http://git-wip-us.apache.org/repos/asf/hbase-site/blob/4f0b7674/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html index 6458b43..23c8bb3 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html @@ -230,843 +230,850 @@ <span class="sourceLineNo">222</span> /**<a name="line.222"></a> <span class="sourceLineNo">223</span> * Let the server manager know a new regionserver has come online<a name="line.223"></a> <span class="sourceLineNo">224</span> * @param request the startup request<a name="line.224"></a> -<span class="sourceLineNo">225</span> * @param ia the InetAddress from which request is received<a name="line.225"></a> -<span class="sourceLineNo">226</span> * @return The ServerName we know this server as.<a name="line.226"></a> -<span class="sourceLineNo">227</span> * @throws IOException<a name="line.227"></a> -<span class="sourceLineNo">228</span> */<a name="line.228"></a> -<span class="sourceLineNo">229</span> ServerName regionServerStartup(RegionServerStartupRequest request, InetAddress ia)<a name="line.229"></a> -<span class="sourceLineNo">230</span> throws IOException {<a name="line.230"></a> -<span class="sourceLineNo">231</span> // Test for case where we get a region startup message from a regionserver<a name="line.231"></a> -<span class="sourceLineNo">232</span> // that has been quickly restarted but whose znode expiration handler has<a name="line.232"></a> -<span class="sourceLineNo">233</span> // not yet run, or from a server whose fail we are currently processing.<a name="line.233"></a> -<span class="sourceLineNo">234</span> // Test its host+port combo is present in serverAddressToServerInfo. If it<a name="line.234"></a> -<span class="sourceLineNo">235</span> // is, reject the server and trigger its expiration. The next time it comes<a name="line.235"></a> -<span class="sourceLineNo">236</span> // in, it should have been removed from serverAddressToServerInfo and queued<a name="line.236"></a> -<span class="sourceLineNo">237</span> // for processing by ProcessServerShutdown.<a name="line.237"></a> -<span class="sourceLineNo">238</span><a name="line.238"></a> -<span class="sourceLineNo">239</span> final String hostname = request.hasUseThisHostnameInstead() ?<a name="line.239"></a> -<span class="sourceLineNo">240</span> request.getUseThisHostnameInstead() :ia.getHostName();<a name="line.240"></a> -<span class="sourceLineNo">241</span> ServerName sn = ServerName.valueOf(hostname, request.getPort(),<a name="line.241"></a> -<span class="sourceLineNo">242</span> request.getServerStartCode());<a name="line.242"></a> -<span class="sourceLineNo">243</span> checkClockSkew(sn, request.getServerCurrentTime());<a name="line.243"></a> -<span class="sourceLineNo">244</span> checkIsDead(sn, "STARTUP");<a name="line.244"></a> -<span class="sourceLineNo">245</span> if (!checkAndRecordNewServer(sn, ServerMetricsBuilder.of(sn))) {<a name="line.245"></a> -<span class="sourceLineNo">246</span> LOG.warn("THIS SHOULD NOT HAPPEN, RegionServerStartup"<a name="line.246"></a> -<span class="sourceLineNo">247</span> + " could not record the server: " + sn);<a name="line.247"></a> -<span class="sourceLineNo">248</span> }<a name="line.248"></a> -<span class="sourceLineNo">249</span> return sn;<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> * Updates last flushed sequence Ids for the regions on server sn<a name="line.253"></a> -<span class="sourceLineNo">254</span> * @param sn<a name="line.254"></a> -<span class="sourceLineNo">255</span> * @param hsl<a name="line.255"></a> -<span class="sourceLineNo">256</span> */<a name="line.256"></a> -<span class="sourceLineNo">257</span> private void updateLastFlushedSequenceIds(ServerName sn, ServerMetrics hsl) {<a name="line.257"></a> -<span class="sourceLineNo">258</span> for (Entry<byte[], RegionMetrics> entry : hsl.getRegionMetrics().entrySet()) {<a name="line.258"></a> -<span class="sourceLineNo">259</span> byte[] encodedRegionName = Bytes.toBytes(RegionInfo.encodeRegionName(entry.getKey()));<a name="line.259"></a> -<span class="sourceLineNo">260</span> Long existingValue = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.260"></a> -<span class="sourceLineNo">261</span> long l = entry.getValue().getCompletedSequenceId();<a name="line.261"></a> -<span class="sourceLineNo">262</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.262"></a> -<span class="sourceLineNo">263</span> if (LOG.isTraceEnabled()) {<a name="line.263"></a> -<span class="sourceLineNo">264</span> LOG.trace(Bytes.toString(encodedRegionName) + ", existingValue=" + existingValue +<a name="line.264"></a> -<span class="sourceLineNo">265</span> ", completeSequenceId=" + l);<a name="line.265"></a> -<span class="sourceLineNo">266</span> }<a name="line.266"></a> -<span class="sourceLineNo">267</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue)) {<a name="line.267"></a> -<span class="sourceLineNo">268</span> flushedSequenceIdByRegion.put(encodedRegionName, l);<a name="line.268"></a> -<span class="sourceLineNo">269</span> } else if (l != HConstants.NO_SEQNUM && l < existingValue) {<a name="line.269"></a> -<span class="sourceLineNo">270</span> LOG.warn("RegionServer " + sn + " indicates a last flushed sequence id ("<a name="line.270"></a> -<span class="sourceLineNo">271</span> + l + ") that is less than the previous last flushed sequence id ("<a name="line.271"></a> -<span class="sourceLineNo">272</span> + existingValue + ") for region " + Bytes.toString(entry.getKey()) + " Ignoring.");<a name="line.272"></a> -<span class="sourceLineNo">273</span> }<a name="line.273"></a> -<span class="sourceLineNo">274</span> ConcurrentNavigableMap<byte[], Long> storeFlushedSequenceId =<a name="line.274"></a> -<span class="sourceLineNo">275</span> computeIfAbsent(storeFlushedSequenceIdsByRegion, encodedRegionName,<a name="line.275"></a> -<span class="sourceLineNo">276</span> () -> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR));<a name="line.276"></a> -<span class="sourceLineNo">277</span> for (Entry<byte[], Long> storeSeqId : entry.getValue().getStoreSequenceId().entrySet()) {<a name="line.277"></a> -<span class="sourceLineNo">278</span> byte[] family = storeSeqId.getKey();<a name="line.278"></a> -<span class="sourceLineNo">279</span> existingValue = storeFlushedSequenceId.get(family);<a name="line.279"></a> -<span class="sourceLineNo">280</span> l = storeSeqId.getValue();<a name="line.280"></a> -<span class="sourceLineNo">281</span> if (LOG.isTraceEnabled()) {<a name="line.281"></a> -<span class="sourceLineNo">282</span> LOG.trace(Bytes.toString(encodedRegionName) + ", family=" + Bytes.toString(family) +<a name="line.282"></a> -<span class="sourceLineNo">283</span> ", existingValue=" + existingValue + ", completeSequenceId=" + l);<a name="line.283"></a> -<span class="sourceLineNo">284</span> }<a name="line.284"></a> -<span class="sourceLineNo">285</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.285"></a> -<span class="sourceLineNo">286</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue.longValue())) {<a name="line.286"></a> -<span class="sourceLineNo">287</span> storeFlushedSequenceId.put(family, l);<a name="line.287"></a> -<span class="sourceLineNo">288</span> }<a name="line.288"></a> -<span class="sourceLineNo">289</span> }<a name="line.289"></a> -<span class="sourceLineNo">290</span> }<a name="line.290"></a> -<span class="sourceLineNo">291</span> }<a name="line.291"></a> -<span class="sourceLineNo">292</span><a name="line.292"></a> -<span class="sourceLineNo">293</span> @VisibleForTesting<a name="line.293"></a> -<span class="sourceLineNo">294</span> public void regionServerReport(ServerName sn,<a name="line.294"></a> -<span class="sourceLineNo">295</span> ServerMetrics sl) throws YouAreDeadException {<a name="line.295"></a> -<span class="sourceLineNo">296</span> checkIsDead(sn, "REPORT");<a name="line.296"></a> -<span class="sourceLineNo">297</span> if (null == this.onlineServers.replace(sn, sl)) {<a name="line.297"></a> -<span class="sourceLineNo">298</span> // Already have this host+port combo and its just different start code?<a name="line.298"></a> -<span class="sourceLineNo">299</span> // Just let the server in. Presume master joining a running cluster.<a name="line.299"></a> -<span class="sourceLineNo">300</span> // recordNewServer is what happens at the end of reportServerStartup.<a name="line.300"></a> -<span class="sourceLineNo">301</span> // The only thing we are skipping is passing back to the regionserver<a name="line.301"></a> -<span class="sourceLineNo">302</span> // the ServerName to use. Here we presume a master has already done<a name="line.302"></a> -<span class="sourceLineNo">303</span> // that so we'll press on with whatever it gave us for ServerName.<a name="line.303"></a> -<span class="sourceLineNo">304</span> if (!checkAndRecordNewServer(sn, sl)) {<a name="line.304"></a> -<span class="sourceLineNo">305</span> LOG.info("RegionServerReport ignored, could not record the server: " + sn);<a name="line.305"></a> -<span class="sourceLineNo">306</span> return; // Not recorded, so no need to move on<a name="line.306"></a> -<span class="sourceLineNo">307</span> }<a name="line.307"></a> -<span class="sourceLineNo">308</span> }<a name="line.308"></a> -<span class="sourceLineNo">309</span> updateLastFlushedSequenceIds(sn, sl);<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> * Check is a server of same host and port already exists,<a name="line.313"></a> -<span class="sourceLineNo">314</span> * if not, or the existed one got a smaller start code, record it.<a name="line.314"></a> -<span class="sourceLineNo">315</span> *<a name="line.315"></a> -<span class="sourceLineNo">316</span> * @param serverName the server to check and record<a name="line.316"></a> -<span class="sourceLineNo">317</span> * @param sl the server load on the server<a name="line.317"></a> -<span class="sourceLineNo">318</span> * @return true if the server is recorded, otherwise, false<a name="line.318"></a> -<span class="sourceLineNo">319</span> */<a name="line.319"></a> -<span class="sourceLineNo">320</span> boolean checkAndRecordNewServer(final ServerName serverName, final ServerMetrics sl) {<a name="line.320"></a> -<span class="sourceLineNo">321</span> ServerName existingServer = null;<a name="line.321"></a> -<span class="sourceLineNo">322</span> synchronized (this.onlineServers) {<a name="line.322"></a> -<span class="sourceLineNo">323</span> existingServer = findServerWithSameHostnamePortWithLock(serverName);<a name="line.323"></a> -<span class="sourceLineNo">324</span> if (existingServer != null && (existingServer.getStartcode() > serverName.getStartcode())) {<a name="line.324"></a> -<span class="sourceLineNo">325</span> LOG.info("Server serverName=" + serverName + " rejected; we already have "<a name="line.325"></a> -<span class="sourceLineNo">326</span> + existingServer.toString() + " registered with same hostname and port");<a name="line.326"></a> -<span class="sourceLineNo">327</span> return false;<a name="line.327"></a> -<span class="sourceLineNo">328</span> }<a name="line.328"></a> -<span class="sourceLineNo">329</span> recordNewServerWithLock(serverName, sl);<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> // Tell our listeners that a server was added<a name="line.332"></a> -<span class="sourceLineNo">333</span> if (!this.listeners.isEmpty()) {<a name="line.333"></a> -<span class="sourceLineNo">334</span> for (ServerListener listener : this.listeners) {<a name="line.334"></a> -<span class="sourceLineNo">335</span> listener.serverAdded(serverName);<a name="line.335"></a> -<span class="sourceLineNo">336</span> }<a name="line.336"></a> -<span class="sourceLineNo">337</span> }<a name="line.337"></a> -<span class="sourceLineNo">338</span><a name="line.338"></a> -<span class="sourceLineNo">339</span> // Note that we assume that same ts means same server, and don't expire in that case.<a name="line.339"></a> -<span class="sourceLineNo">340</span> // TODO: ts can theoretically collide due to clock shifts, so this is a bit hacky.<a name="line.340"></a> -<span class="sourceLineNo">341</span> if (existingServer != null &&<a name="line.341"></a> -<span class="sourceLineNo">342</span> (existingServer.getStartcode() < serverName.getStartcode())) {<a name="line.342"></a> -<span class="sourceLineNo">343</span> LOG.info("Triggering server recovery; existingServer " +<a name="line.343"></a> -<span class="sourceLineNo">344</span> existingServer + " looks stale, new server:" + serverName);<a name="line.344"></a> -<span class="sourceLineNo">345</span> expireServer(existingServer);<a name="line.345"></a> -<span class="sourceLineNo">346</span> }<a name="line.346"></a> -<span class="sourceLineNo">347</span> return true;<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> * Checks if the clock skew between the server and the master. If the clock skew exceeds the<a name="line.351"></a> -<span class="sourceLineNo">352</span> * configured max, it will throw an exception; if it exceeds the configured warning threshold,<a name="line.352"></a> -<span class="sourceLineNo">353</span> * it will log a warning but start normally.<a name="line.353"></a> -<span class="sourceLineNo">354</span> * @param serverName Incoming servers's name<a name="line.354"></a> -<span class="sourceLineNo">355</span> * @param serverCurrentTime<a name="line.355"></a> -<span class="sourceLineNo">356</span> * @throws ClockOutOfSyncException if the skew exceeds the configured max value<a name="line.356"></a> -<span class="sourceLineNo">357</span> */<a name="line.357"></a> -<span class="sourceLineNo">358</span> private void checkClockSkew(final ServerName serverName, final long serverCurrentTime)<a name="line.358"></a> -<span class="sourceLineNo">359</span> throws ClockOutOfSyncException {<a name="line.359"></a> -<span class="sourceLineNo">360</span> long skew = Math.abs(System.currentTimeMillis() - serverCurrentTime);<a name="line.360"></a> -<span class="sourceLineNo">361</span> if (skew > maxSkew) {<a name="line.361"></a> -<span class="sourceLineNo">362</span> String message = "Server " + serverName + " has been " +<a name="line.362"></a> -<span class="sourceLineNo">363</span> "rejected; Reported time is too far out of sync with master. " +<a name="line.363"></a> -<span class="sourceLineNo">364</span> "Time difference of " + skew + "ms > max allowed of " + maxSkew + "ms";<a name="line.364"></a> -<span class="sourceLineNo">365</span> LOG.warn(message);<a name="line.365"></a> -<span class="sourceLineNo">366</span> throw new ClockOutOfSyncException(message);<a name="line.366"></a> -<span class="sourceLineNo">367</span> } else if (skew > warningSkew){<a name="line.367"></a> -<span class="sourceLineNo">368</span> String message = "Reported time for server " + serverName + " is out of sync with master " +<a name="line.368"></a> -<span class="sourceLineNo">369</span> "by " + skew + "ms. (Warning threshold is " + warningSkew + "ms; " +<a name="line.369"></a> -<span class="sourceLineNo">370</span> "error threshold is " + maxSkew + "ms)";<a name="line.370"></a> -<span class="sourceLineNo">371</span> LOG.warn(message);<a name="line.371"></a> -<span class="sourceLineNo">372</span> }<a name="line.372"></a> -<span class="sourceLineNo">373</span> }<a name="line.373"></a> -<span class="sourceLineNo">374</span><a name="line.374"></a> -<span class="sourceLineNo">375</span> /**<a name="line.375"></a> -<span class="sourceLineNo">376</span> * If this server is on the dead list, reject it with a YouAreDeadException.<a name="line.376"></a> -<span class="sourceLineNo">377</span> * If it was dead but came back with a new start code, remove the old entry<a name="line.377"></a> -<span class="sourceLineNo">378</span> * from the dead list.<a name="line.378"></a> -<span class="sourceLineNo">379</span> * @param serverName<a name="line.379"></a> -<span class="sourceLineNo">380</span> * @param what START or REPORT<a name="line.380"></a> -<span class="sourceLineNo">381</span> * @throws org.apache.hadoop.hbase.YouAreDeadException<a name="line.381"></a> -<span class="sourceLineNo">382</span> */<a name="line.382"></a> -<span class="sourceLineNo">383</span> private void checkIsDead(final ServerName serverName, final String what)<a name="line.383"></a> -<span class="sourceLineNo">384</span> throws YouAreDeadException {<a name="line.384"></a> -<span class="sourceLineNo">385</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.385"></a> -<span class="sourceLineNo">386</span> // host name, port and start code all match with existing one of the<a name="line.386"></a> -<span class="sourceLineNo">387</span> // dead servers. So, this server must be dead.<a name="line.387"></a> -<span class="sourceLineNo">388</span> String message = "Server " + what + " rejected; currently processing " +<a name="line.388"></a> -<span class="sourceLineNo">389</span> serverName + " as dead server";<a name="line.389"></a> -<span class="sourceLineNo">390</span> LOG.debug(message);<a name="line.390"></a> -<span class="sourceLineNo">391</span> throw new YouAreDeadException(message);<a name="line.391"></a> -<span class="sourceLineNo">392</span> }<a name="line.392"></a> -<span class="sourceLineNo">393</span> // remove dead server with same hostname and port of newly checking in rs after master<a name="line.393"></a> -<span class="sourceLineNo">394</span> // initialization.See HBASE-5916 for more information.<a name="line.394"></a> -<span class="sourceLineNo">395</span> if ((this.master == null || this.master.isInitialized())<a name="line.395"></a> -<span class="sourceLineNo">396</span> && this.deadservers.cleanPreviousInstance(serverName)) {<a name="line.396"></a> -<span class="sourceLineNo">397</span> // This server has now become alive after we marked it as dead.<a name="line.397"></a> -<span class="sourceLineNo">398</span> // We removed it's previous entry from the dead list to reflect it.<a name="line.398"></a> -<span class="sourceLineNo">399</span> LOG.debug(what + ":" + " Server " + serverName + " came back up," +<a name="line.399"></a> -<span class="sourceLineNo">400</span> " removed it from the dead servers list");<a name="line.400"></a> -<span class="sourceLineNo">401</span> }<a name="line.401"></a> -<span class="sourceLineNo">402</span> }<a name="line.402"></a> -<span class="sourceLineNo">403</span><a name="line.403"></a> -<span class="sourceLineNo">404</span> /**<a name="line.404"></a> -<span class="sourceLineNo">405</span> * Assumes onlineServers is locked.<a name="line.405"></a> -<span class="sourceLineNo">406</span> * @return ServerName with matching hostname and port.<a name="line.406"></a> -<span class="sourceLineNo">407</span> */<a name="line.407"></a> -<span class="sourceLineNo">408</span> private ServerName findServerWithSameHostnamePortWithLock(<a name="line.408"></a> -<span class="sourceLineNo">409</span> final ServerName serverName) {<a name="line.409"></a> -<span class="sourceLineNo">410</span> ServerName end = ServerName.valueOf(serverName.getHostname(), serverName.getPort(),<a name="line.410"></a> -<span class="sourceLineNo">411</span> Long.MAX_VALUE);<a name="line.411"></a> -<span class="sourceLineNo">412</span><a name="line.412"></a> -<span class="sourceLineNo">413</span> ServerName r = onlineServers.lowerKey(end);<a name="line.413"></a> -<span class="sourceLineNo">414</span> if (r != null) {<a name="line.414"></a> -<span class="sourceLineNo">415</span> if (ServerName.isSameAddress(r, serverName)) {<a name="line.415"></a> -<span class="sourceLineNo">416</span> return r;<a name="line.416"></a> -<span class="sourceLineNo">417</span> }<a name="line.417"></a> -<span class="sourceLineNo">418</span> }<a name="line.418"></a> -<span class="sourceLineNo">419</span> return null;<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> /**<a name="line.422"></a> -<span class="sourceLineNo">423</span> * Adds the onlineServers list. onlineServers should be locked.<a name="line.423"></a> -<span class="sourceLineNo">424</span> * @param serverName The remote servers name.<a name="line.424"></a> -<span class="sourceLineNo">425</span> */<a name="line.425"></a> -<span class="sourceLineNo">426</span> @VisibleForTesting<a name="line.426"></a> -<span class="sourceLineNo">427</span> void recordNewServerWithLock(final ServerName serverName, final ServerMetrics sl) {<a name="line.427"></a> -<span class="sourceLineNo">428</span> LOG.info("Registering regionserver=" + serverName);<a name="line.428"></a> -<span class="sourceLineNo">429</span> this.onlineServers.put(serverName, sl);<a name="line.429"></a> -<span class="sourceLineNo">430</span> this.rsAdmins.remove(serverName);<a name="line.430"></a> -<span class="sourceLineNo">431</span> }<a name="line.431"></a> -<span class="sourceLineNo">432</span><a name="line.432"></a> -<span class="sourceLineNo">433</span> public RegionStoreSequenceIds getLastFlushedSequenceId(byte[] encodedRegionName) {<a name="line.433"></a> -<span class="sourceLineNo">434</span> RegionStoreSequenceIds.Builder builder = RegionStoreSequenceIds.newBuilder();<a name="line.434"></a> -<span class="sourceLineNo">435</span> Long seqId = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.435"></a> -<span class="sourceLineNo">436</span> builder.setLastFlushedSequenceId(seqId != null ? seqId.longValue() : HConstants.NO_SEQNUM);<a name="line.436"></a> -<span class="sourceLineNo">437</span> Map<byte[], Long> storeFlushedSequenceId =<a name="line.437"></a> -<span class="sourceLineNo">438</span> storeFlushedSequenceIdsByRegion.get(encodedRegionName);<a name="line.438"></a> -<span class="sourceLineNo">439</span> if (storeFlushedSequenceId != null) {<a name="line.439"></a> -<span class="sourceLineNo">440</span> for (Map.Entry<byte[], Long> entry : storeFlushedSequenceId.entrySet()) {<a name="line.440"></a> -<span class="sourceLineNo">441</span> builder.addStoreSequenceId(StoreSequenceId.newBuilder()<a name="line.441"></a> -<span class="sourceLineNo">442</span> .setFamilyName(UnsafeByteOperations.unsafeWrap(entry.getKey()))<a name="line.442"></a> -<span class="sourceLineNo">443</span> .setSequenceId(entry.getValue().longValue()).build());<a name="line.443"></a> -<span class="sourceLineNo">444</span> }<a name="line.444"></a> -<span class="sourceLineNo">445</span> }<a name="line.445"></a> -<span class="sourceLineNo">446</span> return builder.build();<a name="line.446"></a> -<span class="sourceLineNo">447</span> }<a name="line.447"></a> -<span class="sourceLineNo">448</span><a name="line.448"></a> -<span class="sourceLineNo">449</span> /**<a name="line.449"></a> -<span class="sourceLineNo">450</span> * @param serverName<a name="line.450"></a> -<span class="sourceLineNo">451</span> * @return ServerMetrics if serverName is known else null<a name="line.451"></a> -<span class="sourceLineNo">452</span> */<a name="line.452"></a> -<span class="sourceLineNo">453</span> public ServerMetrics getLoad(final ServerName serverName) {<a name="line.453"></a> -<span class="sourceLineNo">454</span> return this.onlineServers.get(serverName);<a name="line.454"></a> -<span class="sourceLineNo">455</span> }<a name="line.455"></a> -<span class="sourceLineNo">456</span><a name="line.456"></a> -<span class="sourceLineNo">457</span> /**<a name="line.457"></a> -<span class="sourceLineNo">458</span> * Compute the average load across all region servers.<a name="line.458"></a> -<span class="sourceLineNo">459</span> * Currently, this uses a very naive computation - just uses the number of<a name="line.459"></a> -<span class="sourceLineNo">460</span> * regions being served, ignoring stats about number of requests.<a name="line.460"></a> -<span class="sourceLineNo">461</span> * @return the average load<a name="line.461"></a> -<span class="sourceLineNo">462</span> */<a name="line.462"></a> -<span class="sourceLineNo">463</span> public double getAverageLoad() {<a name="line.463"></a> -<span class="sourceLineNo">464</span> int totalLoad = 0;<a name="line.464"></a> -<span class="sourceLineNo">465</span> int numServers = 0;<a name="line.465"></a> -<span class="sourceLineNo">466</span> for (ServerMetrics sl : this.onlineServers.values()) {<a name="line.466"></a> -<span class="sourceLineNo">467</span> numServers++;<a name="line.467"></a> -<span class="sourceLineNo">468</span> totalLoad += sl.getRegionMetrics().size();<a name="line.468"></a> -<span class="sourceLineNo">469</span> }<a name="line.469"></a> -<span class="sourceLineNo">470</span> return numServers == 0 ? 0 :<a name="line.470"></a> -<span class="sourceLineNo">471</span> (double)totalLoad / (double)numServers;<a name="line.471"></a> -<span class="sourceLineNo">472</span> }<a name="line.472"></a> -<span class="sourceLineNo">473</span><a name="line.473"></a> -<span class="sourceLineNo">474</span> /** @return the count of active regionservers */<a name="line.474"></a> -<span class="sourceLineNo">475</span> public int countOfRegionServers() {<a name="line.475"></a> -<span class="sourceLineNo">476</span> // Presumes onlineServers is a concurrent map<a name="line.476"></a> -<span class="sourceLineNo">477</span> return this.onlineServers.size();<a name="line.477"></a> -<span class="sourceLineNo">478</span> }<a name="line.478"></a> -<span class="sourceLineNo">479</span><a name="line.479"></a> -<span class="sourceLineNo">480</span> /**<a name="line.480"></a> -<span class="sourceLineNo">481</span> * @return Read-only map of servers to serverinfo<a name="line.481"></a> -<span class="sourceLineNo">482</span> */<a name="line.482"></a> -<span class="sourceLineNo">483</span> public Map<ServerName, ServerMetrics> getOnlineServers() {<a name="line.483"></a> -<span class="sourceLineNo">484</span> // Presumption is that iterating the returned Map is OK.<a name="line.484"></a> -<span class="sourceLineNo">485</span> synchronized (this.onlineServers) {<a name="line.485"></a> -<span class="sourceLineNo">486</span> return Collections.unmodifiableMap(this.onlineServers);<a name="line.486"></a> -<span class="sourceLineNo">487</span> }<a name="line.487"></a> -<span class="sourceLineNo">488</span> }<a name="line.488"></a> -<span class="sourceLineNo">489</span><a name="line.489"></a> -<span class="sourceLineNo">490</span> public DeadServer getDeadServers() {<a name="line.490"></a> -<span class="sourceLineNo">491</span> return this.deadservers;<a name="line.491"></a> -<span class="sourceLineNo">492</span> }<a name="line.492"></a> -<span class="sourceLineNo">493</span><a name="line.493"></a> -<span class="sourceLineNo">494</span> /**<a name="line.494"></a> -<span class="sourceLineNo">495</span> * Checks if any dead servers are currently in progress.<a name="line.495"></a> -<span class="sourceLineNo">496</span> * @return true if any RS are being processed as dead, false if not<a name="line.496"></a> -<span class="sourceLineNo">497</span> */<a name="line.497"></a> -<span class="sourceLineNo">498</span> public boolean areDeadServersInProgress() {<a name="line.498"></a> -<span class="sourceLineNo">499</span> return this.deadservers.areDeadServersInProgress();<a name="line.499"></a> -<span class="sourceLineNo">500</span> }<a name="line.500"></a> -<span class="sourceLineNo">501</span><a name="line.501"></a> -<span class="sourceLineNo">502</span> void letRegionServersShutdown() {<a name="line.502"></a> -<span class="sourceLineNo">503</span> long previousLogTime = 0;<a name="line.503"></a> -<span class="sourceLineNo">504</span> ServerName sn = master.getServerName();<a name="line.504"></a> -<span class="sourceLineNo">505</span> ZKWatcher zkw = master.getZooKeeper();<a name="line.505"></a> -<span class="sourceLineNo">506</span> int onlineServersCt;<a name="line.506"></a> -<span class="sourceLineNo">507</span> while ((onlineServersCt = onlineServers.size()) > 0){<a name="line.507"></a> -<span class="sourceLineNo">508</span><a name="line.508"></a> -<span class="sourceLineNo">509</span> if (System.currentTimeMillis() > (previousLogTime + 1000)) {<a name="line.509"></a> -<span class="sourceLineNo">510</span> Set<ServerName> remainingServers = onlineServers.keySet();<a name="line.510"></a> -<span class="sourceLineNo">511</span> synchronized (onlineServers) {<a name="line.511"></a> -<span class="sourceLineNo">512</span> if (remainingServers.size() == 1 && remainingServers.contains(sn)) {<a name="line.512"></a> -<span class="sourceLineNo">513</span> // Master will delete itself later.<a name="line.513"></a> -<span class="sourceLineNo">514</span> return;<a name="line.514"></a> -<span class="sourceLineNo">515</span> }<a name="line.515"></a> -<span class="sourceLineNo">516</span> }<a name="line.516"></a> -<span class="sourceLineNo">517</span> StringBuilder sb = new StringBuilder();<a name="line.517"></a> -<span class="sourceLineNo">518</span> // It's ok here to not sync on onlineServers - merely logging<a name="line.518"></a> -<span class="sourceLineNo">519</span> for (ServerName key : remainingServers) {<a name="line.519"></a> -<span class="sourceLineNo">520</span> if (sb.length() > 0) {<a name="line.520"></a> -<span class="sourceLineNo">521</span> sb.append(", ");<a name="line.521"></a> -<span class="sourceLineNo">522</span> }<a name="line.522"></a> -<span class="sourceLineNo">523</span> sb.append(key);<a name="line.523"></a> -<span class="sourceLineNo">524</span> }<a name="line.524"></a> -<span class="sourceLineNo">525</span> LOG.info("Waiting on regionserver(s) " + sb.toString());<a name="line.525"></a> -<span class="sourceLineNo">526</span> previousLogTime = System.currentTimeMillis();<a name="line.526"></a> -<span class="sourceLineNo">527</span> }<a name="line.527"></a> -<span class="sourceLineNo">528</span><a name="line.528"></a> -<span class="sourceLineNo">529</span> try {<a name="line.529"></a> -<span class="sourceLineNo">530</span> List<String> servers = getRegionServersInZK(zkw);<a name="line.530"></a> -<span class="sourceLineNo">531</span> if (servers == null || servers.isEmpty() || (servers.size() == 1<a name="line.531"></a> -<span class="sourceLineNo">532</span> && servers.contains(sn.toString()))) {<a name="line.532"></a> -<span class="sourceLineNo">533</span> LOG.info("ZK shows there is only the master self online, exiting now");<a name="line.533"></a> -<span class="sourceLineNo">534</span> // Master could have lost some ZK events, no need to wait more.<a name="line.534"></a> -<span class="sourceLineNo">535</span> break;<a name="line.535"></a> -<span class="sourceLineNo">536</span> }<a name="line.536"></a> -<span class="sourceLineNo">537</span> } catch (KeeperException ke) {<a name="line.537"></a> -<span class="sourceLineNo">538</span> LOG.warn("Failed to list regionservers", ke);<a name="line.538"></a> -<span class="sourceLineNo">539</span> // ZK is malfunctioning, don't hang here<a name="line.539"></a> -<span class="sourceLineNo">540</span> break;<a name="line.540"></a> -<span class="sourceLineNo">541</span> }<a name="line.541"></a> -<span class="sourceLineNo">542</span> synchronized (onlineServers) {<a name="line.542"></a> -<span class="sourceLineNo">543</span> try {<a name="line.543"></a> -<span class="sourceLineNo">544</span> if (onlineServersCt == onlineServers.size()) onlineServers.wait(100);<a name="line.544"></a> -<span class="sourceLineNo">545</span> } catch (InterruptedException ignored) {<a name="line.545"></a> -<span class="sourceLineNo">546</span> // continue<a name="line.546"></a> -<span class="sourceLineNo">547</span> }<a name="line.547"></a> -<span class="sourceLineNo">548</span> }<a name="line.548"></a> -<span class="sourceLineNo">549</span> }<a name="line.549"></a> -<span class="sourceLineNo">550</span> }<a name="line.550"></a> -<span class="sourceLineNo">551</span><a name="line.551"></a> -<span class="sourceLineNo">552</span> private List<String> getRegionServersInZK(final ZKWatcher zkw)<a name="line.552"></a> -<span class="sourceLineNo">553</span> throws KeeperException {<a name="line.553"></a> -<span class="sourceLineNo">554</span> return ZKUtil.listChildrenNoWatch(zkw, zkw.getZNodePaths().rsZNode);<a name="line.554"></a> -<span class="sourceLineNo">555</span> }<a name="line.555"></a> -<span class="sourceLineNo">556</span><a name="line.556"></a> -<span class="sourceLineNo">557</span> /*<a name="line.557"></a> -<span class="sourceLineNo">558</span> * Expire the passed server. Add it to list of dead servers and queue a<a name="line.558"></a> -<span class="sourceLineNo">559</span> * shutdown processing.<a name="line.559"></a> -<span class="sourceLineNo">560</span> * @return True if we queued a ServerCrashProcedure else false if we did not (could happen<a name="line.560"></a> -<span class="sourceLineNo">561</span> * for many reasons including the fact that its this server that is going down or we already<a name="line.561"></a> -<span class="sourceLineNo">562</span> * have queued an SCP for this server or SCP processing is currently disabled because we are<a name="line.562"></a> -<span class="sourceLineNo">563</span> * in startup phase).<a name="line.563"></a> -<span class="sourceLineNo">564</span> */<a name="line.564"></a> -<span class="sourceLineNo">565</span> public synchronized boolean expireServer(final ServerName serverName) {<a name="line.565"></a> -<span class="sourceLineNo">566</span> // THIS server is going down... can't handle our own expiration.<a name="line.566"></a> -<span class="sourceLineNo">567</span> if (serverName.equals(master.getServerName())) {<a name="line.567"></a> -<span class="sourceLineNo">568</span> if (!(master.isAborted() || master.isStopped())) {<a name="line.568"></a> -<span class="sourceLineNo">569</span> master.stop("We lost our znode?");<a name="line.569"></a> -<span class="sourceLineNo">570</span> }<a name="line.570"></a> -<span class="sourceLineNo">571</span> return false;<a name="line.571"></a> -<span class="sourceLineNo">572</span> }<a name="line.572"></a> -<span class="sourceLineNo">573</span> // No SCP handling during startup.<a name="line.573"></a> -<span class="sourceLineNo">574</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.574"></a> -<span class="sourceLineNo">575</span> LOG.info("Master doesn't enable ServerShutdownHandler during initialization, "<a name="line.575"></a> -<span class="sourceLineNo">576</span> + "delay expiring server " + serverName);<a name="line.576"></a> -<span class="sourceLineNo">577</span> // Even though we delay expire of this server, we still need to handle Meta's RIT<a name="line.577"></a> -<span class="sourceLineNo">578</span> // that are against the crashed server; since when we do RecoverMetaProcedure,<a name="line.578"></a> -<span class="sourceLineNo">579</span> // the SCP is not enabled yet and Meta's RIT may be suspend forever. See HBase-19287<a name="line.579"></a> -<span class="sourceLineNo">580</span> master.getAssignmentManager().handleMetaRITOnCrashedServer(serverName);<a name="line.580"></a> -<span class="sourceLineNo">581</span> this.queuedDeadServers.add(serverName);<a name="line.581"></a> -<span class="sourceLineNo">582</span> // Return true because though on SCP queued, there will be one queued later.<a name="line.582"></a> -<span class="sourceLineNo">583</span> return true;<a name="line.583"></a> -<span class="sourceLineNo">584</span> }<a name="line.584"></a> -<span class="sourceLineNo">585</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.585"></a> -<span class="sourceLineNo">586</span> LOG.warn("Expiration called on {} but crash processing already in progress", serverName);<a name="line.586"></a> -<span class="sourceLineNo">587</span> return false;<a name="line.587"></a> -<span class="sourceLineNo">588</span> }<a name="line.588"></a> -<span class="sourceLineNo">589</span> moveFromOnlineToDeadServers(serverName);<a name="line.589"></a> -<span class="sourceLineNo">590</span><a name="line.590"></a> -<span class="sourceLineNo">591</span> // If cluster is going down, yes, servers are going to be expiring; don't<a name="line.591"></a> -<span class="sourceLineNo">592</span> // process as a dead server<a name="line.592"></a> -<span class="sourceLineNo">593</span> if (isClusterShutdown()) {<a name="line.593"></a> -<span class="sourceLineNo">594</span> LOG.info("Cluster shutdown set; " + serverName +<a name="line.594"></a> -<span class="sourceLineNo">595</span> " expired; onlineServers=" + this.onlineServers.size());<a name="line.595"></a> -<span class="sourceLineNo">596</span> if (this.onlineServers.isEmpty()) {<a name="line.596"></a> -<span class="sourceLineNo">597</span> master.stop("Cluster shutdown set; onlineServer=0");<a name="line.597"></a> -<span class="sourceLineNo">598</span> }<a name="line.598"></a> -<span class="sourceLineNo">599</span> return false;<a name="line.599"></a> -<span class="sourceLineNo">600</span> }<a name="line.600"></a> -<span class="sourceLineNo">601</span> LOG.info("Processing expiration of " + serverName + " on " + this.master.getServerName());<a name="line.601"></a> -<span class="sourceLineNo">602</span> master.getAssignmentManager().submitServerCrash(serverName, true);<a name="line.602"></a> -<span class="sourceLineNo">603</span><a name="line.603"></a> -<span class="sourceLineNo">604</span> // Tell our listeners that a server was removed<a name="line.604"></a> -<span class="sourceLineNo">605</span> if (!this.listeners.isEmpty()) {<a name="line.605"></a> -<span class="sourceLineNo">606</span> for (ServerListener listener : this.listeners) {<a name="line.606"></a> -<span class="sourceLineNo">607</span> listener.serverRemoved(serverName);<a name="line.607"></a> -<span class="sourceLineNo">608</span> }<a name="line.608"></a> -<span class="sourceLineNo">609</span> }<a name="line.609"></a> -<span class="sourceLineNo">610</span> return true;<a name="line.610"></a> -<span class="sourceLineNo">611</span> }<a name="line.611"></a> -<span class="sourceLineNo">612</span><a name="line.612"></a> -<span class="sourceLineNo">613</span> @VisibleForTesting<a name="line.613"></a> -<span class="sourceLineNo">614</span> public void moveFromOnlineToDeadServers(final ServerName sn) {<a name="line.614"></a> -<span class="sourceLineNo">615</span> synchronized (onlineServers) {<a name="line.615"></a> -<span class="sourceLineNo">616</span> if (!this.onlineServers.containsKey(sn)) {<a name="line.616"></a> -<span class="sourceLineNo">617</span> LOG.warn("Expiration of " + sn + " but server not online");<a name="line.617"></a> -<span class="sourceLineNo">618</span> }<a name="line.618"></a> -<span class="sourceLineNo">619</span> // Remove the server from the known servers lists and update load info BUT<a name="line.619"></a> -<span class="sourceLineNo">620</span> // add to deadservers first; do this so it'll show in dead servers list if<a name="line.620"></a> -<span class="sourceLineNo">621</span> // not in online servers list.<a name="line.621"></a> -<span class="sourceLineNo">622</span> this.deadservers.add(sn);<a name="line.622"></a> -<span class="sourceLineNo">623</span> this.onlineServers.remove(sn);<a name="line.623"></a> -<span class="sourceLineNo">624</span> onlineServers.notifyAll();<a name="line.624"></a> -<span class="sourceLineNo">625</span> }<a name="line.625"></a> -<span class="sourceLineNo">626</span> this.rsAdmins.remove(sn);<a name="line.626"></a> -<span class="sourceLineNo">627</span> }<a name="line.627"></a> -<span class="sourceLineNo">628</span><a name="line.628"></a> -<span class="sourceLineNo">629</span> public synchronized void processDeadServer(final ServerName serverName, boolean shouldSplitWal) {<a name="line.629"></a> -<span class="sourceLineNo">630</span> // When assignment manager is cleaning up the zookeeper nodes and rebuilding the<a name="line.630"></a> -<span class="sourceLineNo">631</span> // in-memory region states, region servers could be down. Meta table can and<a name="line.631"></a> -<span class="sourceLineNo">632</span> // should be re-assigned, log splitting can be done too. However, it is better to<a name="line.632"></a> -<span class="sourceLineNo">633</span> // wait till the cleanup is done before re-assigning user regions.<a name="line.633"></a> -<span class="sourceLineNo">634</span> //<a name="line.634"></a> -<span class="sourceLineNo">635</span> // We should not wait in the server shutdown handler thread since it can clog<a name="line.635"></a> -<span class="sourceLineNo">636</span> // the handler threads and meta table could not be re-assigned in case<a name="line.636"></a> -<span class="sourceLineNo">637</span> // the corresponding server is down. So we queue them up here instead.<a name="line.637"></a> -<span class="sourceLineNo">638</span> if (!master.getAssignmentManager().isFailoverCleanupDone()) {<a name="line.638"></a> -<span class="sourceLineNo">639</span> requeuedDeadServers.put(serverName, shouldSplitWal);<a name="line.639"></a> -<span class="sourceLineNo">640</span> return;<a name="line.640"></a> -<span class="sourceLineNo">641</span> }<a name="line.641"></a> -<span class="sourceLineNo">642</span><a name="line.642"></a> -<span class="sourceLineNo">643</span> this.deadservers.add(serverName);<a name="line.643"></a> -<span class="sourceLineNo">644</span> master.getAssignmentManager().submitServerCrash(serverName, shouldSplitWal);<a name="line.644"></a> -<span class="sourceLineNo">645</span> }<a name="line.645"></a> -<span class="sourceLineNo">646</span><a name="line.646"></a> -<span class="sourceLineNo">647</span> /**<a name="line.647"></a> -<span class="sourceLineNo">648</span> * Process the servers which died during master's initialization. It will be<a name="line.648"></a> -<span class="sourceLineNo">649</span> * called after HMaster#assignMeta and AssignmentManager#joinCluster.<a name="line.649"></a> -<span class="sourceLineNo">650</span> * */<a name="line.650"></a> -<span class="sourceLineNo">651</span> synchronized void processQueuedDeadServers() {<a name="line.651"></a> -<span class="sourceLineNo">652</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.652"></a> -<span class="sourceLineNo">653</span> LOG.info("Master hasn't enabled ServerShutdownHandler");<a name="line.653"></a> -<span class="sourceLineNo">654</span> }<a name="line.654"></a> -<span class="sourceLineNo">655</span> Iterator<ServerName> serverIterator = queuedDeadServers.iterator();<a name="line.655"></a> -<span class="sourceLineNo">656</span> while (serverIterator.hasNext()) {<a name="line.656"></a> -<span class="sourceLineNo">657</span> ServerName tmpServerName = serverIterator.next();<a name="line.657"></a> -<span class="sourceLineNo">658</span> expireServer(tmpServerName);<a name="line.658"></a> -<span class="sourceLineNo">659</span> serverIterator.remove();<a name="line.659"></a> -<span class="sourceLineNo">660</span> requeuedDeadServers.remove(tmpServerName);<a name="line.660"></a> -<span class="sourceLineNo">661</span> }<a name="line.661"></a> -<span class="sourceLineNo">662</span><a name="line.662"></a> -<span class="sourceLineNo">663</span> if (!master.getAssignmentManager().isFailoverCleanupDone()) {<a name="line.663"></a> -<span class="sourceLineNo">664</span> if (LOG.isTraceEnabled()) {<a name="line.664"></a> -<span class="sourceLineNo">665</span> LOG.trace("AssignmentManager failover cleanup not done.");<a name="line.665"></a> -<span class="sourceLineNo">666</span> }<a name="line.666"></a> -<span class="sourceLineNo">667</span> }<a name="line.667"></a> -<span class="sourceLineNo">668</span><a name="line.668"></a> -<span class="sourceLineNo">669</span> for (Map.Entry<ServerName, Boolean> entry : requeuedDeadServers.entrySet()) {<a name="line.669"></a> -<span class="sourceLineNo">670</span> processDeadServer(entry.getKey(), entry.getValue());<a name="line.670"></a> -<span class="sourceLineNo">671</span> }<a name="line.671"></a> -<span class="sourceLineNo">672</span> requeuedDeadServers.clear();<a name="line.672"></a> -<span class="sourceLineNo">673</span> }<a name="line.673"></a> -<span class="sourceLineNo">674</span><a name="line.674"></a> -<span class="sourceLineNo">675</span> /*<a name="line.675"></a> -<span class="sourceLineNo">676</span> * Remove the server from the drain list.<a name="line.676"></a> -<span class="sourceLineNo">677</span> */<a name="line.677"></a> -<span class="sourceLineNo">678</span> public synchronized boolean removeServerFromDrainList(final ServerName sn) {<a name="line.678"></a> -<span class="sourceLineNo">679</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.679"></a> -<span class="sourceLineNo">680</span> // <hostname> , <port> , <startcode><a name="line.680"></a> -<span class="sourceLineNo">681</span><a name="line.681"></a> -<span class="sourceLineNo">682</span> if (!this.isServerOnline(sn)) {<a name="line.682"></a> -<span class="sourceLineNo">683</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.683"></a> -<span class="sourceLineNo">684</span> "Removing from draining list anyway, as requested.");<a name="line.684"></a> -<span class="sourceLineNo">685</span> }<a name="line.685"></a> -<span class="sourceLineNo">686</span> // Remove the server from the draining servers lists.<a name="line.686"></a> -<span class="sourceLineNo">687</span> return this.drainingServers.remove(sn);<a name="line.687"></a> -<span class="sourceLineNo">688</span> }<a name="line.688"></a> -<span class="sourceLineNo">689</span><a name="line.689"></a> -<span class="sourceLineNo">690</span> /**<a name="line.690"></a> -<span class="sourceLineNo">691</span> * Add the server to the drain list.<a name="line.691"></a> -<span class="sourceLineNo">692</span> * @param sn<a name="line.692"></a> -<span class="sourceLineNo">693</span> * @return True if the server is added or the server is already on the drain list.<a name="line.693"></a> -<span class="sourceLineNo">694</span> */<a name="line.694"></a> -<span class="sourceLineNo">695</span> public synchronized boolean addServerToDrainList(final ServerName sn) {<a name="line.695"></a> -<span class="sourceLineNo">696</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.696"></a> -<span class="sourceLineNo">697</span> // <hostname> , <port> , <startcode><a name="line.697"></a> -<span class="sourceLineNo">698</span><a name="line.698"></a> -<span class="sourceLineNo">699</span> if (!this.isServerOnline(sn)) {<a name="line.699"></a> -<span class="sourceLineNo">700</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.700"></a> -<span class="sourceLineNo">701</span> "Ignoring request to add it to draining list.");<a name="line.701"></a> -<span class="sourceLineNo">702</span> return false;<a name="line.702"></a> -<span class="sourceLineNo">703</span> }<a name="line.703"></a> -<span class="sourceLineNo">704</span> // Add the server to the draining servers lists, if it's not already in<a name="line.704"></a> -<span class="sourceLineNo">705</span> // it.<a name="line.705"></a> -<span class="sourceLineNo">706</span> if (this.drainingServers.contains(sn)) {<a name="line.706"></a> -<span class="sourceLineNo">707</span> LOG.warn("Server " + sn + " is already in the draining server list." +<a name="line.707"></a> -<span class="sourceLineNo">708</span> "Ignoring request to add it again.");<a name="line.708"></a> -<span class="sourceLineNo">709</span> return true;<a name="line.709"></a> -<span class="sourceLineNo">710</span> }<a name="line.710"></a> -<span class="sourceLineNo">711</span> LOG.info("Server " + sn + " added to draining server list.");<a name="line.711"></a> -<span class="sourceLineNo">712</span> return this.drainingServers.add(sn);<a name="line.712"></a> -<span class="sourceLineNo">713</span> }<a name="line.713"></a> -<span class="sourceLineNo">714</span><a name="line.714"></a> -<span class="sourceLineNo">715</span> // RPC methods to region servers<a name="line.715"></a> +<span class="sourceLineNo">225</span> * @param versionNumber the version of the new regionserver<a name="line.225"></a> +<span class="sourceLineNo">226</span> * @param ia the InetAddress from which request is received<a name="line.226"></a> +<span class="sourceLineNo">227</span> * @return The ServerName we know this server as.<a name="line.227"></a> +<span class="sourceLineNo">228</span> * @throws IOException<a name="line.228"></a> +<span class="sourceLineNo">229</span> */<a name="line.229"></a> +<span class="sourceLineNo">230</span> ServerName regionServerStartup(RegionServerStartupRequest request, int versionNumber,<a name="line.230"></a> +<span class="sourceLineNo">231</span> InetAddress ia)<a name="line.231"></a> +<span class="sourceLineNo">232</span> throws IOException {<a name="line.232"></a> +<span class="sourceLineNo">233</span> // Test for case where we get a region startup message from a regionserver<a name="line.233"></a> +<span class="sourceLineNo">234</span> // that has been quickly restarted but whose znode expiration handler has<a name="line.234"></a> +<span class="sourceLineNo">235</span> // not yet run, or from a server whose fail we are currently processing.<a name="line.235"></a> +<span class="sourceLineNo">236</span> // Test its host+port combo is present in serverAddressToServerInfo. If it<a name="line.236"></a> +<span class="sourceLineNo">237</span> // is, reject the server and trigger its expiration. The next time it comes<a name="line.237"></a> +<span class="sourceLineNo">238</span> // in, it should have been removed from serverAddressToServerInfo and queued<a name="line.238"></a> +<span class="sourceLineNo">239</span> // for processing by ProcessServerShutdown.<a name="line.239"></a> +<span class="sourceLineNo">240</span><a name="line.240"></a> +<span class="sourceLineNo">241</span> final String hostname = request.hasUseThisHostnameInstead() ?<a name="line.241"></a> +<span class="sourceLineNo">242</span> request.getUseThisHostnameInstead() :ia.getHostName();<a name="line.242"></a> +<span class="sourceLineNo">243</span> ServerName sn = ServerName.valueOf(hostname, request.getPort(),<a name="line.243"></a> +<span class="sourceLineNo">244</span> request.getServerStartCode());<a name="line.244"></a> +<span class="sourceLineNo">245</span> checkClockSkew(sn, request.getServerCurrentTime());<a name="line.245"></a> +<span class="sourceLineNo">246</span> checkIsDead(sn, "STARTUP");<a name="line.246"></a> +<span class="sourceLineNo">247</span> if (!checkAndRecordNewServer(sn, ServerMetricsBuilder.of(sn, versionNumber))) {<a name="line.247"></a> +<span class="sourceLineNo">248</span> LOG.warn("THIS SHOULD NOT HAPPEN, RegionServerStartup"<a name="line.248"></a> +<span class="sourceLineNo">249</span> + " could not record the server: " + sn);<a name="line.249"></a> +<span class="sourceLineNo">250</span> }<a name="line.250"></a> +<span class="sourceLineNo">251</span> return sn;<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> * Updates last flushed sequence Ids for the regions on server sn<a name="line.255"></a> +<span class="sourceLineNo">256</span> * @param sn<a name="line.256"></a> +<span class="sourceLineNo">257</span> * @param hsl<a name="line.257"></a> +<span class="sourceLineNo">258</span> */<a name="line.258"></a> +<span class="sourceLineNo">259</span> private void updateLastFlushedSequenceIds(ServerName sn, ServerMetrics hsl) {<a name="line.259"></a> +<span class="sourceLineNo">260</span> for (Entry<byte[], RegionMetrics> entry : hsl.getRegionMetrics().entrySet()) {<a name="line.260"></a> +<span class="sourceLineNo">261</span> byte[] encodedRegionName = Bytes.toBytes(RegionInfo.encodeRegionName(entry.getKey()));<a name="line.261"></a> +<span class="sourceLineNo">262</span> Long existingValue = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.262"></a> +<span class="sourceLineNo">263</span> long l = entry.getValue().getCompletedSequenceId();<a name="line.263"></a> +<span class="sourceLineNo">264</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.264"></a> +<span class="sourceLineNo">265</span> if (LOG.isTraceEnabled()) {<a name="line.265"></a> +<span class="sourceLineNo">266</span> LOG.trace(Bytes.toString(encodedRegionName) + ", existingValue=" + existingValue +<a name="line.266"></a> +<span class="sourceLineNo">267</span> ", completeSequenceId=" + l);<a name="line.267"></a> +<span class="sourceLineNo">268</span> }<a name="line.268"></a> +<span class="sourceLineNo">269</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue)) {<a name="line.269"></a> +<span class="sourceLineNo">270</span> flushedSequenceIdByRegion.put(encodedRegionName, l);<a name="line.270"></a> +<span class="sourceLineNo">271</span> } else if (l != HConstants.NO_SEQNUM && l < existingValue) {<a name="line.271"></a> +<span class="sourceLineNo">272</span> LOG.warn("RegionServer " + sn + " indicates a last flushed sequence id ("<a name="line.272"></a> +<span class="sourceLineNo">273</span> + l + ") that is less than the previous last flushed sequence id ("<a name="line.273"></a> +<span class="sourceLineNo">274</span> + existingValue + ") for region " + Bytes.toString(entry.getKey()) + " Ignoring.");<a name="line.274"></a> +<span class="sourceLineNo">275</span> }<a name="line.275"></a> +<span class="sourceLineNo">276</span> ConcurrentNavigableMap<byte[], Long> storeFlushedSequenceId =<a name="line.276"></a> +<span class="sourceLineNo">277</span> computeIfAbsent(storeFlushedSequenceIdsByRegion, encodedRegionName,<a name="line.277"></a> +<span class="sourceLineNo">278</span> () -> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR));<a name="line.278"></a> +<span class="sourceLineNo">279</span> for (Entry<byte[], Long> storeSeqId : entry.getValue().getStoreSequenceId().entrySet()) {<a name="line.279"></a> +<span class="sourceLineNo">280</span> byte[] family = storeSeqId.getKey();<a name="line.280"></a> +<span class="sourceLineNo">281</span> existingValue = storeFlushedSequenceId.get(family);<a name="line.281"></a> +<span class="sourceLineNo">282</span> l = storeSeqId.getValue();<a name="line.282"></a> +<span class="sourceLineNo">283</span> if (LOG.isTraceEnabled()) {<a name="line.283"></a> +<span class="sourceLineNo">284</span> LOG.trace(Bytes.toString(encodedRegionName) + ", family=" + Bytes.toString(family) +<a name="line.284"></a> +<span class="sourceLineNo">285</span> ", existingValue=" + existingValue + ", completeSequenceId=" + l);<a name="line.285"></a> +<span class="sourceLineNo">286</span> }<a name="line.286"></a> +<span class="sourceLineNo">287</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.287"></a> +<span class="sourceLineNo">288</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue.longValue())) {<a name="line.288"></a> +<span class="sourceLineNo">289</span> storeFlushedSequenceId.put(family, l);<a name="line.289"></a> +<span class="sourceLineNo">290</span> }<a name="line.290"></a> +<span class="sourceLineNo">291</span> }<a name="line.291"></a> +<span class="sourceLineNo">292</span> }<a name="line.292"></a> +<span class="sourceLineNo">293</span> }<a name="line.293"></a> +<span class="sourceLineNo">294</span><a name="line.294"></a> +<span class="sourceLineNo">295</span> @VisibleForTesting<a name="line.295"></a> +<span class="sourceLineNo">296</span> public void regionServerReport(ServerName sn,<a name="line.296"></a> +<span class="sourceLineNo">297</span> ServerMetrics sl) throws YouAreDeadException {<a name="line.297"></a> +<span class="sourceLineNo">298</span> checkIsDead(sn, "REPORT");<a name="line.298"></a> +<span class="sourceLineNo">299</span> if (null == this.onlineServers.replace(sn, sl)) {<a name="line.299"></a> +<span class="sourceLineNo">300</span> // Already have this host+port combo and its just different start code?<a name="line.300"></a> +<span class="sourceLineNo">301</span> // Just let the server in. Presume master joining a running cluster.<a name="line.301"></a> +<span class="sourceLineNo">302</span> // recordNewServer is what happens at the end of reportServerStartup.<a name="line.302"></a> +<span class="sourceLineNo">303</span> // The only thing we are skipping is passing back to the regionserver<a name="line.303"></a> +<span class="sourceLineNo">304</span> // the ServerName to use. Here we presume a master has already done<a name="line.304"></a> +<span class="sourceLineNo">305</span> // that so we'll press on with whatever it gave us for ServerName.<a name="line.305"></a> +<span class="sourceLineNo">306</span> if (!checkAndRecordNewServer(sn, sl)) {<a name="line.306"></a> +<span class="sourceLineNo">307</span> LOG.info("RegionServerReport ignored, could not record the server: " + sn);<a name="line.307"></a> +<span class="sourceLineNo">308</span> return; // Not recorded, so no need to move on<a name="line.308"></a> +<span class="sourceLineNo">309</span> }<a name="line.309"></a> +<span class="sourceLineNo">310</span> }<a name="line.310"></a> +<span class="sourceLineNo">311</span> updateLastFlushedSequenceIds(sn, sl);<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> /**<a name="line.314"></a> +<span class="sourceLineNo">315</span> * Check is a server of same host and port already exists,<a name="line.315"></a> +<span class="sourceLineNo">316</span> * if not, or the existed one got a smaller start code, record it.<a name="line.316"></a> +<span class="sourceLineNo">317</span> *<a name="line.317"></a> +<span class="sourceLineNo">318</span> * @param serverName the server to check and record<a name="line.318"></a> +<span class="sourceLineNo">319</span> * @param sl the server load on the server<a name="line.319"></a> +<span class="sourceLineNo">320</span> * @return true if the server is recorded, otherwise, false<a name="line.320"></a> +<span class="sourceLineNo">321</span> */<a name="line.321"></a> +<span class="sourceLineNo">322</span> boolean checkAndRecordNewServer(final ServerName serverName, final ServerMetrics sl) {<a name="line.322"></a> +<span class="sourceLineNo">323</span> ServerName existingServer = null;<a name="line.323"></a> +<span class="sourceLineNo">324</span> synchronized (this.onlineServers) {<a name="line.324"></a> +<span class="sourceLineNo">325</span> existingServer = findServerWithSameHostnamePortWithLock(serverName);<a name="line.325"></a> +<span class="sourceLineNo">326</span> if (existingServer != null && (existingServer.getStartcode() > serverName.getStartcode())) {<a name="line.326"></a> +<span class="sourceLineNo">327</span> LOG.info("Server serverName=" + serverName + " rejected; we already have "<a name="line.327"></a> +<span class="sourceLineNo">328</span> + existingServer.toString() + " registered with same hostname and port");<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> recordNewServerWithLock(serverName, sl);<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> // Tell our listeners that a server was added<a name="line.334"></a> +<span class="sourceLineNo">335</span> if (!this.listeners.isEmpty()) {<a name="line.335"></a> +<span class="sourceLineNo">336</span> for (ServerListener listener : this.listeners) {<a name="line.336"></a> +<span class="sourceLineNo">337</span> listener.serverAdded(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><a name="line.340"></a> +<span class="sourceLineNo">341</span> // Note that we assume that same ts means same server, and don't expire in that case.<a name="line.341"></a> +<span class="sourceLineNo">342</span> // TODO: ts can theoretically collide due to clock shifts, so this is a bit hacky.<a name="line.342"></a> +<span class="sourceLineNo">343</span> if (existingServer != null &&<a name="line.343"></a> +<span class="sourceLineNo">344</span> (existingServer.getStartcode() < serverName.getStartcode())) {<a name="line.344"></a> +<span class="sourceLineNo">345</span> LOG.info("Triggering server recovery; existingServer " +<a name="line.345"></a> +<span class="sourceLineNo">346</span> existingServer + " looks stale, new server:" + serverName);<a name="line.346"></a> +<span class="sourceLineNo">347</span> expireServer(existingServer);<a name="line.347"></a> +<span class="sourceLineNo">348</span> }<a name="line.348"></a> +<span class="sourceLineNo">349</span> return true;<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> * Checks if the clock skew between the server and the master. If the clock skew exceeds the<a name="line.353"></a> +<span class="sourceLineNo">354</span> * configured max, it will throw an exception; if it exceeds the configured warning threshold,<a name="line.354"></a> +<span class="sourceLineNo">355</span> * it will log a warning but start normally.<a name="line.355"></a> +<span class="sourceLineNo">356</span> * @param serverName Incoming servers's name<a name="line.356"></a> +<span class="sourceLineNo">357</span> * @param serverCurrentTime<a name="line.357"></a> +<span class="sourceLineNo">358</span> * @throws ClockOutOfSyncException if the skew exceeds the configured max value<a name="line.358"></a> +<span class="sourceLineNo">359</span> */<a name="line.359"></a> +<span class="sourceLineNo">360</span> private void checkClockSkew(final ServerName serverName, final long serverCurrentTime)<a name="line.360"></a> +<span class="sourceLineNo">361</span> throws ClockOutOfSyncException {<a name="line.361"></a> +<span class="sourceLineNo">362</span> long skew = Math.abs(System.currentTimeMillis() - serverCurrentTime);<a name="line.362"></a> +<span class="sourceLineNo">363</span> if (skew > maxSkew) {<a name="line.363"></a> +<span class="sourceLineNo">364</span> String message = "Server " + serverName + " has been " +<a name="line.364"></a> +<span class="sourceLineNo">365</span> "rejected; Reported time is too far out of sync with master. " +<a name="line.365"></a> +<span class="sourceLineNo">366</span> "Time difference of " + skew + "ms > max allowed of " + maxSkew + "ms";<a name="line.366"></a> +<span class="sourceLineNo">367</span> LOG.warn(message);<a name="line.367"></a> +<span class="sourceLineNo">368</span> throw new ClockOutOfSyncException(message);<a name="line.368"></a> +<span class="sourceLineNo">369</span> } else if (skew > warningSkew){<a name="line.369"></a> +<span class="sourceLineNo">370</span> String message = "Reported time for server " + serverName + " is out of sync with master " +<a name="line.370"></a> +<span class="sourceLineNo">371</span> "by " + skew + "ms. (Warning threshold is " + warningSkew + "ms; " +<a name="line.371"></a> +<span class="sourceLineNo">372</span> "error threshold is " + maxSkew + "ms)";<a name="line.372"></a> +<span class="sourceLineNo">373</span> LOG.warn(message);<a name="line.373"></a> +<span class="sourceLineNo">374</span> }<a name="line.374"></a> +<span class="sourceLineNo">375</span> }<a name="line.375"></a> +<span class="sourceLineNo">376</span><a name="line.376"></a> +<span class="sourceLineNo">377</span> /**<a name="line.377"></a> +<span class="sourceLineNo">378</span> * If this server is on the dead list, reject it with a YouAreDeadException.<a name="line.378"></a> +<span class="sourceLineNo">379</span> * If it was dead but came back with a new start code, remove the old entry<a name="line.379"></a> +<span class="sourceLineNo">380</span> * from the dead list.<a name="line.380"></a> +<span class="sourceLineNo">381</span> * @param serverName<a name="line.381"></a> +<span class="sourceLineNo">382</span> * @param what START or REPORT<a name="line.382"></a> +<span class="sourceLineNo">383</span> * @throws org.apache.hadoop.hbase.YouAreDeadException<a name="line.383"></a> +<span class="sourceLineNo">384</span> */<a name="line.384"></a> +<span class="sourceLineNo">385</span> private void checkIsDead(final ServerName serverName, final String what)<a name="line.385"></a> +<span class="sourceLineNo">386</span> throws YouAreDeadException {<a name="line.386"></a> +<span class="sourceLineNo">387</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.387"></a> +<span class="sourceLineNo">388</span> // host name, port and start code all match with existing one of the<a name="line.388"></a> +<span class="sourceLineNo">389</span> // dead servers. So, this server must be dead.<a name="line.389"></a> +<span class="sourceLineNo">390</span> String message = "Server " + what + " rejected; currently processing " +<a name="line.390"></a> +<span class="sourceLineNo">391</span> serverName + " as dead server";<a name="line.391"></a> +<span class="sourceLineNo">392</span> LOG.debug(message);<a name="line.392"></a> +<span class="sourceLineNo">393</span> throw new YouAreDeadException(message);<a name="line.393"></a> +<span class="sourceLineNo">394</span> }<a name="line.394"></a> +<span class="sourceLineNo">395</span> // remove dead server with same hostname and port of newly checking in rs after master<a name="line.395"></a> +<span class="sourceLineNo">396</span> // initialization.See HBASE-5916 for more information.<a name="line.396"></a> +<span class="sourceLineNo">397</span> if ((this.master == null || this.master.isInitialized())<a name="line.397"></a> +<span class="sourceLineNo">398</span> && this.deadservers.cleanPreviousInstance(serverName)) {<a name="line.398"></a> +<span class="sourceLineNo">399</span> // This server has now become alive after we marked it as dead.<a name="line.399"></a> +<span class="sourceLineNo">400</span> // We removed it's previous entry from the dead list to reflect it.<a name="line.400"></a> +<span class="sourceLineNo">401</span> LOG.debug(what + ":" + " Server " + serverName + " came back up," +<a name="line.401"></a> +<span class="sourceLineNo">402</span> " removed it from the dead servers list");<a name="line.402"></a> +<span class="sourceLineNo">403</span> }<a name="line.403"></a> +<span class="sourceLineNo">404</span> }<a name="line.404"></a> +<span class="sourceLineNo">405</span><a name="line.405"></a> +<span class="sourceLineNo">406</span> /**<a name="line.406"></a> +<span class="sourceLineNo">407</span> * Assumes onlineServers is locked.<a name="line.407"></a> +<span class="sourceLineNo">408</span> * @return ServerName with matching hostname and port.<a name="line.408"></a> +<span class="sourceLineNo">409</span> */<a name="line.409"></a> +<span class="sourceLineNo">410</span> private ServerName findServerWithSameHostnamePortWithLock(<a name="line.410"></a> +<span class="sourceLineNo">411</span> final ServerName serverName) {<a name="line.411"></a> +<span class="sourceLineNo">412</span> ServerName end = ServerName.valueOf(serverName.getHostname(), serverName.getPort(),<a name="line.412"></a> +<span class="sourceLineNo">413</span> Long.MAX_VALUE);<a name="line.413"></a> +<span class="sourceLineNo">414</span><a name="line.414"></a> +<span class="sourceLineNo">415</span> ServerName r = onlineServers.lowerKey(end);<a name="line.415"></a> +<span class="sourceLineNo">416</span> if (r != null) {<a name="line.416"></a> +<span class="sourceLineNo">417</span> if (ServerName.isSameAddress(r, serverName)) {<a name="line.417"></a> +<span class="sourceLineNo">418</span> return r;<a name="line.418"></a> +<span class="sourceLineNo">419</span> }<a name="line.419"></a> +<span class="sourceLineNo">420</span> }<a name="line.420"></a> +<span class="sourceLineNo">421</span> return null;<a name="line.421"></a> +<span class="sourceLineNo">422</span> }<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">425</span> * Adds the onlineServers list. onlineServers should be locked.<a name="line.425"></a> +<span class="sourceLineNo">426</span> * @param serverName The remote servers name.<a name="line.426"></a> +<span class="sourceLineNo">427</span> */<a name="line.427"></a> +<span class="sourceLineNo">428</span> @VisibleForTesting<a name="line.428"></a> +<span class="sourceLineNo">429</span> void recordNewServerWithLock(final ServerName serverName, final ServerMetrics sl) {<a name="line.429"></a> +<span class="sourceLineNo">430</span> LOG.info("Registering regionserver=" + serverName);<a name="line.430"></a> +<span class="sourceLineNo">431</span> this.onlineServers.put(serverName, sl);<a name="line.431"></a> +<span class="sourceLineNo">432</span> this.rsAdmins.remove(serverName);<a name="line.432"></a> +<span class="sourceLineNo">433</span> }<a name="line.433"></a> +<span class="sourceLineNo">434</span><a name="line.434"></a> +<span class="sourceLineNo">435</span> public RegionStoreSequenceIds getLastFlushedSequenceId(byte[] encodedRegionName) {<a name="line.435"></a> +<span class="sourceLineNo">436</span> RegionStoreSequenceIds.Builder builder = RegionStoreSequenceIds.newBuilder();<a name="line.436"></a> +<span class="sourceLineNo">437</span> Long seqId = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.437"></a> +<span class="sourceLineNo">438</span> builder.setLastFlushedSequenceId(seqId != null ? seqId.longValue() : HConstants.NO_SEQNUM);<a name="line.438"></a> +<span class="sourceLineNo">439</span> Map<byte[], Long> storeFlushedSequenceId =<a name="line.439"></a> +<span class="sourceLineNo">440</span> storeFlushedSequenceIdsByRegion.get(encodedRegionName);<a name="line.440"></a> +<span class="sourceLineNo">441</span> if (storeFlushedSequenceId != null) {<a name="line.441"></a> +<span class="sourceLineNo">442</span> for (Map.Entry<byte[], Long> entry : storeFlushedSequenceId.entrySet()) {<a name="line.442"></a> +<span class="sourceLineNo">443</span> builder.addStoreSequenceId(StoreSequenceId.newBuilder()<a name="line.443"></a> +<span class="sourceLineNo">444</span> .setFamilyName(UnsafeByteOperations.unsafeWrap(entry.getKey()))<a name="line.444"></a> +<span class="sourceLineNo">445</span> .setSequenceId(entry.getValue().longValue()).build());<a name="line.445"></a> +<span class="sourceLineNo">446</span> }<a name="line.446"></a> +<span class="sourceLineNo">447</span> }<a name="line.447"></a> +<span class="sourceLineNo">448</span> return builder.build();<a name="line.448"></a> +<span class="sourceLineNo">449</span> }<a name="line.449"></a> +<span class="sourceLineNo">450</span><a name="line.450"></a> +<span class="sourceLineNo">451</span> /**<a name="line.451"></a> +<span class="sourceLineNo">452</span> * @param serverName<a name="line.452"></a> +<span class="sourceLineNo">453</span> * @return ServerMetrics if serverName is known else null<a name="line.453"></a> +<span class="sourceLineNo">454</span> */<a name="line.454"></a> +<span class="sourceLineNo">455</span> public ServerMetrics getLoad(final ServerName serverName) {<a name="line.455"></a> +<span class="sourceLineNo">456</span> return this.onlineServers.get(serverName);<a name="line.456"></a> +<span class="sourceLineNo">457</span> }<a name="line.457"></a> +<span class="sourceLineNo">458</span><a name="line.458"></a> +<span class="sourceLineNo">459</span> /**<a name="line.459"></a> +<span class="sourceLineNo">460</span> * Compute the average load across all region servers.<a name="line.460"></a> +<span class="sourceLineNo">461</span> * Currently, this uses a very naive computation - just uses the number of<a name="line.461"></a> +<span class="sourceLineNo">462</span> * regions being served, ignoring stats about number of requests.<a name="line.462"></a> +<span class="sourceLineNo">463</span> * @return the average load<a name="line.463"></a> +<span class="sourceLineNo">464</span> */<a name="line.464"></a> +<span class="sourceLineNo">465</span> public double getAverageLoad() {<a name="line.465"></a> +<span class="sourceLineNo">466</span> int totalLoad = 0;<a name="line.466"></a> +<span class="sourceLineNo">467</span> int numServers = 0;<a name="line.467"></a> +<span class="sourceLineNo">468</span> for (ServerMetrics sl : this.onlineServers.values()) {<a name="line.468"></a> +<span class="sourceLineNo">469</span> numServers++;<a name="line.469"></a> +<span class="sourceLineNo">470</span> totalLoad += sl.getRegionMetrics().size();<a name="line.470"></a> +<span class="sourceLineNo">471</span> }<a name="line.471"></a> +<span class="sourceLineNo">472</span> return numServers == 0 ? 0 :<a name="line.472"></a> +<span class="sourceLineNo">473</span> (double)totalLoad / (double)numServers;<a name="line.473"></a> +<span class="sourceLineNo">474</span> }<a name="line.474"></a> +<span class="sourceLineNo">475</span><a name="line.475"></a> +<span class="sourceLineNo">476</span> /** @return the count of active regionservers */<a name="line.476"></a> +<span class="sourceLineNo">477</span> public int countOfRegionServers() {<a name="line.477"></a> +<span class="sourceLineNo">478</span> // Presumes onlineServers is a concurrent map<a name="line.478"></a> +<span class="sourceLineNo">479</span> return this.onlineServers.size();<a name="line.479"></a> +<span class="sourceLineNo">480</span> }<a name="line.480"></a> +<span class="sourceLineNo">481</span><a name="line.481"></a> +<span class="sourceLineNo">482</span> /**<a name="line.482"></a> +<span class="sourceLineNo">483</span> * @return Read-only map of servers to serverinfo<a name="line.483"></a> +<span class="sourceLineNo">484</span> */<a name="line.484"></a> +<span class="sourceLineNo">485</span> public Map<ServerName, ServerMetrics> getOnlineServers() {<a name="line.485"></a> +<span class="sourceLineNo">486</span> // Presumption is that iterating the returned Map is OK.<a name="line.486"></a> +<span class="sourceLineNo">487</span> synchronized (this.onlineServers) {<a name="line.487"></a> +<span class="sourceLineNo">488</span> return Collections.unmodifiableMap(this.onlineServers);<a name="line.488"></a> +<span class="sourceLineNo">489</span> }<a name="line.489"></a> +<span class="sourceLineNo">490</span> }<a name="line.490"></a> +<span class="sourceLineNo">491</span><a name="line.491"></a> +<span class="sourceLineNo">492</span> public DeadServer getDeadServers() {<a name="line.492"></a> +<span class="sourceLineNo">493</span> return this.deadservers;<a name="line.493"></a> +<span class="sourceLineNo">494</span> }<a name="line.494"></a> +<span class="sourceLineNo">495</span><a name="line.495"></a> +<span class="sourceLineNo">496</span> /**<a name="line.496"></a> +<span class="sourceLineNo">497</span> * Checks if any dead servers are currently in progress.<a name="line.497"></a> +<span class="sourceLineNo">498</span> * @return true if any RS are being processed as dead, false if not<a name="line.498"></a> +<span class="sourceLineNo">499</span> */<a name="line.499"></a> +<span class="sourceLineNo">500</span> public boolean areDeadServersInProgress() {<a name="line.500"></a> +<span class="sourceLineNo">501</span> return this.deadservers.areDeadServersInProgress();<a name="line.501"></a> +<span class="sourceLineNo">502</span> }<a name="line.502"></a> +<span class="sourceLineNo">503</span><a name="line.503"></a> +<span class="sourceLineNo">504</span> void letRegionServersShutdown() {<a name="line.504"></a> +<span class="sourceLineNo">505</span> long previousLogTime = 0;<a name="line.505"></a> +<span class="sourceLineNo">506</span> ServerName sn = master.getServerName();<a name="line.506"></a> +<span class="sourceLineNo">507</span> ZKWatcher zkw = master.getZooKeeper();<a name="line.507"></a> +<span class="sourceLineNo">508</span> int onlineServersCt;<a name="line.508"></a> +<span class="sourceLineNo">509</span> while ((onlineServersCt = onlineServers.size()) > 0){<a name="line.509"></a> +<span class="sourceLineNo">510</span><a name="line.510"></a> +<span class="sourceLineNo">511</span> if (System.currentTimeMillis() > (previousLogTime + 1000)) {<a name="line.511"></a> +<span class="sourceLineNo">512</span> Set<ServerName> remainingServers = onlineServers.keySet();<a name="line.512"></a> +<span class="sourceLineNo">513</span> synchronized (onlineServers) {<a name="line.513"></a> +<span class="sourceLineNo">514</span> if (remainingServers.size() == 1 && remainingServers.contains(sn)) {<a name="line.514"></a> +<span class="sourceLineNo">515</span> // Master will delete itself later.<a name="line.515"></a> +<span class="sourceLineNo">516</span> return;<a name="line.516"></a> +<span class="sourceLineNo">517</span> }<a name="line.517"></a> +<span class="sourceLineNo">518</span> }<a name="line.518"></a> +<span class="sourceLineNo">519</span> StringBuilder sb = new StringBuilder();<a name="line.519"></a> +<span class="sourceLineNo">520</span> // It's ok here to not sync on onlineServers - merely logging<a name="line.520"></a> +<span class="sourceLineNo">521</span> for (ServerName key : remainingServers) {<a name="line.521"></a> +<span class="sourceLineNo">522</span> if (sb.length() > 0) {<a name="line.522"></a> +<span class="sourceLineNo">523</span> sb.append(", ");<a name="line.523"></a> +<span class="sourceLineNo">524</span> }<a name="line.524"></a> +<span class="sourceLineNo">525</span> sb.append(key);<a name="line.525"></a> +<span class="sourceLineNo">526</span> }<a name="line.526"></a> +<span class="sourceLineNo">527</span> LOG.info("Waiting on regionserver(s) " + sb.toString());<a name="line.527"></a> +<span class="sourceLineNo">528</span> previousLogTime = System.currentTimeMillis();<a name="line.528"></a> +<span class="sourceLineNo">529</span> }<a name="line.529"></a> +<span class="sourceLineNo">530</span><a name="line.530"></a> +<span class="sourceLineNo">531</span> try {<a name="line.531"></a> +<span class="sourceLineNo">532</span> List<String> servers = getRegionServersInZK(zkw);<a name="line.532"></a> +<span class="sourceLineNo">533</span> if (servers == null || servers.isEmpty() || (servers.size() == 1<a name="line.533"></a> +<span class="sourceLineNo">534</span> && servers.contains(sn.toString()))) {<a name="line.534"></a> +<span class="sourceLineNo">535</span> LOG.info("ZK shows there is only the master self online, exiting now");<a name="line.535"></a> +<span class="sourceLineNo">536</span> // Master could have lost some ZK events, no need to wait more.<a name="line.536"></a> +<span class="sourceLineNo">537</span> break;<a name="line.537"></a> +<span class="sourceLineNo">538</span> }<a name="line.538"></a> +<span class="sourceLineNo">539</span> } catch (KeeperException ke) {<a name="line.539"></a> +<span class="sourceLineNo">540</span> LOG.warn("Failed to list regionservers", ke);<a name="line.540"></a> +<span class="sourceLineNo">541</span> // ZK is malfunctioning, don't hang here<a name="line.541"></a> +<span class="sourceLineNo">542</span> break;<a name="line.542"></a> +<span class="sourceLineNo">543</span> }<a name="line.543"></a> +<span class="sourceLineNo">544</span> synchronized (onlineServers) {<a name="line.544"></a> +<span class="sourceLineNo">545</span> try {<a name="line.545"></a> +<span class="sourceLineNo">546</span> if (onlineServersCt == onlineServers.size()) onlineServers.wait(100);<a name="line.546"></a> +<span class="sourceLineNo">547</span> } catch (InterruptedException ignored) {<a name="line.547"></a> +<span class="sourceLineNo">548</span> // continue<a name="line.548"></a> +<span class="sourceLineNo">549</span> }<a name="line.549"></a> +<span class="sourceLineNo">550</span> }<a name="line.550"></a> +<span class="sourceLineNo">551</span> }<a name="line.551"></a> +<span class="sourceLineNo">552</span> }<a name="line.552"></a> +<span class="sourceLineNo">553</span><a name="line.553"></a> +<span class="sourceLineNo">554</span> private List<String> getRegionServersInZK(final ZKWatcher zkw)<a name="line.554"></a> +<span class="sourceLineNo">555</span> throws KeeperException {<a name="line.555"></a> +<span class="sourceLineNo">556</span> return ZKUtil.listChildrenNoWatch(zkw, zkw.getZNodePaths().rsZNode);<a name="line.556"></a> +<span class="sourceLineNo">557</span> }<a name="line.557"></a> +<span class="sourceLineNo">558</span><a name="line.558"></a> +<span class="sourceLineNo">559</span> /*<a name="line.559"></a> +<span class="sourceLineNo">560</span> * Expire the passed server. Add it to list of dead servers and queue a<a name="line.560"></a> +<span class="sourceLineNo">561</span> * shutdown processing.<a name="line.561"></a> +<span class="sourceLineNo">562</span> * @return True if we queued a ServerCrashProcedure else false if we did not (could happen<a name="line.562"></a> +<span class="sourceLineNo">563</span> * for many reasons including the fact that its this server that is going down or we already<a name="line.563"></a> +<span class="sourceLineNo">564</span> * have queued an SCP for this server or SCP processing is currently disabled because we are<a name="line.564"></a> +<span class="sourceLineNo">565</span> * in startup phase).<a name="line.565"></a> +<span class="sourceLineNo">566</span> */<a name="line.566"></a> +<span class="sourceLineNo">567</span> public synchronized boolean expireServer(final ServerName serverName) {<a name="line.567"></a> +<span class="sourceLineNo">568</span> // THIS server is going down... can't handle our own expiration.<a name="line.568"></a> +<span class="sourceLineNo">569</span> if (serverName.equals(master.getServerName())) {<a name="line.569"></a> +<span class="sourceLineNo">570</span> if (!(master.isAborted() || master.isStopped())) {<a name="line.570"></a> +<span class="sourceLineNo">571</span> master.stop("We lost our znode?");<a name="line.571"></a> +<span class="sourceLineNo">572</span> }<a name="line.572"></a> +<span class="sourceLineNo">573</span> return false;<a name="line.573"></a> +<span class="sourceLineNo">574</span> }<a name="line.574"></a> +<span class="sourceLineNo">575</span> // No SCP handling during startup.<a name="line.575"></a> +<span class="sourceLineNo">576</span> if (!master.isServerCrashProcessingEnabled()) {<a name="line.576"></a> +<span class="sourceLineNo">577</span> LOG.info("Master doesn't enable ServerShutdownHandler during initialization, "<a name="line.577"></a> +<span class="sourceLineNo">578</span> + "delay expiring server " + serverName);<a name="line.578"></a> +<span class="sourceLineNo">579</span> // Even though we delay expire of this server, we still need to handle Meta's RIT<a name="line.579"></a> +<span class="sourceLineNo">580</span> // that are against the crashed server; since when we do RecoverMetaProcedure,<a name="line.580"></a> +<span class="sourceLineNo">581</span> // the SCP is not enabled yet and Meta's RIT may be suspend forever. See HBase-19287<a name="line.581"></a> +<span class="sourceLineNo">582</span> master.getAssignmentManager().handleMetaRITOnCrashedServer(serverName);<a name="line.582"></a> +<span class="sourceLineNo">583</span> this.queuedDeadServers.add(serverName);<a name="line.583"></a> +<span class="sourceLineNo">584</span> // Return true because though on SCP queued, there will be one queued later.<a name="line.584"></a> +<span class="sourceLineNo">585</span> return true;<a name="line.585"></a> +<span class="sourceLineNo">586</span> }<a name="line.586"></a> +<span class="sourceLineNo">587</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.587"></a> +<span class="sourceLineNo">588</span> LOG.warn("Expiration called on {} but crash processing already in progress", serverName);<a name="line.588"></a> +<span class="sourceLineNo">589</span> return false;<a name="line.589"></a> +<span class="sourceLineNo">590</span> }<a name="line.590"></a> +<span class="sourceLineNo">591</span> moveFromOnlineToDeadServers(serverName);<a name="line.591"></a> +<span class="sourceLineNo">592</span><a name="line.592"></a> +<span class="sourceLineNo">593</span> // If cluster is going down, yes, servers are going to be expiring; don't<a name="line.593"></a> +<span class="sourceLineNo">594</span> // pro
<TRUNCATED>
