http://git-wip-us.apache.org/repos/asf/hbase-site/blob/738e976e/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html index a7df46f..f0b26f3 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/MasterWalManager.html @@ -152,195 +152,219 @@ <span class="sourceLineNo">144</span> }<a name="line.144"></a> <span class="sourceLineNo">145</span><a name="line.145"></a> <span class="sourceLineNo">146</span> /**<a name="line.146"></a> -<span class="sourceLineNo">147</span> * @return listing of ServerNames found in the filesystem under the WAL directory<a name="line.147"></a> -<span class="sourceLineNo">148</span> * that COULD BE 'alive'; excludes those that have a '-splitting' suffix as these are already<a name="line.148"></a> -<span class="sourceLineNo">149</span> * being split -- they cannot be 'alive'.<a name="line.149"></a> +<span class="sourceLineNo">147</span> * Get Servernames which are currently splitting; paths have a '-splitting' suffix.<a name="line.147"></a> +<span class="sourceLineNo">148</span> * @return ServerName<a name="line.148"></a> +<span class="sourceLineNo">149</span> * @throws IOException IOException<a name="line.149"></a> <span class="sourceLineNo">150</span> */<a name="line.150"></a> -<span class="sourceLineNo">151</span> public Set<ServerName> getLiveServersFromWALDir() throws IOException {<a name="line.151"></a> -<span class="sourceLineNo">152</span> Path walDirPath = new Path(rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.152"></a> -<span class="sourceLineNo">153</span> FileStatus[] walDirForLiveServers = FSUtils.listStatus(fs, walDirPath,<a name="line.153"></a> -<span class="sourceLineNo">154</span> p -> !p.getName().endsWith(AbstractFSWALProvider.SPLITTING_EXT));<a name="line.154"></a> -<span class="sourceLineNo">155</span> if (walDirForLiveServers == null) {<a name="line.155"></a> -<span class="sourceLineNo">156</span> return Collections.emptySet();<a name="line.156"></a> -<span class="sourceLineNo">157</span> }<a name="line.157"></a> -<span class="sourceLineNo">158</span> return Stream.of(walDirForLiveServers).map(s -> {<a name="line.158"></a> -<span class="sourceLineNo">159</span> ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(s.getPath());<a name="line.159"></a> -<span class="sourceLineNo">160</span> if (serverName == null) {<a name="line.160"></a> -<span class="sourceLineNo">161</span> LOG.warn("Log folder {} doesn't look like its name includes a " +<a name="line.161"></a> -<span class="sourceLineNo">162</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.162"></a> -<span class="sourceLineNo">163</span> "write ahead logs they may be saved in this location.", s.getPath());<a name="line.163"></a> -<span class="sourceLineNo">164</span> return null;<a name="line.164"></a> -<span class="sourceLineNo">165</span> }<a name="line.165"></a> -<span class="sourceLineNo">166</span> return serverName;<a name="line.166"></a> -<span class="sourceLineNo">167</span> }).filter(s -> s != null).collect(Collectors.toSet());<a name="line.167"></a> -<span class="sourceLineNo">168</span> }<a name="line.168"></a> -<span class="sourceLineNo">169</span><a name="line.169"></a> -<span class="sourceLineNo">170</span> /**<a name="line.170"></a> -<span class="sourceLineNo">171</span> * Inspect the log directory to find dead servers which need recovery work<a name="line.171"></a> -<span class="sourceLineNo">172</span> * @return A set of ServerNames which aren't running but still have WAL files left in file system<a name="line.172"></a> -<span class="sourceLineNo">173</span> * @deprecated With proc-v2, we can record the crash server with procedure store, so do not need<a name="line.173"></a> -<span class="sourceLineNo">174</span> * to scan the wal directory to find out the splitting wal directory any more. Leave<a name="line.174"></a> -<span class="sourceLineNo">175</span> * it here only because {@code RecoverMetaProcedure}(which is also deprecated) uses<a name="line.175"></a> -<span class="sourceLineNo">176</span> * it.<a name="line.176"></a> -<span class="sourceLineNo">177</span> */<a name="line.177"></a> -<span class="sourceLineNo">178</span> @Deprecated<a name="line.178"></a> -<span class="sourceLineNo">179</span> public Set<ServerName> getFailedServersFromLogFolders() {<a name="line.179"></a> -<span class="sourceLineNo">180</span> boolean retrySplitting = !conf.getBoolean("hbase.hlog.split.skip.errors",<a name="line.180"></a> -<span class="sourceLineNo">181</span> WALSplitter.SPLIT_SKIP_ERRORS_DEFAULT);<a name="line.181"></a> -<span class="sourceLineNo">182</span><a name="line.182"></a> -<span class="sourceLineNo">183</span> Set<ServerName> serverNames = new HashSet<>();<a name="line.183"></a> -<span class="sourceLineNo">184</span> Path logsDirPath = new Path(this.rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.184"></a> -<span class="sourceLineNo">185</span><a name="line.185"></a> -<span class="sourceLineNo">186</span> do {<a name="line.186"></a> -<span class="sourceLineNo">187</span> if (services.isStopped()) {<a name="line.187"></a> -<span class="sourceLineNo">188</span> LOG.warn("Master stopped while trying to get failed servers.");<a name="line.188"></a> -<span class="sourceLineNo">189</span> break;<a name="line.189"></a> -<span class="sourceLineNo">190</span> }<a name="line.190"></a> -<span class="sourceLineNo">191</span> try {<a name="line.191"></a> -<span class="sourceLineNo">192</span> if (!this.fs.exists(logsDirPath)) return serverNames;<a name="line.192"></a> -<span class="sourceLineNo">193</span> FileStatus[] logFolders = FSUtils.listStatus(this.fs, logsDirPath, null);<a name="line.193"></a> -<span class="sourceLineNo">194</span> // Get online servers after getting log folders to avoid log folder deletion of newly<a name="line.194"></a> -<span class="sourceLineNo">195</span> // checked in region servers . see HBASE-5916<a name="line.195"></a> -<span class="sourceLineNo">196</span> Set<ServerName> onlineServers = services.getServerManager().getOnlineServers().keySet();<a name="line.196"></a> -<span class="sourceLineNo">197</span><a name="line.197"></a> -<span class="sourceLineNo">198</span> if (logFolders == null || logFolders.length == 0) {<a name="line.198"></a> -<span class="sourceLineNo">199</span> LOG.debug("No log files to split, proceeding...");<a name="line.199"></a> -<span class="sourceLineNo">200</span> return serverNames;<a name="line.200"></a> -<span class="sourceLineNo">201</span> }<a name="line.201"></a> -<span class="sourceLineNo">202</span> for (FileStatus status : logFolders) {<a name="line.202"></a> -<span class="sourceLineNo">203</span> FileStatus[] curLogFiles = FSUtils.listStatus(this.fs, status.getPath(), null);<a name="line.203"></a> -<span class="sourceLineNo">204</span> if (curLogFiles == null || curLogFiles.length == 0) {<a name="line.204"></a> -<span class="sourceLineNo">205</span> // Empty log folder. No recovery needed<a name="line.205"></a> -<span class="sourceLineNo">206</span> continue;<a name="line.206"></a> -<span class="sourceLineNo">207</span> }<a name="line.207"></a> -<span class="sourceLineNo">208</span> final ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(<a name="line.208"></a> -<span class="sourceLineNo">209</span> status.getPath());<a name="line.209"></a> -<span class="sourceLineNo">210</span> if (null == serverName) {<a name="line.210"></a> -<span class="sourceLineNo">211</span> LOG.warn("Log folder " + status.getPath() + " doesn't look like its name includes a " +<a name="line.211"></a> -<span class="sourceLineNo">212</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.212"></a> -<span class="sourceLineNo">213</span> "write ahead logs they may be saved in this location.");<a name="line.213"></a> -<span class="sourceLineNo">214</span> } else if (!onlineServers.contains(serverName)) {<a name="line.214"></a> -<span class="sourceLineNo">215</span> LOG.info("Log folder " + status.getPath() + " doesn't belong "<a name="line.215"></a> -<span class="sourceLineNo">216</span> + "to a known region server, splitting");<a name="line.216"></a> -<span class="sourceLineNo">217</span> serverNames.add(serverName);<a name="line.217"></a> -<span class="sourceLineNo">218</span> } else {<a name="line.218"></a> -<span class="sourceLineNo">219</span> LOG.info("Log folder " + status.getPath() + " belongs to an existing region server");<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> retrySplitting = false;<a name="line.222"></a> -<span class="sourceLineNo">223</span> } catch (IOException ioe) {<a name="line.223"></a> -<span class="sourceLineNo">224</span> LOG.warn("Failed getting failed servers to be recovered.", ioe);<a name="line.224"></a> -<span class="sourceLineNo">225</span> if (!checkFileSystem()) {<a name="line.225"></a> -<span class="sourceLineNo">226</span> LOG.warn("Bad Filesystem, exiting");<a name="line.226"></a> -<span class="sourceLineNo">227</span> Runtime.getRuntime().halt(1);<a name="line.227"></a> -<span class="sourceLineNo">228</span> }<a name="line.228"></a> -<span class="sourceLineNo">229</span> try {<a name="line.229"></a> -<span class="sourceLineNo">230</span> if (retrySplitting) {<a name="line.230"></a> -<span class="sourceLineNo">231</span> Thread.sleep(conf.getInt("hbase.hlog.split.failure.retry.interval", 30 * 1000));<a name="line.231"></a> -<span class="sourceLineNo">232</span> }<a name="line.232"></a> -<span class="sourceLineNo">233</span> } catch (InterruptedException e) {<a name="line.233"></a> -<span class="sourceLineNo">234</span> LOG.warn("Interrupted, aborting since cannot return w/o splitting");<a name="line.234"></a> -<span class="sourceLineNo">235</span> Thread.currentThread().interrupt();<a name="line.235"></a> -<span class="sourceLineNo">236</span> retrySplitting = false;<a name="line.236"></a> -<span class="sourceLineNo">237</span> Runtime.getRuntime().halt(1);<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> } while (retrySplitting);<a name="line.240"></a> -<span class="sourceLineNo">241</span><a name="line.241"></a> -<span class="sourceLineNo">242</span> return serverNames;<a name="line.242"></a> -<span class="sourceLineNo">243</span> }<a name="line.243"></a> -<span class="sourceLineNo">244</span><a name="line.244"></a> -<span class="sourceLineNo">245</span> public void splitLog(final ServerName serverName) throws IOException {<a name="line.245"></a> -<span class="sourceLineNo">246</span> splitLog(Collections.<ServerName>singleton(serverName));<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> /**<a name="line.249"></a> -<span class="sourceLineNo">250</span> * Specialized method to handle the splitting for meta WAL<a name="line.250"></a> -<span class="sourceLineNo">251</span> * @param serverName logs belonging to this server will be split<a name="line.251"></a> -<span class="sourceLineNo">252</span> */<a name="line.252"></a> -<span class="sourceLineNo">253</span> public void splitMetaLog(final ServerName serverName) throws IOException {<a name="line.253"></a> -<span class="sourceLineNo">254</span> splitMetaLog(Collections.<ServerName>singleton(serverName));<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> /**<a name="line.257"></a> -<span class="sourceLineNo">258</span> * Specialized method to handle the splitting for meta WAL<a name="line.258"></a> -<span class="sourceLineNo">259</span> * @param serverNames logs belonging to these servers will be split<a name="line.259"></a> -<span class="sourceLineNo">260</span> */<a name="line.260"></a> -<span class="sourceLineNo">261</span> public void splitMetaLog(final Set<ServerName> serverNames) throws IOException {<a name="line.261"></a> -<span class="sourceLineNo">262</span> splitLog(serverNames, META_FILTER);<a name="line.262"></a> -<span class="sourceLineNo">263</span> }<a name="line.263"></a> -<span class="sourceLineNo">264</span><a name="line.264"></a> -<span class="sourceLineNo">265</span> @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="UL_UNRELEASED_LOCK", justification=<a name="line.265"></a> -<span class="sourceLineNo">266</span> "We only release this lock when we set it. Updates to code that uses it should verify use " +<a name="line.266"></a> -<span class="sourceLineNo">267</span> "of the guard boolean.")<a name="line.267"></a> -<span class="sourceLineNo">268</span> private List<Path> getLogDirs(final Set<ServerName> serverNames) throws IOException {<a name="line.268"></a> -<span class="sourceLineNo">269</span> List<Path> logDirs = new ArrayList<>();<a name="line.269"></a> -<span class="sourceLineNo">270</span> boolean needReleaseLock = false;<a name="line.270"></a> -<span class="sourceLineNo">271</span> if (!this.services.isInitialized()) {<a name="line.271"></a> -<span class="sourceLineNo">272</span> // during master initialization, we could have multiple places splitting a same wal<a name="line.272"></a> -<span class="sourceLineNo">273</span> // XXX: Does this still exist after we move to proc-v2?<a name="line.273"></a> -<span class="sourceLineNo">274</span> this.splitLogLock.lock();<a name="line.274"></a> -<span class="sourceLineNo">275</span> needReleaseLock = true;<a name="line.275"></a> -<span class="sourceLineNo">276</span> }<a name="line.276"></a> -<span class="sourceLineNo">277</span> try {<a name="line.277"></a> -<span class="sourceLineNo">278</span> for (ServerName serverName : serverNames) {<a name="line.278"></a> -<span class="sourceLineNo">279</span> Path logDir = new Path(this.rootDir,<a name="line.279"></a> -<span class="sourceLineNo">280</span> AbstractFSWALProvider.getWALDirectoryName(serverName.toString()));<a name="line.280"></a> -<span class="sourceLineNo">281</span> Path splitDir = logDir.suffix(AbstractFSWALProvider.SPLITTING_EXT);<a name="line.281"></a> -<span class="sourceLineNo">282</span> // Rename the directory so a rogue RS doesn't create more WALs<a name="line.282"></a> -<span class="sourceLineNo">283</span> if (fs.exists(logDir)) {<a name="line.283"></a> -<span class="sourceLineNo">284</span> if (!this.fs.rename(logDir, splitDir)) {<a name="line.284"></a> -<span class="sourceLineNo">285</span> throw new IOException("Failed fs.rename for log split: " + logDir);<a name="line.285"></a> -<span class="sourceLineNo">286</span> }<a name="line.286"></a> -<span class="sourceLineNo">287</span> logDir = splitDir;<a name="line.287"></a> -<span class="sourceLineNo">288</span> LOG.debug("Renamed region directory: " + splitDir);<a name="line.288"></a> -<span class="sourceLineNo">289</span> } else if (!fs.exists(splitDir)) {<a name="line.289"></a> -<span class="sourceLineNo">290</span> LOG.info("Log dir for server " + serverName + " does not exist");<a name="line.290"></a> -<span class="sourceLineNo">291</span> continue;<a name="line.291"></a> -<span class="sourceLineNo">292</span> }<a name="line.292"></a> -<span class="sourceLineNo">293</span> logDirs.add(splitDir);<a name="line.293"></a> -<span class="sourceLineNo">294</span> }<a name="line.294"></a> -<span class="sourceLineNo">295</span> } catch (IOException ioe) {<a name="line.295"></a> -<span class="sourceLineNo">296</span> if (!checkFileSystem()) {<a name="line.296"></a> -<span class="sourceLineNo">297</span> this.services.abort("Aborting due to filesystem unavailable", ioe);<a name="line.297"></a> -<span class="sourceLineNo">298</span> throw ioe;<a name="line.298"></a> -<span class="sourceLineNo">299</span> }<a name="line.299"></a> -<span class="sourceLineNo">300</span> } finally {<a name="line.300"></a> -<span class="sourceLineNo">301</span> if (needReleaseLock) {<a name="line.301"></a> -<span class="sourceLineNo">302</span> this.splitLogLock.unlock();<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> return logDirs;<a name="line.305"></a> -<span class="sourceLineNo">306</span> }<a name="line.306"></a> -<span class="sourceLineNo">307</span><a name="line.307"></a> -<span class="sourceLineNo">308</span> public void splitLog(final Set<ServerName> serverNames) throws IOException {<a name="line.308"></a> -<span class="sourceLineNo">309</span> splitLog(serverNames, NON_META_FILTER);<a name="line.309"></a> -<span class="sourceLineNo">310</span> }<a name="line.310"></a> -<span class="sourceLineNo">311</span><a name="line.311"></a> -<span class="sourceLineNo">312</span> /**<a name="line.312"></a> -<span class="sourceLineNo">313</span> * This method is the base split method that splits WAL files matching a filter. Callers should<a name="line.313"></a> -<span class="sourceLineNo">314</span> * pass the appropriate filter for meta and non-meta WALs.<a name="line.314"></a> -<span class="sourceLineNo">315</span> * @param serverNames logs belonging to these servers will be split; this will rename the log<a name="line.315"></a> -<span class="sourceLineNo">316</span> * directory out from under a soft-failed server<a name="line.316"></a> -<span class="sourceLineNo">317</span> */<a name="line.317"></a> -<span class="sourceLineNo">318</span> public void splitLog(final Set<ServerName> serverNames, PathFilter filter) throws IOException {<a name="line.318"></a> -<span class="sourceLineNo">319</span> long splitTime = 0, splitLogSize = 0;<a name="line.319"></a> -<span class="sourceLineNo">320</span> List<Path> logDirs = getLogDirs(serverNames);<a name="line.320"></a> -<span class="sourceLineNo">321</span><a name="line.321"></a> -<span class="sourceLineNo">322</span> splitLogManager.handleDeadWorkers(serverNames);<a name="line.322"></a> -<span class="sourceLineNo">323</span> splitTime = EnvironmentEdgeManager.currentTime();<a name="line.323"></a> -<span class="sourceLineNo">324</span> splitLogSize = splitLogManager.splitLogDistributed(serverNames, logDirs, filter);<a name="line.324"></a> -<span class="sourceLineNo">325</span> splitTime = EnvironmentEdgeManager.currentTime() - splitTime;<a name="line.325"></a> -<span class="sourceLineNo">326</span><a name="line.326"></a> -<span class="sourceLineNo">327</span> if (this.metricsMasterFilesystem != null) {<a name="line.327"></a> -<span class="sourceLineNo">328</span> if (filter == META_FILTER) {<a name="line.328"></a> -<span class="sourceLineNo">329</span> this.metricsMasterFilesystem.addMetaWALSplit(splitTime, splitLogSize);<a name="line.329"></a> -<span class="sourceLineNo">330</span> } else {<a name="line.330"></a> -<span class="sourceLineNo">331</span> this.metricsMasterFilesystem.addSplit(splitTime, splitLogSize);<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">151</span> public Set<ServerName> getSplittingServersFromWALDir() throws IOException {<a name="line.151"></a> +<span class="sourceLineNo">152</span> return getServerNamesFromWALDirPath(<a name="line.152"></a> +<span class="sourceLineNo">153</span> p -> p.getName().endsWith(AbstractFSWALProvider.SPLITTING_EXT));<a name="line.153"></a> +<span class="sourceLineNo">154</span> }<a name="line.154"></a> +<span class="sourceLineNo">155</span><a name="line.155"></a> +<span class="sourceLineNo">156</span> /**<a name="line.156"></a> +<span class="sourceLineNo">157</span> * Get Servernames that COULD BE 'alive'; excludes those that have a '-splitting' suffix as these<a name="line.157"></a> +<span class="sourceLineNo">158</span> * are already being split -- they cannot be 'alive'.<a name="line.158"></a> +<span class="sourceLineNo">159</span> * @return ServerName<a name="line.159"></a> +<span class="sourceLineNo">160</span> * @throws IOException IOException<a name="line.160"></a> +<span class="sourceLineNo">161</span> */<a name="line.161"></a> +<span class="sourceLineNo">162</span> public Set<ServerName> getLiveServersFromWALDir() throws IOException {<a name="line.162"></a> +<span class="sourceLineNo">163</span> return getServerNamesFromWALDirPath(<a name="line.163"></a> +<span class="sourceLineNo">164</span> p -> !p.getName().endsWith(AbstractFSWALProvider.SPLITTING_EXT));<a name="line.164"></a> +<span class="sourceLineNo">165</span> }<a name="line.165"></a> +<span class="sourceLineNo">166</span><a name="line.166"></a> +<span class="sourceLineNo">167</span> /**<a name="line.167"></a> +<span class="sourceLineNo">168</span> * @return listing of ServerNames found by parsing WAL directory paths in FS.<a name="line.168"></a> +<span class="sourceLineNo">169</span> *<a name="line.169"></a> +<span class="sourceLineNo">170</span> */<a name="line.170"></a> +<span class="sourceLineNo">171</span> public Set<ServerName> getServerNamesFromWALDirPath(final PathFilter filter) throws IOException {<a name="line.171"></a> +<span class="sourceLineNo">172</span> FileStatus[] walDirForServerNames = getWALDirPaths(filter);<a name="line.172"></a> +<span class="sourceLineNo">173</span> return Stream.of(walDirForServerNames).map(s -> {<a name="line.173"></a> +<span class="sourceLineNo">174</span> ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(s.getPath());<a name="line.174"></a> +<span class="sourceLineNo">175</span> if (serverName == null) {<a name="line.175"></a> +<span class="sourceLineNo">176</span> LOG.warn("Log folder {} doesn't look like its name includes a " +<a name="line.176"></a> +<span class="sourceLineNo">177</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.177"></a> +<span class="sourceLineNo">178</span> "write ahead logs they may be saved in this location.", s.getPath());<a name="line.178"></a> +<span class="sourceLineNo">179</span> return null;<a name="line.179"></a> +<span class="sourceLineNo">180</span> }<a name="line.180"></a> +<span class="sourceLineNo">181</span> return serverName;<a name="line.181"></a> +<span class="sourceLineNo">182</span> }).filter(s -> s != null).collect(Collectors.toSet());<a name="line.182"></a> +<span class="sourceLineNo">183</span> }<a name="line.183"></a> +<span class="sourceLineNo">184</span><a name="line.184"></a> +<span class="sourceLineNo">185</span> /**<a name="line.185"></a> +<span class="sourceLineNo">186</span> * @return List of all RegionServer WAL dirs; i.e. this.rootDir/HConstants.HREGION_LOGDIR_NAME.<a name="line.186"></a> +<span class="sourceLineNo">187</span> */<a name="line.187"></a> +<span class="sourceLineNo">188</span> public FileStatus[] getWALDirPaths(final PathFilter filter) throws IOException {<a name="line.188"></a> +<span class="sourceLineNo">189</span> Path walDirPath = new Path(rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.189"></a> +<span class="sourceLineNo">190</span> FileStatus[] walDirForServerNames = FSUtils.listStatus(fs, walDirPath, filter);<a name="line.190"></a> +<span class="sourceLineNo">191</span> return walDirForServerNames == null? new FileStatus[0]: walDirForServerNames;<a name="line.191"></a> +<span class="sourceLineNo">192</span> }<a name="line.192"></a> +<span class="sourceLineNo">193</span><a name="line.193"></a> +<span class="sourceLineNo">194</span> /**<a name="line.194"></a> +<span class="sourceLineNo">195</span> * Inspect the log directory to find dead servers which need recovery work<a name="line.195"></a> +<span class="sourceLineNo">196</span> * @return A set of ServerNames which aren't running but still have WAL files left in file system<a name="line.196"></a> +<span class="sourceLineNo">197</span> * @deprecated With proc-v2, we can record the crash server with procedure store, so do not need<a name="line.197"></a> +<span class="sourceLineNo">198</span> * to scan the wal directory to find out the splitting wal directory any more. Leave<a name="line.198"></a> +<span class="sourceLineNo">199</span> * it here only because {@code RecoverMetaProcedure}(which is also deprecated) uses<a name="line.199"></a> +<span class="sourceLineNo">200</span> * it.<a name="line.200"></a> +<span class="sourceLineNo">201</span> */<a name="line.201"></a> +<span class="sourceLineNo">202</span> @Deprecated<a name="line.202"></a> +<span class="sourceLineNo">203</span> public Set<ServerName> getFailedServersFromLogFolders() {<a name="line.203"></a> +<span class="sourceLineNo">204</span> boolean retrySplitting = !conf.getBoolean("hbase.hlog.split.skip.errors",<a name="line.204"></a> +<span class="sourceLineNo">205</span> WALSplitter.SPLIT_SKIP_ERRORS_DEFAULT);<a name="line.205"></a> +<span class="sourceLineNo">206</span><a name="line.206"></a> +<span class="sourceLineNo">207</span> Set<ServerName> serverNames = new HashSet<>();<a name="line.207"></a> +<span class="sourceLineNo">208</span> Path logsDirPath = new Path(this.rootDir, HConstants.HREGION_LOGDIR_NAME);<a name="line.208"></a> +<span class="sourceLineNo">209</span><a name="line.209"></a> +<span class="sourceLineNo">210</span> do {<a name="line.210"></a> +<span class="sourceLineNo">211</span> if (services.isStopped()) {<a name="line.211"></a> +<span class="sourceLineNo">212</span> LOG.warn("Master stopped while trying to get failed servers.");<a name="line.212"></a> +<span class="sourceLineNo">213</span> break;<a name="line.213"></a> +<span class="sourceLineNo">214</span> }<a name="line.214"></a> +<span class="sourceLineNo">215</span> try {<a name="line.215"></a> +<span class="sourceLineNo">216</span> if (!this.fs.exists(logsDirPath)) return serverNames;<a name="line.216"></a> +<span class="sourceLineNo">217</span> FileStatus[] logFolders = FSUtils.listStatus(this.fs, logsDirPath, null);<a name="line.217"></a> +<span class="sourceLineNo">218</span> // Get online servers after getting log folders to avoid log folder deletion of newly<a name="line.218"></a> +<span class="sourceLineNo">219</span> // checked in region servers . see HBASE-5916<a name="line.219"></a> +<span class="sourceLineNo">220</span> Set<ServerName> onlineServers = services.getServerManager().getOnlineServers().keySet();<a name="line.220"></a> +<span class="sourceLineNo">221</span><a name="line.221"></a> +<span class="sourceLineNo">222</span> if (logFolders == null || logFolders.length == 0) {<a name="line.222"></a> +<span class="sourceLineNo">223</span> LOG.debug("No log files to split, proceeding...");<a name="line.223"></a> +<span class="sourceLineNo">224</span> return serverNames;<a name="line.224"></a> +<span class="sourceLineNo">225</span> }<a name="line.225"></a> +<span class="sourceLineNo">226</span> for (FileStatus status : logFolders) {<a name="line.226"></a> +<span class="sourceLineNo">227</span> FileStatus[] curLogFiles = FSUtils.listStatus(this.fs, status.getPath(), null);<a name="line.227"></a> +<span class="sourceLineNo">228</span> if (curLogFiles == null || curLogFiles.length == 0) {<a name="line.228"></a> +<span class="sourceLineNo">229</span> // Empty log folder. No recovery needed<a name="line.229"></a> +<span class="sourceLineNo">230</span> continue;<a name="line.230"></a> +<span class="sourceLineNo">231</span> }<a name="line.231"></a> +<span class="sourceLineNo">232</span> final ServerName serverName = AbstractFSWALProvider.getServerNameFromWALDirectoryName(<a name="line.232"></a> +<span class="sourceLineNo">233</span> status.getPath());<a name="line.233"></a> +<span class="sourceLineNo">234</span> if (null == serverName) {<a name="line.234"></a> +<span class="sourceLineNo">235</span> LOG.warn("Log folder " + status.getPath() + " doesn't look like its name includes a " +<a name="line.235"></a> +<span class="sourceLineNo">236</span> "region server name; leaving in place. If you see later errors about missing " +<a name="line.236"></a> +<span class="sourceLineNo">237</span> "write ahead logs they may be saved in this location.");<a name="line.237"></a> +<span class="sourceLineNo">238</span> } else if (!onlineServers.contains(serverName)) {<a name="line.238"></a> +<span class="sourceLineNo">239</span> LOG.info("Log folder " + status.getPath() + " doesn't belong "<a name="line.239"></a> +<span class="sourceLineNo">240</span> + "to a known region server, splitting");<a name="line.240"></a> +<span class="sourceLineNo">241</span> serverNames.add(serverName);<a name="line.241"></a> +<span class="sourceLineNo">242</span> } else {<a name="line.242"></a> +<span class="sourceLineNo">243</span> LOG.info("Log folder " + status.getPath() + " belongs to an existing region server");<a name="line.243"></a> +<span class="sourceLineNo">244</span> }<a name="line.244"></a> +<span class="sourceLineNo">245</span> }<a name="line.245"></a> +<span class="sourceLineNo">246</span> retrySplitting = false;<a name="line.246"></a> +<span class="sourceLineNo">247</span> } catch (IOException ioe) {<a name="line.247"></a> +<span class="sourceLineNo">248</span> LOG.warn("Failed getting failed servers to be recovered.", ioe);<a name="line.248"></a> +<span class="sourceLineNo">249</span> if (!checkFileSystem()) {<a name="line.249"></a> +<span class="sourceLineNo">250</span> LOG.warn("Bad Filesystem, exiting");<a name="line.250"></a> +<span class="sourceLineNo">251</span> Runtime.getRuntime().halt(1);<a name="line.251"></a> +<span class="sourceLineNo">252</span> }<a name="line.252"></a> +<span class="sourceLineNo">253</span> try {<a name="line.253"></a> +<span class="sourceLineNo">254</span> if (retrySplitting) {<a name="line.254"></a> +<span class="sourceLineNo">255</span> Thread.sleep(conf.getInt("hbase.hlog.split.failure.retry.interval", 30 * 1000));<a name="line.255"></a> +<span class="sourceLineNo">256</span> }<a name="line.256"></a> +<span class="sourceLineNo">257</span> } catch (InterruptedException e) {<a name="line.257"></a> +<span class="sourceLineNo">258</span> LOG.warn("Interrupted, aborting since cannot return w/o splitting");<a name="line.258"></a> +<span class="sourceLineNo">259</span> Thread.currentThread().interrupt();<a name="line.259"></a> +<span class="sourceLineNo">260</span> retrySplitting = false;<a name="line.260"></a> +<span class="sourceLineNo">261</span> Runtime.getRuntime().halt(1);<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> } while (retrySplitting);<a name="line.264"></a> +<span class="sourceLineNo">265</span><a name="line.265"></a> +<span class="sourceLineNo">266</span> return serverNames;<a name="line.266"></a> +<span class="sourceLineNo">267</span> }<a name="line.267"></a> +<span class="sourceLineNo">268</span><a name="line.268"></a> +<span class="sourceLineNo">269</span> public void splitLog(final ServerName serverName) throws IOException {<a name="line.269"></a> +<span class="sourceLineNo">270</span> splitLog(Collections.<ServerName>singleton(serverName));<a name="line.270"></a> +<span class="sourceLineNo">271</span> }<a name="line.271"></a> +<span class="sourceLineNo">272</span><a name="line.272"></a> +<span class="sourceLineNo">273</span> /**<a name="line.273"></a> +<span class="sourceLineNo">274</span> * Specialized method to handle the splitting for meta WAL<a name="line.274"></a> +<span class="sourceLineNo">275</span> * @param serverName logs belonging to this server will be split<a name="line.275"></a> +<span class="sourceLineNo">276</span> */<a name="line.276"></a> +<span class="sourceLineNo">277</span> public void splitMetaLog(final ServerName serverName) throws IOException {<a name="line.277"></a> +<span class="sourceLineNo">278</span> splitMetaLog(Collections.<ServerName>singleton(serverName));<a name="line.278"></a> +<span class="sourceLineNo">279</span> }<a name="line.279"></a> +<span class="sourceLineNo">280</span><a name="line.280"></a> +<span class="sourceLineNo">281</span> /**<a name="line.281"></a> +<span class="sourceLineNo">282</span> * Specialized method to handle the splitting for meta WAL<a name="line.282"></a> +<span class="sourceLineNo">283</span> * @param serverNames logs belonging to these servers will be split<a name="line.283"></a> +<span class="sourceLineNo">284</span> */<a name="line.284"></a> +<span class="sourceLineNo">285</span> public void splitMetaLog(final Set<ServerName> serverNames) throws IOException {<a name="line.285"></a> +<span class="sourceLineNo">286</span> splitLog(serverNames, META_FILTER);<a name="line.286"></a> +<span class="sourceLineNo">287</span> }<a name="line.287"></a> +<span class="sourceLineNo">288</span><a name="line.288"></a> +<span class="sourceLineNo">289</span> @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="UL_UNRELEASED_LOCK", justification=<a name="line.289"></a> +<span class="sourceLineNo">290</span> "We only release this lock when we set it. Updates to code that uses it should verify use " +<a name="line.290"></a> +<span class="sourceLineNo">291</span> "of the guard boolean.")<a name="line.291"></a> +<span class="sourceLineNo">292</span> private List<Path> getLogDirs(final Set<ServerName> serverNames) throws IOException {<a name="line.292"></a> +<span class="sourceLineNo">293</span> List<Path> logDirs = new ArrayList<>();<a name="line.293"></a> +<span class="sourceLineNo">294</span> boolean needReleaseLock = false;<a name="line.294"></a> +<span class="sourceLineNo">295</span> if (!this.services.isInitialized()) {<a name="line.295"></a> +<span class="sourceLineNo">296</span> // during master initialization, we could have multiple places splitting a same wal<a name="line.296"></a> +<span class="sourceLineNo">297</span> // XXX: Does this still exist after we move to proc-v2?<a name="line.297"></a> +<span class="sourceLineNo">298</span> this.splitLogLock.lock();<a name="line.298"></a> +<span class="sourceLineNo">299</span> needReleaseLock = true;<a name="line.299"></a> +<span class="sourceLineNo">300</span> }<a name="line.300"></a> +<span class="sourceLineNo">301</span> try {<a name="line.301"></a> +<span class="sourceLineNo">302</span> for (ServerName serverName : serverNames) {<a name="line.302"></a> +<span class="sourceLineNo">303</span> Path logDir = new Path(this.rootDir,<a name="line.303"></a> +<span class="sourceLineNo">304</span> AbstractFSWALProvider.getWALDirectoryName(serverName.toString()));<a name="line.304"></a> +<span class="sourceLineNo">305</span> Path splitDir = logDir.suffix(AbstractFSWALProvider.SPLITTING_EXT);<a name="line.305"></a> +<span class="sourceLineNo">306</span> // Rename the directory so a rogue RS doesn't create more WALs<a name="line.306"></a> +<span class="sourceLineNo">307</span> if (fs.exists(logDir)) {<a name="line.307"></a> +<span class="sourceLineNo">308</span> if (!this.fs.rename(logDir, splitDir)) {<a name="line.308"></a> +<span class="sourceLineNo">309</span> throw new IOException("Failed fs.rename for log split: " + logDir);<a name="line.309"></a> +<span class="sourceLineNo">310</span> }<a name="line.310"></a> +<span class="sourceLineNo">311</span> logDir = splitDir;<a name="line.311"></a> +<span class="sourceLineNo">312</span> LOG.debug("Renamed region directory: " + splitDir);<a name="line.312"></a> +<span class="sourceLineNo">313</span> } else if (!fs.exists(splitDir)) {<a name="line.313"></a> +<span class="sourceLineNo">314</span> LOG.info("Log dir for server " + serverName + " does not exist");<a name="line.314"></a> +<span class="sourceLineNo">315</span> continue;<a name="line.315"></a> +<span class="sourceLineNo">316</span> }<a name="line.316"></a> +<span class="sourceLineNo">317</span> logDirs.add(splitDir);<a name="line.317"></a> +<span class="sourceLineNo">318</span> }<a name="line.318"></a> +<span class="sourceLineNo">319</span> } catch (IOException ioe) {<a name="line.319"></a> +<span class="sourceLineNo">320</span> if (!checkFileSystem()) {<a name="line.320"></a> +<span class="sourceLineNo">321</span> this.services.abort("Aborting due to filesystem unavailable", ioe);<a name="line.321"></a> +<span class="sourceLineNo">322</span> throw ioe;<a name="line.322"></a> +<span class="sourceLineNo">323</span> }<a name="line.323"></a> +<span class="sourceLineNo">324</span> } finally {<a name="line.324"></a> +<span class="sourceLineNo">325</span> if (needReleaseLock) {<a name="line.325"></a> +<span class="sourceLineNo">326</span> this.splitLogLock.unlock();<a name="line.326"></a> +<span class="sourceLineNo">327</span> }<a name="line.327"></a> +<span class="sourceLineNo">328</span> }<a name="line.328"></a> +<span class="sourceLineNo">329</span> return logDirs;<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> public void splitLog(final Set<ServerName> serverNames) throws IOException {<a name="line.332"></a> +<span class="sourceLineNo">333</span> splitLog(serverNames, NON_META_FILTER);<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">335</span><a name="line.335"></a> +<span class="sourceLineNo">336</span> /**<a name="line.336"></a> +<span class="sourceLineNo">337</span> * This method is the base split method that splits WAL files matching a filter. Callers should<a name="line.337"></a> +<span class="sourceLineNo">338</span> * pass the appropriate filter for meta and non-meta WALs.<a name="line.338"></a> +<span class="sourceLineNo">339</span> * @param serverNames logs belonging to these servers will be split; this will rename the log<a name="line.339"></a> +<span class="sourceLineNo">340</span> * directory out from under a soft-failed server<a name="line.340"></a> +<span class="sourceLineNo">341</span> */<a name="line.341"></a> +<span class="sourceLineNo">342</span> public void splitLog(final Set<ServerName> serverNames, PathFilter filter) throws IOException {<a name="line.342"></a> +<span class="sourceLineNo">343</span> long splitTime = 0, splitLogSize = 0;<a name="line.343"></a> +<span class="sourceLineNo">344</span> List<Path> logDirs = getLogDirs(serverNames);<a name="line.344"></a> +<span class="sourceLineNo">345</span><a name="line.345"></a> +<span class="sourceLineNo">346</span> splitLogManager.handleDeadWorkers(serverNames);<a name="line.346"></a> +<span class="sourceLineNo">347</span> splitTime = EnvironmentEdgeManager.currentTime();<a name="line.347"></a> +<span class="sourceLineNo">348</span> splitLogSize = splitLogManager.splitLogDistributed(serverNames, logDirs, filter);<a name="line.348"></a> +<span class="sourceLineNo">349</span> splitTime = EnvironmentEdgeManager.currentTime() - splitTime;<a name="line.349"></a> +<span class="sourceLineNo">350</span><a name="line.350"></a> +<span class="sourceLineNo">351</span> if (this.metricsMasterFilesystem != null) {<a name="line.351"></a> +<span class="sourceLineNo">352</span> if (filter == META_FILTER) {<a name="line.352"></a> +<span class="sourceLineNo">353</span> this.metricsMasterFilesystem.addMetaWALSplit(splitTime, splitLogSize);<a name="line.353"></a> +<span class="sourceLineNo">354</span> } else {<a name="line.354"></a> +<span class="sourceLineNo">355</span> this.metricsMasterFilesystem.addSplit(splitTime, splitLogSize);<a name="line.355"></a> +<span class="sourceLineNo">356</span> }<a name="line.356"></a> +<span class="sourceLineNo">357</span> }<a name="line.357"></a> +<span class="sourceLineNo">358</span> }<a name="line.358"></a> +<span class="sourceLineNo">359</span>}<a name="line.359"></a>
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/738e976e/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html index 1b8c609..b5a9f93 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/RegionServerTracker.html @@ -123,77 +123,85 @@ <span class="sourceLineNo">115</span> * protection to prevent concurrency issues with server expiration operation.<a name="line.115"></a> <span class="sourceLineNo">116</span> * @param deadServersFromPE the region servers which already have SCP associated.<a name="line.116"></a> <span class="sourceLineNo">117</span> * @param liveServersFromWALDir the live region servers from wal directory.<a name="line.117"></a> -<span class="sourceLineNo">118</span> */<a name="line.118"></a> -<span class="sourceLineNo">119</span> public void start(Set<ServerName> deadServersFromPE, Set<ServerName> liveServersFromWALDir)<a name="line.119"></a> -<span class="sourceLineNo">120</span> throws KeeperException, IOException {<a name="line.120"></a> -<span class="sourceLineNo">121</span> LOG.info("Starting RegionServerTracker; {} have existing ServerCrashProcedures, {} " +<a name="line.121"></a> -<span class="sourceLineNo">122</span> "possibly 'live' servers.", deadServersFromPE.size(), liveServersFromWALDir.size());<a name="line.122"></a> -<span class="sourceLineNo">123</span> watcher.registerListener(this);<a name="line.123"></a> -<span class="sourceLineNo">124</span> synchronized (this) {<a name="line.124"></a> -<span class="sourceLineNo">125</span> List<String> servers =<a name="line.125"></a> -<span class="sourceLineNo">126</span> ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.126"></a> -<span class="sourceLineNo">127</span> for (String n : servers) {<a name="line.127"></a> -<span class="sourceLineNo">128</span> Pair<ServerName, RegionServerInfo> pair = getServerInfo(n);<a name="line.128"></a> -<span class="sourceLineNo">129</span> ServerName serverName = pair.getFirst();<a name="line.129"></a> -<span class="sourceLineNo">130</span> RegionServerInfo info = pair.getSecond();<a name="line.130"></a> -<span class="sourceLineNo">131</span> regionServers.add(serverName);<a name="line.131"></a> -<span class="sourceLineNo">132</span> ServerMetrics serverMetrics = info != null ? ServerMetricsBuilder.of(serverName,<a name="line.132"></a> -<span class="sourceLineNo">133</span> VersionInfoUtil.getVersionNumber(info.getVersionInfo()),<a name="line.133"></a> -<span class="sourceLineNo">134</span> info.getVersionInfo().getVersion()) : ServerMetricsBuilder.of(serverName);<a name="line.134"></a> -<span class="sourceLineNo">135</span> serverManager.checkAndRecordNewServer(serverName, serverMetrics);<a name="line.135"></a> -<span class="sourceLineNo">136</span> }<a name="line.136"></a> -<span class="sourceLineNo">137</span> serverManager.findDeadServersAndProcess(deadServersFromPE, liveServersFromWALDir);<a name="line.137"></a> -<span class="sourceLineNo">138</span> }<a name="line.138"></a> -<span class="sourceLineNo">139</span> }<a name="line.139"></a> -<span class="sourceLineNo">140</span><a name="line.140"></a> -<span class="sourceLineNo">141</span> public void stop() {<a name="line.141"></a> -<span class="sourceLineNo">142</span> executor.shutdownNow();<a name="line.142"></a> -<span class="sourceLineNo">143</span> }<a name="line.143"></a> -<span class="sourceLineNo">144</span><a name="line.144"></a> -<span class="sourceLineNo">145</span> private synchronized void refresh() {<a name="line.145"></a> -<span class="sourceLineNo">146</span> List<String> names;<a name="line.146"></a> -<span class="sourceLineNo">147</span> try {<a name="line.147"></a> -<span class="sourceLineNo">148</span> names = ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.148"></a> -<span class="sourceLineNo">149</span> } catch (KeeperException e) {<a name="line.149"></a> -<span class="sourceLineNo">150</span> // here we need to abort as we failed to set watcher on the rs node which means that we can<a name="line.150"></a> -<span class="sourceLineNo">151</span> // not track the node deleted evetnt any more.<a name="line.151"></a> -<span class="sourceLineNo">152</span> server.abort("Unexpected zk exception getting RS nodes", e);<a name="line.152"></a> -<span class="sourceLineNo">153</span> return;<a name="line.153"></a> -<span class="sourceLineNo">154</span> }<a name="line.154"></a> -<span class="sourceLineNo">155</span> Set<ServerName> servers =<a name="line.155"></a> -<span class="sourceLineNo">156</span> names.stream().map(ServerName::parseServerName).collect(Collectors.toSet());<a name="line.156"></a> -<span class="sourceLineNo">157</span> for (Iterator<ServerName> iter = regionServers.iterator(); iter.hasNext();) {<a name="line.157"></a> -<span class="sourceLineNo">158</span> ServerName sn = iter.next();<a name="line.158"></a> -<span class="sourceLineNo">159</span> if (!servers.contains(sn)) {<a name="line.159"></a> -<span class="sourceLineNo">160</span> LOG.info("RegionServer ephemeral node deleted, processing expiration [{}]", sn);<a name="line.160"></a> -<span class="sourceLineNo">161</span> serverManager.expireServer(sn);<a name="line.161"></a> -<span class="sourceLineNo">162</span> iter.remove();<a name="line.162"></a> -<span class="sourceLineNo">163</span> }<a name="line.163"></a> -<span class="sourceLineNo">164</span> }<a name="line.164"></a> -<span class="sourceLineNo">165</span> // here we do not need to parse the region server info as it is useless now, we only need the<a name="line.165"></a> -<span class="sourceLineNo">166</span> // server name.<a name="line.166"></a> -<span class="sourceLineNo">167</span> boolean newServerAdded = false;<a name="line.167"></a> -<span class="sourceLineNo">168</span> for (ServerName sn : servers) {<a name="line.168"></a> -<span class="sourceLineNo">169</span> if (regionServers.add(sn)) {<a name="line.169"></a> -<span class="sourceLineNo">170</span> newServerAdded = true;<a name="line.170"></a> -<span class="sourceLineNo">171</span> LOG.info("RegionServer ephemeral node created, adding [" + sn + "]");<a name="line.171"></a> -<span class="sourceLineNo">172</span> }<a name="line.172"></a> -<span class="sourceLineNo">173</span> }<a name="line.173"></a> -<span class="sourceLineNo">174</span> if (newServerAdded && server.isInitialized()) {<a name="line.174"></a> -<span class="sourceLineNo">175</span> // Only call the check to move servers if a RegionServer was added to the cluster; in this<a name="line.175"></a> -<span class="sourceLineNo">176</span> // case it could be a server with a new version so it makes sense to run the check.<a name="line.176"></a> -<span class="sourceLineNo">177</span> server.checkIfShouldMoveSystemRegionAsync();<a name="line.177"></a> -<span class="sourceLineNo">178</span> }<a name="line.178"></a> -<span class="sourceLineNo">179</span> }<a name="line.179"></a> -<span class="sourceLineNo">180</span><a name="line.180"></a> -<span class="sourceLineNo">181</span> @Override<a name="line.181"></a> -<span class="sourceLineNo">182</span> public void nodeChildrenChanged(String path) {<a name="line.182"></a> -<span class="sourceLineNo">183</span> if (path.equals(watcher.getZNodePaths().rsZNode) && !server.isAborted() &&<a name="line.183"></a> -<span class="sourceLineNo">184</span> !server.isStopped()) {<a name="line.184"></a> -<span class="sourceLineNo">185</span> executor.execute(this::refresh);<a name="line.185"></a> +<span class="sourceLineNo">118</span> * @param splittingServersFromWALDir Servers whose WALs are being actively 'split'.<a name="line.118"></a> +<span class="sourceLineNo">119</span> */<a name="line.119"></a> +<span class="sourceLineNo">120</span> public void start(Set<ServerName> deadServersFromPE, Set<ServerName> liveServersFromWALDir,<a name="line.120"></a> +<span class="sourceLineNo">121</span> Set<ServerName> splittingServersFromWALDir)<a name="line.121"></a> +<span class="sourceLineNo">122</span> throws KeeperException, IOException {<a name="line.122"></a> +<span class="sourceLineNo">123</span> LOG.info("Starting RegionServerTracker; {} have existing ServerCrashProcedures, {} " +<a name="line.123"></a> +<span class="sourceLineNo">124</span> "possibly 'live' servers, and {} 'splitting'.", deadServersFromPE.size(),<a name="line.124"></a> +<span class="sourceLineNo">125</span> liveServersFromWALDir.size(), splittingServersFromWALDir.size());<a name="line.125"></a> +<span class="sourceLineNo">126</span> // deadServersFromPE is made from a list of outstanding ServerCrashProcedures.<a name="line.126"></a> +<span class="sourceLineNo">127</span> // splittingServersFromWALDir are being actively split -- the directory in the FS ends in<a name="line.127"></a> +<span class="sourceLineNo">128</span> // '-SPLITTING'. Each splitting server should have a corresponding SCP. Log if not.<a name="line.128"></a> +<span class="sourceLineNo">129</span> splittingServersFromWALDir.stream().map(s -> !deadServersFromPE.contains(s)).<a name="line.129"></a> +<span class="sourceLineNo">130</span> forEach(s -> LOG.error("{} has no matching ServerCrashProcedure", s));<a name="line.130"></a> +<span class="sourceLineNo">131</span> watcher.registerListener(this);<a name="line.131"></a> +<span class="sourceLineNo">132</span> synchronized (this) {<a name="line.132"></a> +<span class="sourceLineNo">133</span> List<String> servers =<a name="line.133"></a> +<span class="sourceLineNo">134</span> ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.134"></a> +<span class="sourceLineNo">135</span> for (String n : servers) {<a name="line.135"></a> +<span class="sourceLineNo">136</span> Pair<ServerName, RegionServerInfo> pair = getServerInfo(n);<a name="line.136"></a> +<span class="sourceLineNo">137</span> ServerName serverName = pair.getFirst();<a name="line.137"></a> +<span class="sourceLineNo">138</span> RegionServerInfo info = pair.getSecond();<a name="line.138"></a> +<span class="sourceLineNo">139</span> regionServers.add(serverName);<a name="line.139"></a> +<span class="sourceLineNo">140</span> ServerMetrics serverMetrics = info != null ? ServerMetricsBuilder.of(serverName,<a name="line.140"></a> +<span class="sourceLineNo">141</span> VersionInfoUtil.getVersionNumber(info.getVersionInfo()),<a name="line.141"></a> +<span class="sourceLineNo">142</span> info.getVersionInfo().getVersion()) : ServerMetricsBuilder.of(serverName);<a name="line.142"></a> +<span class="sourceLineNo">143</span> serverManager.checkAndRecordNewServer(serverName, serverMetrics);<a name="line.143"></a> +<span class="sourceLineNo">144</span> }<a name="line.144"></a> +<span class="sourceLineNo">145</span> serverManager.findDeadServersAndProcess(deadServersFromPE, liveServersFromWALDir);<a name="line.145"></a> +<span class="sourceLineNo">146</span> }<a name="line.146"></a> +<span class="sourceLineNo">147</span> }<a name="line.147"></a> +<span class="sourceLineNo">148</span><a name="line.148"></a> +<span class="sourceLineNo">149</span> public void stop() {<a name="line.149"></a> +<span class="sourceLineNo">150</span> executor.shutdownNow();<a name="line.150"></a> +<span class="sourceLineNo">151</span> }<a name="line.151"></a> +<span class="sourceLineNo">152</span><a name="line.152"></a> +<span class="sourceLineNo">153</span> private synchronized void refresh() {<a name="line.153"></a> +<span class="sourceLineNo">154</span> List<String> names;<a name="line.154"></a> +<span class="sourceLineNo">155</span> try {<a name="line.155"></a> +<span class="sourceLineNo">156</span> names = ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.156"></a> +<span class="sourceLineNo">157</span> } catch (KeeperException e) {<a name="line.157"></a> +<span class="sourceLineNo">158</span> // here we need to abort as we failed to set watcher on the rs node which means that we can<a name="line.158"></a> +<span class="sourceLineNo">159</span> // not track the node deleted evetnt any more.<a name="line.159"></a> +<span class="sourceLineNo">160</span> server.abort("Unexpected zk exception getting RS nodes", e);<a name="line.160"></a> +<span class="sourceLineNo">161</span> return;<a name="line.161"></a> +<span class="sourceLineNo">162</span> }<a name="line.162"></a> +<span class="sourceLineNo">163</span> Set<ServerName> servers =<a name="line.163"></a> +<span class="sourceLineNo">164</span> names.stream().map(ServerName::parseServerName).collect(Collectors.toSet());<a name="line.164"></a> +<span class="sourceLineNo">165</span> for (Iterator<ServerName> iter = regionServers.iterator(); iter.hasNext();) {<a name="line.165"></a> +<span class="sourceLineNo">166</span> ServerName sn = iter.next();<a name="line.166"></a> +<span class="sourceLineNo">167</span> if (!servers.contains(sn)) {<a name="line.167"></a> +<span class="sourceLineNo">168</span> LOG.info("RegionServer ephemeral node deleted, processing expiration [{}]", sn);<a name="line.168"></a> +<span class="sourceLineNo">169</span> serverManager.expireServer(sn);<a name="line.169"></a> +<span class="sourceLineNo">170</span> iter.remove();<a name="line.170"></a> +<span class="sourceLineNo">171</span> }<a name="line.171"></a> +<span class="sourceLineNo">172</span> }<a name="line.172"></a> +<span class="sourceLineNo">173</span> // here we do not need to parse the region server info as it is useless now, we only need the<a name="line.173"></a> +<span class="sourceLineNo">174</span> // server name.<a name="line.174"></a> +<span class="sourceLineNo">175</span> boolean newServerAdded = false;<a name="line.175"></a> +<span class="sourceLineNo">176</span> for (ServerName sn : servers) {<a name="line.176"></a> +<span class="sourceLineNo">177</span> if (regionServers.add(sn)) {<a name="line.177"></a> +<span class="sourceLineNo">178</span> newServerAdded = true;<a name="line.178"></a> +<span class="sourceLineNo">179</span> LOG.info("RegionServer ephemeral node created, adding [" + sn + "]");<a name="line.179"></a> +<span class="sourceLineNo">180</span> }<a name="line.180"></a> +<span class="sourceLineNo">181</span> }<a name="line.181"></a> +<span class="sourceLineNo">182</span> if (newServerAdded && server.isInitialized()) {<a name="line.182"></a> +<span class="sourceLineNo">183</span> // Only call the check to move servers if a RegionServer was added to the cluster; in this<a name="line.183"></a> +<span class="sourceLineNo">184</span> // case it could be a server with a new version so it makes sense to run the check.<a name="line.184"></a> +<span class="sourceLineNo">185</span> server.checkIfShouldMoveSystemRegionAsync();<a name="line.185"></a> <span class="sourceLineNo">186</span> }<a name="line.186"></a> <span class="sourceLineNo">187</span> }<a name="line.187"></a> -<span class="sourceLineNo">188</span>}<a name="line.188"></a> +<span class="sourceLineNo">188</span><a name="line.188"></a> +<span class="sourceLineNo">189</span> @Override<a name="line.189"></a> +<span class="sourceLineNo">190</span> public void nodeChildrenChanged(String path) {<a name="line.190"></a> +<span class="sourceLineNo">191</span> if (path.equals(watcher.getZNodePaths().rsZNode) && !server.isAborted() &&<a name="line.191"></a> +<span class="sourceLineNo">192</span> !server.isStopped()) {<a name="line.192"></a> +<span class="sourceLineNo">193</span> executor.execute(this::refresh);<a name="line.193"></a> +<span class="sourceLineNo">194</span> }<a name="line.194"></a> +<span class="sourceLineNo">195</span> }<a name="line.195"></a> +<span class="sourceLineNo">196</span>}<a name="line.196"></a> http://git-wip-us.apache.org/repos/asf/hbase-site/blob/738e976e/devapidocs/src-html/org/apache/hadoop/hbase/master/TableNamespaceManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/TableNamespaceManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/TableNamespaceManager.html index a1f5165..73a8028 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/TableNamespaceManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/TableNamespaceManager.html @@ -391,18 +391,22 @@ <span class="sourceLineNo">383</span> return;<a name="line.383"></a> <span class="sourceLineNo">384</span> }<a name="line.384"></a> <span class="sourceLineNo">385</span> try {<a name="line.385"></a> -<span class="sourceLineNo">386</span> this.zkNamespaceManager.stop();<a name="line.386"></a> -<span class="sourceLineNo">387</span> } catch (IOException ioe) {<a name="line.387"></a> -<span class="sourceLineNo">388</span> LOG.warn("Failed NamespaceManager close", ioe);<a name="line.388"></a> -<span class="sourceLineNo">389</span> }<a name="line.389"></a> -<span class="sourceLineNo">390</span> try {<a name="line.390"></a> -<span class="sourceLineNo">391</span> this.nsTable.close();<a name="line.391"></a> -<span class="sourceLineNo">392</span> } catch (IOException ioe) {<a name="line.392"></a> -<span class="sourceLineNo">393</span> LOG.warn("Failed Namespace Table close", ioe);<a name="line.393"></a> -<span class="sourceLineNo">394</span> }<a name="line.394"></a> -<span class="sourceLineNo">395</span> this.stopped = true;<a name="line.395"></a> -<span class="sourceLineNo">396</span> }<a name="line.396"></a> -<span class="sourceLineNo">397</span>}<a name="line.397"></a> +<span class="sourceLineNo">386</span> if (this.zkNamespaceManager != null) {<a name="line.386"></a> +<span class="sourceLineNo">387</span> this.zkNamespaceManager.stop();<a name="line.387"></a> +<span class="sourceLineNo">388</span> }<a name="line.388"></a> +<span class="sourceLineNo">389</span> } catch (IOException ioe) {<a name="line.389"></a> +<span class="sourceLineNo">390</span> LOG.warn("Failed NamespaceManager close", ioe);<a name="line.390"></a> +<span class="sourceLineNo">391</span> }<a name="line.391"></a> +<span class="sourceLineNo">392</span> try {<a name="line.392"></a> +<span class="sourceLineNo">393</span> if (this.nsTable != null) {<a name="line.393"></a> +<span class="sourceLineNo">394</span> this.nsTable.close();<a name="line.394"></a> +<span class="sourceLineNo">395</span> }<a name="line.395"></a> +<span class="sourceLineNo">396</span> } catch (IOException ioe) {<a name="line.396"></a> +<span class="sourceLineNo">397</span> LOG.warn("Failed Namespace Table close", ioe);<a name="line.397"></a> +<span class="sourceLineNo">398</span> }<a name="line.398"></a> +<span class="sourceLineNo">399</span> this.stopped = true;<a name="line.399"></a> +<span class="sourceLineNo">400</span> }<a name="line.400"></a> +<span class="sourceLineNo">401</span>}<a name="line.401"></a>
