</span> * resource handling rather than observing in a #join is unexpected).<a name="line.283"></a> -<span class="sourceLineNo">284</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.284"></a> -<span class="sourceLineNo">285</span> * (Should be ok).<a name="line.285"></a> -<span class="sourceLineNo">286</span> */<a name="line.286"></a> -<span class="sourceLineNo">287</span> private CopyOnWriteArrayList<WorkerThread> workerThreads;<a name="line.287"></a> -<span class="sourceLineNo">288</span><a name="line.288"></a> -<span class="sourceLineNo">289</span> /**<a name="line.289"></a> -<span class="sourceLineNo">290</span> * Created in the {@link #start(int, boolean)} method. Terminated in {@link #join()} (FIX! Doing<a name="line.290"></a> -<span class="sourceLineNo">291</span> * resource handling rather than observing in a #join is unexpected).<a name="line.291"></a> -<span class="sourceLineNo">292</span> * Overridden when we do the ProcedureTestingUtility.testRecoveryAndDoubleExecution trickery<a name="line.292"></a> -<span class="sourceLineNo">293</span> * (Should be ok).<a name="line.293"></a> -<span class="sourceLineNo">294</span> */<a name="line.294"></a> -<span class="sourceLineNo">295</span> private TimeoutExecutorThread<TEnvironment> timeoutExecutor;<a name="line.295"></a> -<span class="sourceLineNo">296</span><a name="line.296"></a> -<span class="sourceLineNo">297</span> private int corePoolSize;<a name="line.297"></a> -<span class="sourceLineNo">298</span> private int maxPoolSize;<a name="line.298"></a> -<span class="sourceLineNo">299</span><a name="line.299"></a> -<span class="sourceLineNo">300</span> private volatile long keepAliveTime;<a name="line.300"></a> -<span class="sourceLineNo">301</span><a name="line.301"></a> -<span class="sourceLineNo">302</span> /**<a name="line.302"></a> -<span class="sourceLineNo">303</span> * Scheduler/Queue that contains runnable procedures.<a name="line.303"></a> -<span class="sourceLineNo">304</span> */<a name="line.304"></a> -<span class="sourceLineNo">305</span> private final ProcedureScheduler scheduler;<a name="line.305"></a> -<span class="sourceLineNo">306</span><a name="line.306"></a> -<span class="sourceLineNo">307</span> private final AtomicLong lastProcId = new AtomicLong(-1);<a name="line.307"></a> -<span class="sourceLineNo">308</span> private final AtomicLong workerId = new AtomicLong(0);<a name="line.308"></a> -<span class="sourceLineNo">309</span> private final AtomicInteger activeExecutorCount = new AtomicInteger(0);<a name="line.309"></a> -<span class="sourceLineNo">310</span> private final AtomicBoolean running = new AtomicBoolean(false);<a name="line.310"></a> -<span class="sourceLineNo">311</span> private final TEnvironment environment;<a name="line.311"></a> -<span class="sourceLineNo">312</span> private final ProcedureStore store;<a name="line.312"></a> -<span class="sourceLineNo">313</span><a name="line.313"></a> -<span class="sourceLineNo">314</span> private final boolean checkOwnerSet;<a name="line.314"></a> -<span class="sourceLineNo">315</span><a name="line.315"></a> -<span class="sourceLineNo">316</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.316"></a> -<span class="sourceLineNo">317</span> final ProcedureStore store) {<a name="line.317"></a> -<span class="sourceLineNo">318</span> this(conf, environment, store, new SimpleProcedureScheduler());<a name="line.318"></a> -<span class="sourceLineNo">319</span> }<a name="line.319"></a> -<span class="sourceLineNo">320</span><a name="line.320"></a> -<span class="sourceLineNo">321</span> public ProcedureExecutor(final Configuration conf, final TEnvironment environment,<a name="line.321"></a> -<span class="sourceLineNo">322</span> final ProcedureStore store, final ProcedureScheduler scheduler) {<a name="line.322"></a> -<span class="sourceLineNo">323</span> this.environment = environment;<a name="line.323"></a> -<span class="sourceLineNo">324</span> this.scheduler = scheduler;<a name="line.324"></a> -<span class="sourceLineNo">325</span> this.store = store;<a name="line.325"></a> -<span class="sourceLineNo">326</span> this.conf = conf;<a name="line.326"></a> -<span class="sourceLineNo">327</span> this.checkOwnerSet = conf.getBoolean(CHECK_OWNER_SET_CONF_KEY, DEFAULT_CHECK_OWNER_SET);<a name="line.327"></a> -<span class="sourceLineNo">328</span> refreshConfiguration(conf);<a name="line.328"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.util.IdLock;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.util.NonceKey;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.slf4j.Logger;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.slf4j.LoggerFactory;<a name="line.55"></a> +<span class="sourceLineNo">056</span><a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;<a name="line.58"></a> +<span class="sourceLineNo">059</span><a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureState;<a name="line.60"></a> +<span class="sourceLineNo">061</span><a name="line.61"></a> +<span class="sourceLineNo">062</span>/**<a name="line.62"></a> +<span class="sourceLineNo">063</span> * Thread Pool that executes the submitted procedures.<a name="line.63"></a> +<span class="sourceLineNo">064</span> * The executor has a ProcedureStore associated.<a name="line.64"></a> +<span class="sourceLineNo">065</span> * Each operation is logged and on restart the pending procedures are resumed.<a name="line.65"></a> +<span class="sourceLineNo">066</span> *<a name="line.66"></a> +<span class="sourceLineNo">067</span> * Unless the Procedure code throws an error (e.g. invalid user input)<a name="line.67"></a> +<span class="sourceLineNo">068</span> * the procedure will complete (at some point in time), On restart the pending<a name="line.68"></a> +<span class="sourceLineNo">069</span> * procedures are resumed and the once failed will be rolledback.<a name="line.69"></a> +<span class="sourceLineNo">070</span> *<a name="line.70"></a> +<span class="sourceLineNo">071</span> * The user can add procedures to the executor via submitProcedure(proc)<a name="line.71"></a> +<span class="sourceLineNo">072</span> * check for the finished state via isFinished(procId)<a name="line.72"></a> +<span class="sourceLineNo">073</span> * and get the result via getResult(procId)<a name="line.73"></a> +<span class="sourceLineNo">074</span> */<a name="line.74"></a> +<span class="sourceLineNo">075</span>@InterfaceAudience.Private<a name="line.75"></a> +<span class="sourceLineNo">076</span>public class ProcedureExecutor<TEnvironment> {<a name="line.76"></a> +<span class="sourceLineNo">077</span> private static final Logger LOG = LoggerFactory.getLogger(ProcedureExecutor.class);<a name="line.77"></a> +<span class="sourceLineNo">078</span><a name="line.78"></a> +<span class="sourceLineNo">079</span> public static final String CHECK_OWNER_SET_CONF_KEY = "hbase.procedure.check.owner.set";<a name="line.79"></a> +<span class="sourceLineNo">080</span> private static final boolean DEFAULT_CHECK_OWNER_SET = false;<a name="line.80"></a> +<span class="sourceLineNo">081</span><a name="line.81"></a> +<span class="sourceLineNo">082</span> public static final String WORKER_KEEP_ALIVE_TIME_CONF_KEY =<a name="line.82"></a> +<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">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">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">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">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">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">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">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">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">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">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">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">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> }<a name="line.330"></a> -<span class="sourceLineNo">331</span><a name="line.331"></a> -<span class="sourceLineNo">332</span> private void load(final boolean abortOnCorruption) throws IOException {<a name="line.332"></a> -<span class="sourceLineNo">333</span> Preconditions.checkArgument(completed.isEmpty(), "completed not empty");<a name="line.333"></a> -<span class="sourceLineNo">334</span> Preconditions.checkArgument(rollbackStack.isEmpty(), "rollback state not empty");<a name="line.334"></a> -<span class="sourceLineNo">335</span> Preconditions.checkArgument(procedures.isEmpty(), "procedure map not empty");<a name="line.335"></a> -<span class="sourceLineNo">336</span> Preconditions.checkArgument(scheduler.size() == 0, "run queue not empty");<a name="line.336"></a> -<span class="sourceLineNo">337</span><a name="line.337"></a> -<span class="sourceLineNo">338</span> store.load(new ProcedureStore.ProcedureLoader() {<a name="line.338"></a> -<span class="sourceLineNo">339</span> @Override<a name="line.339"></a> -<span class="sourceLineNo">340</span> public void setMaxProcId(long maxProcId) {<a name="line.340"></a> -<span class="sourceLineNo">341</span> assert lastProcId.get() < 0 : "expected only one call to setMaxProcId()";<a name="line.341"></a> -<span class="sourceLineNo">342</span> lastProcId.set(maxProcId);<a name="line.342"></a> -<span class="sourceLineNo">343</span> }<a name="line.343"></a> -<span class="sourceLineNo">344</span><a name="line.344"></a> -<span class="sourceLineNo">345</span> @Override<a name="line.345"></a> -<span class="sourceLineNo">346</span> public void load(ProcedureIterator procIter) throws IOException {<a name="line.346"></a> -<span class="sourceLineNo">347</span> loadProcedures(procIter, abortOnCorruption);<a name="line.347"></a> -<span class="sourceLineNo">348</span> }<a name="line.348"></a> -<span class="sourceLineNo">349</span><a name="line.349"></a> -<span class="sourceLineNo">350</span> @Override<a name="line.350"></a> -<span class="sourceLineNo">351</span> public void handleCorrupted(ProcedureIterator procIter) throws IOException {<a name="line.351"></a> -<span class="sourceLineNo">352</span> int corruptedCount = 0;<a name="line.352"></a> -<span class="sourceLineNo">353</span> while (procIter.hasNext()) {<a name="line.353"></a> -<span class="sourceLineNo">354</span> Procedure<?> proc = procIter.next();<a name="line.354"></a> -<span class="sourceLineNo">355</span> LOG.error("Corrupt " + proc);<a name="line.355"></a> -<span class="sourceLineNo">356</span> corruptedCount++;<a name="line.356"></a> -<span class="sourceLineNo">357</span> }<a name="line.357"></a> -<span class="sourceLineNo">358</span> if (abortOnCorruption && corruptedCount > 0) {<a name="line.358"></a> -<span class="sourceLineNo">359</span> throw new IOException("found " + corruptedCount + " corrupted procedure(s) on replay");<a name="line.359"></a> -<span class="sourceLineNo">360</span> }<a name="line.360"></a> -<span class="sourceLineNo">361</span> }<a name="line.361"></a> -<span class="sourceLineNo">362</span> });<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> private void restoreLock(Procedure<TEnvironment> proc, Set<Long> restored) {<a name="line.365"></a> -<span class="sourceLineNo">366</span> proc.restoreLock(getEnvironment());<a name="line.366"></a> -<span class="sourceLineNo">367</span> restored.add(proc.getProcId());<a name="line.367"></a> -<span class="sourceLineNo">368</span> }<a name="line.368"></a> -<span class="sourceLineNo">369</span><a name="line.369"></a> -<span class="sourceLineNo">370</span> private void restoreLocks(Deque<Procedure<TEnvironment>> stack, Set<Long> restored) {<a name="line.370"></a> -<span class="sourceLineNo">371</span> while (!stack.isEmpty()) {<a name="line.371"></a> -<span class="sourceLineNo">372</span> restoreLock(stack.pop(), restored);<a name="line.372"></a> -<span class="sourceLineNo">373</span> }<a name="line.373"></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> // Restore the locks for all the procedures.<a name="line.376"></a> -<span class="sourceLineNo">377</span> // Notice that we need to restore the locks starting from the root proc, otherwise there will be<a name="line.377"></a> -<span class="sourceLineNo">378</span> // problem that a sub procedure may hold the exclusive lock first and then we are stuck when<a name="line.378"></a> -<span class="sourceLineNo">379</span> // calling the acquireLock method for the parent procedure.<a name="line.379"></a> -<span class="sourceLineNo">380</span> // The algorithm is straight-forward:<a name="line.380"></a> -<span class="sourceLineNo">381</span> // 1. Use a set to record the procedures which locks have already been restored.<a name="line.381"></a> -<span class="sourceLineNo">382</span> // 2. Use a stack to store the hierarchy of the procedures<a name="line.382"></a> -<span class="sourceLineNo">383</span> // 3. For all the procedure, we will first try to find its parent and push it into the stack,<a name="line.383"></a> -<span class="sourceLineNo">384</span> // unless<a name="line.384"></a> -<span class="sourceLineNo">385</span> // a. We have no parent, i.e, we are the root procedure<a name="line.385"></a> -<span class="sourceLineNo">386</span> // b. The lock has already been restored(by checking the set introduced in #1)<a name="line.386"></a> -<span class="sourceLineNo">387</span> // then we start to pop the stack and call acquireLock for each procedure.<a name="line.387"></a> -<span class="sourceLineNo">388</span> // Notice that this should be done for all procedures, not only the ones in runnableList.<a name="line.388"></a> -<span class="sourceLineNo">389</span> private void restoreLocks() {<a name="line.389"></a> -<span class="sourceLineNo">390</span> Set<Long> restored = new HashSet<>();<a name="line.390"></a> -<span class="sourceLineNo">391</span> Deque<Procedure<TEnvironment>> stack = new ArrayDeque<>();<a name="line.391"></a> -<span class="sourceLineNo">392</span> procedures.values().forEach(proc -> {<a name="line.392"></a> -<span class="sourceLineNo">393</span> for (;;) {<a name="line.393"></a> -<span class="sourceLineNo">394</span> if (restored.contains(proc.getProcId())) {<a name="line.394"></a> -<span class="sourceLineNo">395</span> restoreLocks(stack, restored);<a name="line.395"></a> -<span class="sourceLineNo">396</span> return;<a name="line.396"></a> -<span class="sourceLineNo">397</span> }<a name="line.397"></a> -<span class="sourceLineNo">398</span> if (!proc.hasParent()) {<a name="line.398"></a> -<span class="sourceLineNo">399</span> restoreLock(proc, restored);<a name="line.399"></a> -<span class="sourceLineNo">400</span> restoreLocks(stack, restored);<a name="line.400"></a> -<span class="sourceLineNo">401</span> return;<a name="line.401"></a> -<span class="sourceLineNo">402</span> }<a name="line.402"></a> -<span class="sourceLineNo">403</span> stack.push(proc);<a name="line.403"></a> -<span class="sourceLineNo">404</span> proc = procedures.get(proc.getParentProcId());<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 loadProcedures(ProcedureIterator procIter, boolean abortOnCorruption)<a name="line.409"></a> -<span class="sourceLineNo">410</span> throws IOException {<a name="line.410"></a> -<span class="sourceLineNo">411</span> // 1. Build the rollback stack<a name="line.411"></a> -<span class="sourceLineNo">412</span> int runnablesCount = 0;<a name="line.412"></a> -<span class="sourceLineNo">413</span> int failedCount = 0;<a name="line.413"></a> -<span class="sourceLineNo">414</span> while (procIter.hasNext()) {<a name="line.414"></a> -<span class="sourceLineNo">415</span> boolean finished = procIter.isNextFinished();<a name="line.415"></a> -<span class="sourceLineNo">416</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.416"></a> -<span class="sourceLineNo">417</span> NonceKey nonceKey = proc.getNonceKey();<a name="line.417"></a> -<span class="sourceLineNo">418</span> long procId = proc.getProcId();<a name="line.418"></a> -<span class="sourceLineNo">419</span><a name="line.419"></a> -<span class="sourceLineNo">420</span> if (finished) {<a name="line.420"></a> -<span class="sourceLineNo">421</span> completed.put(proc.getProcId(), new CompletedProcedureRetainer<>(proc));<a name="line.421"></a> -<span class="sourceLineNo">422</span> LOG.debug("Completed {}", proc);<a name="line.422"></a> -<span class="sourceLineNo">423</span> } else {<a name="line.423"></a> -<span class="sourceLineNo">424</span> if (!proc.hasParent()) {<a name="line.424"></a> -<span class="sourceLineNo">425</span> assert !proc.isFinished() : "unexpected finished procedure";<a name="line.425"></a> -<span class="sourceLineNo">426</span> rollbackStack.put(proc.getProcId(), new RootProcedureState<>());<a name="line.426"></a> -<span class="sourceLineNo">427</span> }<a name="line.427"></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">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> // add the procedure to the map<a name="line.429"></a> -<span class="sourceLineNo">430</span> proc.beforeReplay(getEnvironment());<a name="line.430"></a> -<span class="sourceLineNo">431</span> procedures.put(proc.getProcId(), proc);<a name="line.431"></a> -<span class="sourceLineNo">432</span><a name="line.432"></a> -<span class="sourceLineNo">433</span> if (proc.getState() == ProcedureState.RUNNABLE) {<a name="line.433"></a> -<span class="sourceLineNo">434</span> runnablesCount++;<a name="line.434"></a> -<span class="sourceLineNo">435</span> } else if (proc.getState() == ProcedureState.FAILED) {<a name="line.435"></a> -<span class="sourceLineNo">436</span> failedCount++;<a name="line.436"></a> -<span class="sourceLineNo">437</span> }<a name="line.437"></a> -<span class="sourceLineNo">438</span> }<a name="line.438"></a> -<span class="sourceLineNo">439</span><a name="line.439"></a> -<span class="sourceLineNo">440</span> // add the nonce to the map<a name="line.440"></a> -<span class="sourceLineNo">441</span> if (nonceKey != null) {<a name="line.441"></a> -<span class="sourceLineNo">442</span> nonceKeysToProcIdsMap.put(nonceKey, procId);<a name="line.442"></a> -<span class="sourceLineNo">443</span> }<a name="line.443"></a> -<span class="sourceLineNo">444</span> }<a name="line.444"></a> -<span class="sourceLineNo">445</span><a name="line.445"></a> -<span class="sourceLineNo">446</span> // 2. Initialize the stacks<a name="line.446"></a> -<span class="sourceLineNo">447</span> // In the old implementation, for procedures in FAILED state, we will push it into the<a name="line.447"></a> -<span class="sourceLineNo">448</span> // ProcedureScheduler directly to execute the rollback. But this does not work after we<a name="line.448"></a> -<span class="sourceLineNo">449</span> // introduce the restore lock stage.<a name="line.449"></a> -<span class="sourceLineNo">450</span> // For now, when we acquire a xlock, we will remove the queue from runQueue in scheduler, and<a name="line.450"></a> -<span class="sourceLineNo">451</span> // then when a procedure which has lock access, for example, a sub procedure of the procedure<a name="line.451"></a> -<span class="sourceLineNo">452</span> // which has the xlock, is pushed into the scheduler, we will add the queue back to let the<a name="line.452"></a> -<span class="sourceLineNo">453</span> // workers poll from it. The assumption here is that, the procedure which has the xlock should<a name="line.453"></a> -<span class="sourceLineNo">454</span> // have been polled out already, so when loading we can not add the procedure to scheduler first<a name="line.454"></a> -<span class="sourceLineNo">455</span> // and then call acquireLock, since the procedure is still in the queue, and since we will<a name="line.455"></a> -<span class="sourceLineNo">456</span> // remove the queue from runQueue, then no one can poll it out, then there is a dead lock<a name="line.456"></a> -<span class="sourceLineNo">457</span> List<Procedure<TEnvironment>> runnableList = new ArrayList<>(runnablesCount);<a name="line.457"></a> -<span class="sourceLineNo">458</span> List<Procedure<TEnvironment>> failedList = new ArrayList<>(failedCount);<a name="line.458"></a> -<span class="sourceLineNo">459</span> Set<Procedure<TEnvironment>> waitingSet = null;<a name="line.459"></a> -<span class="sourceLineNo">460</span> procIter.reset();<a name="line.460"></a> -<span class="sourceLineNo">461</span> while (procIter.hasNext()) {<a name="line.461"></a> -<span class="sourceLineNo">462</span> if (procIter.isNextFinished()) {<a name="line.462"></a> -<span class="sourceLineNo">463</span> procIter.skipNext();<a name="line.463"></a> -<span class="sourceLineNo">464</span> continue;<a name="line.464"></a> -<span class="sourceLineNo">465</span> }<a name="line.465"></a> -<span class="sourceLineNo">466</span><a name="line.466"></a> -<span class="sourceLineNo">467</span> Procedure<TEnvironment> proc = procIter.next();<a name="line.467"></a> -<span class="sourceLineNo">468</span> assert !(proc.isFinished() && !proc.hasParent()) : "unexpected completed proc=" + proc;<a name="line.468"></a> -<span class="sourceLineNo">469</span><a name="line.469"></a> -<span class="sourceLineNo">470</span> LOG.debug("Loading {}", proc);<a name="line.470"></a> -<span class="sourceLineNo">471</span><a name="line.471"></a> -<span class="sourceLineNo">472</span> Long rootProcId = getRootProcedureId(proc);<a name="line.472"></a> -<span class="sourceLineNo">473</span> if (rootProcId == null) {<a name="line.473"></a> -<span class="sourceLineNo">474</span> // The 'proc' was ready to run but the root procedure was rolledback?<a name="line.474"></a> -<span class="sourceLineNo">475</span> scheduler.addBack(proc);<a name="line.475"></a> -<span class="sourceLineNo">476</span> continue;<a name="line.476"></a> -<span class="sourceLineNo">477</span> }<a name="line.477"></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">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> if (proc.hasParent()) {<a name="line.479"></a> -<span class="sourceLineNo">480</span> Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());<a name="line.480"></a> -<span class="sourceLineNo">481</span> // corrupted procedures are handled later at step 3<a name="line.481"></a> -<span class="sourceLineNo">482</span> if (parent != null && !proc.isFinished()) {<a name="line.482"></a> -<span class="sourc
<TRUNCATED>