| <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"> |
| <html lang="en"> |
| <head> |
| <title>Source code</title> |
| <link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style"> |
| </head> |
| <body> |
| <div class="sourceContainer"> |
| <pre><span class="sourceLineNo">001</span>/*<a name="line.1"></a> |
| <span class="sourceLineNo">002</span> *<a name="line.2"></a> |
| <span class="sourceLineNo">003</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.3"></a> |
| <span class="sourceLineNo">004</span> * or more contributor license agreements. See the NOTICE file<a name="line.4"></a> |
| <span class="sourceLineNo">005</span> * distributed with this work for additional information<a name="line.5"></a> |
| <span class="sourceLineNo">006</span> * regarding copyright ownership. The ASF licenses this file<a name="line.6"></a> |
| <span class="sourceLineNo">007</span> * to you under the Apache License, Version 2.0 (the<a name="line.7"></a> |
| <span class="sourceLineNo">008</span> * "License"); you may not use this file except in compliance<a name="line.8"></a> |
| <span class="sourceLineNo">009</span> * with the License. You may obtain a copy of the License at<a name="line.9"></a> |
| <span class="sourceLineNo">010</span> *<a name="line.10"></a> |
| <span class="sourceLineNo">011</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.11"></a> |
| <span class="sourceLineNo">012</span> *<a name="line.12"></a> |
| <span class="sourceLineNo">013</span> * Unless required by applicable law or agreed to in writing, software<a name="line.13"></a> |
| <span class="sourceLineNo">014</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.14"></a> |
| <span class="sourceLineNo">015</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.15"></a> |
| <span class="sourceLineNo">016</span> * See the License for the specific language governing permissions and<a name="line.16"></a> |
| <span class="sourceLineNo">017</span> * limitations under the License.<a name="line.17"></a> |
| <span class="sourceLineNo">018</span> */<a name="line.18"></a> |
| <span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.master;<a name="line.19"></a> |
| <span class="sourceLineNo">020</span>import java.io.IOException;<a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import java.io.InterruptedIOException;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span>import java.util.List;<a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.util.Optional;<a name="line.23"></a> |
| <span class="sourceLineNo">024</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.24"></a> |
| <span class="sourceLineNo">025</span>import org.apache.hadoop.hbase.Server;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.ServerName;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.ZNodeClearer;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.exceptions.DeserializationException;<a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.monitoring.MonitoredTask;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.zookeeper.MasterAddressTracker;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.zookeeper.ZKListener;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.zookeeper.ZNodePaths;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import org.apache.zookeeper.KeeperException;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import org.slf4j.Logger;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import org.slf4j.LoggerFactory;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import org.apache.hbase.thirdparty.com.google.common.collect.ImmutableList;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span><a name="line.41"></a> |
| <span class="sourceLineNo">042</span>/**<a name="line.42"></a> |
| <span class="sourceLineNo">043</span> * Handles everything on master-side related to master election. Keeps track of<a name="line.43"></a> |
| <span class="sourceLineNo">044</span> * currently active master and registered backup masters.<a name="line.44"></a> |
| <span class="sourceLineNo">045</span> *<a name="line.45"></a> |
| <span class="sourceLineNo">046</span> * <p>Listens and responds to ZooKeeper notifications on the master znodes,<a name="line.46"></a> |
| <span class="sourceLineNo">047</span> * both <code>nodeCreated</code> and <code>nodeDeleted</code>.<a name="line.47"></a> |
| <span class="sourceLineNo">048</span> *<a name="line.48"></a> |
| <span class="sourceLineNo">049</span> * <p>Contains blocking methods which will hold up backup masters, waiting<a name="line.49"></a> |
| <span class="sourceLineNo">050</span> * for the active master to fail.<a name="line.50"></a> |
| <span class="sourceLineNo">051</span> *<a name="line.51"></a> |
| <span class="sourceLineNo">052</span> * <p>This class is instantiated in the HMaster constructor and the method<a name="line.52"></a> |
| <span class="sourceLineNo">053</span> * #blockUntilBecomingActiveMaster() is called to wait until becoming<a name="line.53"></a> |
| <span class="sourceLineNo">054</span> * the active master of the cluster.<a name="line.54"></a> |
| <span class="sourceLineNo">055</span> */<a name="line.55"></a> |
| <span class="sourceLineNo">056</span>@InterfaceAudience.Private<a name="line.56"></a> |
| <span class="sourceLineNo">057</span>public class ActiveMasterManager extends ZKListener {<a name="line.57"></a> |
| <span class="sourceLineNo">058</span> private static final Logger LOG = LoggerFactory.getLogger(ActiveMasterManager.class);<a name="line.58"></a> |
| <span class="sourceLineNo">059</span><a name="line.59"></a> |
| <span class="sourceLineNo">060</span> final AtomicBoolean clusterHasActiveMaster = new AtomicBoolean(false);<a name="line.60"></a> |
| <span class="sourceLineNo">061</span> final AtomicBoolean clusterShutDown = new AtomicBoolean(false);<a name="line.61"></a> |
| <span class="sourceLineNo">062</span><a name="line.62"></a> |
| <span class="sourceLineNo">063</span> // This server's information. Package-private for child implementations.<a name="line.63"></a> |
| <span class="sourceLineNo">064</span> int infoPort;<a name="line.64"></a> |
| <span class="sourceLineNo">065</span> final ServerName sn;<a name="line.65"></a> |
| <span class="sourceLineNo">066</span> final Server master;<a name="line.66"></a> |
| <span class="sourceLineNo">067</span><a name="line.67"></a> |
| <span class="sourceLineNo">068</span> // Active master's server name. Invalidated anytime active master changes (based on ZK<a name="line.68"></a> |
| <span class="sourceLineNo">069</span> // notifications) and lazily fetched on-demand.<a name="line.69"></a> |
| <span class="sourceLineNo">070</span> // ServerName is immutable, so we don't need heavy synchronization around it.<a name="line.70"></a> |
| <span class="sourceLineNo">071</span> volatile ServerName activeMasterServerName;<a name="line.71"></a> |
| <span class="sourceLineNo">072</span> // Registered backup masters. List is kept up to date based on ZK change notifications to<a name="line.72"></a> |
| <span class="sourceLineNo">073</span> // backup znode.<a name="line.73"></a> |
| <span class="sourceLineNo">074</span> private volatile ImmutableList<ServerName> backupMasters;<a name="line.74"></a> |
| <span class="sourceLineNo">075</span><a name="line.75"></a> |
| <span class="sourceLineNo">076</span> /**<a name="line.76"></a> |
| <span class="sourceLineNo">077</span> * @param watcher ZK watcher<a name="line.77"></a> |
| <span class="sourceLineNo">078</span> * @param sn ServerName<a name="line.78"></a> |
| <span class="sourceLineNo">079</span> * @param master In an instance of a Master.<a name="line.79"></a> |
| <span class="sourceLineNo">080</span> */<a name="line.80"></a> |
| <span class="sourceLineNo">081</span> ActiveMasterManager(ZKWatcher watcher, ServerName sn, Server master)<a name="line.81"></a> |
| <span class="sourceLineNo">082</span> throws InterruptedIOException {<a name="line.82"></a> |
| <span class="sourceLineNo">083</span> super(watcher);<a name="line.83"></a> |
| <span class="sourceLineNo">084</span> watcher.registerListener(this);<a name="line.84"></a> |
| <span class="sourceLineNo">085</span> this.sn = sn;<a name="line.85"></a> |
| <span class="sourceLineNo">086</span> this.master = master;<a name="line.86"></a> |
| <span class="sourceLineNo">087</span> updateBackupMasters();<a name="line.87"></a> |
| <span class="sourceLineNo">088</span> }<a name="line.88"></a> |
| <span class="sourceLineNo">089</span><a name="line.89"></a> |
| <span class="sourceLineNo">090</span> // will be set after jetty server is started<a name="line.90"></a> |
| <span class="sourceLineNo">091</span> public void setInfoPort(int infoPort) {<a name="line.91"></a> |
| <span class="sourceLineNo">092</span> this.infoPort = infoPort;<a name="line.92"></a> |
| <span class="sourceLineNo">093</span> }<a name="line.93"></a> |
| <span class="sourceLineNo">094</span><a name="line.94"></a> |
| <span class="sourceLineNo">095</span> @Override<a name="line.95"></a> |
| <span class="sourceLineNo">096</span> public void nodeCreated(String path) {<a name="line.96"></a> |
| <span class="sourceLineNo">097</span> handle(path);<a name="line.97"></a> |
| <span class="sourceLineNo">098</span> }<a name="line.98"></a> |
| <span class="sourceLineNo">099</span><a name="line.99"></a> |
| <span class="sourceLineNo">100</span> @Override<a name="line.100"></a> |
| <span class="sourceLineNo">101</span> public void nodeChildrenChanged(String path) {<a name="line.101"></a> |
| <span class="sourceLineNo">102</span> if (path.equals(watcher.getZNodePaths().backupMasterAddressesZNode)) {<a name="line.102"></a> |
| <span class="sourceLineNo">103</span> try {<a name="line.103"></a> |
| <span class="sourceLineNo">104</span> updateBackupMasters();<a name="line.104"></a> |
| <span class="sourceLineNo">105</span> } catch (InterruptedIOException ioe) {<a name="line.105"></a> |
| <span class="sourceLineNo">106</span> LOG.error("Error updating backup masters", ioe);<a name="line.106"></a> |
| <span class="sourceLineNo">107</span> }<a name="line.107"></a> |
| <span class="sourceLineNo">108</span> }<a name="line.108"></a> |
| <span class="sourceLineNo">109</span> }<a name="line.109"></a> |
| <span class="sourceLineNo">110</span><a name="line.110"></a> |
| <span class="sourceLineNo">111</span> @Override<a name="line.111"></a> |
| <span class="sourceLineNo">112</span> public void nodeDeleted(String path) {<a name="line.112"></a> |
| <span class="sourceLineNo">113</span> // We need to keep track of the cluster's shutdown status while<a name="line.113"></a> |
| <span class="sourceLineNo">114</span> // we wait on the current master. We consider that, if the cluster<a name="line.114"></a> |
| <span class="sourceLineNo">115</span> // was already in a "shutdown" state when we started, that this master<a name="line.115"></a> |
| <span class="sourceLineNo">116</span> // is part of a new cluster that was started shortly after the old cluster<a name="line.116"></a> |
| <span class="sourceLineNo">117</span> // shut down, so that state is now irrelevant. This means that the shutdown<a name="line.117"></a> |
| <span class="sourceLineNo">118</span> // state must be set while we wait on the active master in order<a name="line.118"></a> |
| <span class="sourceLineNo">119</span> // to shutdown this master. See HBASE-8519.<a name="line.119"></a> |
| <span class="sourceLineNo">120</span> if(path.equals(watcher.getZNodePaths().clusterStateZNode) && !master.isStopped()) {<a name="line.120"></a> |
| <span class="sourceLineNo">121</span> clusterShutDown.set(true);<a name="line.121"></a> |
| <span class="sourceLineNo">122</span> }<a name="line.122"></a> |
| <span class="sourceLineNo">123</span> handle(path);<a name="line.123"></a> |
| <span class="sourceLineNo">124</span> }<a name="line.124"></a> |
| <span class="sourceLineNo">125</span><a name="line.125"></a> |
| <span class="sourceLineNo">126</span> void handle(final String path) {<a name="line.126"></a> |
| <span class="sourceLineNo">127</span> if (path.equals(watcher.getZNodePaths().masterAddressZNode) && !master.isStopped()) {<a name="line.127"></a> |
| <span class="sourceLineNo">128</span> handleMasterNodeChange();<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><a name="line.131"></a> |
| <span class="sourceLineNo">132</span> private void updateBackupMasters() throws InterruptedIOException {<a name="line.132"></a> |
| <span class="sourceLineNo">133</span> backupMasters =<a name="line.133"></a> |
| <span class="sourceLineNo">134</span> ImmutableList.copyOf(MasterAddressTracker.getBackupMastersAndRenewWatch(watcher));<a name="line.134"></a> |
| <span class="sourceLineNo">135</span> }<a name="line.135"></a> |
| <span class="sourceLineNo">136</span><a name="line.136"></a> |
| <span class="sourceLineNo">137</span> /**<a name="line.137"></a> |
| <span class="sourceLineNo">138</span> * Fetches the active master's ServerName from zookeeper.<a name="line.138"></a> |
| <span class="sourceLineNo">139</span> */<a name="line.139"></a> |
| <span class="sourceLineNo">140</span> private void fetchAndSetActiveMasterServerName() {<a name="line.140"></a> |
| <span class="sourceLineNo">141</span> LOG.debug("Attempting to fetch active master sn from zk");<a name="line.141"></a> |
| <span class="sourceLineNo">142</span> try {<a name="line.142"></a> |
| <span class="sourceLineNo">143</span> activeMasterServerName = MasterAddressTracker.getMasterAddress(watcher);<a name="line.143"></a> |
| <span class="sourceLineNo">144</span> } catch (IOException | KeeperException e) {<a name="line.144"></a> |
| <span class="sourceLineNo">145</span> // Log and ignore for now and re-fetch later if needed.<a name="line.145"></a> |
| <span class="sourceLineNo">146</span> LOG.error("Error fetching active master information", e);<a name="line.146"></a> |
| <span class="sourceLineNo">147</span> }<a name="line.147"></a> |
| <span class="sourceLineNo">148</span> }<a name="line.148"></a> |
| <span class="sourceLineNo">149</span><a name="line.149"></a> |
| <span class="sourceLineNo">150</span> public Optional<ServerName> getActiveMasterServerName() {<a name="line.150"></a> |
| <span class="sourceLineNo">151</span> if (!clusterHasActiveMaster.get()) {<a name="line.151"></a> |
| <span class="sourceLineNo">152</span> return Optional.empty();<a name="line.152"></a> |
| <span class="sourceLineNo">153</span> }<a name="line.153"></a> |
| <span class="sourceLineNo">154</span> if (activeMasterServerName == null) {<a name="line.154"></a> |
| <span class="sourceLineNo">155</span> fetchAndSetActiveMasterServerName();<a name="line.155"></a> |
| <span class="sourceLineNo">156</span> }<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> // It could still be null, but return whatever we have.<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> return Optional.ofNullable(activeMasterServerName);<a name="line.158"></a> |
| <span class="sourceLineNo">159</span> }<a name="line.159"></a> |
| <span class="sourceLineNo">160</span><a name="line.160"></a> |
| <span class="sourceLineNo">161</span> /**<a name="line.161"></a> |
| <span class="sourceLineNo">162</span> * Handle a change in the master node. Doesn't matter whether this was called<a name="line.162"></a> |
| <span class="sourceLineNo">163</span> * from a nodeCreated or nodeDeleted event because there are no guarantees<a name="line.163"></a> |
| <span class="sourceLineNo">164</span> * that the current state of the master node matches the event at the time of<a name="line.164"></a> |
| <span class="sourceLineNo">165</span> * our next ZK request.<a name="line.165"></a> |
| <span class="sourceLineNo">166</span> *<a name="line.166"></a> |
| <span class="sourceLineNo">167</span> * <p>Uses the watchAndCheckExists method which watches the master address node<a name="line.167"></a> |
| <span class="sourceLineNo">168</span> * regardless of whether it exists or not. If it does exist (there is an<a name="line.168"></a> |
| <span class="sourceLineNo">169</span> * active master), it returns true. Otherwise it returns false.<a name="line.169"></a> |
| <span class="sourceLineNo">170</span> *<a name="line.170"></a> |
| <span class="sourceLineNo">171</span> * <p>A watcher is set which guarantees that this method will get called again if<a name="line.171"></a> |
| <span class="sourceLineNo">172</span> * there is another change in the master node.<a name="line.172"></a> |
| <span class="sourceLineNo">173</span> */<a name="line.173"></a> |
| <span class="sourceLineNo">174</span> private void handleMasterNodeChange() {<a name="line.174"></a> |
| <span class="sourceLineNo">175</span> // Watch the node and check if it exists.<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> try {<a name="line.176"></a> |
| <span class="sourceLineNo">177</span> synchronized(clusterHasActiveMaster) {<a name="line.177"></a> |
| <span class="sourceLineNo">178</span> if (ZKUtil.watchAndCheckExists(watcher, watcher.getZNodePaths().masterAddressZNode)) {<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> // A master node exists, there is an active master<a name="line.179"></a> |
| <span class="sourceLineNo">180</span> LOG.trace("A master is now available");<a name="line.180"></a> |
| <span class="sourceLineNo">181</span> clusterHasActiveMaster.set(true);<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> } else {<a name="line.182"></a> |
| <span class="sourceLineNo">183</span> // Node is no longer there, cluster does not have an active master<a name="line.183"></a> |
| <span class="sourceLineNo">184</span> LOG.debug("No master available. Notifying waiting threads");<a name="line.184"></a> |
| <span class="sourceLineNo">185</span> clusterHasActiveMaster.set(false);<a name="line.185"></a> |
| <span class="sourceLineNo">186</span> // Notify any thread waiting to become the active master<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> clusterHasActiveMaster.notifyAll();<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> }<a name="line.188"></a> |
| <span class="sourceLineNo">189</span> // Reset the active master sn. Will be re-fetched later if needed.<a name="line.189"></a> |
| <span class="sourceLineNo">190</span> // We don't want to make a synchronous RPC under a monitor.<a name="line.190"></a> |
| <span class="sourceLineNo">191</span> activeMasterServerName = null;<a name="line.191"></a> |
| <span class="sourceLineNo">192</span> }<a name="line.192"></a> |
| <span class="sourceLineNo">193</span> } catch (KeeperException ke) {<a name="line.193"></a> |
| <span class="sourceLineNo">194</span> master.abort("Received an unexpected KeeperException, aborting", ke);<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><a name="line.197"></a> |
| <span class="sourceLineNo">198</span> /**<a name="line.198"></a> |
| <span class="sourceLineNo">199</span> * Block until becoming the active master.<a name="line.199"></a> |
| <span class="sourceLineNo">200</span> *<a name="line.200"></a> |
| <span class="sourceLineNo">201</span> * Method blocks until there is not another active master and our attempt<a name="line.201"></a> |
| <span class="sourceLineNo">202</span> * to become the new active master is successful.<a name="line.202"></a> |
| <span class="sourceLineNo">203</span> *<a name="line.203"></a> |
| <span class="sourceLineNo">204</span> * This also makes sure that we are watching the master znode so will be<a name="line.204"></a> |
| <span class="sourceLineNo">205</span> * notified if another master dies.<a name="line.205"></a> |
| <span class="sourceLineNo">206</span> * @param checkInterval the interval to check if the master is stopped<a name="line.206"></a> |
| <span class="sourceLineNo">207</span> * @param startupStatus the monitor status to track the progress<a name="line.207"></a> |
| <span class="sourceLineNo">208</span> * @return True if no issue becoming active master else false if another<a name="line.208"></a> |
| <span class="sourceLineNo">209</span> * master was running or if some other problem (zookeeper, stop flag has been<a name="line.209"></a> |
| <span class="sourceLineNo">210</span> * set on this Master)<a name="line.210"></a> |
| <span class="sourceLineNo">211</span> */<a name="line.211"></a> |
| <span class="sourceLineNo">212</span> boolean blockUntilBecomingActiveMaster(<a name="line.212"></a> |
| <span class="sourceLineNo">213</span> int checkInterval, MonitoredTask startupStatus) {<a name="line.213"></a> |
| <span class="sourceLineNo">214</span> String backupZNode = ZNodePaths.joinZNode(<a name="line.214"></a> |
| <span class="sourceLineNo">215</span> this.watcher.getZNodePaths().backupMasterAddressesZNode, this.sn.toString());<a name="line.215"></a> |
| <span class="sourceLineNo">216</span> while (!(master.isAborted() || master.isStopped())) {<a name="line.216"></a> |
| <span class="sourceLineNo">217</span> startupStatus.setStatus("Trying to register in ZK as active master");<a name="line.217"></a> |
| <span class="sourceLineNo">218</span> // Try to become the active master, watch if there is another master.<a name="line.218"></a> |
| <span class="sourceLineNo">219</span> // Write out our ServerName as versioned bytes.<a name="line.219"></a> |
| <span class="sourceLineNo">220</span> try {<a name="line.220"></a> |
| <span class="sourceLineNo">221</span> if (MasterAddressTracker.setMasterAddress(this.watcher,<a name="line.221"></a> |
| <span class="sourceLineNo">222</span> this.watcher.getZNodePaths().masterAddressZNode, this.sn, infoPort)) {<a name="line.222"></a> |
| <span class="sourceLineNo">223</span><a name="line.223"></a> |
| <span class="sourceLineNo">224</span> // If we were a backup master before, delete our ZNode from the backup<a name="line.224"></a> |
| <span class="sourceLineNo">225</span> // master directory since we are the active now)<a name="line.225"></a> |
| <span class="sourceLineNo">226</span> if (ZKUtil.checkExists(this.watcher, backupZNode) != -1) {<a name="line.226"></a> |
| <span class="sourceLineNo">227</span> LOG.info("Deleting ZNode for " + backupZNode + " from backup master directory");<a name="line.227"></a> |
| <span class="sourceLineNo">228</span> ZKUtil.deleteNodeFailSilent(this.watcher, backupZNode);<a name="line.228"></a> |
| <span class="sourceLineNo">229</span> }<a name="line.229"></a> |
| <span class="sourceLineNo">230</span> // Save the znode in a file, this will allow to check if we crash in the launch scripts<a name="line.230"></a> |
| <span class="sourceLineNo">231</span> ZNodeClearer.writeMyEphemeralNodeOnDisk(this.sn.toString());<a name="line.231"></a> |
| <span class="sourceLineNo">232</span><a name="line.232"></a> |
| <span class="sourceLineNo">233</span> // We are the master, return<a name="line.233"></a> |
| <span class="sourceLineNo">234</span> startupStatus.setStatus("Successfully registered as active master.");<a name="line.234"></a> |
| <span class="sourceLineNo">235</span> this.clusterHasActiveMaster.set(true);<a name="line.235"></a> |
| <span class="sourceLineNo">236</span> activeMasterServerName = sn;<a name="line.236"></a> |
| <span class="sourceLineNo">237</span> LOG.info("Registered as active master=" + this.sn);<a name="line.237"></a> |
| <span class="sourceLineNo">238</span> return true;<a name="line.238"></a> |
| <span class="sourceLineNo">239</span> }<a name="line.239"></a> |
| <span class="sourceLineNo">240</span><a name="line.240"></a> |
| <span class="sourceLineNo">241</span> // Invalidate the active master name so that subsequent requests do not get any stale<a name="line.241"></a> |
| <span class="sourceLineNo">242</span> // master information. Will be re-fetched if needed.<a name="line.242"></a> |
| <span class="sourceLineNo">243</span> activeMasterServerName = null;<a name="line.243"></a> |
| <span class="sourceLineNo">244</span> // There is another active master running elsewhere or this is a restart<a name="line.244"></a> |
| <span class="sourceLineNo">245</span> // and the master ephemeral node has not expired yet.<a name="line.245"></a> |
| <span class="sourceLineNo">246</span> this.clusterHasActiveMaster.set(true);<a name="line.246"></a> |
| <span class="sourceLineNo">247</span><a name="line.247"></a> |
| <span class="sourceLineNo">248</span> String msg;<a name="line.248"></a> |
| <span class="sourceLineNo">249</span> byte[] bytes =<a name="line.249"></a> |
| <span class="sourceLineNo">250</span> ZKUtil.getDataAndWatch(this.watcher, this.watcher.getZNodePaths().masterAddressZNode);<a name="line.250"></a> |
| <span class="sourceLineNo">251</span> if (bytes == null) {<a name="line.251"></a> |
| <span class="sourceLineNo">252</span> msg = ("A master was detected, but went down before its address " +<a name="line.252"></a> |
| <span class="sourceLineNo">253</span> "could be read. Attempting to become the next active master");<a name="line.253"></a> |
| <span class="sourceLineNo">254</span> } else {<a name="line.254"></a> |
| <span class="sourceLineNo">255</span> ServerName currentMaster;<a name="line.255"></a> |
| <span class="sourceLineNo">256</span> try {<a name="line.256"></a> |
| <span class="sourceLineNo">257</span> currentMaster = ProtobufUtil.parseServerNameFrom(bytes);<a name="line.257"></a> |
| <span class="sourceLineNo">258</span> } catch (DeserializationException e) {<a name="line.258"></a> |
| <span class="sourceLineNo">259</span> LOG.warn("Failed parse", e);<a name="line.259"></a> |
| <span class="sourceLineNo">260</span> // Hopefully next time around we won't fail the parse. Dangerous.<a name="line.260"></a> |
| <span class="sourceLineNo">261</span> continue;<a name="line.261"></a> |
| <span class="sourceLineNo">262</span> }<a name="line.262"></a> |
| <span class="sourceLineNo">263</span> if (ServerName.isSameAddress(currentMaster, this.sn)) {<a name="line.263"></a> |
| <span class="sourceLineNo">264</span> msg = ("Current master has this master's address, " +<a name="line.264"></a> |
| <span class="sourceLineNo">265</span> currentMaster + "; master was restarted? Deleting node.");<a name="line.265"></a> |
| <span class="sourceLineNo">266</span> // Hurry along the expiration of the znode.<a name="line.266"></a> |
| <span class="sourceLineNo">267</span> ZKUtil.deleteNode(this.watcher, this.watcher.getZNodePaths().masterAddressZNode);<a name="line.267"></a> |
| <span class="sourceLineNo">268</span><a name="line.268"></a> |
| <span class="sourceLineNo">269</span> // We may have failed to delete the znode at the previous step, but<a name="line.269"></a> |
| <span class="sourceLineNo">270</span> // we delete the file anyway: a second attempt to delete the znode is likely to fail<a name="line.270"></a> |
| <span class="sourceLineNo">271</span> // again.<a name="line.271"></a> |
| <span class="sourceLineNo">272</span> ZNodeClearer.deleteMyEphemeralNodeOnDisk();<a name="line.272"></a> |
| <span class="sourceLineNo">273</span> } else {<a name="line.273"></a> |
| <span class="sourceLineNo">274</span> msg = "Another master is the active master, " + currentMaster +<a name="line.274"></a> |
| <span class="sourceLineNo">275</span> "; waiting to become the next active master";<a name="line.275"></a> |
| <span class="sourceLineNo">276</span> }<a name="line.276"></a> |
| <span class="sourceLineNo">277</span> }<a name="line.277"></a> |
| <span class="sourceLineNo">278</span> LOG.info(msg);<a name="line.278"></a> |
| <span class="sourceLineNo">279</span> startupStatus.setStatus(msg);<a name="line.279"></a> |
| <span class="sourceLineNo">280</span> } catch (KeeperException ke) {<a name="line.280"></a> |
| <span class="sourceLineNo">281</span> master.abort("Received an unexpected KeeperException, aborting", ke);<a name="line.281"></a> |
| <span class="sourceLineNo">282</span> return false;<a name="line.282"></a> |
| <span class="sourceLineNo">283</span> }<a name="line.283"></a> |
| <span class="sourceLineNo">284</span> synchronized (this.clusterHasActiveMaster) {<a name="line.284"></a> |
| <span class="sourceLineNo">285</span> while (clusterHasActiveMaster.get() && !master.isStopped()) {<a name="line.285"></a> |
| <span class="sourceLineNo">286</span> try {<a name="line.286"></a> |
| <span class="sourceLineNo">287</span> clusterHasActiveMaster.wait(checkInterval);<a name="line.287"></a> |
| <span class="sourceLineNo">288</span> } catch (InterruptedException e) {<a name="line.288"></a> |
| <span class="sourceLineNo">289</span> // We expect to be interrupted when a master dies,<a name="line.289"></a> |
| <span class="sourceLineNo">290</span> // will fall out if so<a name="line.290"></a> |
| <span class="sourceLineNo">291</span> LOG.debug("Interrupted waiting for master to die", e);<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> if (clusterShutDown.get()) {<a name="line.294"></a> |
| <span class="sourceLineNo">295</span> this.master.stop(<a name="line.295"></a> |
| <span class="sourceLineNo">296</span> "Cluster went down before this master became active");<a name="line.296"></a> |
| <span class="sourceLineNo">297</span> }<a name="line.297"></a> |
| <span class="sourceLineNo">298</span> }<a name="line.298"></a> |
| <span class="sourceLineNo">299</span> }<a name="line.299"></a> |
| <span class="sourceLineNo">300</span> return false;<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> /**<a name="line.303"></a> |
| <span class="sourceLineNo">304</span> * @return True if cluster has an active master.<a name="line.304"></a> |
| <span class="sourceLineNo">305</span> */<a name="line.305"></a> |
| <span class="sourceLineNo">306</span> boolean hasActiveMaster() {<a name="line.306"></a> |
| <span class="sourceLineNo">307</span> try {<a name="line.307"></a> |
| <span class="sourceLineNo">308</span> if (ZKUtil.checkExists(watcher, watcher.getZNodePaths().masterAddressZNode) >= 0) {<a name="line.308"></a> |
| <span class="sourceLineNo">309</span> return true;<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> catch (KeeperException ke) {<a name="line.312"></a> |
| <span class="sourceLineNo">313</span> LOG.info("Received an unexpected KeeperException when checking " +<a name="line.313"></a> |
| <span class="sourceLineNo">314</span> "isActiveMaster : "+ ke);<a name="line.314"></a> |
| <span class="sourceLineNo">315</span> }<a name="line.315"></a> |
| <span class="sourceLineNo">316</span> return false;<a name="line.316"></a> |
| <span class="sourceLineNo">317</span> }<a name="line.317"></a> |
| <span class="sourceLineNo">318</span><a name="line.318"></a> |
| <span class="sourceLineNo">319</span> public void stop() {<a name="line.319"></a> |
| <span class="sourceLineNo">320</span> try {<a name="line.320"></a> |
| <span class="sourceLineNo">321</span> synchronized (clusterHasActiveMaster) {<a name="line.321"></a> |
| <span class="sourceLineNo">322</span> // Master is already stopped, wake up the manager<a name="line.322"></a> |
| <span class="sourceLineNo">323</span> // thread so that it can shutdown soon.<a name="line.323"></a> |
| <span class="sourceLineNo">324</span> clusterHasActiveMaster.notifyAll();<a name="line.324"></a> |
| <span class="sourceLineNo">325</span> }<a name="line.325"></a> |
| <span class="sourceLineNo">326</span> // If our address is in ZK, delete it on our way out<a name="line.326"></a> |
| <span class="sourceLineNo">327</span> ServerName activeMaster = null;<a name="line.327"></a> |
| <span class="sourceLineNo">328</span> try {<a name="line.328"></a> |
| <span class="sourceLineNo">329</span> activeMaster = MasterAddressTracker.getMasterAddress(this.watcher);<a name="line.329"></a> |
| <span class="sourceLineNo">330</span> } catch (IOException e) {<a name="line.330"></a> |
| <span class="sourceLineNo">331</span> LOG.warn("Failed get of master address: " + e.toString());<a name="line.331"></a> |
| <span class="sourceLineNo">332</span> }<a name="line.332"></a> |
| <span class="sourceLineNo">333</span> if (activeMaster != null && activeMaster.equals(this.sn)) {<a name="line.333"></a> |
| <span class="sourceLineNo">334</span> ZKUtil.deleteNode(watcher, watcher.getZNodePaths().masterAddressZNode);<a name="line.334"></a> |
| <span class="sourceLineNo">335</span> // We may have failed to delete the znode at the previous step, but<a name="line.335"></a> |
| <span class="sourceLineNo">336</span> // we delete the file anyway: a second attempt to delete the znode is likely to fail again.<a name="line.336"></a> |
| <span class="sourceLineNo">337</span> ZNodeClearer.deleteMyEphemeralNodeOnDisk();<a name="line.337"></a> |
| <span class="sourceLineNo">338</span> }<a name="line.338"></a> |
| <span class="sourceLineNo">339</span> } catch (KeeperException e) {<a name="line.339"></a> |
| <span class="sourceLineNo">340</span> LOG.debug(this.watcher.prefix("Failed delete of our master address node; " +<a name="line.340"></a> |
| <span class="sourceLineNo">341</span> e.getMessage()));<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> /**<a name="line.345"></a> |
| <span class="sourceLineNo">346</span> * @return list of registered backup masters.<a name="line.346"></a> |
| <span class="sourceLineNo">347</span> */<a name="line.347"></a> |
| <span class="sourceLineNo">348</span> public List<ServerName> getBackupMasters() {<a name="line.348"></a> |
| <span class="sourceLineNo">349</span> return backupMasters;<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> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |