http://git-wip-us.apache.org/repos/asf/hbase-site/blob/e11cf2cb/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 51fcd6e..6b9e2a8 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.html @@ -54,1029 +54,1026 @@ <span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.ServerMetricsBuilder;<a name="line.46"></a> <span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.ServerName;<a name="line.47"></a> <span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.YouAreDeadException;<a name="line.48"></a> -<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.ZooKeeperConnectionException;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.client.RetriesExhaustedException;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.ipc.HBaseRpcController;<a name="line.53"></a> -<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.ipc.RpcControllerFactory;<a name="line.54"></a> -<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.monitoring.MonitoredTask;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.57"></a> -<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.59"></a> -<span class="sourceLineNo">060</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.60"></a> -<span class="sourceLineNo">061</span>import org.apache.zookeeper.KeeperException;<a name="line.61"></a> -<span class="sourceLineNo">062</span>import org.slf4j.Logger;<a name="line.62"></a> -<span class="sourceLineNo">063</span>import org.slf4j.LoggerFactory;<a name="line.63"></a> -<span class="sourceLineNo">064</span><a name="line.64"></a> -<span class="sourceLineNo">065</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.65"></a> -<span class="sourceLineNo">066</span>import org.apache.hbase.thirdparty.com.google.protobuf.UnsafeByteOperations;<a name="line.66"></a> -<span class="sourceLineNo">067</span><a name="line.67"></a> -<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.68"></a> -<span class="sourceLineNo">069</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService;<a name="line.69"></a> -<span class="sourceLineNo">070</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.RegionStoreSequenceIds;<a name="line.70"></a> -<span class="sourceLineNo">071</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.StoreSequenceId;<a name="line.71"></a> -<span class="sourceLineNo">072</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionServerStartupRequest;<a name="line.72"></a> -<span class="sourceLineNo">073</span><a name="line.73"></a> -<span class="sourceLineNo">074</span>/**<a name="line.74"></a> -<span class="sourceLineNo">075</span> * The ServerManager class manages info about region servers.<a name="line.75"></a> -<span class="sourceLineNo">076</span> * <p><a name="line.76"></a> -<span class="sourceLineNo">077</span> * Maintains lists of online and dead servers. Processes the startups,<a name="line.77"></a> -<span class="sourceLineNo">078</span> * shutdowns, and deaths of region servers.<a name="line.78"></a> -<span class="sourceLineNo">079</span> * <p><a name="line.79"></a> -<span class="sourceLineNo">080</span> * Servers are distinguished in two different ways. A given server has a<a name="line.80"></a> -<span class="sourceLineNo">081</span> * location, specified by hostname and port, and of which there can only be one<a name="line.81"></a> -<span class="sourceLineNo">082</span> * online at any given time. A server instance is specified by the location<a name="line.82"></a> -<span class="sourceLineNo">083</span> * (hostname and port) as well as the startcode (timestamp from when the server<a name="line.83"></a> -<span class="sourceLineNo">084</span> * was started). This is used to differentiate a restarted instance of a given<a name="line.84"></a> -<span class="sourceLineNo">085</span> * server from the original instance.<a name="line.85"></a> -<span class="sourceLineNo">086</span> * <p><a name="line.86"></a> -<span class="sourceLineNo">087</span> * If a sever is known not to be running any more, it is called dead. The dead<a name="line.87"></a> -<span class="sourceLineNo">088</span> * server needs to be handled by a ServerShutdownHandler. If the handler is not<a name="line.88"></a> -<span class="sourceLineNo">089</span> * enabled yet, the server can't be handled right away so it is queued up.<a name="line.89"></a> -<span class="sourceLineNo">090</span> * After the handler is enabled, the server will be submitted to a handler to handle.<a name="line.90"></a> -<span class="sourceLineNo">091</span> * However, the handler may be just partially enabled. If so,<a name="line.91"></a> -<span class="sourceLineNo">092</span> * the server cannot be fully processed, and be queued up for further processing.<a name="line.92"></a> -<span class="sourceLineNo">093</span> * A server is fully processed only after the handler is fully enabled<a name="line.93"></a> -<span class="sourceLineNo">094</span> * and has completed the handling.<a name="line.94"></a> -<span class="sourceLineNo">095</span> */<a name="line.95"></a> -<span class="sourceLineNo">096</span>@InterfaceAudience.Private<a name="line.96"></a> -<span class="sourceLineNo">097</span>public class ServerManager {<a name="line.97"></a> -<span class="sourceLineNo">098</span> public static final String WAIT_ON_REGIONSERVERS_MAXTOSTART =<a name="line.98"></a> -<span class="sourceLineNo">099</span> "hbase.master.wait.on.regionservers.maxtostart";<a name="line.99"></a> -<span class="sourceLineNo">100</span><a name="line.100"></a> -<span class="sourceLineNo">101</span> public static final String WAIT_ON_REGIONSERVERS_MINTOSTART =<a name="line.101"></a> -<span class="sourceLineNo">102</span> "hbase.master.wait.on.regionservers.mintostart";<a name="line.102"></a> -<span class="sourceLineNo">103</span><a name="line.103"></a> -<span class="sourceLineNo">104</span> public static final String WAIT_ON_REGIONSERVERS_TIMEOUT =<a name="line.104"></a> -<span class="sourceLineNo">105</span> "hbase.master.wait.on.regionservers.timeout";<a name="line.105"></a> -<span class="sourceLineNo">106</span><a name="line.106"></a> -<span class="sourceLineNo">107</span> public static final String WAIT_ON_REGIONSERVERS_INTERVAL =<a name="line.107"></a> -<span class="sourceLineNo">108</span> "hbase.master.wait.on.regionservers.interval";<a name="line.108"></a> -<span class="sourceLineNo">109</span><a name="line.109"></a> -<span class="sourceLineNo">110</span> private static final Logger LOG = LoggerFactory.getLogger(ServerManager.class);<a name="line.110"></a> -<span class="sourceLineNo">111</span><a name="line.111"></a> -<span class="sourceLineNo">112</span> // Set if we are to shutdown the cluster.<a name="line.112"></a> -<span class="sourceLineNo">113</span> private AtomicBoolean clusterShutdown = new AtomicBoolean(false);<a name="line.113"></a> -<span class="sourceLineNo">114</span><a name="line.114"></a> -<span class="sourceLineNo">115</span> /**<a name="line.115"></a> -<span class="sourceLineNo">116</span> * The last flushed sequence id for a region.<a name="line.116"></a> -<span class="sourceLineNo">117</span> */<a name="line.117"></a> -<span class="sourceLineNo">118</span> private final ConcurrentNavigableMap<byte[], Long> flushedSequenceIdByRegion =<a name="line.118"></a> -<span class="sourceLineNo">119</span> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.119"></a> -<span class="sourceLineNo">120</span><a name="line.120"></a> -<span class="sourceLineNo">121</span> /**<a name="line.121"></a> -<span class="sourceLineNo">122</span> * The last flushed sequence id for a store in a region.<a name="line.122"></a> -<span class="sourceLineNo">123</span> */<a name="line.123"></a> -<span class="sourceLineNo">124</span> private final ConcurrentNavigableMap<byte[], ConcurrentNavigableMap<byte[], Long>><a name="line.124"></a> -<span class="sourceLineNo">125</span> storeFlushedSequenceIdsByRegion = new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.125"></a> -<span class="sourceLineNo">126</span><a name="line.126"></a> -<span class="sourceLineNo">127</span> /** Map of registered servers to their current load */<a name="line.127"></a> -<span class="sourceLineNo">128</span> private final ConcurrentNavigableMap<ServerName, ServerMetrics> onlineServers =<a name="line.128"></a> -<span class="sourceLineNo">129</span> new ConcurrentSkipListMap<>();<a name="line.129"></a> -<span class="sourceLineNo">130</span><a name="line.130"></a> -<span class="sourceLineNo">131</span> /**<a name="line.131"></a> -<span class="sourceLineNo">132</span> * Map of admin interfaces per registered regionserver; these interfaces we use to control<a name="line.132"></a> -<span class="sourceLineNo">133</span> * regionservers out on the cluster<a name="line.133"></a> -<span class="sourceLineNo">134</span> */<a name="line.134"></a> -<span class="sourceLineNo">135</span> private final Map<ServerName, AdminService.BlockingInterface> rsAdmins = new HashMap<>();<a name="line.135"></a> -<span class="sourceLineNo">136</span><a name="line.136"></a> -<span class="sourceLineNo">137</span> /** List of region servers that should not get any more new regions. */<a name="line.137"></a> -<span class="sourceLineNo">138</span> private final ArrayList<ServerName> drainingServers = new ArrayList<>();<a name="line.138"></a> -<span class="sourceLineNo">139</span><a name="line.139"></a> -<span class="sourceLineNo">140</span> private final MasterServices master;<a name="line.140"></a> -<span class="sourceLineNo">141</span> private final ClusterConnection connection;<a name="line.141"></a> -<span class="sourceLineNo">142</span><a name="line.142"></a> -<span class="sourceLineNo">143</span> private final DeadServer deadservers = new DeadServer();<a name="line.143"></a> -<span class="sourceLineNo">144</span><a name="line.144"></a> -<span class="sourceLineNo">145</span> private final long maxSkew;<a name="line.145"></a> -<span class="sourceLineNo">146</span> private final long warningSkew;<a name="line.146"></a> -<span class="sourceLineNo">147</span><a name="line.147"></a> -<span class="sourceLineNo">148</span> private final RpcControllerFactory rpcControllerFactory;<a name="line.148"></a> -<span class="sourceLineNo">149</span><a name="line.149"></a> -<span class="sourceLineNo">150</span> /**<a name="line.150"></a> -<span class="sourceLineNo">151</span> * Set of region servers which are dead but not processed immediately. If one<a name="line.151"></a> -<span class="sourceLineNo">152</span> * server died before master enables ServerShutdownHandler, the server will be<a name="line.152"></a> -<span class="sourceLineNo">153</span> * added to this set and will be processed through calling<a name="line.153"></a> -<span class="sourceLineNo">154</span> * {@link ServerManager#processQueuedDeadServers()} by master.<a name="line.154"></a> -<span class="sourceLineNo">155</span> * <p><a name="line.155"></a> -<span class="sourceLineNo">156</span> * A dead server is a server instance known to be dead, not listed in the /hbase/rs<a name="line.156"></a> -<span class="sourceLineNo">157</span> * znode any more. It may have not been submitted to ServerShutdownHandler yet<a name="line.157"></a> -<span class="sourceLineNo">158</span> * because the handler is not enabled.<a name="line.158"></a> -<span class="sourceLineNo">159</span> * <p><a name="line.159"></a> -<span class="sourceLineNo">160</span> * A dead server, which has been submitted to ServerShutdownHandler while the<a name="line.160"></a> -<span class="sourceLineNo">161</span> * handler is not enabled, is queued up.<a name="line.161"></a> -<span class="sourceLineNo">162</span> * <p><a name="line.162"></a> -<span class="sourceLineNo">163</span> * So this is a set of region servers known to be dead but not submitted to<a name="line.163"></a> -<span class="sourceLineNo">164</span> * ServerShutdownHandler for processing yet.<a name="line.164"></a> -<span class="sourceLineNo">165</span> */<a name="line.165"></a> -<span class="sourceLineNo">166</span> private Set<ServerName> queuedDeadServers = new HashSet<>();<a name="line.166"></a> -<span class="sourceLineNo">167</span><a name="line.167"></a> -<span class="sourceLineNo">168</span> /**<a name="line.168"></a> -<span class="sourceLineNo">169</span> * Set of region servers which are dead and submitted to ServerShutdownHandler to process but not<a name="line.169"></a> -<span class="sourceLineNo">170</span> * fully processed immediately.<a name="line.170"></a> -<span class="sourceLineNo">171</span> * <p><a name="line.171"></a> -<span class="sourceLineNo">172</span> * If one server died before assignment manager finished the failover cleanup, the server will be<a name="line.172"></a> -<span class="sourceLineNo">173</span> * added to this set and will be processed through calling<a name="line.173"></a> -<span class="sourceLineNo">174</span> * {@link ServerManager#processQueuedDeadServers()} by assignment manager.<a name="line.174"></a> -<span class="sourceLineNo">175</span> * <p><a name="line.175"></a> -<span class="sourceLineNo">176</span> * The Boolean value indicates whether log split is needed inside ServerShutdownHandler<a name="line.176"></a> -<span class="sourceLineNo">177</span> * <p><a name="line.177"></a> -<span class="sourceLineNo">178</span> * ServerShutdownHandler processes a dead server submitted to the handler after the handler is<a name="line.178"></a> -<span class="sourceLineNo">179</span> * enabled. It may not be able to complete the processing because meta is not yet online or master<a name="line.179"></a> -<span class="sourceLineNo">180</span> * is currently in startup mode. In this case, the dead server will be parked in this set<a name="line.180"></a> -<span class="sourceLineNo">181</span> * temporarily.<a name="line.181"></a> -<span class="sourceLineNo">182</span> */<a name="line.182"></a> -<span class="sourceLineNo">183</span> private Map<ServerName, Boolean> requeuedDeadServers = new ConcurrentHashMap<>();<a name="line.183"></a> -<span class="sourceLineNo">184</span><a name="line.184"></a> -<span class="sourceLineNo">185</span> /** Listeners that are called on server events. */<a name="line.185"></a> -<span class="sourceLineNo">186</span> private List<ServerListener> listeners = new CopyOnWriteArrayList<>();<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">189</span> * Constructor.<a name="line.189"></a> -<span class="sourceLineNo">190</span> * @param master<a name="line.190"></a> -<span class="sourceLineNo">191</span> * @throws ZooKeeperConnectionException<a name="line.191"></a> -<span class="sourceLineNo">192</span> */<a name="line.192"></a> -<span class="sourceLineNo">193</span> public ServerManager(final MasterServices master) {<a name="line.193"></a> -<span class="sourceLineNo">194</span> this(master, true);<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> -<span class="sourceLineNo">197</span> ServerManager(final MasterServices master, final boolean connect) {<a name="line.197"></a> -<span class="sourceLineNo">198</span> this.master = master;<a name="line.198"></a> -<span class="sourceLineNo">199</span> Configuration c = master.getConfiguration();<a name="line.199"></a> -<span class="sourceLineNo">200</span> maxSkew = c.getLong("hbase.master.maxclockskew", 30000);<a name="line.200"></a> -<span class="sourceLineNo">201</span> warningSkew = c.getLong("hbase.master.warningclockskew", 10000);<a name="line.201"></a> -<span class="sourceLineNo">202</span> this.connection = connect? master.getClusterConnection(): null;<a name="line.202"></a> -<span class="sourceLineNo">203</span> this.rpcControllerFactory = this.connection == null? null: connection.getRpcControllerFactory();<a name="line.203"></a> -<span class="sourceLineNo">204</span> }<a name="line.204"></a> -<span class="sourceLineNo">205</span><a name="line.205"></a> -<span class="sourceLineNo">206</span> /**<a name="line.206"></a> -<span class="sourceLineNo">207</span> * Add the listener to the notification list.<a name="line.207"></a> -<span class="sourceLineNo">208</span> * @param listener The ServerListener to register<a name="line.208"></a> -<span class="sourceLineNo">209</span> */<a name="line.209"></a> -<span class="sourceLineNo">210</span> public void registerListener(final ServerListener listener) {<a name="line.210"></a> -<span class="sourceLineNo">211</span> this.listeners.add(listener);<a name="line.211"></a> -<span class="sourceLineNo">212</span> }<a name="line.212"></a> -<span class="sourceLineNo">213</span><a name="line.213"></a> -<span class="sourceLineNo">214</span> /**<a name="line.214"></a> -<span class="sourceLineNo">215</span> * Remove the listener from the notification list.<a name="line.215"></a> -<span class="sourceLineNo">216</span> * @param listener The ServerListener to unregister<a name="line.216"></a> -<span class="sourceLineNo">217</span> */<a name="line.217"></a> -<span class="sourceLineNo">218</span> public boolean unregisterListener(final ServerListener listener) {<a name="line.218"></a> -<span class="sourceLineNo">219</span> return this.listeners.remove(listener);<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> /**<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 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">049</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.client.RetriesExhaustedException;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.ipc.HBaseRpcController;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.ipc.RpcControllerFactory;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.monitoring.MonitoredTask;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.zookeeper.KeeperException;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.slf4j.Logger;<a name="line.61"></a> +<span class="sourceLineNo">062</span>import org.slf4j.LoggerFactory;<a name="line.62"></a> +<span class="sourceLineNo">063</span><a name="line.63"></a> +<span class="sourceLineNo">064</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.64"></a> +<span class="sourceLineNo">065</span>import org.apache.hbase.thirdparty.com.google.protobuf.UnsafeByteOperations;<a name="line.65"></a> +<span class="sourceLineNo">066</span><a name="line.66"></a> +<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.67"></a> +<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService;<a name="line.68"></a> +<span class="sourceLineNo">069</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.RegionStoreSequenceIds;<a name="line.69"></a> +<span class="sourceLineNo">070</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.StoreSequenceId;<a name="line.70"></a> +<span class="sourceLineNo">071</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionServerStartupRequest;<a name="line.71"></a> +<span class="sourceLineNo">072</span><a name="line.72"></a> +<span class="sourceLineNo">073</span>/**<a name="line.73"></a> +<span class="sourceLineNo">074</span> * The ServerManager class manages info about region servers.<a name="line.74"></a> +<span class="sourceLineNo">075</span> * <p><a name="line.75"></a> +<span class="sourceLineNo">076</span> * Maintains lists of online and dead servers. Processes the startups,<a name="line.76"></a> +<span class="sourceLineNo">077</span> * shutdowns, and deaths of region servers.<a name="line.77"></a> +<span class="sourceLineNo">078</span> * <p><a name="line.78"></a> +<span class="sourceLineNo">079</span> * Servers are distinguished in two different ways. A given server has a<a name="line.79"></a> +<span class="sourceLineNo">080</span> * location, specified by hostname and port, and of which there can only be one<a name="line.80"></a> +<span class="sourceLineNo">081</span> * online at any given time. A server instance is specified by the location<a name="line.81"></a> +<span class="sourceLineNo">082</span> * (hostname and port) as well as the startcode (timestamp from when the server<a name="line.82"></a> +<span class="sourceLineNo">083</span> * was started). This is used to differentiate a restarted instance of a given<a name="line.83"></a> +<span class="sourceLineNo">084</span> * server from the original instance.<a name="line.84"></a> +<span class="sourceLineNo">085</span> * <p><a name="line.85"></a> +<span class="sourceLineNo">086</span> * If a sever is known not to be running any more, it is called dead. The dead<a name="line.86"></a> +<span class="sourceLineNo">087</span> * server needs to be handled by a ServerShutdownHandler. If the handler is not<a name="line.87"></a> +<span class="sourceLineNo">088</span> * enabled yet, the server can't be handled right away so it is queued up.<a name="line.88"></a> +<span class="sourceLineNo">089</span> * After the handler is enabled, the server will be submitted to a handler to handle.<a name="line.89"></a> +<span class="sourceLineNo">090</span> * However, the handler may be just partially enabled. If so,<a name="line.90"></a> +<span class="sourceLineNo">091</span> * the server cannot be fully processed, and be queued up for further processing.<a name="line.91"></a> +<span class="sourceLineNo">092</span> * A server is fully processed only after the handler is fully enabled<a name="line.92"></a> +<span class="sourceLineNo">093</span> * and has completed the handling.<a name="line.93"></a> +<span class="sourceLineNo">094</span> */<a name="line.94"></a> +<span class="sourceLineNo">095</span>@InterfaceAudience.Private<a name="line.95"></a> +<span class="sourceLineNo">096</span>public class ServerManager {<a name="line.96"></a> +<span class="sourceLineNo">097</span> public static final String WAIT_ON_REGIONSERVERS_MAXTOSTART =<a name="line.97"></a> +<span class="sourceLineNo">098</span> "hbase.master.wait.on.regionservers.maxtostart";<a name="line.98"></a> +<span class="sourceLineNo">099</span><a name="line.99"></a> +<span class="sourceLineNo">100</span> public static final String WAIT_ON_REGIONSERVERS_MINTOSTART =<a name="line.100"></a> +<span class="sourceLineNo">101</span> "hbase.master.wait.on.regionservers.mintostart";<a name="line.101"></a> +<span class="sourceLineNo">102</span><a name="line.102"></a> +<span class="sourceLineNo">103</span> public static final String WAIT_ON_REGIONSERVERS_TIMEOUT =<a name="line.103"></a> +<span class="sourceLineNo">104</span> "hbase.master.wait.on.regionservers.timeout";<a name="line.104"></a> +<span class="sourceLineNo">105</span><a name="line.105"></a> +<span class="sourceLineNo">106</span> public static final String WAIT_ON_REGIONSERVERS_INTERVAL =<a name="line.106"></a> +<span class="sourceLineNo">107</span> "hbase.master.wait.on.regionservers.interval";<a name="line.107"></a> +<span class="sourceLineNo">108</span><a name="line.108"></a> +<span class="sourceLineNo">109</span> private static final Logger LOG = LoggerFactory.getLogger(ServerManager.class);<a name="line.109"></a> +<span class="sourceLineNo">110</span><a name="line.110"></a> +<span class="sourceLineNo">111</span> // Set if we are to shutdown the cluster.<a name="line.111"></a> +<span class="sourceLineNo">112</span> private AtomicBoolean clusterShutdown = new AtomicBoolean(false);<a name="line.112"></a> +<span class="sourceLineNo">113</span><a name="line.113"></a> +<span class="sourceLineNo">114</span> /**<a name="line.114"></a> +<span class="sourceLineNo">115</span> * The last flushed sequence id for a region.<a name="line.115"></a> +<span class="sourceLineNo">116</span> */<a name="line.116"></a> +<span class="sourceLineNo">117</span> private final ConcurrentNavigableMap<byte[], Long> flushedSequenceIdByRegion =<a name="line.117"></a> +<span class="sourceLineNo">118</span> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.118"></a> +<span class="sourceLineNo">119</span><a name="line.119"></a> +<span class="sourceLineNo">120</span> /**<a name="line.120"></a> +<span class="sourceLineNo">121</span> * The last flushed sequence id for a store in a region.<a name="line.121"></a> +<span class="sourceLineNo">122</span> */<a name="line.122"></a> +<span class="sourceLineNo">123</span> private final ConcurrentNavigableMap<byte[], ConcurrentNavigableMap<byte[], Long>><a name="line.123"></a> +<span class="sourceLineNo">124</span> storeFlushedSequenceIdsByRegion = new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.124"></a> +<span class="sourceLineNo">125</span><a name="line.125"></a> +<span class="sourceLineNo">126</span> /** Map of registered servers to their current load */<a name="line.126"></a> +<span class="sourceLineNo">127</span> private final ConcurrentNavigableMap<ServerName, ServerMetrics> onlineServers =<a name="line.127"></a> +<span class="sourceLineNo">128</span> new ConcurrentSkipListMap<>();<a name="line.128"></a> +<span class="sourceLineNo">129</span><a name="line.129"></a> +<span class="sourceLineNo">130</span> /**<a name="line.130"></a> +<span class="sourceLineNo">131</span> * Map of admin interfaces per registered regionserver; these interfaces we use to control<a name="line.131"></a> +<span class="sourceLineNo">132</span> * regionservers out on the cluster<a name="line.132"></a> +<span class="sourceLineNo">133</span> */<a name="line.133"></a> +<span class="sourceLineNo">134</span> private final Map<ServerName, AdminService.BlockingInterface> rsAdmins = new HashMap<>();<a name="line.134"></a> +<span class="sourceLineNo">135</span><a name="line.135"></a> +<span class="sourceLineNo">136</span> /** List of region servers that should not get any more new regions. */<a name="line.136"></a> +<span class="sourceLineNo">137</span> private final ArrayList<ServerName> drainingServers = new ArrayList<>();<a name="line.137"></a> +<span class="sourceLineNo">138</span><a name="line.138"></a> +<span class="sourceLineNo">139</span> private final MasterServices master;<a name="line.139"></a> +<span class="sourceLineNo">140</span> private final ClusterConnection connection;<a name="line.140"></a> +<span class="sourceLineNo">141</span><a name="line.141"></a> +<span class="sourceLineNo">142</span> private final DeadServer deadservers = new DeadServer();<a name="line.142"></a> +<span class="sourceLineNo">143</span><a name="line.143"></a> +<span class="sourceLineNo">144</span> private final long maxSkew;<a name="line.144"></a> +<span class="sourceLineNo">145</span> private final long warningSkew;<a name="line.145"></a> +<span class="sourceLineNo">146</span><a name="line.146"></a> +<span class="sourceLineNo">147</span> private final RpcControllerFactory rpcControllerFactory;<a name="line.147"></a> +<span class="sourceLineNo">148</span><a name="line.148"></a> +<span class="sourceLineNo">149</span> /**<a name="line.149"></a> +<span class="sourceLineNo">150</span> * Set of region servers which are dead but not processed immediately. If one<a name="line.150"></a> +<span class="sourceLineNo">151</span> * server died before master enables ServerShutdownHandler, the server will be<a name="line.151"></a> +<span class="sourceLineNo">152</span> * added to this set and will be processed through calling<a name="line.152"></a> +<span class="sourceLineNo">153</span> * {@link ServerManager#processQueuedDeadServers()} by master.<a name="line.153"></a> +<span class="sourceLineNo">154</span> * <p><a name="line.154"></a> +<span class="sourceLineNo">155</span> * A dead server is a server instance known to be dead, not listed in the /hbase/rs<a name="line.155"></a> +<span class="sourceLineNo">156</span> * znode any more. It may have not been submitted to ServerShutdownHandler yet<a name="line.156"></a> +<span class="sourceLineNo">157</span> * because the handler is not enabled.<a name="line.157"></a> +<span class="sourceLineNo">158</span> * <p><a name="line.158"></a> +<span class="sourceLineNo">159</span> * A dead server, which has been submitted to ServerShutdownHandler while the<a name="line.159"></a> +<span class="sourceLineNo">160</span> * handler is not enabled, is queued up.<a name="line.160"></a> +<span class="sourceLineNo">161</span> * <p><a name="line.161"></a> +<span class="sourceLineNo">162</span> * So this is a set of region servers known to be dead but not submitted to<a name="line.162"></a> +<span class="sourceLineNo">163</span> * ServerShutdownHandler for processing yet.<a name="line.163"></a> +<span class="sourceLineNo">164</span> */<a name="line.164"></a> +<span class="sourceLineNo">165</span> private Set<ServerName> queuedDeadServers = new HashSet<>();<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> * Set of region servers which are dead and submitted to ServerShutdownHandler to process but not<a name="line.168"></a> +<span class="sourceLineNo">169</span> * fully processed immediately.<a name="line.169"></a> +<span class="sourceLineNo">170</span> * <p><a name="line.170"></a> +<span class="sourceLineNo">171</span> * If one server died before assignment manager finished the failover cleanup, the server will be<a name="line.171"></a> +<span class="sourceLineNo">172</span> * added to this set and will be processed through calling<a name="line.172"></a> +<span class="sourceLineNo">173</span> * {@link ServerManager#processQueuedDeadServers()} by assignment manager.<a name="line.173"></a> +<span class="sourceLineNo">174</span> * <p><a name="line.174"></a> +<span class="sourceLineNo">175</span> * The Boolean value indicates whether log split is needed inside ServerShutdownHandler<a name="line.175"></a> +<span class="sourceLineNo">176</span> * <p><a name="line.176"></a> +<span class="sourceLineNo">177</span> * ServerShutdownHandler processes a dead server submitted to the handler after the handler is<a name="line.177"></a> +<span class="sourceLineNo">178</span> * enabled. It may not be able to complete the processing because meta is not yet online or master<a name="line.178"></a> +<span class="sourceLineNo">179</span> * is currently in startup mode. In this case, the dead server will be parked in this set<a name="line.179"></a> +<span class="sourceLineNo">180</span> * temporarily.<a name="line.180"></a> +<span class="sourceLineNo">181</span> */<a name="line.181"></a> +<span class="sourceLineNo">182</span> private Map<ServerName, Boolean> requeuedDeadServers = new ConcurrentHashMap<>();<a name="line.182"></a> +<span class="sourceLineNo">183</span><a name="line.183"></a> +<span class="sourceLineNo">184</span> /** Listeners that are called on server events. */<a name="line.184"></a> +<span class="sourceLineNo">185</span> private List<ServerListener> listeners = new CopyOnWriteArrayList<>();<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> * Constructor.<a name="line.188"></a> +<span class="sourceLineNo">189</span> */<a name="line.189"></a> +<span class="sourceLineNo">190</span> public ServerManager(final MasterServices master) {<a name="line.190"></a> +<span class="sourceLineNo">191</span> this.master = master;<a name="line.191"></a> +<span class="sourceLineNo">192</span> Configuration c = master.getConfiguration();<a name="line.192"></a> +<span class="sourceLineNo">193</span> maxSkew = c.getLong("hbase.master.maxclockskew", 30000);<a name="line.193"></a> +<span class="sourceLineNo">194</span> warningSkew = c.getLong("hbase.master.warningclockskew", 10000);<a name="line.194"></a> +<span class="sourceLineNo">195</span> this.connection = master.getClusterConnection();<a name="line.195"></a> +<span class="sourceLineNo">196</span> this.rpcControllerFactory = this.connection == null? null: connection.getRpcControllerFactory();<a name="line.196"></a> +<span class="sourceLineNo">197</span> }<a name="line.197"></a> +<span class="sourceLineNo">198</span><a name="line.198"></a> +<span class="sourceLineNo">199</span> /**<a name="line.199"></a> +<span class="sourceLineNo">200</span> * Add the listener to the notification list.<a name="line.200"></a> +<span class="sourceLineNo">201</span> * @param listener The ServerListener to register<a name="line.201"></a> +<span class="sourceLineNo">202</span> */<a name="line.202"></a> +<span class="sourceLineNo">203</span> public void registerListener(final ServerListener listener) {<a name="line.203"></a> +<span class="sourceLineNo">204</span> this.listeners.add(listener);<a name="line.204"></a> +<span class="sourceLineNo">205</span> }<a name="line.205"></a> +<span class="sourceLineNo">206</span><a name="line.206"></a> +<span class="sourceLineNo">207</span> /**<a name="line.207"></a> +<span class="sourceLineNo">208</span> * Remove the listener from the notification list.<a name="line.208"></a> +<span class="sourceLineNo">209</span> * @param listener The ServerListener to unregister<a name="line.209"></a> +<span class="sourceLineNo">210</span> */<a name="line.210"></a> +<span class="sourceLineNo">211</span> public boolean unregisterListener(final ServerListener listener) {<a name="line.211"></a> +<span class="sourceLineNo">212</span> return this.listeners.remove(listener);<a name="line.212"></a> +<span class="sourceLineNo">213</span> }<a name="line.213"></a> +<span class="sourceLineNo">214</span><a name="line.214"></a> +<span class="sourceLineNo">215</span> /**<a name="line.215"></a> +<span class="sourceLineNo">216</span> * Let the server manager know a new regionserver has come online<a name="line.216"></a> +<span class="sourceLineNo">217</span> * @param request the startup request<a name="line.217"></a> +<span class="sourceLineNo">218</span> * @param versionNumber the version of the new regionserver<a name="line.218"></a> +<span class="sourceLineNo">219</span> * @param ia the InetAddress from which request is received<a name="line.219"></a> +<span class="sourceLineNo">220</span> * @return The ServerName we know this server as.<a name="line.220"></a> +<span class="sourceLineNo">221</span> * @throws IOException<a name="line.221"></a> +<span class="sourceLineNo">222</span> */<a name="line.222"></a> +<span class="sourceLineNo">223</span> ServerName regionServerStartup(RegionServerStartupRequest request, int versionNumber,<a name="line.223"></a> +<span class="sourceLineNo">224</span> InetAddress ia) throws IOException {<a name="line.224"></a> +<span class="sourceLineNo">225</span> // Test for case where we get a region startup message from a regionserver<a name="line.225"></a> +<span class="sourceLineNo">226</span> // that has been quickly restarted but whose znode expiration handler has<a name="line.226"></a> +<span class="sourceLineNo">227</span> // not yet run, or from a server whose fail we are currently processing.<a name="line.227"></a> +<span class="sourceLineNo">228</span> // Test its host+port combo is present in serverAddressToServerInfo. If it<a name="line.228"></a> +<span class="sourceLineNo">229</span> // is, reject the server and trigger its expiration. The next time it comes<a name="line.229"></a> +<span class="sourceLineNo">230</span> // in, it should have been removed from serverAddressToServerInfo and queued<a name="line.230"></a> +<span class="sourceLineNo">231</span> // for processing by ProcessServerShutdown.<a name="line.231"></a> +<span class="sourceLineNo">232</span><a name="line.232"></a> +<span class="sourceLineNo">233</span> final String hostname = request.hasUseThisHostnameInstead() ?<a name="line.233"></a> +<span class="sourceLineNo">234</span> request.getUseThisHostnameInstead() :ia.getHostName();<a name="line.234"></a> +<span class="sourceLineNo">235</span> ServerName sn = ServerName.valueOf(hostname, request.getPort(),<a name="line.235"></a> +<span class="sourceLineNo">236</span> request.getServerStartCode());<a name="line.236"></a> +<span class="sourceLineNo">237</span> checkClockSkew(sn, request.getServerCurrentTime());<a name="line.237"></a> +<span class="sourceLineNo">238</span> checkIsDead(sn, "STARTUP");<a name="line.238"></a> +<span class="sourceLineNo">239</span> if (!checkAndRecordNewServer(sn, ServerMetricsBuilder.of(sn, versionNumber))) {<a name="line.239"></a> +<span class="sourceLineNo">240</span> LOG.warn("THIS SHOULD NOT HAPPEN, RegionServerStartup"<a name="line.240"></a> +<span class="sourceLineNo">241</span> + " could not record the server: " + sn);<a name="line.241"></a> +<span class="sourceLineNo">242</span> }<a name="line.242"></a> +<span class="sourceLineNo">243</span> return sn;<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> /**<a name="line.246"></a> +<span class="sourceLineNo">247</span> * Updates last flushed sequence Ids for the regions on server sn<a name="line.247"></a> +<span class="sourceLineNo">248</span> * @param sn<a name="line.248"></a> +<span class="sourceLineNo">249</span> * @param hsl<a name="line.249"></a> +<span class="sourceLineNo">250</span> */<a name="line.250"></a> +<span class="sourceLineNo">251</span> private void updateLastFlushedSequenceIds(ServerName sn, ServerMetrics hsl) {<a name="line.251"></a> +<span class="sourceLineNo">252</span> for (Entry<byte[], RegionMetrics> entry : hsl.getRegionMetrics().entrySet()) {<a name="line.252"></a> +<span class="sourceLineNo">253</span> byte[] encodedRegionName = Bytes.toBytes(RegionInfo.encodeRegionName(entry.getKey()));<a name="line.253"></a> +<span class="sourceLineNo">254</span> Long existingValue = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.254"></a> +<span class="sourceLineNo">255</span> long l = entry.getValue().getCompletedSequenceId();<a name="line.255"></a> +<span class="sourceLineNo">256</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.256"></a> +<span class="sourceLineNo">257</span> if (LOG.isTraceEnabled()) {<a name="line.257"></a> +<span class="sourceLineNo">258</span> LOG.trace(Bytes.toString(encodedRegionName) + ", existingValue=" + existingValue +<a name="line.258"></a> +<span class="sourceLineNo">259</span> ", completeSequenceId=" + l);<a name="line.259"></a> +<span class="sourceLineNo">260</span> }<a name="line.260"></a> +<span class="sourceLineNo">261</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue)) {<a name="line.261"></a> +<span class="sourceLineNo">262</span> flushedSequenceIdByRegion.put(encodedRegionName, l);<a name="line.262"></a> +<span class="sourceLineNo">263</span> } else if (l != HConstants.NO_SEQNUM && l < existingValue) {<a name="line.263"></a> +<span class="sourceLineNo">264</span> LOG.warn("RegionServer " + sn + " indicates a last flushed sequence id ("<a name="line.264"></a> +<span class="sourceLineNo">265</span> + l + ") that is less than the previous last flushed sequence id ("<a name="line.265"></a> +<span class="sourceLineNo">266</span> + existingValue + ") for region " + Bytes.toString(entry.getKey()) + " Ignoring.");<a name="line.266"></a> +<span class="sourceLineNo">267</span> }<a name="line.267"></a> +<span class="sourceLineNo">268</span> ConcurrentNavigableMap<byte[], Long> storeFlushedSequenceId =<a name="line.268"></a> +<span class="sourceLineNo">269</span> computeIfAbsent(storeFlushedSequenceIdsByRegion, encodedRegionName,<a name="line.269"></a> +<span class="sourceLineNo">270</span> () -> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR));<a name="line.270"></a> +<span class="sourceLineNo">271</span> for (Entry<byte[], Long> storeSeqId : entry.getValue().getStoreSequenceId().entrySet()) {<a name="line.271"></a> +<span class="sourceLineNo">272</span> byte[] family = storeSeqId.getKey();<a name="line.272"></a> +<span class="sourceLineNo">273</span> existingValue = storeFlushedSequenceId.get(family);<a name="line.273"></a> +<span class="sourceLineNo">274</span> l = storeSeqId.getValue();<a name="line.274"></a> +<span class="sourceLineNo">275</span> if (LOG.isTraceEnabled()) {<a name="line.275"></a> +<span class="sourceLineNo">276</span> LOG.trace(Bytes.toString(encodedRegionName) + ", family=" + Bytes.toString(family) +<a name="line.276"></a> +<span class="sourceLineNo">277</span> ", existingValue=" + existingValue + ", completeSequenceId=" + l);<a name="line.277"></a> +<span class="sourceLineNo">278</span> }<a name="line.278"></a> +<span class="sourceLineNo">279</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.279"></a> +<span class="sourceLineNo">280</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue.longValue())) {<a name="line.280"></a> +<span class="sourceLineNo">281</span> storeFlushedSequenceId.put(family, l);<a name="line.281"></a> +<span class="sourceLineNo">282</span> }<a name="line.282"></a> +<span class="sourceLineNo">283</span> }<a name="line.283"></a> +<span class="sourceLineNo">284</span> }<a name="line.284"></a> +<span class="sourceLineNo">285</span> }<a name="line.285"></a> +<span class="sourceLineNo">286</span><a name="line.286"></a> +<span class="sourceLineNo">287</span> @VisibleForTesting<a name="line.287"></a> +<span class="sourceLineNo">288</span> public void regionServerReport(ServerName sn,<a name="line.288"></a> +<span class="sourceLineNo">289</span> ServerMetrics sl) throws YouAreDeadException {<a name="line.289"></a> +<span class="sourceLineNo">290</span> checkIsDead(sn, "REPORT");<a name="line.290"></a> +<span class="sourceLineNo">291</span> if (null == this.onlineServers.replace(sn, sl)) {<a name="line.291"></a> +<span class="sourceLineNo">292</span> // Already have this host+port combo and its just different start code?<a name="line.292"></a> +<span class="sourceLineNo">293</span> // Just let the server in. Presume master joining a running cluster.<a name="line.293"></a> +<span class="sourceLineNo">294</span> // recordNewServer is what happens at the end of reportServerStartup.<a name="line.294"></a> +<span class="sourceLineNo">295</span> // The only thing we are skipping is passing back to the regionserver<a name="line.295"></a> +<span class="sourceLineNo">296</span> // the ServerName to use. Here we presume a master has already done<a name="line.296"></a> +<span class="sourceLineNo">297</span> // that so we'll press on with whatever it gave us for ServerName.<a name="line.297"></a> +<span class="sourceLineNo">298</span> if (!checkAndRecordNewServer(sn, sl)) {<a name="line.298"></a> +<span class="sourceLineNo">299</span> LOG.info("RegionServerReport ignored, could not record the server: " + sn);<a name="line.299"></a> +<span class="sourceLineNo">300</span> return; // Not recorded, so no need to move on<a name="line.300"></a> +<span class="sourceLineNo">301</span> }<a name="line.301"></a> +<span class="sourceLineNo">302</span> }<a name="line.302"></a> +<span class="sourceLineNo">303</span> updateLastFlushedSequenceIds(sn, sl);<a name="line.303"></a> +<span class="sourceLineNo">304</span> }<a name="line.304"></a> +<span class="sourceLineNo">305</span><a name="line.305"></a> +<span class="sourceLineNo">306</span> /**<a name="line.306"></a> +<span class="sourceLineNo">307</span> * Check is a server of same host and port already exists,<a name="line.307"></a> +<span class="sourceLineNo">308</span> * if not, or the existed one got a smaller start code, record it.<a name="line.308"></a> +<span class="sourceLineNo">309</span> *<a name="line.309"></a> +<span class="sourceLineNo">310</span> * @param serverName the server to check and record<a name="line.310"></a> +<span class="sourceLineNo">311</span> * @param sl the server load on the server<a name="line.311"></a> +<span class="sourceLineNo">312</span> * @return true if the server is recorded, otherwise, false<a name="line.312"></a> +<span class="sourceLineNo">313</span> */<a name="line.313"></a> +<span class="sourceLineNo">314</span> boolean checkAndRecordNewServer(final ServerName serverName, final ServerMetrics sl) {<a name="line.314"></a> +<span class="sourceLineNo">315</span> ServerName existingServer = null;<a name="line.315"></a> +<span class="sourceLineNo">316</span> synchronized (this.onlineServers) {<a name="line.316"></a> +<span class="sourceLineNo">317</span> existingServer = findServerWithSameHostnamePortWithLock(serverName);<a name="line.317"></a> +<span class="sourceLineNo">318</span> if (existingServer != null && (existingServer.getStartcode() > serverName.getStartcode())) {<a name="line.318"></a> +<span class="sourceLineNo">319</span> LOG.info("Server serverName=" + serverName + " rejected; we already have "<a name="line.319"></a> +<span class="sourceLineNo">320</span> + existingServer.toString() + " registered with same hostname and port");<a name="line.320"></a> +<span class="sourceLineNo">321</span> return false;<a name="line.321"></a> +<span class="sourceLineNo">322</span> }<a name="line.322"></a> +<span class="sourceLineNo">323</span> recordNewServerWithLock(serverName, sl);<a name="line.323"></a> +<span class="sourceLineNo">324</span> }<a name="line.324"></a> +<span class="sourceLineNo">325</span><a name="line.325"></a> +<span class="sourceLineNo">326</span> // Tell our listeners that a server was added<a name="line.326"></a> +<span class="sourceLineNo">327</span> if (!this.listeners.isEmpty()) {<a name="line.327"></a> +<span class="sourceLineNo">328</span> for (ServerListener listener : this.listeners) {<a name="line.328"></a> +<span class="sourceLineNo">329</span> listener.serverAdded(serverName);<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">331</span> }<a name="line.331"></a> +<span class="sourceLineNo">332</span><a name="line.332"></a> +<span class="sourceLineNo">333</span> // Note that we assume that same ts means same server, and don't expire in that case.<a name="line.333"></a> +<span class="sourceLineNo">334</span> // TODO: ts can theoretically collide due to clock shifts, so this is a bit hacky.<a name="line.334"></a> +<span class="sourceLineNo">335</span> if (existingServer != null &&<a name="line.335"></a> +<span class="sourceLineNo">336</span> (existingServer.getStartcode() < serverName.getStartcode())) {<a name="line.336"></a> +<span class="sourceLineNo">337</span> LOG.info("Triggering server recovery; existingServer " +<a name="line.337"></a> +<span class="sourceLineNo">338</span> existingServer + " looks stale, new server:" + serverName);<a name="line.338"></a> +<span class="sourceLineNo">339</span> expireServer(existingServer);<a name="line.339"></a> +<span class="sourceLineNo">340</span> }<a name="line.340"></a> +<span class="sourceLineNo">341</span> return true;<a name="line.341"></a> +<span class="sourceLineNo">342</span> }<a name="line.342"></a> +<span class="sourceLineNo">343</span><a name="line.343"></a> +<span class="sourceLineNo">344</span> /**<a name="line.344"></a> +<span class="sourceLineNo">345</span> * Checks if the clock skew between the server and the master. If the clock skew exceeds the<a name="line.345"></a> +<span class="sourceLineNo">346</span> * configured max, it will throw an exception; if it exceeds the configured warning threshold,<a name="line.346"></a> +<span class="sourceLineNo">347</span> * it will log a warning but start normally.<a name="line.347"></a> +<span class="sourceLineNo">348</span> * @param serverName Incoming servers's name<a name="line.348"></a> +<span class="sourceLineNo">349</span> * @param serverCurrentTime<a name="line.349"></a> +<span class="sourceLineNo">350</span> * @throws ClockOutOfSyncException if the skew exceeds the configured max value<a name="line.350"></a> +<span class="sourceLineNo">351</span> */<a name="line.351"></a> +<span class="sourceLineNo">352</span> private void checkClockSkew(final ServerName serverName, final long serverCurrentTime)<a name="line.352"></a> +<span class="sourceLineNo">353</span> throws ClockOutOfSyncException {<a name="line.353"></a> +<span class="sourceLineNo">354</span> long skew = Math.abs(System.currentTimeMillis() - serverCurrentTime);<a name="line.354"></a> +<span class="sourceLineNo">355</span> if (skew > maxSkew) {<a name="line.355"></a> +<span class="sourceLineNo">356</span> String message = "Server " + serverName + " has been " +<a name="line.356"></a> +<span class="sourceLineNo">357</span> "rejected; Reported time is too far out of sync with master. " +<a name="line.357"></a> +<span class="sourceLineNo">358</span> "Time difference of " + skew + "ms > max allowed of " + maxSkew + "ms";<a name="line.358"></a> +<span class="sourceLineNo">359</span> LOG.warn(message);<a name="line.359"></a> +<span class="sourceLineNo">360</span> throw new ClockOutOfSyncException(message);<a name="line.360"></a> +<span class="sourceLineNo">361</span> } else if (skew > warningSkew){<a name="line.361"></a> +<span class="sourceLineNo">362</span> String message = "Reported time for server " + serverName + " is out of sync with master " +<a name="line.362"></a> +<span class="sourceLineNo">363</span> "by " + skew + "ms. (Warning threshold is " + warningSkew + "ms; " +<a name="line.363"></a> +<span class="sourceLineNo">364</span> "error threshold is " + 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> }<a name="line.366"></a> +<span class="sourceLineNo">367</span> }<a name="line.367"></a> +<span class="sourceLineNo">368</span><a name="line.368"></a> +<span class="sourceLineNo">369</span> /**<a name="line.369"></a> +<span class="sourceLineNo">370</span> * If this server is on the dead list, reject it with a YouAreDeadException.<a name="line.370"></a> +<span class="sourceLineNo">371</span> * If it was dead but came back with a new start code, remove the old entry<a name="line.371"></a> +<span class="sourceLineNo">372</span> * from the dead list.<a name="line.372"></a> +<span class="sourceLineNo">373</span> * @param serverName<a name="line.373"></a> +<span class="sourceLineNo">374</span> * @param what START or REPORT<a name="line.374"></a> +<span class="sourceLineNo">375</span> * @throws org.apache.hadoop.hbase.YouAreDeadException<a name="line.375"></a> +<span class="sourceLineNo">376</span> */<a name="line.376"></a> +<span class="sourceLineNo">377</span> private void checkIsDead(final ServerName serverName, final String what)<a name="line.377"></a> +<span class="sourceLineNo">378</span> throws YouAreDeadException {<a name="line.378"></a> +<span class="sourceLineNo">379</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.379"></a> +<span class="sourceLineNo">380</span> // host name, port and start code all match with existing one of the<a name="line.380"></a> +<span class="sourceLineNo">381</span> // dead servers. So, this server must be dead.<a name="line.381"></a> +<span class="sourceLineNo">382</span> String message = "Server " + what + " rejected; currently processing " +<a name="line.382"></a> +<span class="sourceLineNo">383</span> serverName + " as dead server";<a name="line.383"></a> +<span class="sourceLineNo">384</span> LOG.debug(message);<a name="line.384"></a> +<span class="sourceLineNo">385</span> throw new YouAreDeadException(message);<a name="line.385"></a> +<span class="sourceLineNo">386</span> }<a name="line.386"></a> +<span class="sourceLineNo">387</span> // remove dead server with same hostname and port of newly checking in rs after master<a name="line.387"></a> +<span class="sourceLineNo">388</span> // initialization.See HBASE-5916 for more information.<a name="line.388"></a> +<span class="sourceLineNo">389</span> if ((this.master == null || this.master.isInitialized())<a name="line.389"></a> +<span class="sourceLineNo">390</span> && this.deadservers.cleanPreviousInstance(serverName)) {<a name="line.390"></a> +<span class="sourceLineNo">391</span> // This server has now become alive after we marked it as dead.<a name="line.391"></a> +<span class="sourceLineNo">392</span> // We removed it's previous entry from the dead list to reflect it.<a name="line.392"></a> +<span class="sourceLineNo">393</span> LOG.debug(what + ":" + " Server " + serverName + " came back up," +<a name="line.393"></a> +<span class="sourceLineNo">394</span> " removed it from the dead servers list");<a name="line.394"></a> +<span class="sourceLineNo">395</span> }<a name="line.395"></a> +<span class="sourceLineNo">396</span> }<a name="line.396"></a> +<span class="sourceLineNo">397</span><a name="line.397"></a> +<span class="sourceLineNo">398</span> /**<a name="line.398"></a> +<span class="sourceLineNo">399</span> * Assumes onlineServers is locked.<a name="line.399"></a> +<span class="sourceLineNo">400</span> * @return ServerName with matching hostname and port.<a name="line.400"></a> +<span class="sourceLineNo">401</span> */<a name="line.401"></a> +<span class="sourceLineNo">402</span> private ServerName findServerWithSameHostnamePortWithLock(<a name="line.402"></a> +<span class="sourceLineNo">403</span> final ServerName serverName) {<a name="line.403"></a> +<span class="sourceLineNo">404</span> ServerName end = ServerName.valueOf(serverName.getHostname(), serverName.getPort(),<a name="line.404"></a> +<span class="sourceLineNo">405</span> Long.MAX_VALUE);<a name="line.405"></a> +<span class="sourceLineNo">406</span><a name="line.406"></a> +<span class="sourceLineNo">407</span> ServerName r = onlineServers.lowerKey(end);<a name="line.407"></a> +<span class="sourceLineNo">408</span> if (r != null) {<a name="line.408"></a> +<span class="sourceLineNo">409</span> if (ServerName.isSameAddress(r, serverName)) {<a name="line.409"></a> +<span class="sourceLineNo">410</span> return r;<a name="line.410"></a> +<span class="sourceLineNo">411</span> }<a name="line.411"></a> +<span class="sourceLineNo">412</span> }<a name="line.412"></a> +<span class="sourceLineNo">413</span> return null;<a name="line.413"></a> +<span class="sourceLineNo">414</span> }<a name="line.414"></a> +<span class="sourceLineNo">415</span><a name="line.415"></a> +<span class="sourceLineNo">416</span> /**<a name="line.416"></a> +<span class="sourceLineNo">417</span> * Adds the onlineServers list. onlineServers should be locked.<a name="line.417"></a> +<span class="sourceLineNo">418</span> * @param serverName The remote servers name.<a name="line.418"></a> +<span class="sourceLineNo">419</span> */<a name="line.419"></a> +<span class="sourceLineNo">420</span> @VisibleForTesting<a name="line.420"></a> +<span class="sourceLineNo">421</span> void recordNewServerWithLock(final ServerName serverName, final ServerMetrics sl) {<a name="line.421"></a> +<span class="sourceLineNo">422</span> LOG.info("Registering regionserver=" + serverName);<a name="line.422"></a> +<span class="sourceLineNo">423</span> this.onlineServers.put(serverName, sl);<a name="line.423"></a> +<span class="sourceLineNo">424</span> this.rsAdmins.remove(serverName);<a name="line.424"></a> +<span class="sourceLineNo">425</span> }<a name="line.425"></a> +<span class="sourceLineNo">426</span><a name="line.426"></a> +<span class="sourceLineNo">427</span> public RegionStoreSequenceIds getLastFlushedSequenceId(byte[] encodedRegionName) {<a name="line.427"></a> +<span class="sourceLineNo">428</span> RegionStoreSequenceIds.Builder builder = RegionStoreSequenceIds.newBuilder();<a name="line.428"></a> +<span class="sourceLineNo">429</span> Long seqId = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.429"></a> +<span class="sourceLineNo">430</span> builder.setLastFlushedSequenceId(seqId != null ? seqId.longValue() : HConstants.NO_SEQNUM);<a name="line.430"></a> +<span class="sourceLineNo">431</span> Map<byte[], Long> storeFlushedSequenceId =<a name="line.431"></a> +<span class="sourceLineNo">432</span> storeFlushedSequenceIdsByRegion.get(encodedRegionName);<a name="line.432"></a> +<span class="sourceLineNo">433</span> if (storeFlushedSequenceId != null) {<a name="line.433"></a> +<span class="sourceLineNo">434</span> for (Map.Entry<byte[], Long> entry : storeFlushedSequenceId.entrySet()) {<a name="line.434"></a> +<span class="sourceLineNo">435</span> builder.addStoreSequenceId(StoreSequenceId.newBuilder()<a name="line.435"></a> +<span class="sourceLineNo">436</span> .setFamilyName(UnsafeByteOperations.unsafeWrap(entry.getKey()))<a name="line.436"></a> +<span class="sourceLineNo">437</span> .setSequenceId(entry.getValue().longValue()).build());<a name="line.437"></a> +<span class="sourceLineNo">438</span> }<a name="line.438"></a> +<span class="sourceLineNo">439</span> }<a name="line.439"></a> +<span class="sourceLineNo">440</span> return builder.build();<a name="line.440"></a> +<span class="sourceLineNo">441</span> }<a name="line.441"></a> +<span class="sourceLineNo">442</span><a name="line.442"></a> +<span class="sourceLineNo">443</span> /**<a name="line.443"></a> +<span class="sourceLineNo">444</span> * @param serverName<a name="line.444"></a> +<span class="sourceLineNo">445</span> * @return ServerMetrics if serverName is known else null<a name="line.445"></a> +<span class="sourceLineNo">446</span> */<a name="line.446"></a> +<span class="sourceLineNo">447</span> public ServerMetrics getLoad(final ServerName serverName) {<a name="line.447"></a> +<span class="sourceLineNo">448</span> return this.onlineServers.get(serverName);<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="sourceL
<TRUNCATED>