http://git-wip-us.apache.org/repos/asf/hbase-site/blob/3469cbc0/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/UnassignProcedure.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/UnassignProcedure.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/UnassignProcedure.html index ca94fbc..269fcc2 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/UnassignProcedure.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/assignment/UnassignProcedure.html @@ -207,114 +207,154 @@ <span class="sourceLineNo">199</span> return false;<a name="line.199"></a> <span class="sourceLineNo">200</span> }<a name="line.200"></a> <span class="sourceLineNo">201</span><a name="line.201"></a> -<span class="sourceLineNo">202</span> // Mark the region as CLOSING.<a name="line.202"></a> -<span class="sourceLineNo">203</span> env.getAssignmentManager().markRegionAsClosing(regionNode);<a name="line.203"></a> -<span class="sourceLineNo">204</span><a name="line.204"></a> -<span class="sourceLineNo">205</span> // Add the close region operation the the server dispatch queue.<a name="line.205"></a> -<span class="sourceLineNo">206</span> if (!addToRemoteDispatcher(env, regionNode.getRegionLocation())) {<a name="line.206"></a> -<span class="sourceLineNo">207</span> // If addToRemoteDispatcher fails, it calls the callback #remoteCallFailed.<a name="line.207"></a> -<span class="sourceLineNo">208</span> }<a name="line.208"></a> -<span class="sourceLineNo">209</span><a name="line.209"></a> -<span class="sourceLineNo">210</span> // Return true to keep the procedure running.<a name="line.210"></a> -<span class="sourceLineNo">211</span> return true;<a name="line.211"></a> -<span class="sourceLineNo">212</span> }<a name="line.212"></a> -<span class="sourceLineNo">213</span><a name="line.213"></a> -<span class="sourceLineNo">214</span> @Override<a name="line.214"></a> -<span class="sourceLineNo">215</span> protected void finishTransition(final MasterProcedureEnv env, final RegionStateNode regionNode)<a name="line.215"></a> -<span class="sourceLineNo">216</span> throws IOException {<a name="line.216"></a> -<span class="sourceLineNo">217</span> AssignmentManager am = env.getAssignmentManager();<a name="line.217"></a> -<span class="sourceLineNo">218</span> RegionInfo regionInfo = getRegionInfo();<a name="line.218"></a> -<span class="sourceLineNo">219</span><a name="line.219"></a> -<span class="sourceLineNo">220</span> if (!removeAfterUnassigning) {<a name="line.220"></a> -<span class="sourceLineNo">221</span> am.markRegionAsClosed(regionNode);<a name="line.221"></a> -<span class="sourceLineNo">222</span> } else {<a name="line.222"></a> -<span class="sourceLineNo">223</span> // Remove from in-memory states<a name="line.223"></a> -<span class="sourceLineNo">224</span> am.getRegionStates().deleteRegion(regionInfo);<a name="line.224"></a> -<span class="sourceLineNo">225</span> env.getMasterServices().getServerManager().removeRegion(regionInfo);<a name="line.225"></a> -<span class="sourceLineNo">226</span> FavoredNodesManager fnm = env.getMasterServices().getFavoredNodesManager();<a name="line.226"></a> -<span class="sourceLineNo">227</span> if (fnm != null) {<a name="line.227"></a> -<span class="sourceLineNo">228</span> fnm.deleteFavoredNodesForRegions(Lists.newArrayList(regionInfo));<a name="line.228"></a> -<span class="sourceLineNo">229</span> }<a name="line.229"></a> -<span class="sourceLineNo">230</span> }<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> @Override<a name="line.233"></a> -<span class="sourceLineNo">234</span> public RemoteOperation remoteCallBuild(final MasterProcedureEnv env, final ServerName serverName) {<a name="line.234"></a> -<span class="sourceLineNo">235</span> assert serverName.equals(getRegionState(env).getRegionLocation());<a name="line.235"></a> -<span class="sourceLineNo">236</span> return new RegionCloseOperation(this, getRegionInfo(), this.destinationServer);<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> @Override<a name="line.239"></a> -<span class="sourceLineNo">240</span> protected void reportTransition(final MasterProcedureEnv env, final RegionStateNode regionNode,<a name="line.240"></a> -<span class="sourceLineNo">241</span> final TransitionCode code, final long seqId) throws UnexpectedStateException {<a name="line.241"></a> -<span class="sourceLineNo">242</span> switch (code) {<a name="line.242"></a> -<span class="sourceLineNo">243</span> case CLOSED:<a name="line.243"></a> -<span class="sourceLineNo">244</span> setTransitionState(RegionTransitionState.REGION_TRANSITION_FINISH);<a name="line.244"></a> -<span class="sourceLineNo">245</span> break;<a name="line.245"></a> -<span class="sourceLineNo">246</span> default:<a name="line.246"></a> -<span class="sourceLineNo">247</span> throw new UnexpectedStateException(String.format(<a name="line.247"></a> -<span class="sourceLineNo">248</span> "Received report unexpected transition state=%s for region=%s server=%s, expected CLOSED.",<a name="line.248"></a> -<span class="sourceLineNo">249</span> code, regionNode.getRegionInfo(), regionNode.getRegionLocation()));<a name="line.249"></a> -<span class="sourceLineNo">250</span> }<a name="line.250"></a> -<span class="sourceLineNo">251</span> }<a name="line.251"></a> -<span class="sourceLineNo">252</span><a name="line.252"></a> -<span class="sourceLineNo">253</span> @Override<a name="line.253"></a> -<span class="sourceLineNo">254</span> protected boolean remoteCallFailed(final MasterProcedureEnv env, final RegionStateNode regionNode,<a name="line.254"></a> -<span class="sourceLineNo">255</span> final IOException exception) {<a name="line.255"></a> -<span class="sourceLineNo">256</span> // TODO: Is there on-going rpc to cleanup?<a name="line.256"></a> -<span class="sourceLineNo">257</span> if (exception instanceof ServerCrashException) {<a name="line.257"></a> -<span class="sourceLineNo">258</span> // This exception comes from ServerCrashProcedure AFTER log splitting.<a name="line.258"></a> -<span class="sourceLineNo">259</span> // SCP found this region as a RIT. Its call into here says it is ok to let this procedure go<a name="line.259"></a> -<span class="sourceLineNo">260</span> // complete. This complete will release lock on this region so subsequent action on region<a name="line.260"></a> -<span class="sourceLineNo">261</span> // can succeed; e.g. the assign that follows this unassign when a move (w/o wait on SCP<a name="line.261"></a> -<span class="sourceLineNo">262</span> // the assign could run w/o logs being split so data loss).<a name="line.262"></a> -<span class="sourceLineNo">263</span> try {<a name="line.263"></a> -<span class="sourceLineNo">264</span> reportTransition(env, regionNode, TransitionCode.CLOSED, HConstants.NO_SEQNUM);<a name="line.264"></a> -<span class="sourceLineNo">265</span> } catch (UnexpectedStateException e) {<a name="line.265"></a> -<span class="sourceLineNo">266</span> // Should never happen.<a name="line.266"></a> -<span class="sourceLineNo">267</span> throw new RuntimeException(e);<a name="line.267"></a> -<span class="sourceLineNo">268</span> }<a name="line.268"></a> -<span class="sourceLineNo">269</span> } else if (exception instanceof RegionServerAbortedException ||<a name="line.269"></a> -<span class="sourceLineNo">270</span> exception instanceof RegionServerStoppedException ||<a name="line.270"></a> -<span class="sourceLineNo">271</span> exception instanceof ServerNotRunningYetException) {<a name="line.271"></a> -<span class="sourceLineNo">272</span> // RS is aborting, we cannot offline the region since the region may need to do WAL<a name="line.272"></a> -<span class="sourceLineNo">273</span> // recovery. Until we see the RS expiration, we should retry.<a name="line.273"></a> -<span class="sourceLineNo">274</span> // TODO: This should be suspend like the below where we call expire on server?<a name="line.274"></a> -<span class="sourceLineNo">275</span> LOG.info("Ignoring; waiting on ServerCrashProcedure", exception);<a name="line.275"></a> -<span class="sourceLineNo">276</span> } else if (exception instanceof NotServingRegionException) {<a name="line.276"></a> -<span class="sourceLineNo">277</span> LOG.info("IS THIS OK? ANY LOGS TO REPLAY; ACTING AS THOUGH ALL GOOD " + regionNode,<a name="line.277"></a> -<span class="sourceLineNo">278</span> exception);<a name="line.278"></a> -<span class="sourceLineNo">279</span> setTransitionState(RegionTransitionState.REGION_TRANSITION_FINISH);<a name="line.279"></a> -<span class="sourceLineNo">280</span> } else {<a name="line.280"></a> -<span class="sourceLineNo">281</span> LOG.warn("Expiring server " + this + "; " + regionNode.toShortString() +<a name="line.281"></a> -<span class="sourceLineNo">282</span> ", exception=" + exception);<a name="line.282"></a> -<span class="sourceLineNo">283</span> env.getMasterServices().getServerManager().expireServer(regionNode.getRegionLocation());<a name="line.283"></a> -<span class="sourceLineNo">284</span> // Return false so this procedure stays in suspended state. It will be woken up by the<a name="line.284"></a> -<span class="sourceLineNo">285</span> // ServerCrashProcedure that was scheduled when we called #expireServer above. SCP calls<a name="line.285"></a> -<span class="sourceLineNo">286</span> // #handleRIT which will call this method only the exception will be a ServerCrashException<a name="line.286"></a> -<span class="sourceLineNo">287</span> // this time around (See above).<a name="line.287"></a> -<span class="sourceLineNo">288</span> // TODO: Add a SCP as a new subprocedure that we now come to depend on.<a name="line.288"></a> -<span class="sourceLineNo">289</span> return false;<a name="line.289"></a> -<span class="sourceLineNo">290</span> }<a name="line.290"></a> -<span class="sourceLineNo">291</span> return true;<a name="line.291"></a> -<span class="sourceLineNo">292</span> }<a name="line.292"></a> -<span class="sourceLineNo">293</span><a name="line.293"></a> -<span class="sourceLineNo">294</span> @Override<a name="line.294"></a> -<span class="sourceLineNo">295</span> public void toStringClassDetails(StringBuilder sb) {<a name="line.295"></a> -<span class="sourceLineNo">296</span> super.toStringClassDetails(sb);<a name="line.296"></a> -<span class="sourceLineNo">297</span> sb.append(", server=").append(this.hostingServer);<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> @Override<a name="line.300"></a> -<span class="sourceLineNo">301</span> public ServerName getServer(final MasterProcedureEnv env) {<a name="line.301"></a> -<span class="sourceLineNo">302</span> return this.hostingServer;<a name="line.302"></a> -<span class="sourceLineNo">303</span> }<a name="line.303"></a> -<span class="sourceLineNo">304</span><a name="line.304"></a> -<span class="sourceLineNo">305</span> @Override<a name="line.305"></a> -<span class="sourceLineNo">306</span> protected ProcedureMetrics getProcedureMetrics(MasterProcedureEnv env) {<a name="line.306"></a> -<span class="sourceLineNo">307</span> return env.getAssignmentManager().getAssignmentManagerMetrics().getUnassignProcMetrics();<a name="line.307"></a> -<span class="sourceLineNo">308</span> }<a name="line.308"></a> -<span class="sourceLineNo">309</span>}<a name="line.309"></a> +<span class="sourceLineNo">202</span><a name="line.202"></a> +<span class="sourceLineNo">203</span> // Mark the region as CLOSING.<a name="line.203"></a> +<span class="sourceLineNo">204</span> env.getAssignmentManager().markRegionAsClosing(regionNode);<a name="line.204"></a> +<span class="sourceLineNo">205</span><a name="line.205"></a> +<span class="sourceLineNo">206</span> // Add the close region operation to the server dispatch queue.<a name="line.206"></a> +<span class="sourceLineNo">207</span> if (!addToRemoteDispatcher(env, regionNode.getRegionLocation())) {<a name="line.207"></a> +<span class="sourceLineNo">208</span> // If addToRemoteDispatcher fails, it calls the callback #remoteCallFailed.<a name="line.208"></a> +<span class="sourceLineNo">209</span> }<a name="line.209"></a> +<span class="sourceLineNo">210</span><a name="line.210"></a> +<span class="sourceLineNo">211</span> // Return true to keep the procedure running.<a name="line.211"></a> +<span class="sourceLineNo">212</span> return true;<a name="line.212"></a> +<span class="sourceLineNo">213</span> }<a name="line.213"></a> +<span class="sourceLineNo">214</span><a name="line.214"></a> +<span class="sourceLineNo">215</span> @Override<a name="line.215"></a> +<span class="sourceLineNo">216</span> protected void finishTransition(final MasterProcedureEnv env, final RegionStateNode regionNode)<a name="line.216"></a> +<span class="sourceLineNo">217</span> throws IOException {<a name="line.217"></a> +<span class="sourceLineNo">218</span> AssignmentManager am = env.getAssignmentManager();<a name="line.218"></a> +<span class="sourceLineNo">219</span> RegionInfo regionInfo = getRegionInfo();<a name="line.219"></a> +<span class="sourceLineNo">220</span><a name="line.220"></a> +<span class="sourceLineNo">221</span> if (!removeAfterUnassigning) {<a name="line.221"></a> +<span class="sourceLineNo">222</span> am.markRegionAsClosed(regionNode);<a name="line.222"></a> +<span class="sourceLineNo">223</span> } else {<a name="line.223"></a> +<span class="sourceLineNo">224</span> // Remove from in-memory states<a name="line.224"></a> +<span class="sourceLineNo">225</span> am.getRegionStates().deleteRegion(regionInfo);<a name="line.225"></a> +<span class="sourceLineNo">226</span> env.getMasterServices().getServerManager().removeRegion(regionInfo);<a name="line.226"></a> +<span class="sourceLineNo">227</span> FavoredNodesManager fnm = env.getMasterServices().getFavoredNodesManager();<a name="line.227"></a> +<span class="sourceLineNo">228</span> if (fnm != null) {<a name="line.228"></a> +<span class="sourceLineNo">229</span> fnm.deleteFavoredNodesForRegions(Lists.newArrayList(regionInfo));<a name="line.229"></a> +<span class="sourceLineNo">230</span> }<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><a name="line.233"></a> +<span class="sourceLineNo">234</span> @Override<a name="line.234"></a> +<span class="sourceLineNo">235</span> public RemoteOperation remoteCallBuild(final MasterProcedureEnv env, final ServerName serverName) {<a name="line.235"></a> +<span class="sourceLineNo">236</span> assert serverName.equals(getRegionState(env).getRegionLocation());<a name="line.236"></a> +<span class="sourceLineNo">237</span> return new RegionCloseOperation(this, getRegionInfo(), this.destinationServer);<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> @Override<a name="line.240"></a> +<span class="sourceLineNo">241</span> protected void reportTransition(final MasterProcedureEnv env, final RegionStateNode regionNode,<a name="line.241"></a> +<span class="sourceLineNo">242</span> final TransitionCode code, final long seqId) throws UnexpectedStateException {<a name="line.242"></a> +<span class="sourceLineNo">243</span> switch (code) {<a name="line.243"></a> +<span class="sourceLineNo">244</span> case CLOSED:<a name="line.244"></a> +<span class="sourceLineNo">245</span> setTransitionState(RegionTransitionState.REGION_TRANSITION_FINISH);<a name="line.245"></a> +<span class="sourceLineNo">246</span> break;<a name="line.246"></a> +<span class="sourceLineNo">247</span> default:<a name="line.247"></a> +<span class="sourceLineNo">248</span> throw new UnexpectedStateException(String.format(<a name="line.248"></a> +<span class="sourceLineNo">249</span> "Received report unexpected transition state=%s for region=%s server=%s, expected CLOSED.",<a name="line.249"></a> +<span class="sourceLineNo">250</span> code, regionNode.getRegionInfo(), regionNode.getRegionLocation()));<a name="line.250"></a> +<span class="sourceLineNo">251</span> }<a name="line.251"></a> +<span class="sourceLineNo">252</span> }<a name="line.252"></a> +<span class="sourceLineNo">253</span><a name="line.253"></a> +<span class="sourceLineNo">254</span> /**<a name="line.254"></a> +<span class="sourceLineNo">255</span> * Our remote call failed but there are a few states where it is safe to proceed with the<a name="line.255"></a> +<span class="sourceLineNo">256</span> * unassign; e.g. if a server crash and it has had all of its WALs processed, then we can allow<a name="line.256"></a> +<span class="sourceLineNo">257</span> * this unassign to go to completion.<a name="line.257"></a> +<span class="sourceLineNo">258</span> * @return True if it is safe to proceed with the unassign.<a name="line.258"></a> +<span class="sourceLineNo">259</span> */<a name="line.259"></a> +<span class="sourceLineNo">260</span> private boolean isSafeToProceed(final MasterProcedureEnv env, final RegionStateNode regionNode,<a name="line.260"></a> +<span class="sourceLineNo">261</span> final IOException exception) {<a name="line.261"></a> +<span class="sourceLineNo">262</span> if (exception instanceof ServerCrashException) {<a name="line.262"></a> +<span class="sourceLineNo">263</span> // This exception comes from ServerCrashProcedure AFTER log splitting. Its a signaling<a name="line.263"></a> +<span class="sourceLineNo">264</span> // exception. SCP found this region as a RIT during its processing of the crash. Its call<a name="line.264"></a> +<span class="sourceLineNo">265</span> // into here says it is ok to let this procedure go complete.<a name="line.265"></a> +<span class="sourceLineNo">266</span> return true;<a name="line.266"></a> +<span class="sourceLineNo">267</span> }<a name="line.267"></a> +<span class="sourceLineNo">268</span> if (exception instanceof NotServingRegionException) {<a name="line.268"></a> +<span class="sourceLineNo">269</span> LOG.warn("IS OK? ANY LOGS TO REPLAY; ACTING AS THOUGH ALL GOOD {}", regionNode, exception);<a name="line.269"></a> +<span class="sourceLineNo">270</span> return true;<a name="line.270"></a> +<span class="sourceLineNo">271</span> }<a name="line.271"></a> +<span class="sourceLineNo">272</span> return false;<a name="line.272"></a> +<span class="sourceLineNo">273</span> }<a name="line.273"></a> +<span class="sourceLineNo">274</span><a name="line.274"></a> +<span class="sourceLineNo">275</span> /**<a name="line.275"></a> +<span class="sourceLineNo">276</span> * Set it up so when procedure is unsuspended, we'll move to the procedure finish.<a name="line.276"></a> +<span class="sourceLineNo">277</span> */<a name="line.277"></a> +<span class="sourceLineNo">278</span> protected void proceed(final MasterProcedureEnv env, final RegionStateNode regionNode) {<a name="line.278"></a> +<span class="sourceLineNo">279</span> try {<a name="line.279"></a> +<span class="sourceLineNo">280</span> reportTransition(env, regionNode, TransitionCode.CLOSED, HConstants.NO_SEQNUM);<a name="line.280"></a> +<span class="sourceLineNo">281</span> } catch (UnexpectedStateException e) {<a name="line.281"></a> +<span class="sourceLineNo">282</span> // Should never happen.<a name="line.282"></a> +<span class="sourceLineNo">283</span> throw new RuntimeException(e);<a name="line.283"></a> +<span class="sourceLineNo">284</span> }<a name="line.284"></a> +<span class="sourceLineNo">285</span> }<a name="line.285"></a> +<span class="sourceLineNo">286</span><a name="line.286"></a> +<span class="sourceLineNo">287</span> /**<a name="line.287"></a> +<span class="sourceLineNo">288</span> * @return If true, we will re-wake up this procedure; if false, the procedure stays suspended.<a name="line.288"></a> +<span class="sourceLineNo">289</span> */<a name="line.289"></a> +<span class="sourceLineNo">290</span> @Override<a name="line.290"></a> +<span class="sourceLineNo">291</span> protected boolean remoteCallFailed(final MasterProcedureEnv env, final RegionStateNode regionNode,<a name="line.291"></a> +<span class="sourceLineNo">292</span> final IOException exception) {<a name="line.292"></a> +<span class="sourceLineNo">293</span> // Be careful reading the below; we do returns in middle of the method a few times.<a name="line.293"></a> +<span class="sourceLineNo">294</span> if (isSafeToProceed(env, regionNode, exception)) {<a name="line.294"></a> +<span class="sourceLineNo">295</span> proceed(env, regionNode);<a name="line.295"></a> +<span class="sourceLineNo">296</span> } else if (exception instanceof RegionServerAbortedException ||<a name="line.296"></a> +<span class="sourceLineNo">297</span> exception instanceof RegionServerStoppedException) {<a name="line.297"></a> +<span class="sourceLineNo">298</span> // RS is aborting/stopping, we cannot offline the region since the region may need to do WAL<a name="line.298"></a> +<span class="sourceLineNo">299</span> // recovery. Until we see the RS expiration, stay suspended; return false.<a name="line.299"></a> +<span class="sourceLineNo">300</span> LOG.info("Ignoring; waiting on ServerCrashProcedure", exception);<a name="line.300"></a> +<span class="sourceLineNo">301</span> return false;<a name="line.301"></a> +<span class="sourceLineNo">302</span> } else if (exception instanceof ServerNotRunningYetException) {<a name="line.302"></a> +<span class="sourceLineNo">303</span> // This should not happen. If it does, procedure will be woken-up and we'll retry.<a name="line.303"></a> +<span class="sourceLineNo">304</span> // TODO: Needs a pause and backoff?<a name="line.304"></a> +<span class="sourceLineNo">305</span> LOG.info("Retry", exception);<a name="line.305"></a> +<span class="sourceLineNo">306</span> } else {<a name="line.306"></a> +<span class="sourceLineNo">307</span> // We failed to RPC this server. Set it as expired.<a name="line.307"></a> +<span class="sourceLineNo">308</span> ServerName serverName = regionNode.getRegionLocation();<a name="line.308"></a> +<span class="sourceLineNo">309</span> LOG.warn("Expiring {}, {} {}; exception={}", serverName, this, regionNode.toShortString(),<a name="line.309"></a> +<span class="sourceLineNo">310</span> exception.getClass().getSimpleName());<a name="line.310"></a> +<span class="sourceLineNo">311</span> if (!env.getMasterServices().getServerManager().expireServer(serverName)) {<a name="line.311"></a> +<span class="sourceLineNo">312</span> // Failed to queue an expire. Lots of possible reasons including it may be already expired.<a name="line.312"></a> +<span class="sourceLineNo">313</span> // If so, is it beyond the state where we will be woken-up if go ahead and suspend the<a name="line.313"></a> +<span class="sourceLineNo">314</span> // procedure. Look for this rare condition.<a name="line.314"></a> +<span class="sourceLineNo">315</span> if (env.getAssignmentManager().isDeadServerProcessed(serverName)) {<a name="line.315"></a> +<span class="sourceLineNo">316</span> // Its ok to proceed with this unassign.<a name="line.316"></a> +<span class="sourceLineNo">317</span> LOG.info("{} is dead and processed; moving procedure to finished state; {}",<a name="line.317"></a> +<span class="sourceLineNo">318</span> serverName, this);<a name="line.318"></a> +<span class="sourceLineNo">319</span> proceed(env, regionNode);<a name="line.319"></a> +<span class="sourceLineNo">320</span> // Return true; wake up the procedure so we can act on proceed.<a name="line.320"></a> +<span class="sourceLineNo">321</span> return true;<a name="line.321"></a> +<span class="sourceLineNo">322</span> }<a name="line.322"></a> +<span class="sourceLineNo">323</span> }<a name="line.323"></a> +<span class="sourceLineNo">324</span> // Return false so this procedure stays in suspended state. It will be woken up by the<a name="line.324"></a> +<span class="sourceLineNo">325</span> // ServerCrashProcedure that was scheduled when we called #expireServer above. SCP calls<a name="line.325"></a> +<span class="sourceLineNo">326</span> // #handleRIT which will call this method only the exception will be a ServerCrashException<a name="line.326"></a> +<span class="sourceLineNo">327</span> // this time around (See above).<a name="line.327"></a> +<span class="sourceLineNo">328</span> // TODO: Add a SCP as a new subprocedure that we now come to depend on.<a name="line.328"></a> +<span class="sourceLineNo">329</span> return false;<a name="line.329"></a> +<span class="sourceLineNo">330</span> }<a name="line.330"></a> +<span class="sourceLineNo">331</span> return true;<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> @Override<a name="line.334"></a> +<span class="sourceLineNo">335</span> public void toStringClassDetails(StringBuilder sb) {<a name="line.335"></a> +<span class="sourceLineNo">336</span> super.toStringClassDetails(sb);<a name="line.336"></a> +<span class="sourceLineNo">337</span> sb.append(", server=").append(this.hostingServer);<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> @Override<a name="line.340"></a> +<span class="sourceLineNo">341</span> public ServerName getServer(final MasterProcedureEnv env) {<a name="line.341"></a> +<span class="sourceLineNo">342</span> return this.hostingServer;<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> protected ProcedureMetrics getProcedureMetrics(MasterProcedureEnv env) {<a name="line.346"></a> +<span class="sourceLineNo">347</span> return env.getAssignmentManager().getAssignmentManagerMetrics().getUnassignProcMetrics();<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>
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/3469cbc0/devapidocs/src-html/org/apache/hadoop/hbase/master/procedure/ServerCrashProcedure.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/procedure/ServerCrashProcedure.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/procedure/ServerCrashProcedure.html index bd8f253..455d1b4 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/procedure/ServerCrashProcedure.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/procedure/ServerCrashProcedure.html @@ -171,265 +171,267 @@ <span class="sourceLineNo">163</span> "; cycles=" + getCycles());<a name="line.163"></a> <span class="sourceLineNo">164</span> }<a name="line.164"></a> <span class="sourceLineNo">165</span> // Handle RIT against crashed server. Will cancel any ongoing assigns/unassigns.<a name="line.165"></a> -<span class="sourceLineNo">166</span> // Returns list of regions we need to reassign.<a name="line.166"></a> -<span class="sourceLineNo">167</span> List<RegionInfo> toAssign = handleRIT(env, regionsOnCrashedServer);<a name="line.167"></a> -<span class="sourceLineNo">168</span> AssignmentManager am = env.getAssignmentManager();<a name="line.168"></a> -<span class="sourceLineNo">169</span> // CreateAssignProcedure will try to use the old location for the region deploy.<a name="line.169"></a> -<span class="sourceLineNo">170</span> addChildProcedure(am.createAssignProcedures(toAssign));<a name="line.170"></a> -<span class="sourceLineNo">171</span> setNextState(ServerCrashState.SERVER_CRASH_HANDLE_RIT2);<a name="line.171"></a> -<span class="sourceLineNo">172</span> } else {<a name="line.172"></a> -<span class="sourceLineNo">173</span> setNextState(ServerCrashState.SERVER_CRASH_FINISH);<a name="line.173"></a> -<span class="sourceLineNo">174</span> }<a name="line.174"></a> -<span class="sourceLineNo">175</span> break;<a name="line.175"></a> -<span class="sourceLineNo">176</span><a name="line.176"></a> -<span class="sourceLineNo">177</span> case SERVER_CRASH_HANDLE_RIT2:<a name="line.177"></a> -<span class="sourceLineNo">178</span> // Run the handleRIT again for case where another procedure managed to grab the lock on<a name="line.178"></a> -<span class="sourceLineNo">179</span> // a region ahead of this crash handling procedure. Can happen in rare case. See<a name="line.179"></a> -<span class="sourceLineNo">180</span> handleRIT(env, regionsOnCrashedServer);<a name="line.180"></a> -<span class="sourceLineNo">181</span> setNextState(ServerCrashState.SERVER_CRASH_FINISH);<a name="line.181"></a> -<span class="sourceLineNo">182</span> break;<a name="line.182"></a> -<span class="sourceLineNo">183</span><a name="line.183"></a> -<span class="sourceLineNo">184</span> case SERVER_CRASH_FINISH:<a name="line.184"></a> -<span class="sourceLineNo">185</span> services.getAssignmentManager().getRegionStates().removeServer(serverName);<a name="line.185"></a> -<span class="sourceLineNo">186</span> services.getServerManager().getDeadServers().finish(serverName);<a name="line.186"></a> -<span class="sourceLineNo">187</span> return Flow.NO_MORE_STATE;<a name="line.187"></a> -<span class="sourceLineNo">188</span><a name="line.188"></a> -<span class="sourceLineNo">189</span> default:<a name="line.189"></a> -<span class="sourceLineNo">190</span> throw new UnsupportedOperationException("unhandled state=" + state);<a name="line.190"></a> -<span class="sourceLineNo">191</span> }<a name="line.191"></a> -<span class="sourceLineNo">192</span> } catch (IOException e) {<a name="line.192"></a> -<span class="sourceLineNo">193</span> LOG.warn("Failed state=" + state + ", retry " + this + "; cycles=" + getCycles(), e);<a name="line.193"></a> -<span class="sourceLineNo">194</span> }<a name="line.194"></a> -<span class="sourceLineNo">195</span> return Flow.HAS_MORE_STATE;<a name="line.195"></a> -<span class="sourceLineNo">196</span> }<a name="line.196"></a> -<span class="sourceLineNo">197</span><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> * @param env<a name="line.200"></a> -<span class="sourceLineNo">201</span> * @throws IOException<a name="line.201"></a> -<span class="sourceLineNo">202</span> */<a name="line.202"></a> -<span class="sourceLineNo">203</span> private void processMeta(final MasterProcedureEnv env) throws IOException {<a name="line.203"></a> -<span class="sourceLineNo">204</span> LOG.debug("{}; processing hbase:meta", this);<a name="line.204"></a> -<span class="sourceLineNo">205</span><a name="line.205"></a> -<span class="sourceLineNo">206</span> // Assign meta if still carrying it. Check again: region may be assigned because of RIT timeout<a name="line.206"></a> -<span class="sourceLineNo">207</span> final AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.207"></a> -<span class="sourceLineNo">208</span> for (RegionInfo hri: am.getRegionStates().getServerRegionInfoSet(serverName)) {<a name="line.208"></a> -<span class="sourceLineNo">209</span> if (!isDefaultMetaRegion(hri)) {<a name="line.209"></a> -<span class="sourceLineNo">210</span> continue;<a name="line.210"></a> -<span class="sourceLineNo">211</span> }<a name="line.211"></a> -<span class="sourceLineNo">212</span> addChildProcedure(new RecoverMetaProcedure(serverName, this.shouldSplitWal));<a name="line.212"></a> -<span class="sourceLineNo">213</span> }<a name="line.213"></a> -<span class="sourceLineNo">214</span> }<a name="line.214"></a> -<span class="sourceLineNo">215</span><a name="line.215"></a> -<span class="sourceLineNo">216</span> private boolean filterDefaultMetaRegions(final List<RegionInfo> regions) {<a name="line.216"></a> -<span class="sourceLineNo">217</span> if (regions == null) return false;<a name="line.217"></a> -<span class="sourceLineNo">218</span> regions.removeIf(this::isDefaultMetaRegion);<a name="line.218"></a> -<span class="sourceLineNo">219</span> return !regions.isEmpty();<a name="line.219"></a> -<span class="sourceLineNo">220</span> }<a name="line.220"></a> -<span class="sourceLineNo">221</span><a name="line.221"></a> -<span class="sourceLineNo">222</span> private boolean isDefaultMetaRegion(final RegionInfo hri) {<a name="line.222"></a> -<span class="sourceLineNo">223</span> return hri.getTable().equals(TableName.META_TABLE_NAME) &&<a name="line.223"></a> -<span class="sourceLineNo">224</span> RegionReplicaUtil.isDefaultReplica(hri);<a name="line.224"></a> -<span class="sourceLineNo">225</span> }<a name="line.225"></a> -<span class="sourceLineNo">226</span><a name="line.226"></a> -<span class="sourceLineNo">227</span> private void splitLogs(final MasterProcedureEnv env) throws IOException {<a name="line.227"></a> -<span class="sourceLineNo">228</span> if (LOG.isDebugEnabled()) {<a name="line.228"></a> -<span class="sourceLineNo">229</span> LOG.debug("Splitting WALs " + this);<a name="line.229"></a> -<span class="sourceLineNo">230</span> }<a name="line.230"></a> -<span class="sourceLineNo">231</span> MasterWalManager mwm = env.getMasterServices().getMasterWalManager();<a name="line.231"></a> -<span class="sourceLineNo">232</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.232"></a> -<span class="sourceLineNo">233</span> // TODO: For Matteo. Below BLOCKs!!!! Redo so can relinquish executor while it is running.<a name="line.233"></a> -<span class="sourceLineNo">234</span> // PROBLEM!!! WE BLOCK HERE.<a name="line.234"></a> -<span class="sourceLineNo">235</span> mwm.splitLog(this.serverName);<a name="line.235"></a> -<span class="sourceLineNo">236</span> if (LOG.isDebugEnabled()) {<a name="line.236"></a> -<span class="sourceLineNo">237</span> LOG.debug("Done splitting WALs " + this);<a name="line.237"></a> -<span class="sourceLineNo">238</span> }<a name="line.238"></a> -<span class="sourceLineNo">239</span> am.getRegionStates().logSplit(this.serverName);<a name="line.239"></a> -<span class="sourceLineNo">240</span> }<a name="line.240"></a> -<span class="sourceLineNo">241</span><a name="line.241"></a> -<span class="sourceLineNo">242</span> @Override<a name="line.242"></a> -<span class="sourceLineNo">243</span> protected void rollbackState(MasterProcedureEnv env, ServerCrashState state)<a name="line.243"></a> -<span class="sourceLineNo">244</span> throws IOException {<a name="line.244"></a> -<span class="sourceLineNo">245</span> // Can't rollback.<a name="line.245"></a> -<span class="sourceLineNo">246</span> throw new UnsupportedOperationException("unhandled state=" + state);<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> @Override<a name="line.249"></a> -<span class="sourceLineNo">250</span> protected ServerCrashState getState(int stateId) {<a name="line.250"></a> -<span class="sourceLineNo">251</span> return ServerCrashState.forNumber(stateId);<a name="line.251"></a> -<span class="sourceLineNo">252</span> }<a name="line.252"></a> -<span class="sourceLineNo">253</span><a name="line.253"></a> -<span class="sourceLineNo">254</span> @Override<a name="line.254"></a> -<span class="sourceLineNo">255</span> protected int getStateId(ServerCrashState state) {<a name="line.255"></a> -<span class="sourceLineNo">256</span> return state.getNumber();<a name="line.256"></a> -<span class="sourceLineNo">257</span> }<a name="line.257"></a> -<span class="sourceLineNo">258</span><a name="line.258"></a> -<span class="sourceLineNo">259</span> @Override<a name="line.259"></a> -<span class="sourceLineNo">260</span> protected ServerCrashState getInitialState() {<a name="line.260"></a> -<span class="sourceLineNo">261</span> return ServerCrashState.SERVER_CRASH_START;<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> @Override<a name="line.264"></a> -<span class="sourceLineNo">265</span> protected boolean abort(MasterProcedureEnv env) {<a name="line.265"></a> -<span class="sourceLineNo">266</span> // TODO<a name="line.266"></a> -<span class="sourceLineNo">267</span> return false;<a name="line.267"></a> -<span class="sourceLineNo">268</span> }<a name="line.268"></a> -<span class="sourceLineNo">269</span><a name="line.269"></a> -<span class="sourceLineNo">270</span> @Override<a name="line.270"></a> -<span class="sourceLineNo">271</span> protected LockState acquireLock(final MasterProcedureEnv env) {<a name="line.271"></a> -<span class="sourceLineNo">272</span> // TODO: Put this BACK AFTER AMv2 goes in!!!!<a name="line.272"></a> -<span class="sourceLineNo">273</span> // if (env.waitFailoverCleanup(this)) return LockState.LOCK_EVENT_WAIT;<a name="line.273"></a> -<span class="sourceLineNo">274</span> if (env.waitServerCrashProcessingEnabled(this)) return LockState.LOCK_EVENT_WAIT;<a name="line.274"></a> -<span class="sourceLineNo">275</span> if (env.getProcedureScheduler().waitServerExclusiveLock(this, getServerName())) {<a name="line.275"></a> -<span class="sourceLineNo">276</span> return LockState.LOCK_EVENT_WAIT;<a name="line.276"></a> -<span class="sourceLineNo">277</span> }<a name="line.277"></a> -<span class="sourceLineNo">278</span> return LockState.LOCK_ACQUIRED;<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> @Override<a name="line.281"></a> -<span class="sourceLineNo">282</span> protected void releaseLock(final MasterProcedureEnv env) {<a name="line.282"></a> -<span class="sourceLineNo">283</span> env.getProcedureScheduler().wakeServerExclusiveLock(this, getServerName());<a name="line.283"></a> -<span class="sourceLineNo">284</span> }<a name="line.284"></a> -<span class="sourceLineNo">285</span><a name="line.285"></a> -<span class="sourceLineNo">286</span> @Override<a name="line.286"></a> -<span class="sourceLineNo">287</span> public void toStringClassDetails(StringBuilder sb) {<a name="line.287"></a> -<span class="sourceLineNo">288</span> sb.append(getClass().getSimpleName());<a name="line.288"></a> -<span class="sourceLineNo">289</span> sb.append(" server=");<a name="line.289"></a> -<span class="sourceLineNo">290</span> sb.append(serverName);<a name="line.290"></a> -<span class="sourceLineNo">291</span> sb.append(", splitWal=");<a name="line.291"></a> -<span class="sourceLineNo">292</span> sb.append(shouldSplitWal);<a name="line.292"></a> -<span class="sourceLineNo">293</span> sb.append(", meta=");<a name="line.293"></a> -<span class="sourceLineNo">294</span> sb.append(carryingMeta);<a name="line.294"></a> -<span class="sourceLineNo">295</span> }<a name="line.295"></a> -<span class="sourceLineNo">296</span><a name="line.296"></a> -<span class="sourceLineNo">297</span> @Override<a name="line.297"></a> -<span class="sourceLineNo">298</span> protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.298"></a> -<span class="sourceLineNo">299</span> throws IOException {<a name="line.299"></a> -<span class="sourceLineNo">300</span> super.serializeStateData(serializer);<a name="line.300"></a> -<span class="sourceLineNo">301</span><a name="line.301"></a> -<span class="sourceLineNo">302</span> MasterProcedureProtos.ServerCrashStateData.Builder state =<a name="line.302"></a> -<span class="sourceLineNo">303</span> MasterProcedureProtos.ServerCrashStateData.newBuilder().<a name="line.303"></a> -<span class="sourceLineNo">304</span> setServerName(ProtobufUtil.toServerName(this.serverName)).<a name="line.304"></a> -<span class="sourceLineNo">305</span> setCarryingMeta(this.carryingMeta).<a name="line.305"></a> -<span class="sourceLineNo">306</span> setShouldSplitWal(this.shouldSplitWal);<a name="line.306"></a> -<span class="sourceLineNo">307</span> if (this.regionsOnCrashedServer != null && !this.regionsOnCrashedServer.isEmpty()) {<a name="line.307"></a> -<span class="sourceLineNo">308</span> for (RegionInfo hri: this.regionsOnCrashedServer) {<a name="line.308"></a> -<span class="sourceLineNo">309</span> state.addRegionsOnCrashedServer(ProtobufUtil.toRegionInfo(hri));<a name="line.309"></a> -<span class="sourceLineNo">310</span> }<a name="line.310"></a> -<span class="sourceLineNo">311</span> }<a name="line.311"></a> -<span class="sourceLineNo">312</span> serializer.serialize(state.build());<a name="line.312"></a> -<span class="sourceLineNo">313</span> }<a name="line.313"></a> -<span class="sourceLineNo">314</span><a name="line.314"></a> -<span class="sourceLineNo">315</span> @Override<a name="line.315"></a> -<span class="sourceLineNo">316</span> protected void deserializeStateData(ProcedureStateSerializer serializer)<a name="line.316"></a> -<span class="sourceLineNo">317</span> throws IOException {<a name="line.317"></a> -<span class="sourceLineNo">318</span> super.deserializeStateData(serializer);<a name="line.318"></a> -<span class="sourceLineNo">319</span><a name="line.319"></a> -<span class="sourceLineNo">320</span> MasterProcedureProtos.ServerCrashStateData state =<a name="line.320"></a> -<span class="sourceLineNo">321</span> serializer.deserialize(MasterProcedureProtos.ServerCrashStateData.class);<a name="line.321"></a> -<span class="sourceLineNo">322</span> this.serverName = ProtobufUtil.toServerName(state.getServerName());<a name="line.322"></a> -<span class="sourceLineNo">323</span> this.carryingMeta = state.hasCarryingMeta()? state.getCarryingMeta(): false;<a name="line.323"></a> -<span class="sourceLineNo">324</span> // shouldSplitWAL has a default over in pb so this invocation will always work.<a name="line.324"></a> -<span class="sourceLineNo">325</span> this.shouldSplitWal = state.getShouldSplitWal();<a name="line.325"></a> -<span class="sourceLineNo">326</span> int size = state.getRegionsOnCrashedServerCount();<a name="line.326"></a> -<span class="sourceLineNo">327</span> if (size > 0) {<a name="line.327"></a> -<span class="sourceLineNo">328</span> this.regionsOnCrashedServer = new ArrayList<>(size);<a name="line.328"></a> -<span class="sourceLineNo">329</span> for (org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionInfo ri: state.getRegionsOnCrashedServerList()) {<a name="line.329"></a> -<span class="sourceLineNo">330</span> this.regionsOnCrashedServer.add(ProtobufUtil.toRegionInfo(ri));<a name="line.330"></a> -<span class="sourceLineNo">331</span> }<a name="line.331"></a> -<span class="sourceLineNo">332</span> }<a name="line.332"></a> -<span class="sourceLineNo">333</span> }<a name="line.333"></a> -<span class="sourceLineNo">334</span><a name="line.334"></a> -<span class="sourceLineNo">335</span> @Override<a name="line.335"></a> -<span class="sourceLineNo">336</span> public ServerName getServerName() {<a name="line.336"></a> -<span class="sourceLineNo">337</span> return this.serverName;<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> @Override<a name="line.340"></a> -<span class="sourceLineNo">341</span> public boolean hasMetaTableRegion() {<a name="line.341"></a> -<span class="sourceLineNo">342</span> return this.carryingMeta;<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 ServerOperationType getServerOperationType() {<a name="line.346"></a> -<span class="sourceLineNo">347</span> return ServerOperationType.CRASH_HANDLER;<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> /**<a name="line.350"></a> -<span class="sourceLineNo">351</span> * For this procedure, yield at end of each successful flow step so that all crashed servers<a name="line.351"></a> -<span class="sourceLineNo">352</span> * can make progress rather than do the default which has each procedure running to completion<a name="line.352"></a> -<span class="sourceLineNo">353</span> * before we move to the next. For crashed servers, especially if running with distributed log<a name="line.353"></a> -<span class="sourceLineNo">354</span> * replay, we will want all servers to come along; we do not want the scenario where a server is<a name="line.354"></a> -<span class="sourceLineNo">355</span> * stuck waiting for regions to online so it can replay edits.<a name="line.355"></a> -<span class="sourceLineNo">356</span> */<a name="line.356"></a> -<span class="sourceLineNo">357</span> @Override<a name="line.357"></a> -<span class="sourceLineNo">358</span> protected boolean isYieldBeforeExecuteFromState(MasterProcedureEnv env, ServerCrashState state) {<a name="line.358"></a> -<span class="sourceLineNo">359</span> return true;<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> @Override<a name="line.362"></a> -<span class="sourceLineNo">363</span> protected boolean shouldWaitClientAck(MasterProcedureEnv env) {<a name="line.363"></a> -<span class="sourceLineNo">364</span> // The operation is triggered internally on the server<a name="line.364"></a> -<span class="sourceLineNo">365</span> // the client does not know about this procedure.<a name="line.365"></a> -<span class="sourceLineNo">366</span> return false;<a name="line.366"></a> -<span class="sourceLineNo">367</span> }<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> * Handle any outstanding RIT that are up against this.serverName, the crashed server.<a name="line.370"></a> -<span class="sourceLineNo">371</span> * Notify them of crash. Remove assign entries from the passed in <code>regions</code><a name="line.371"></a> -<span class="sourceLineNo">372</span> * otherwise we have two assigns going on and they will fight over who has lock.<a name="line.372"></a> -<span class="sourceLineNo">373</span> * Notify Unassigns. If unable to unassign because server went away, unassigns block waiting<a name="line.373"></a> -<span class="sourceLineNo">374</span> * on the below callback from a ServerCrashProcedure before proceeding.<a name="line.374"></a> -<span class="sourceLineNo">375</span> * @param regions Regions on the Crashed Server.<a name="line.375"></a> -<span class="sourceLineNo">376</span> * @return List of regions we should assign to new homes (not same as regions on crashed server).<a name="line.376"></a> -<span class="sourceLineNo">377</span> */<a name="line.377"></a> -<span class="sourceLineNo">378</span> private List<RegionInfo> handleRIT(final MasterProcedureEnv env, List<RegionInfo> regions) {<a name="line.378"></a> -<span class="sourceLineNo">379</span> if (regions == null || regions.isEmpty()) {<a name="line.379"></a> -<span class="sourceLineNo">380</span> return Collections.emptyList();<a name="line.380"></a> -<span class="sourceLineNo">381</span> }<a name="line.381"></a> -<span class="sourceLineNo">382</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.382"></a> -<span class="sourceLineNo">383</span> List<RegionInfo> toAssign = new ArrayList<RegionInfo>(regions);<a name="line.383"></a> -<span class="sourceLineNo">384</span> // Get an iterator so can remove items.<a name="line.384"></a> -<span class="sourceLineNo">385</span> final Iterator<RegionInfo> it = toAssign.iterator();<a name="line.385"></a> -<span class="sourceLineNo">386</span> ServerCrashException sce = null;<a name="line.386"></a> -<span class="sourceLineNo">387</span> while (it.hasNext()) {<a name="line.387"></a> -<span class="sourceLineNo">388</span> final RegionInfo hri = it.next();<a name="line.388"></a> -<span class="sourceLineNo">389</span> RegionTransitionProcedure rtp = am.getRegionStates().getRegionTransitionProcedure(hri);<a name="line.389"></a> -<span class="sourceLineNo">390</span> if (rtp == null) {<a name="line.390"></a> -<span class="sourceLineNo">391</span> continue;<a name="line.391"></a> -<span class="sourceLineNo">392</span> }<a name="line.392"></a> -<span class="sourceLineNo">393</span> // Make sure the RIT is against this crashed server. In the case where there are many<a name="line.393"></a> -<span class="sourceLineNo">394</span> // processings of a crashed server -- backed up for whatever reason (slow WAL split) --<a name="line.394"></a> -<span class="sourceLineNo">395</span> // then a previous SCP may have already failed an assign, etc., and it may have a new<a name="line.395"></a> -<span class="sourceLineNo">396</span> // location target; DO NOT fail these else we make for assign flux.<a name="line.396"></a> -<span class="sourceLineNo">397</span> ServerName rtpServerName = rtp.getServer(env);<a name="line.397"></a> -<span class="sourceLineNo">398</span> if (rtpServerName == null) {<a name="line.398"></a> -<span class="sourceLineNo">399</span> LOG.warn("RIT with ServerName null! " + rtp);<a name="line.399"></a> -<span class="sourceLineNo">400</span> continue;<a name="line.400"></a> -<span class="sourceLineNo">401</span> }<a name="line.401"></a> -<span class="sourceLineNo">402</span> if (!rtpServerName.equals(this.serverName)) continue;<a name="line.402"></a> -<span class="sourceLineNo">403</span> LOG.info("pid=" + getProcId() + " found RIT " + rtp + "; " +<a name="line.403"></a> -<span class="sourceLineNo">404</span> rtp.getRegionState(env).toShortString());<a name="line.404"></a> -<span class="sourceLineNo">405</span> // Notify RIT on server crash.<a name="line.405"></a> -<span class="sourceLineNo">406</span> if (sce == null) {<a name="line.406"></a> -<span class="sourceLineNo">407</span> sce = new ServerCrashException(getProcId(), getServerName());<a name="line.407"></a> -<span class="sourceLineNo">408</span> }<a name="line.408"></a> -<span class="sourceLineNo">409</span> rtp.remoteCallFailed(env, this.serverName, sce);<a name="line.409"></a> -<span class="sourceLineNo">410</span> // If an assign, remove from passed-in list of regions so we subsequently do not create<a name="line.410"></a> -<span class="sourceLineNo">411</span> // a new assign; the exisitng assign after the call to remoteCallFailed will recalibrate<a name="line.411"></a> -<span class="sourceLineNo">412</span> // and assign to a server other than the crashed one; no need to create new assign.<a name="line.412"></a> -<span class="sourceLineNo">413</span> // If an unassign, do not return this region; the above cancel will wake up the unassign and<a name="line.413"></a> -<span class="sourceLineNo">414</span> // it will complete. Done.<a name="line.414"></a> -<span class="sourceLineNo">415</span> it.remove();<a name="line.415"></a> -<span class="sourceLineNo">416</span> }<a name="line.416"></a> -<span class="sourceLineNo">417</span> return toAssign;<a name="line.417"></a> -<span class="sourceLineNo">418</span> }<a name="line.418"></a> -<span class="sourceLineNo">419</span><a name="line.419"></a> -<span class="sourceLineNo">420</span> @Override<a name="line.420"></a> -<span class="sourceLineNo">421</span> protected ProcedureMetrics getProcedureMetrics(MasterProcedureEnv env) {<a name="line.421"></a> -<span class="sourceLineNo">422</span> return env.getMasterServices().getMasterMetrics().getServerCrashProcMetrics();<a name="line.422"></a> -<span class="sourceLineNo">423</span> }<a name="line.423"></a> -<span class="sourceLineNo">424</span>}<a name="line.424"></a> +<span class="sourceLineNo">166</span> // Returns list of regions we need to reassign. NOTE: there is nothing to stop a<a name="line.166"></a> +<span class="sourceLineNo">167</span> // dispatch happening AFTER this point. Check for the condition if a dispatch RPC fails<a name="line.167"></a> +<span class="sourceLineNo">168</span> // inside in AssignProcedure/UnassignProcedure. AssignProcedure just keeps retrying.<a name="line.168"></a> +<span class="sourceLineNo">169</span> // UnassignProcedure is more complicated. See where it does the check by calling<a name="line.169"></a> +<span class="sourceLineNo">170</span> // am#isDeadServerProcessed.<a name="line.170"></a> +<span class="sourceLineNo">171</span> List<RegionInfo> toAssign = handleRIT(env, regionsOnCrashedServer);<a name="line.171"></a> +<span class="sourceLineNo">172</span> AssignmentManager am = env.getAssignmentManager();<a name="line.172"></a> +<span class="sourceLineNo">173</span> // CreateAssignProcedure will try to use the old location for the region deploy.<a name="line.173"></a> +<span class="sourceLineNo">174</span> addChildProcedure(am.createAssignProcedures(toAssign));<a name="line.174"></a> +<span class="sourceLineNo">175</span> setNextState(ServerCrashState.SERVER_CRASH_HANDLE_RIT2);<a name="line.175"></a> +<span class="sourceLineNo">176</span> } else {<a name="line.176"></a> +<span class="sourceLineNo">177</span> setNextState(ServerCrashState.SERVER_CRASH_FINISH);<a name="line.177"></a> +<span class="sourceLineNo">178</span> }<a name="line.178"></a> +<span class="sourceLineNo">179</span> break;<a name="line.179"></a> +<span class="sourceLineNo">180</span><a name="line.180"></a> +<span class="sourceLineNo">181</span> case SERVER_CRASH_HANDLE_RIT2:<a name="line.181"></a> +<span class="sourceLineNo">182</span> // Noop. Left in place because we used to call handleRIT here for a second time<a name="line.182"></a> +<span class="sourceLineNo">183</span> // but no longer necessary since HBASE-20634.<a name="line.183"></a> +<span class="sourceLineNo">184</span> setNextState(ServerCrashState.SERVER_CRASH_FINISH);<a name="line.184"></a> +<span class="sourceLineNo">185</span> break;<a name="line.185"></a> +<span class="sourceLineNo">186</span><a name="line.186"></a> +<span class="sourceLineNo">187</span> case SERVER_CRASH_FINISH:<a name="line.187"></a> +<span class="sourceLineNo">188</span> services.getAssignmentManager().getRegionStates().removeServer(serverName);<a name="line.188"></a> +<span class="sourceLineNo">189</span> services.getServerManager().getDeadServers().finish(serverName);<a name="line.189"></a> +<span class="sourceLineNo">190</span> return Flow.NO_MORE_STATE;<a name="line.190"></a> +<span class="sourceLineNo">191</span><a name="line.191"></a> +<span class="sourceLineNo">192</span> default:<a name="line.192"></a> +<span class="sourceLineNo">193</span> throw new UnsupportedOperationException("unhandled state=" + state);<a name="line.193"></a> +<span class="sourceLineNo">194</span> }<a name="line.194"></a> +<span class="sourceLineNo">195</span> } catch (IOException e) {<a name="line.195"></a> +<span class="sourceLineNo">196</span> LOG.warn("Failed state=" + state + ", retry " + this + "; cycles=" + getCycles(), e);<a name="line.196"></a> +<span class="sourceLineNo">197</span> }<a name="line.197"></a> +<span class="sourceLineNo">198</span> return Flow.HAS_MORE_STATE;<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><a name="line.201"></a> +<span class="sourceLineNo">202</span> /**<a name="line.202"></a> +<span class="sourceLineNo">203</span> * @param env<a name="line.203"></a> +<span class="sourceLineNo">204</span> * @throws IOException<a name="line.204"></a> +<span class="sourceLineNo">205</span> */<a name="line.205"></a> +<span class="sourceLineNo">206</span> private void processMeta(final MasterProcedureEnv env) throws IOException {<a name="line.206"></a> +<span class="sourceLineNo">207</span> LOG.debug("{}; processing hbase:meta", this);<a name="line.207"></a> +<span class="sourceLineNo">208</span><a name="line.208"></a> +<span class="sourceLineNo">209</span> // Assign meta if still carrying it. Check again: region may be assigned because of RIT timeout<a name="line.209"></a> +<span class="sourceLineNo">210</span> final AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.210"></a> +<span class="sourceLineNo">211</span> for (RegionInfo hri: am.getRegionStates().getServerRegionInfoSet(serverName)) {<a name="line.211"></a> +<span class="sourceLineNo">212</span> if (!isDefaultMetaRegion(hri)) {<a name="line.212"></a> +<span class="sourceLineNo">213</span> continue;<a name="line.213"></a> +<span class="sourceLineNo">214</span> }<a name="line.214"></a> +<span class="sourceLineNo">215</span> addChildProcedure(new RecoverMetaProcedure(serverName, this.shouldSplitWal));<a name="line.215"></a> +<span class="sourceLineNo">216</span> }<a name="line.216"></a> +<span class="sourceLineNo">217</span> }<a name="line.217"></a> +<span class="sourceLineNo">218</span><a name="line.218"></a> +<span class="sourceLineNo">219</span> private boolean filterDefaultMetaRegions(final List<RegionInfo> regions) {<a name="line.219"></a> +<span class="sourceLineNo">220</span> if (regions == null) return false;<a name="line.220"></a> +<span class="sourceLineNo">221</span> regions.removeIf(this::isDefaultMetaRegion);<a name="line.221"></a> +<span class="sourceLineNo">222</span> return !regions.isEmpty();<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> private boolean isDefaultMetaRegion(final RegionInfo hri) {<a name="line.225"></a> +<span class="sourceLineNo">226</span> return hri.getTable().equals(TableName.META_TABLE_NAME) &&<a name="line.226"></a> +<span class="sourceLineNo">227</span> RegionReplicaUtil.isDefaultReplica(hri);<a name="line.227"></a> +<span class="sourceLineNo">228</span> }<a name="line.228"></a> +<span class="sourceLineNo">229</span><a name="line.229"></a> +<span class="sourceLineNo">230</span> private void splitLogs(final MasterProcedureEnv env) throws IOException {<a name="line.230"></a> +<span class="sourceLineNo">231</span> if (LOG.isDebugEnabled()) {<a name="line.231"></a> +<span class="sourceLineNo">232</span> LOG.debug("Splitting WALs " + this);<a name="line.232"></a> +<span class="sourceLineNo">233</span> }<a name="line.233"></a> +<span class="sourceLineNo">234</span> MasterWalManager mwm = env.getMasterServices().getMasterWalManager();<a name="line.234"></a> +<span class="sourceLineNo">235</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.235"></a> +<span class="sourceLineNo">236</span> // TODO: For Matteo. Below BLOCKs!!!! Redo so can relinquish executor while it is running.<a name="line.236"></a> +<span class="sourceLineNo">237</span> // PROBLEM!!! WE BLOCK HERE.<a name="line.237"></a> +<span class="sourceLineNo">238</span> am.getRegionStates().logSplitting(this.serverName);<a name="line.238"></a> +<span class="sourceLineNo">239</span> mwm.splitLog(this.serverName);<a name="line.239"></a> +<span class="sourceLineNo">240</span> am.getRegionStates().logSplit(this.serverName);<a name="line.240"></a> +<span class="sourceLineNo">241</span> LOG.debug("Done splitting WALs {}", this);<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> @Override<a name="line.244"></a> +<span class="sourceLineNo">245</span> protected void rollbackState(MasterProcedureEnv env, ServerCrashState state)<a name="line.245"></a> +<span class="sourceLineNo">246</span> throws IOException {<a name="line.246"></a> +<span class="sourceLineNo">247</span> // Can't rollback.<a name="line.247"></a> +<span class="sourceLineNo">248</span> throw new UnsupportedOperationException("unhandled state=" + state);<a name="line.248"></a> +<span class="sourceLineNo">249</span> }<a name="line.249"></a> +<span class="sourceLineNo">250</span><a name="line.250"></a> +<span class="sourceLineNo">251</span> @Override<a name="line.251"></a> +<span class="sourceLineNo">252</span> protected ServerCrashState getState(int stateId) {<a name="line.252"></a> +<span class="sourceLineNo">253</span> return ServerCrashState.forNumber(stateId);<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> @Override<a name="line.256"></a> +<span class="sourceLineNo">257</span> protected int getStateId(ServerCrashState state) {<a name="line.257"></a> +<span class="sourceLineNo">258</span> return state.getNumber();<a name="line.258"></a> +<span class="sourceLineNo">259</span> }<a name="line.259"></a> +<span class="sourceLineNo">260</span><a name="line.260"></a> +<span class="sourceLineNo">261</span> @Override<a name="line.261"></a> +<span class="sourceLineNo">262</span> protected ServerCrashState getInitialState() {<a name="line.262"></a> +<span class="sourceLineNo">263</span> return ServerCrashState.SERVER_CRASH_START;<a name="line.263"></a> +<span class="sourceLineNo">264</span> }<a name="line.264"></a> +<span class="sourceLineNo">265</span><a name="line.265"></a> +<span class="sourceLineNo">266</span> @Override<a name="line.266"></a> +<span class="sourceLineNo">267</span> protected boolean abort(MasterProcedureEnv env) {<a name="line.267"></a> +<span class="sourceLineNo">268</span> // TODO<a name="line.268"></a> +<span class="sourceLineNo">269</span> return false;<a name="line.269"></a> +<span class="sourceLineNo">270</span> }<a name="line.270"></a> +<span class="sourceLineNo">271</span><a name="line.271"></a> +<span class="sourceLineNo">272</span> @Override<a name="line.272"></a> +<span class="sourceLineNo">273</span> protected LockState acquireLock(final MasterProcedureEnv env) {<a name="line.273"></a> +<span class="sourceLineNo">274</span> // TODO: Put this BACK AFTER AMv2 goes in!!!!<a name="line.274"></a> +<span class="sourceLineNo">275</span> // if (env.waitFailoverCleanup(this)) return LockState.LOCK_EVENT_WAIT;<a name="line.275"></a> +<span class="sourceLineNo">276</span> if (env.waitServerCrashProcessingEnabled(this)) return LockState.LOCK_EVENT_WAIT;<a name="line.276"></a> +<span class="sourceLineNo">277</span> if (env.getProcedureScheduler().waitServerExclusiveLock(this, getServerName())) {<a name="line.277"></a> +<span class="sourceLineNo">278</span> return LockState.LOCK_EVENT_WAIT;<a name="line.278"></a> +<span class="sourceLineNo">279</span> }<a name="line.279"></a> +<span class="sourceLineNo">280</span> return LockState.LOCK_ACQUIRED;<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> @Override<a name="line.283"></a> +<span class="sourceLineNo">284</span> protected void releaseLock(final MasterProcedureEnv env) {<a name="line.284"></a> +<span class="sourceLineNo">285</span> env.getProcedureScheduler().wakeServerExclusiveLock(this, getServerName());<a name="line.285"></a> +<span class="sourceLineNo">286</span> }<a name="line.286"></a> +<span class="sourceLineNo">287</span><a name="line.287"></a> +<span class="sourceLineNo">288</span> @Override<a name="line.288"></a> +<span class="sourceLineNo">289</span> public void toStringClassDetails(StringBuilder sb) {<a name="line.289"></a> +<span class="sourceLineNo">290</span> sb.append(getClass().getSimpleName());<a name="line.290"></a> +<span class="sourceLineNo">291</span> sb.append(" server=");<a name="line.291"></a> +<span class="sourceLineNo">292</span> sb.append(serverName);<a name="line.292"></a> +<span class="sourceLineNo">293</span> sb.append(", splitWal=");<a name="line.293"></a> +<span class="sourceLineNo">294</span> sb.append(shouldSplitWal);<a name="line.294"></a> +<span class="sourceLineNo">295</span> sb.append(", meta=");<a name="line.295"></a> +<span class="sourceLineNo">296</span> sb.append(carryingMeta);<a name="line.296"></a> +<span class="sourceLineNo">297</span> }<a name="line.297"></a> +<span class="sourceLineNo">298</span><a name="line.298"></a> +<span class="sourceLineNo">299</span> @Override<a name="line.299"></a> +<span class="sourceLineNo">300</span> protected void serializeStateData(ProcedureStateSerializer serializer)<a name="line.300"></a> +<span class="sourceLineNo">301</span> throws IOException {<a name="line.301"></a> +<span class="sourceLineNo">302</span> super.serializeStateData(serializer);<a name="line.302"></a> +<span class="sourceLineNo">303</span><a name="line.303"></a> +<span class="sourceLineNo">304</span> MasterProcedureProtos.ServerCrashStateData.Builder state =<a name="line.304"></a> +<span class="sourceLineNo">305</span> MasterProcedureProtos.ServerCrashStateData.newBuilder().<a name="line.305"></a> +<span class="sourceLineNo">306</span> setServerName(ProtobufUtil.toServerName(this.serverName)).<a name="line.306"></a> +<span class="sourceLineNo">307</span> setCarryingMeta(this.carryingMeta).<a name="line.307"></a> +<span class="sourceLineNo">308</span> setShouldSplitWal(this.shouldSplitWal);<a name="line.308"></a> +<span class="sourceLineNo">309</span> if (this.regionsOnCrashedServer != null && !this.regionsOnCrashedServer.isEmpty()) {<a name="line.309"></a> +<span class="sourceLineNo">310</span> for (RegionInfo hri: this.regionsOnCrashedServer) {<a name="line.310"></a> +<span class="sourceLineNo">311</span> state.addRegionsOnCrashedServer(ProtobufUtil.toRegionInfo(hri));<a name="line.311"></a> +<span class="sourceLineNo">312</span> }<a name="line.312"></a> +<span class="sourceLineNo">313</span> }<a name="line.313"></a> +<span class="sourceLineNo">314</span> serializer.serialize(state.build());<a name="line.314"></a> +<span class="sourceLineNo">315</span> }<a name="line.315"></a> +<span class="sourceLineNo">316</span><a name="line.316"></a> +<span class="sourceLineNo">317</span> @Override<a name="line.317"></a> +<span class="sourceLineNo">318</span> protected void deserializeStateData(ProcedureStateSerializer serializer)<a name="line.318"></a> +<span class="sourceLineNo">319</span> throws IOException {<a name="line.319"></a> +<span class="sourceLineNo">320</span> super.deserializeStateData(serializer);<a name="line.320"></a> +<span class="sourceLineNo">321</span><a name="line.321"></a> +<span class="sourceLineNo">322</span> MasterProcedureProtos.ServerCrashStateData state =<a name="line.322"></a> +<span class="sourceLineNo">323</span> serializer.deserialize(MasterProcedureProtos.ServerCrashStateData.class);<a name="line.323"></a> +<span class="sourceLineNo">324</span> this.serverName = ProtobufUtil.toServerName(state.getServerName());<a name="line.324"></a> +<span class="sourceLineNo">325</span> this.carryingMeta = state.hasCarryingMeta()? state.getCarryingMeta(): false;<a name="line.325"></a> +<span class="sourceLineNo">326</span> // shouldSplitWAL has a default over in pb so this invocation will always work.<a name="line.326"></a> +<span class="sourceLineNo">327</span> this.shouldSplitWal = state.getShouldSplitWal();<a name="line.327"></a> +<span class="sourceLineNo">328</span> int size = state.getRegionsOnCrashedServerCount();<a name="line.328"></a> +<span class="sourceLineNo">329</span> if (size > 0) {<a name="line.329"></a> +<span class="sourceLineNo">330</span> this.regionsOnCrashedServer = new ArrayList<>(size);<a name="line.330"></a> +<span class="sourceLineNo">331</span> for (org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionInfo ri: state.getRegionsOnCrashedServerList()) {<a name="line.331"></a> +<span class="sourceLineNo">332</span> this.regionsOnCrashedServer.add(ProtobufUtil.toRegionInfo(ri));<a name="line.332"></a> +<span class="sourceLineNo">333</span> }<a name="line.333"></a> +<span class="sourceLineNo">334</span> }<a name="line.334"></a> +<span class="sourceLineNo">335</span> }<a name="line.335"></a> +<span class="sourceLineNo">336</span><a name="line.336"></a> +<span class="sourceLineNo">337</span> @Override<a name="line.337"></a> +<span class="sourceLineNo">338</span> public ServerName getServerName() {<a name="line.338"></a> +<span class="sourceLineNo">339</span> return this.serverName;<a name="line.339"></a> +<span class="sourceLineNo">340</span> }<a name="line.340"></a> +<span class="sourceLineNo">341</span><a name="line.341"></a> +<span class="sourceLineNo">342</span> @Override<a name="line.342"></a> +<span class="sourceLineNo">343</span> public boolean hasMetaTableRegion() {<a name="line.343"></a> +<span class="sourceLineNo">344</span> return this.carryingMeta;<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> @Override<a name="line.347"></a> +<span class="sourceLineNo">348</span> public ServerOperationType getServerOperationType() {<a name="line.348"></a> +<span class="sourceLineNo">349</span> return ServerOperationType.CRASH_HANDLER;<a name="line.349"></a> +<span class="sourceLineNo">350</span> }<a name="line.350"></a> +<span class="sourceLineNo">351</span><a name="line.351"></a> +<span class="sourceLineNo">352</span> /**<a name="line.352"></a> +<span class="sourceLineNo">353</span> * For this procedure, yield at end of each successful flow step so that all crashed servers<a name="line.353"></a> +<span class="sourceLineNo">354</span> * can make progress rather than do the default which has each procedure running to completion<a name="line.354"></a> +<span class="sourceLineNo">355</span> * before we move to the next. For crashed servers, especially if running with distributed log<a name="line.355"></a> +<span class="sourceLineNo">356</span> * replay, we will want all servers to come along; we do not want the scenario where a server is<a name="line.356"></a> +<span class="sourceLineNo">357</span> * stuck waiting for regions to online so it can replay edits.<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> protected boolean isYieldBeforeExecuteFromState(MasterProcedureEnv env, ServerCrashState state) {<a name="line.360"></a> +<span class="sourceLineNo">361</span> return true;<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> @Override<a name="line.364"></a> +<span class="sourceLineNo">365</span> protected boolean shouldWaitClientAck(MasterProcedureEnv env) {<a name="line.365"></a> +<span class="sourceLineNo">366</span> // The operation is triggered internally on the server<a name="line.366"></a> +<span class="sourceLineNo">367</span> // the client does not know about this procedure.<a name="line.367"></a> +<span class="sourceLineNo">368</span> return false;<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> * Handle any outstanding RIT that are up against this.serverName, the crashed server.<a name="line.372"></a> +<span class="sourceLineNo">373</span> * Notify them of crash. Remove assign entries from the passed in <code>regions</code><a name="line.373"></a> +<span class="sourceLineNo">374</span> * otherwise we have two assigns going on and they will fight over who has lock.<a name="line.374"></a> +<span class="sourceLineNo">375</span> * Notify Unassigns. If unable to unassign because server went away, unassigns block waiting<a name="line.375"></a> +<span class="sourceLineNo">376</span> * on the below callback from a ServerCrashProcedure before proceeding.<a name="line.376"></a> +<span class="sourceLineNo">377</span> * @param regions Regions on the Crashed Server.<a name="line.377"></a> +<span class="sourceLineNo">378</span> * @return List of regions we should assign to new homes (not same as regions on crashed server).<a name="line.378"></a> +<span class="sourceLineNo">379</span> */<a name="line.379"></a> +<span class="sourceLineNo">380</span> private List<RegionInfo> handleRIT(final MasterProcedureEnv env, List<RegionInfo> regions) {<a name="line.380"></a> +<span class="sourceLineNo">381</span> if (regions == null || regions.isEmpty()) {<a name="line.381"></a> +<span class="sourceLineNo">382</span> return Collections.emptyList();<a name="line.382"></a> +<span class="sourceLineNo">383</span> }<a name="line.383"></a> +<span class="sourceLineNo">384</span> AssignmentManager am = env.getMasterServices().getAssignmentManager();<a name="line.384"></a> +<span class="sourceLineNo">385</span> List<RegionInfo> toAssign = new ArrayList<RegionInfo>(regions);<a name="line.385"></a> +<span class="sourceLineNo">386</span> // Get an iterator so can remove items.<a name="line.386"></a> +<span class="sourceLineNo">387</span> final Iterator<RegionInfo> it = toAssign.iterator();<a name="line.387"></a> +<span class="sourceLineNo">388</span> ServerCrashException sce = null;<a name="line.388"></a> +<span class="sourceLineNo">389</span> while (it.hasNext()) {<a name="line.389"></a> +<span class="sourceLineNo">390</span> final RegionInfo hri = it.next();<a name="line.390"></a> +<span class="sourceLineNo">391</span> RegionTransitionProcedure rtp = am.getRegionStates().getRegionTransitionProcedure(hri);<a name="line.391"></a> +<span class="sourceLineNo">392</span> if (rtp == null) {<a name="line.392"></a> +<span class="sourceLineNo">393</span> continue;<a name="line.393"></a> +<span class="sourceLineNo">394</span> }<a name="line.394"></a> +<span class="sourceLineNo">395</span> // Make sure the RIT is against this crashed server. In the case where there are many<a name="line.395"></a> +<span class="sourceLineNo">396</span> // processings of a crashed server -- backed up for whatever reason (slow WAL split) --<a name="line.396"></a> +<span class="sourceLineNo">397</span> // then a previous SCP may have already failed an assign, etc., and it may have a new<a name="line.397"></a> +<span class="sourceLineNo">398</span> // location target; DO NOT fail these else we make for assign flux.<a name="line.398"></a> +<span class="sourceLineNo">399</span> ServerName rtpServerName = rtp.getServer(env);<a name="line.399"></a> +<span class="sourceLineNo">400</span> if (rtpServerName == null) {<a name="line.400"></a> +<span class="sourceLineNo">401</span> LOG.warn("RIT with ServerName null! " + rtp);<a name="line.401"></a> +<span class="sourceLineNo">402</span> continue;<a name="line.402"></a> +<span class="sourceLineNo">403</span> }<a name="line.403"></a> +<span class="sourceLineNo">404</span> if (!rtpServerName.equals(this.serverName)) continue;<a name="line.404"></a> +<span class="sourceLineNo">405</span> LOG.info("pid=" + getProcId() + " found RIT " + rtp + "; " +<a name="line.405"></a> +<span class="sourceLineNo">406</span> rtp.getRegionState(env).toShortString());<a name="line.406"></a> +<span class="sourceLineNo">407</span> // Notify RIT on server crash.<a name="line.407"></a> +<span class="sourceLineNo">408</span> if (sce == null) {<a name="line.408"></a> +<span class="sourceLineNo">409</span> sce = new ServerCrashException(getProcId(), getServerName());<a name="line.409"></a> +<span class="sourceLineNo">410</span> }<a name="line.410"></a> +<span class="sourceLineNo">411</span> rtp.remoteCallFailed(env, this.serverName, sce);<a name="line.411"></a> +<span class="sourceLineNo">412</span> // If an assign, remove from passed-in list of regions so we subsequently do not create<a name="line.412"></a> +<span class="sourceLineNo">413</span> // a new assign; the exisitng assign after the call to remoteCallFailed will recalibrate<a name="line.413"></a> +<span class="sourceLineNo">414</span> // and assign to a server other than the crashed one; no need to create new assign.<a name="line.414"></a> +<span class="sourceLineNo">415</span> // If an unassign, do not return this region; the above cancel will wake up the unassign and<a name="line.415"></a> +<span class="sourceLineNo">416</span> // it will complete. Done.<a name="line.416"></a> +<span class="sourceLineNo">417</span> it.remove();<a name="line.417"></a> +<span class="sourceLineNo">418</span> }<a name="line.418"></a> +<span class="sourceLineNo">419</span> return toAssign;<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> @Override<a name="line.422"></a> +<span class="sourceLineNo">423</span> protected ProcedureMetrics getProcedureMetrics(MasterProcedureEnv env) {<a name="line.423"></a> +<span class="sourceLineNo">424</span> return env.getMasterServices().getMasterMetrics().getServerCrashProcMetrics();<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>