blob: 3ceff590cd7baf2981d3ed39ebf36514560e05a2 [file] [log] [blame]
<!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> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a>
<span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a>
<span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a>
<span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a>
<span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a>
<span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a>
<span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a>
<span class="sourceLineNo">009</span> *<a name="line.9"></a>
<span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a>
<span class="sourceLineNo">011</span> *<a name="line.11"></a>
<span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a>
<span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a>
<span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a>
<span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a>
<span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a>
<span class="sourceLineNo">017</span> */<a name="line.17"></a>
<span class="sourceLineNo">018</span>package org.apache.hadoop.hbase;<a name="line.18"></a>
<span class="sourceLineNo">019</span><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.util.ArrayList;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import java.util.Comparator;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import java.util.EnumSet;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.HashSet;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.List;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import java.util.Objects;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import java.util.Set;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import java.util.TreeSet;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.conf.Configuration;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.ClusterManager.ServiceType;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.client.Admin;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.client.RegionLocator;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.util.Threads;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.slf4j.Logger;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.slf4j.LoggerFactory;<a name="line.41"></a>
<span class="sourceLineNo">042</span><a name="line.42"></a>
<span class="sourceLineNo">043</span>/**<a name="line.43"></a>
<span class="sourceLineNo">044</span> * Manages the interactions with an already deployed distributed cluster (as opposed to a<a name="line.44"></a>
<span class="sourceLineNo">045</span> * pseudo-distributed, or mini/local cluster). This is used by integration and system tests.<a name="line.45"></a>
<span class="sourceLineNo">046</span> */<a name="line.46"></a>
<span class="sourceLineNo">047</span>@InterfaceAudience.Private<a name="line.47"></a>
<span class="sourceLineNo">048</span>public class DistributedHBaseCluster extends HBaseClusterInterface {<a name="line.48"></a>
<span class="sourceLineNo">049</span><a name="line.49"></a>
<span class="sourceLineNo">050</span> private static final Logger LOG = LoggerFactory.getLogger(DistributedHBaseCluster.class);<a name="line.50"></a>
<span class="sourceLineNo">051</span><a name="line.51"></a>
<span class="sourceLineNo">052</span> private Admin admin;<a name="line.52"></a>
<span class="sourceLineNo">053</span> private final Connection connection;<a name="line.53"></a>
<span class="sourceLineNo">054</span><a name="line.54"></a>
<span class="sourceLineNo">055</span> private ClusterManager clusterManager;<a name="line.55"></a>
<span class="sourceLineNo">056</span> /**<a name="line.56"></a>
<span class="sourceLineNo">057</span> * List of RegionServers killed so far. ServerName also comprises startCode of a server, so any<a name="line.57"></a>
<span class="sourceLineNo">058</span> * restarted instances of the same server will have different ServerName and will not coincide<a name="line.58"></a>
<span class="sourceLineNo">059</span> * with past dead ones. So there's no need to cleanup this list.<a name="line.59"></a>
<span class="sourceLineNo">060</span> */<a name="line.60"></a>
<span class="sourceLineNo">061</span> private final Set&lt;ServerName&gt; killedRegionServers = new HashSet&lt;&gt;();<a name="line.61"></a>
<span class="sourceLineNo">062</span><a name="line.62"></a>
<span class="sourceLineNo">063</span> public DistributedHBaseCluster(Configuration conf, ClusterManager clusterManager)<a name="line.63"></a>
<span class="sourceLineNo">064</span> throws IOException {<a name="line.64"></a>
<span class="sourceLineNo">065</span> super(conf);<a name="line.65"></a>
<span class="sourceLineNo">066</span> this.clusterManager = clusterManager;<a name="line.66"></a>
<span class="sourceLineNo">067</span> this.connection = ConnectionFactory.createConnection(conf);<a name="line.67"></a>
<span class="sourceLineNo">068</span> this.admin = this.connection.getAdmin();<a name="line.68"></a>
<span class="sourceLineNo">069</span> this.initialClusterStatus = getClusterMetrics();<a name="line.69"></a>
<span class="sourceLineNo">070</span> }<a name="line.70"></a>
<span class="sourceLineNo">071</span><a name="line.71"></a>
<span class="sourceLineNo">072</span> public void setClusterManager(ClusterManager clusterManager) {<a name="line.72"></a>
<span class="sourceLineNo">073</span> this.clusterManager = clusterManager;<a name="line.73"></a>
<span class="sourceLineNo">074</span> }<a name="line.74"></a>
<span class="sourceLineNo">075</span><a name="line.75"></a>
<span class="sourceLineNo">076</span> public ClusterManager getClusterManager() {<a name="line.76"></a>
<span class="sourceLineNo">077</span> return clusterManager;<a name="line.77"></a>
<span class="sourceLineNo">078</span> }<a name="line.78"></a>
<span class="sourceLineNo">079</span><a name="line.79"></a>
<span class="sourceLineNo">080</span> /**<a name="line.80"></a>
<span class="sourceLineNo">081</span> * Returns a ClusterStatus for this HBase cluster<a name="line.81"></a>
<span class="sourceLineNo">082</span> */<a name="line.82"></a>
<span class="sourceLineNo">083</span> @Override<a name="line.83"></a>
<span class="sourceLineNo">084</span> public ClusterMetrics getClusterMetrics() throws IOException {<a name="line.84"></a>
<span class="sourceLineNo">085</span> return admin.getClusterMetrics();<a name="line.85"></a>
<span class="sourceLineNo">086</span> }<a name="line.86"></a>
<span class="sourceLineNo">087</span><a name="line.87"></a>
<span class="sourceLineNo">088</span> @Override<a name="line.88"></a>
<span class="sourceLineNo">089</span> public ClusterMetrics getInitialClusterMetrics() throws IOException {<a name="line.89"></a>
<span class="sourceLineNo">090</span> return initialClusterStatus;<a name="line.90"></a>
<span class="sourceLineNo">091</span> }<a name="line.91"></a>
<span class="sourceLineNo">092</span><a name="line.92"></a>
<span class="sourceLineNo">093</span> @Override<a name="line.93"></a>
<span class="sourceLineNo">094</span> public void close() throws IOException {<a name="line.94"></a>
<span class="sourceLineNo">095</span> if (this.admin != null) {<a name="line.95"></a>
<span class="sourceLineNo">096</span> admin.close();<a name="line.96"></a>
<span class="sourceLineNo">097</span> }<a name="line.97"></a>
<span class="sourceLineNo">098</span> if (this.connection != null &amp;&amp; !this.connection.isClosed()) {<a name="line.98"></a>
<span class="sourceLineNo">099</span> this.connection.close();<a name="line.99"></a>
<span class="sourceLineNo">100</span> }<a name="line.100"></a>
<span class="sourceLineNo">101</span> }<a name="line.101"></a>
<span class="sourceLineNo">102</span><a name="line.102"></a>
<span class="sourceLineNo">103</span> @Override<a name="line.103"></a>
<span class="sourceLineNo">104</span> public void startRegionServer(String hostname, int port) throws IOException {<a name="line.104"></a>
<span class="sourceLineNo">105</span> LOG.info("Starting RS on: {}", hostname);<a name="line.105"></a>
<span class="sourceLineNo">106</span> clusterManager.start(ServiceType.HBASE_REGIONSERVER, hostname, port);<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> @Override<a name="line.109"></a>
<span class="sourceLineNo">110</span> public void killRegionServer(ServerName serverName) throws IOException {<a name="line.110"></a>
<span class="sourceLineNo">111</span> LOG.info("Aborting RS: {}", serverName.getServerName());<a name="line.111"></a>
<span class="sourceLineNo">112</span> killedRegionServers.add(serverName);<a name="line.112"></a>
<span class="sourceLineNo">113</span> clusterManager.kill(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),<a name="line.113"></a>
<span class="sourceLineNo">114</span> serverName.getPort());<a name="line.114"></a>
<span class="sourceLineNo">115</span> }<a name="line.115"></a>
<span class="sourceLineNo">116</span><a name="line.116"></a>
<span class="sourceLineNo">117</span> @Override<a name="line.117"></a>
<span class="sourceLineNo">118</span> public boolean isKilledRS(ServerName serverName) {<a name="line.118"></a>
<span class="sourceLineNo">119</span> return killedRegionServers.contains(serverName);<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> @Override<a name="line.122"></a>
<span class="sourceLineNo">123</span> public void stopRegionServer(ServerName serverName) throws IOException {<a name="line.123"></a>
<span class="sourceLineNo">124</span> LOG.info("Stopping RS: {}", serverName.getServerName());<a name="line.124"></a>
<span class="sourceLineNo">125</span> clusterManager.stop(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),<a name="line.125"></a>
<span class="sourceLineNo">126</span> serverName.getPort());<a name="line.126"></a>
<span class="sourceLineNo">127</span> }<a name="line.127"></a>
<span class="sourceLineNo">128</span><a name="line.128"></a>
<span class="sourceLineNo">129</span> @Override<a name="line.129"></a>
<span class="sourceLineNo">130</span> public void waitForRegionServerToStop(ServerName serverName, long timeout) throws IOException {<a name="line.130"></a>
<span class="sourceLineNo">131</span> waitForServiceToStop(ServiceType.HBASE_REGIONSERVER, serverName, timeout);<a name="line.131"></a>
<span class="sourceLineNo">132</span> }<a name="line.132"></a>
<span class="sourceLineNo">133</span><a name="line.133"></a>
<span class="sourceLineNo">134</span> @Override<a name="line.134"></a>
<span class="sourceLineNo">135</span> public void suspendRegionServer(ServerName serverName) throws IOException {<a name="line.135"></a>
<span class="sourceLineNo">136</span> LOG.info("Suspend RS: {}", serverName.getServerName());<a name="line.136"></a>
<span class="sourceLineNo">137</span> clusterManager.suspend(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),<a name="line.137"></a>
<span class="sourceLineNo">138</span> serverName.getPort());<a name="line.138"></a>
<span class="sourceLineNo">139</span> }<a name="line.139"></a>
<span class="sourceLineNo">140</span><a name="line.140"></a>
<span class="sourceLineNo">141</span> @Override<a name="line.141"></a>
<span class="sourceLineNo">142</span> public void resumeRegionServer(ServerName serverName) throws IOException {<a name="line.142"></a>
<span class="sourceLineNo">143</span> LOG.info("Resume RS: {}", serverName.getServerName());<a name="line.143"></a>
<span class="sourceLineNo">144</span> clusterManager.resume(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),<a name="line.144"></a>
<span class="sourceLineNo">145</span> serverName.getPort());<a name="line.145"></a>
<span class="sourceLineNo">146</span> }<a name="line.146"></a>
<span class="sourceLineNo">147</span><a name="line.147"></a>
<span class="sourceLineNo">148</span> @Override<a name="line.148"></a>
<span class="sourceLineNo">149</span> public void startZkNode(String hostname, int port) throws IOException {<a name="line.149"></a>
<span class="sourceLineNo">150</span> LOG.info("Starting ZooKeeper node on: {}", hostname);<a name="line.150"></a>
<span class="sourceLineNo">151</span> clusterManager.start(ServiceType.ZOOKEEPER_SERVER, hostname, port);<a name="line.151"></a>
<span class="sourceLineNo">152</span> }<a name="line.152"></a>
<span class="sourceLineNo">153</span><a name="line.153"></a>
<span class="sourceLineNo">154</span> @Override<a name="line.154"></a>
<span class="sourceLineNo">155</span> public void killZkNode(ServerName serverName) throws IOException {<a name="line.155"></a>
<span class="sourceLineNo">156</span> LOG.info("Aborting ZooKeeper node on: {}", serverName.getServerName());<a name="line.156"></a>
<span class="sourceLineNo">157</span> clusterManager.kill(ServiceType.ZOOKEEPER_SERVER, serverName.getHostname(),<a name="line.157"></a>
<span class="sourceLineNo">158</span> serverName.getPort());<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> @Override<a name="line.161"></a>
<span class="sourceLineNo">162</span> public void stopZkNode(ServerName serverName) throws IOException {<a name="line.162"></a>
<span class="sourceLineNo">163</span> LOG.info("Stopping ZooKeeper node: {}", serverName.getServerName());<a name="line.163"></a>
<span class="sourceLineNo">164</span> clusterManager.stop(ServiceType.ZOOKEEPER_SERVER, serverName.getHostname(),<a name="line.164"></a>
<span class="sourceLineNo">165</span> serverName.getPort());<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> @Override<a name="line.168"></a>
<span class="sourceLineNo">169</span> public void waitForZkNodeToStart(ServerName serverName, long timeout) throws IOException {<a name="line.169"></a>
<span class="sourceLineNo">170</span> waitForServiceToStart(ServiceType.ZOOKEEPER_SERVER, serverName, timeout);<a name="line.170"></a>
<span class="sourceLineNo">171</span> }<a name="line.171"></a>
<span class="sourceLineNo">172</span><a name="line.172"></a>
<span class="sourceLineNo">173</span> @Override<a name="line.173"></a>
<span class="sourceLineNo">174</span> public void waitForZkNodeToStop(ServerName serverName, long timeout) throws IOException {<a name="line.174"></a>
<span class="sourceLineNo">175</span> waitForServiceToStop(ServiceType.ZOOKEEPER_SERVER, serverName, timeout);<a name="line.175"></a>
<span class="sourceLineNo">176</span> }<a name="line.176"></a>
<span class="sourceLineNo">177</span><a name="line.177"></a>
<span class="sourceLineNo">178</span> @Override<a name="line.178"></a>
<span class="sourceLineNo">179</span> public void startDataNode(ServerName serverName) throws IOException {<a name="line.179"></a>
<span class="sourceLineNo">180</span> LOG.info("Starting data node on: {}", serverName.getServerName());<a name="line.180"></a>
<span class="sourceLineNo">181</span> clusterManager.start(ServiceType.HADOOP_DATANODE, serverName.getHostname(),<a name="line.181"></a>
<span class="sourceLineNo">182</span> serverName.getPort());<a name="line.182"></a>
<span class="sourceLineNo">183</span> }<a name="line.183"></a>
<span class="sourceLineNo">184</span><a name="line.184"></a>
<span class="sourceLineNo">185</span> @Override<a name="line.185"></a>
<span class="sourceLineNo">186</span> public void killDataNode(ServerName serverName) throws IOException {<a name="line.186"></a>
<span class="sourceLineNo">187</span> LOG.info("Aborting data node on: {}", serverName.getServerName());<a name="line.187"></a>
<span class="sourceLineNo">188</span> clusterManager.kill(ServiceType.HADOOP_DATANODE, serverName.getHostname(),<a name="line.188"></a>
<span class="sourceLineNo">189</span> serverName.getPort());<a name="line.189"></a>
<span class="sourceLineNo">190</span> }<a name="line.190"></a>
<span class="sourceLineNo">191</span><a name="line.191"></a>
<span class="sourceLineNo">192</span> @Override<a name="line.192"></a>
<span class="sourceLineNo">193</span> public void stopDataNode(ServerName serverName) throws IOException {<a name="line.193"></a>
<span class="sourceLineNo">194</span> LOG.info("Stopping data node on: {}", serverName.getServerName());<a name="line.194"></a>
<span class="sourceLineNo">195</span> clusterManager.stop(ServiceType.HADOOP_DATANODE, serverName.getHostname(),<a name="line.195"></a>
<span class="sourceLineNo">196</span> serverName.getPort());<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> @Override<a name="line.199"></a>
<span class="sourceLineNo">200</span> public void waitForDataNodeToStart(ServerName serverName, long timeout) throws IOException {<a name="line.200"></a>
<span class="sourceLineNo">201</span> waitForServiceToStart(ServiceType.HADOOP_DATANODE, serverName, timeout);<a name="line.201"></a>
<span class="sourceLineNo">202</span> }<a name="line.202"></a>
<span class="sourceLineNo">203</span><a name="line.203"></a>
<span class="sourceLineNo">204</span> @Override<a name="line.204"></a>
<span class="sourceLineNo">205</span> public void waitForDataNodeToStop(ServerName serverName, long timeout) throws IOException {<a name="line.205"></a>
<span class="sourceLineNo">206</span> waitForServiceToStop(ServiceType.HADOOP_DATANODE, serverName, timeout);<a name="line.206"></a>
<span class="sourceLineNo">207</span> }<a name="line.207"></a>
<span class="sourceLineNo">208</span><a name="line.208"></a>
<span class="sourceLineNo">209</span> @Override<a name="line.209"></a>
<span class="sourceLineNo">210</span> public void startNameNode(ServerName serverName) throws IOException {<a name="line.210"></a>
<span class="sourceLineNo">211</span> LOG.info("Starting name node on: {}", serverName.getServerName());<a name="line.211"></a>
<span class="sourceLineNo">212</span> clusterManager.start(ServiceType.HADOOP_NAMENODE, serverName.getHostname(),<a name="line.212"></a>
<span class="sourceLineNo">213</span> serverName.getPort());<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> @Override<a name="line.216"></a>
<span class="sourceLineNo">217</span> public void killNameNode(ServerName serverName) throws IOException {<a name="line.217"></a>
<span class="sourceLineNo">218</span> LOG.info("Aborting name node on: {}", serverName.getServerName());<a name="line.218"></a>
<span class="sourceLineNo">219</span> clusterManager.kill(ServiceType.HADOOP_NAMENODE, serverName.getHostname(),<a name="line.219"></a>
<span class="sourceLineNo">220</span> serverName.getPort());<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> @Override<a name="line.223"></a>
<span class="sourceLineNo">224</span> public void stopNameNode(ServerName serverName) throws IOException {<a name="line.224"></a>
<span class="sourceLineNo">225</span> LOG.info("Stopping name node on: {}", serverName.getServerName());<a name="line.225"></a>
<span class="sourceLineNo">226</span> clusterManager.stop(ServiceType.HADOOP_NAMENODE, serverName.getHostname(),<a name="line.226"></a>
<span class="sourceLineNo">227</span> serverName.getPort());<a name="line.227"></a>
<span class="sourceLineNo">228</span> }<a name="line.228"></a>
<span class="sourceLineNo">229</span><a name="line.229"></a>
<span class="sourceLineNo">230</span> @Override<a name="line.230"></a>
<span class="sourceLineNo">231</span> public void waitForNameNodeToStart(ServerName serverName, long timeout) throws IOException {<a name="line.231"></a>
<span class="sourceLineNo">232</span> waitForServiceToStart(ServiceType.HADOOP_NAMENODE, serverName, timeout);<a name="line.232"></a>
<span class="sourceLineNo">233</span> }<a name="line.233"></a>
<span class="sourceLineNo">234</span><a name="line.234"></a>
<span class="sourceLineNo">235</span> @Override<a name="line.235"></a>
<span class="sourceLineNo">236</span> public void waitForNameNodeToStop(ServerName serverName, long timeout) throws IOException {<a name="line.236"></a>
<span class="sourceLineNo">237</span> waitForServiceToStop(ServiceType.HADOOP_NAMENODE, serverName, timeout);<a name="line.237"></a>
<span class="sourceLineNo">238</span> }<a name="line.238"></a>
<span class="sourceLineNo">239</span><a name="line.239"></a>
<span class="sourceLineNo">240</span> @Override<a name="line.240"></a>
<span class="sourceLineNo">241</span> public void startJournalNode(ServerName serverName) throws IOException {<a name="line.241"></a>
<span class="sourceLineNo">242</span> LOG.info("Starting journal node on: {}", serverName.getServerName());<a name="line.242"></a>
<span class="sourceLineNo">243</span> clusterManager.start(ServiceType.HADOOP_JOURNALNODE, serverName.getHostname(),<a name="line.243"></a>
<span class="sourceLineNo">244</span> serverName.getPort());<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> @Override<a name="line.247"></a>
<span class="sourceLineNo">248</span> public void killJournalNode(ServerName serverName) throws IOException {<a name="line.248"></a>
<span class="sourceLineNo">249</span> LOG.info("Aborting journal node on: {}", serverName.getServerName());<a name="line.249"></a>
<span class="sourceLineNo">250</span> clusterManager.kill(ServiceType.HADOOP_JOURNALNODE, serverName.getHostname(),<a name="line.250"></a>
<span class="sourceLineNo">251</span> serverName.getPort());<a name="line.251"></a>
<span class="sourceLineNo">252</span> }<a name="line.252"></a>
<span class="sourceLineNo">253</span><a name="line.253"></a>
<span class="sourceLineNo">254</span> @Override<a name="line.254"></a>
<span class="sourceLineNo">255</span> public void stopJournalNode(ServerName serverName) throws IOException {<a name="line.255"></a>
<span class="sourceLineNo">256</span> LOG.info("Stopping journal node on: {}", serverName.getServerName());<a name="line.256"></a>
<span class="sourceLineNo">257</span> clusterManager.stop(ServiceType.HADOOP_JOURNALNODE, serverName.getHostname(),<a name="line.257"></a>
<span class="sourceLineNo">258</span> serverName.getPort());<a name="line.258"></a>
<span class="sourceLineNo">259</span> }<a name="line.259"></a>
<span class="sourceLineNo">260</span><a name="line.260"></a>
<span class="sourceLineNo">261</span> @Override<a name="line.261"></a>
<span class="sourceLineNo">262</span> public void waitForJournalNodeToStart(ServerName serverName, long timeout) throws IOException {<a name="line.262"></a>
<span class="sourceLineNo">263</span> waitForServiceToStart(ServiceType.HADOOP_JOURNALNODE, serverName, timeout);<a name="line.263"></a>
<span class="sourceLineNo">264</span> }<a name="line.264"></a>
<span class="sourceLineNo">265</span><a name="line.265"></a>
<span class="sourceLineNo">266</span> @Override<a name="line.266"></a>
<span class="sourceLineNo">267</span> public void waitForJournalNodeToStop(ServerName serverName, long timeout) throws IOException {<a name="line.267"></a>
<span class="sourceLineNo">268</span> waitForServiceToStop(ServiceType.HADOOP_JOURNALNODE, serverName, timeout);<a name="line.268"></a>
<span class="sourceLineNo">269</span> }<a name="line.269"></a>
<span class="sourceLineNo">270</span><a name="line.270"></a>
<span class="sourceLineNo">271</span> private void waitForServiceToStop(ServiceType service, ServerName serverName, long timeout)<a name="line.271"></a>
<span class="sourceLineNo">272</span> throws IOException {<a name="line.272"></a>
<span class="sourceLineNo">273</span> LOG.info("Waiting for service: {} to stop: {}", service, serverName.getServerName());<a name="line.273"></a>
<span class="sourceLineNo">274</span> long start = EnvironmentEdgeManager.currentTime();<a name="line.274"></a>
<span class="sourceLineNo">275</span><a name="line.275"></a>
<span class="sourceLineNo">276</span> while ((EnvironmentEdgeManager.currentTime() - start) &lt; timeout) {<a name="line.276"></a>
<span class="sourceLineNo">277</span> if (!clusterManager.isRunning(service, serverName.getHostname(), serverName.getPort())) {<a name="line.277"></a>
<span class="sourceLineNo">278</span> return;<a name="line.278"></a>
<span class="sourceLineNo">279</span> }<a name="line.279"></a>
<span class="sourceLineNo">280</span> Threads.sleep(100);<a name="line.280"></a>
<span class="sourceLineNo">281</span> }<a name="line.281"></a>
<span class="sourceLineNo">282</span> throw new IOException("Timed-out waiting for service to stop: " + serverName);<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> private void waitForServiceToStart(ServiceType service, ServerName serverName, long timeout)<a name="line.285"></a>
<span class="sourceLineNo">286</span> throws IOException {<a name="line.286"></a>
<span class="sourceLineNo">287</span> LOG.info("Waiting for service: {} to start: {}", service, serverName.getServerName());<a name="line.287"></a>
<span class="sourceLineNo">288</span> long start = EnvironmentEdgeManager.currentTime();<a name="line.288"></a>
<span class="sourceLineNo">289</span><a name="line.289"></a>
<span class="sourceLineNo">290</span> while ((EnvironmentEdgeManager.currentTime() - start) &lt; timeout) {<a name="line.290"></a>
<span class="sourceLineNo">291</span> if (clusterManager.isRunning(service, serverName.getHostname(), serverName.getPort())) {<a name="line.291"></a>
<span class="sourceLineNo">292</span> return;<a name="line.292"></a>
<span class="sourceLineNo">293</span> }<a name="line.293"></a>
<span class="sourceLineNo">294</span> Threads.sleep(100);<a name="line.294"></a>
<span class="sourceLineNo">295</span> }<a name="line.295"></a>
<span class="sourceLineNo">296</span> throw new IOException("Timed-out waiting for service to start: " + serverName);<a name="line.296"></a>
<span class="sourceLineNo">297</span> }<a name="line.297"></a>
<span class="sourceLineNo">298</span><a name="line.298"></a>
<span class="sourceLineNo">299</span> @Override<a name="line.299"></a>
<span class="sourceLineNo">300</span> public void startMaster(String hostname, int port) throws IOException {<a name="line.300"></a>
<span class="sourceLineNo">301</span> LOG.info("Starting Master on: {}:{}", hostname, port);<a name="line.301"></a>
<span class="sourceLineNo">302</span> clusterManager.start(ServiceType.HBASE_MASTER, hostname, port);<a name="line.302"></a>
<span class="sourceLineNo">303</span> }<a name="line.303"></a>
<span class="sourceLineNo">304</span><a name="line.304"></a>
<span class="sourceLineNo">305</span> @Override<a name="line.305"></a>
<span class="sourceLineNo">306</span> public void killMaster(ServerName serverName) throws IOException {<a name="line.306"></a>
<span class="sourceLineNo">307</span> LOG.info("Aborting Master: {}", serverName.getServerName());<a name="line.307"></a>
<span class="sourceLineNo">308</span> clusterManager.kill(ServiceType.HBASE_MASTER, serverName.getHostname(), serverName.getPort());<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> @Override<a name="line.311"></a>
<span class="sourceLineNo">312</span> public void stopMaster(ServerName serverName) throws IOException {<a name="line.312"></a>
<span class="sourceLineNo">313</span> LOG.info("Stopping Master: {}", serverName.getServerName());<a name="line.313"></a>
<span class="sourceLineNo">314</span> clusterManager.stop(ServiceType.HBASE_MASTER, serverName.getHostname(), serverName.getPort());<a name="line.314"></a>
<span class="sourceLineNo">315</span> }<a name="line.315"></a>
<span class="sourceLineNo">316</span><a name="line.316"></a>
<span class="sourceLineNo">317</span> @Override<a name="line.317"></a>
<span class="sourceLineNo">318</span> public void waitForMasterToStop(ServerName serverName, long timeout) throws IOException {<a name="line.318"></a>
<span class="sourceLineNo">319</span> waitForServiceToStop(ServiceType.HBASE_MASTER, serverName, timeout);<a name="line.319"></a>
<span class="sourceLineNo">320</span> }<a name="line.320"></a>
<span class="sourceLineNo">321</span><a name="line.321"></a>
<span class="sourceLineNo">322</span> @Override<a name="line.322"></a>
<span class="sourceLineNo">323</span> public boolean waitForActiveAndReadyMaster(long timeout) throws IOException {<a name="line.323"></a>
<span class="sourceLineNo">324</span> long start = EnvironmentEdgeManager.currentTime();<a name="line.324"></a>
<span class="sourceLineNo">325</span> while (EnvironmentEdgeManager.currentTime() - start &lt; timeout) {<a name="line.325"></a>
<span class="sourceLineNo">326</span> try {<a name="line.326"></a>
<span class="sourceLineNo">327</span> connection.getAdmin().getClusterMetrics(EnumSet.of(ClusterMetrics.Option.HBASE_VERSION));<a name="line.327"></a>
<span class="sourceLineNo">328</span> return true;<a name="line.328"></a>
<span class="sourceLineNo">329</span> } catch (MasterNotRunningException m) {<a name="line.329"></a>
<span class="sourceLineNo">330</span> LOG.warn("Master not started yet " + m);<a name="line.330"></a>
<span class="sourceLineNo">331</span> } catch (ZooKeeperConnectionException e) {<a name="line.331"></a>
<span class="sourceLineNo">332</span> LOG.warn("Failed to connect to ZK " + e);<a name="line.332"></a>
<span class="sourceLineNo">333</span> }<a name="line.333"></a>
<span class="sourceLineNo">334</span> Threads.sleep(1000);<a name="line.334"></a>
<span class="sourceLineNo">335</span> }<a name="line.335"></a>
<span class="sourceLineNo">336</span> return false;<a name="line.336"></a>
<span class="sourceLineNo">337</span> }<a name="line.337"></a>
<span class="sourceLineNo">338</span><a name="line.338"></a>
<span class="sourceLineNo">339</span> @Override<a name="line.339"></a>
<span class="sourceLineNo">340</span> public ServerName getServerHoldingRegion(TableName tn, byte[] regionName) throws IOException {<a name="line.340"></a>
<span class="sourceLineNo">341</span> byte[] startKey = RegionInfo.getStartKey(regionName);<a name="line.341"></a>
<span class="sourceLineNo">342</span> HRegionLocation regionLoc = null;<a name="line.342"></a>
<span class="sourceLineNo">343</span> try (RegionLocator locator = connection.getRegionLocator(tn)) {<a name="line.343"></a>
<span class="sourceLineNo">344</span> regionLoc = locator.getRegionLocation(startKey, true);<a name="line.344"></a>
<span class="sourceLineNo">345</span> }<a name="line.345"></a>
<span class="sourceLineNo">346</span> if (regionLoc == null) {<a name="line.346"></a>
<span class="sourceLineNo">347</span> LOG.warn("Cannot find region server holding region {}", Bytes.toStringBinary(regionName));<a name="line.347"></a>
<span class="sourceLineNo">348</span> return null;<a name="line.348"></a>
<span class="sourceLineNo">349</span> }<a name="line.349"></a>
<span class="sourceLineNo">350</span> return regionLoc.getServerName();<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> @Override<a name="line.353"></a>
<span class="sourceLineNo">354</span> public void waitUntilShutDown() {<a name="line.354"></a>
<span class="sourceLineNo">355</span> // Simply wait for a few seconds for now (after issuing serverManager.kill<a name="line.355"></a>
<span class="sourceLineNo">356</span> throw new RuntimeException(HConstants.NOT_IMPLEMENTED);<a name="line.356"></a>
<span class="sourceLineNo">357</span> }<a name="line.357"></a>
<span class="sourceLineNo">358</span><a name="line.358"></a>
<span class="sourceLineNo">359</span> @Override<a name="line.359"></a>
<span class="sourceLineNo">360</span> public void shutdown() throws IOException {<a name="line.360"></a>
<span class="sourceLineNo">361</span> // not sure we want this<a name="line.361"></a>
<span class="sourceLineNo">362</span> throw new RuntimeException(HConstants.NOT_IMPLEMENTED);<a name="line.362"></a>
<span class="sourceLineNo">363</span> }<a name="line.363"></a>
<span class="sourceLineNo">364</span><a name="line.364"></a>
<span class="sourceLineNo">365</span> @Override<a name="line.365"></a>
<span class="sourceLineNo">366</span> public boolean isDistributedCluster() {<a name="line.366"></a>
<span class="sourceLineNo">367</span> return true;<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> @Override<a name="line.370"></a>
<span class="sourceLineNo">371</span> public boolean restoreClusterMetrics(ClusterMetrics initial) throws IOException {<a name="line.371"></a>
<span class="sourceLineNo">372</span> ClusterMetrics current = getClusterMetrics();<a name="line.372"></a>
<span class="sourceLineNo">373</span><a name="line.373"></a>
<span class="sourceLineNo">374</span> LOG.info("Restoring cluster - started");<a name="line.374"></a>
<span class="sourceLineNo">375</span><a name="line.375"></a>
<span class="sourceLineNo">376</span> // do a best effort restore<a name="line.376"></a>
<span class="sourceLineNo">377</span> boolean success = restoreMasters(initial, current);<a name="line.377"></a>
<span class="sourceLineNo">378</span> success = restoreRegionServers(initial, current) &amp;&amp; success;<a name="line.378"></a>
<span class="sourceLineNo">379</span> success = restoreAdmin() &amp;&amp; success;<a name="line.379"></a>
<span class="sourceLineNo">380</span><a name="line.380"></a>
<span class="sourceLineNo">381</span> LOG.info("Restoring cluster - done");<a name="line.381"></a>
<span class="sourceLineNo">382</span> return success;<a name="line.382"></a>
<span class="sourceLineNo">383</span> }<a name="line.383"></a>
<span class="sourceLineNo">384</span><a name="line.384"></a>
<span class="sourceLineNo">385</span> protected boolean restoreMasters(ClusterMetrics initial, ClusterMetrics current) {<a name="line.385"></a>
<span class="sourceLineNo">386</span> List&lt;IOException&gt; deferred = new ArrayList&lt;&gt;();<a name="line.386"></a>
<span class="sourceLineNo">387</span> // check whether current master has changed<a name="line.387"></a>
<span class="sourceLineNo">388</span> final ServerName initMaster = initial.getMasterName();<a name="line.388"></a>
<span class="sourceLineNo">389</span> if (!ServerName.isSameAddress(initMaster, current.getMasterName())) {<a name="line.389"></a>
<span class="sourceLineNo">390</span> LOG.info("Restoring cluster - Initial active master : {} has changed to : {}",<a name="line.390"></a>
<span class="sourceLineNo">391</span> initMaster.getAddress(), current.getMasterName().getAddress());<a name="line.391"></a>
<span class="sourceLineNo">392</span> // If initial master is stopped, start it, before restoring the state.<a name="line.392"></a>
<span class="sourceLineNo">393</span> // It will come up as a backup master, if there is already an active master.<a name="line.393"></a>
<span class="sourceLineNo">394</span> try {<a name="line.394"></a>
<span class="sourceLineNo">395</span> if (<a name="line.395"></a>
<span class="sourceLineNo">396</span> !clusterManager.isRunning(ServiceType.HBASE_MASTER, initMaster.getHostname(),<a name="line.396"></a>
<span class="sourceLineNo">397</span> initMaster.getPort())<a name="line.397"></a>
<span class="sourceLineNo">398</span> ) {<a name="line.398"></a>
<span class="sourceLineNo">399</span> LOG.info("Restoring cluster - starting initial active master at:{}",<a name="line.399"></a>
<span class="sourceLineNo">400</span> initMaster.getAddress());<a name="line.400"></a>
<span class="sourceLineNo">401</span> startMaster(initMaster.getHostname(), initMaster.getPort());<a name="line.401"></a>
<span class="sourceLineNo">402</span> }<a name="line.402"></a>
<span class="sourceLineNo">403</span><a name="line.403"></a>
<span class="sourceLineNo">404</span> // master has changed, we would like to undo this.<a name="line.404"></a>
<span class="sourceLineNo">405</span> // 1. Kill the current backups<a name="line.405"></a>
<span class="sourceLineNo">406</span> // 2. Stop current master<a name="line.406"></a>
<span class="sourceLineNo">407</span> // 3. Start backup masters<a name="line.407"></a>
<span class="sourceLineNo">408</span> for (ServerName currentBackup : current.getBackupMasterNames()) {<a name="line.408"></a>
<span class="sourceLineNo">409</span> if (!ServerName.isSameAddress(currentBackup, initMaster)) {<a name="line.409"></a>
<span class="sourceLineNo">410</span> LOG.info("Restoring cluster - stopping backup master: {}", currentBackup);<a name="line.410"></a>
<span class="sourceLineNo">411</span> stopMaster(currentBackup);<a name="line.411"></a>
<span class="sourceLineNo">412</span> }<a name="line.412"></a>
<span class="sourceLineNo">413</span> }<a name="line.413"></a>
<span class="sourceLineNo">414</span> LOG.info("Restoring cluster - stopping active master: {}", current.getMasterName());<a name="line.414"></a>
<span class="sourceLineNo">415</span> stopMaster(current.getMasterName());<a name="line.415"></a>
<span class="sourceLineNo">416</span> waitForActiveAndReadyMaster(); // wait so that active master takes over<a name="line.416"></a>
<span class="sourceLineNo">417</span> } catch (IOException ex) {<a name="line.417"></a>
<span class="sourceLineNo">418</span> // if we fail to start the initial active master, we do not want to continue stopping<a name="line.418"></a>
<span class="sourceLineNo">419</span> // backup masters. Just keep what we have now<a name="line.419"></a>
<span class="sourceLineNo">420</span> deferred.add(ex);<a name="line.420"></a>
<span class="sourceLineNo">421</span> }<a name="line.421"></a>
<span class="sourceLineNo">422</span><a name="line.422"></a>
<span class="sourceLineNo">423</span> // start backup masters<a name="line.423"></a>
<span class="sourceLineNo">424</span> for (ServerName backup : initial.getBackupMasterNames()) {<a name="line.424"></a>
<span class="sourceLineNo">425</span> try {<a name="line.425"></a>
<span class="sourceLineNo">426</span> // these are not started in backup mode, but we should already have an active master<a name="line.426"></a>
<span class="sourceLineNo">427</span> if (<a name="line.427"></a>
<span class="sourceLineNo">428</span> !clusterManager.isRunning(ServiceType.HBASE_MASTER, backup.getHostname(),<a name="line.428"></a>
<span class="sourceLineNo">429</span> backup.getPort())<a name="line.429"></a>
<span class="sourceLineNo">430</span> ) {<a name="line.430"></a>
<span class="sourceLineNo">431</span> LOG.info("Restoring cluster - starting initial backup master: {}", backup.getAddress());<a name="line.431"></a>
<span class="sourceLineNo">432</span> startMaster(backup.getHostname(), backup.getPort());<a name="line.432"></a>
<span class="sourceLineNo">433</span> }<a name="line.433"></a>
<span class="sourceLineNo">434</span> } catch (IOException ex) {<a name="line.434"></a>
<span class="sourceLineNo">435</span> deferred.add(ex);<a name="line.435"></a>
<span class="sourceLineNo">436</span> }<a name="line.436"></a>
<span class="sourceLineNo">437</span> }<a name="line.437"></a>
<span class="sourceLineNo">438</span> } else {<a name="line.438"></a>
<span class="sourceLineNo">439</span> // current master has not changed, match up backup masters<a name="line.439"></a>
<span class="sourceLineNo">440</span> Set&lt;ServerName&gt; toStart = new TreeSet&lt;&gt;(new ServerNameIgnoreStartCodeComparator());<a name="line.440"></a>
<span class="sourceLineNo">441</span> Set&lt;ServerName&gt; toKill = new TreeSet&lt;&gt;(new ServerNameIgnoreStartCodeComparator());<a name="line.441"></a>
<span class="sourceLineNo">442</span> toStart.addAll(initial.getBackupMasterNames());<a name="line.442"></a>
<span class="sourceLineNo">443</span> toKill.addAll(current.getBackupMasterNames());<a name="line.443"></a>
<span class="sourceLineNo">444</span><a name="line.444"></a>
<span class="sourceLineNo">445</span> for (ServerName server : current.getBackupMasterNames()) {<a name="line.445"></a>
<span class="sourceLineNo">446</span> toStart.remove(server);<a name="line.446"></a>
<span class="sourceLineNo">447</span> }<a name="line.447"></a>
<span class="sourceLineNo">448</span> for (ServerName server : initial.getBackupMasterNames()) {<a name="line.448"></a>
<span class="sourceLineNo">449</span> toKill.remove(server);<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> for (ServerName sn : toStart) {<a name="line.452"></a>
<span class="sourceLineNo">453</span> try {<a name="line.453"></a>
<span class="sourceLineNo">454</span> if (!clusterManager.isRunning(ServiceType.HBASE_MASTER, sn.getHostname(), sn.getPort())) {<a name="line.454"></a>
<span class="sourceLineNo">455</span> LOG.info("Restoring cluster - starting initial backup master: {}", sn.getAddress());<a name="line.455"></a>
<span class="sourceLineNo">456</span> startMaster(sn.getHostname(), sn.getPort());<a name="line.456"></a>
<span class="sourceLineNo">457</span> }<a name="line.457"></a>
<span class="sourceLineNo">458</span> } catch (IOException ex) {<a name="line.458"></a>
<span class="sourceLineNo">459</span> deferred.add(ex);<a name="line.459"></a>
<span class="sourceLineNo">460</span> }<a name="line.460"></a>
<span class="sourceLineNo">461</span> }<a name="line.461"></a>
<span class="sourceLineNo">462</span><a name="line.462"></a>
<span class="sourceLineNo">463</span> for (ServerName sn : toKill) {<a name="line.463"></a>
<span class="sourceLineNo">464</span> try {<a name="line.464"></a>
<span class="sourceLineNo">465</span> if (clusterManager.isRunning(ServiceType.HBASE_MASTER, sn.getHostname(), sn.getPort())) {<a name="line.465"></a>
<span class="sourceLineNo">466</span> LOG.info("Restoring cluster - stopping backup master: {}", sn.getAddress());<a name="line.466"></a>
<span class="sourceLineNo">467</span> stopMaster(sn);<a name="line.467"></a>
<span class="sourceLineNo">468</span> }<a name="line.468"></a>
<span class="sourceLineNo">469</span> } catch (IOException ex) {<a name="line.469"></a>
<span class="sourceLineNo">470</span> deferred.add(ex);<a name="line.470"></a>
<span class="sourceLineNo">471</span> }<a name="line.471"></a>
<span class="sourceLineNo">472</span> }<a name="line.472"></a>
<span class="sourceLineNo">473</span> }<a name="line.473"></a>
<span class="sourceLineNo">474</span> if (!deferred.isEmpty()) {<a name="line.474"></a>
<span class="sourceLineNo">475</span> LOG.warn("Restoring cluster - restoring region servers reported {} errors:", deferred.size());<a name="line.475"></a>
<span class="sourceLineNo">476</span> for (int i = 0; i &lt; deferred.size() &amp;&amp; i &lt; 3; i++) {<a name="line.476"></a>
<span class="sourceLineNo">477</span> LOG.warn(Objects.toString(deferred.get(i)));<a name="line.477"></a>
<span class="sourceLineNo">478</span> }<a name="line.478"></a>
<span class="sourceLineNo">479</span> }<a name="line.479"></a>
<span class="sourceLineNo">480</span><a name="line.480"></a>
<span class="sourceLineNo">481</span> return deferred.isEmpty();<a name="line.481"></a>
<span class="sourceLineNo">482</span> }<a name="line.482"></a>
<span class="sourceLineNo">483</span><a name="line.483"></a>
<span class="sourceLineNo">484</span> private static class ServerNameIgnoreStartCodeComparator implements Comparator&lt;ServerName&gt; {<a name="line.484"></a>
<span class="sourceLineNo">485</span> @Override<a name="line.485"></a>
<span class="sourceLineNo">486</span> public int compare(ServerName o1, ServerName o2) {<a name="line.486"></a>
<span class="sourceLineNo">487</span> int compare = o1.getHostname().compareToIgnoreCase(o2.getHostname());<a name="line.487"></a>
<span class="sourceLineNo">488</span> if (compare != 0) return compare;<a name="line.488"></a>
<span class="sourceLineNo">489</span> compare = o1.getPort() - o2.getPort();<a name="line.489"></a>
<span class="sourceLineNo">490</span> if (compare != 0) return compare;<a name="line.490"></a>
<span class="sourceLineNo">491</span> return 0;<a name="line.491"></a>
<span class="sourceLineNo">492</span> }<a name="line.492"></a>
<span class="sourceLineNo">493</span> }<a name="line.493"></a>
<span class="sourceLineNo">494</span><a name="line.494"></a>
<span class="sourceLineNo">495</span> protected boolean restoreRegionServers(ClusterMetrics initial, ClusterMetrics current) {<a name="line.495"></a>
<span class="sourceLineNo">496</span> Set&lt;ServerName&gt; toStart = new TreeSet&lt;&gt;(new ServerNameIgnoreStartCodeComparator());<a name="line.496"></a>
<span class="sourceLineNo">497</span> Set&lt;ServerName&gt; toKill = new TreeSet&lt;&gt;(new ServerNameIgnoreStartCodeComparator());<a name="line.497"></a>
<span class="sourceLineNo">498</span> toStart.addAll(initial.getLiveServerMetrics().keySet());<a name="line.498"></a>
<span class="sourceLineNo">499</span> toKill.addAll(current.getLiveServerMetrics().keySet());<a name="line.499"></a>
<span class="sourceLineNo">500</span><a name="line.500"></a>
<span class="sourceLineNo">501</span> ServerName master = initial.getMasterName();<a name="line.501"></a>
<span class="sourceLineNo">502</span><a name="line.502"></a>
<span class="sourceLineNo">503</span> for (ServerName server : current.getLiveServerMetrics().keySet()) {<a name="line.503"></a>
<span class="sourceLineNo">504</span> toStart.remove(server);<a name="line.504"></a>
<span class="sourceLineNo">505</span> }<a name="line.505"></a>
<span class="sourceLineNo">506</span> for (ServerName server : initial.getLiveServerMetrics().keySet()) {<a name="line.506"></a>
<span class="sourceLineNo">507</span> toKill.remove(server);<a name="line.507"></a>
<span class="sourceLineNo">508</span> }<a name="line.508"></a>
<span class="sourceLineNo">509</span><a name="line.509"></a>
<span class="sourceLineNo">510</span> List&lt;IOException&gt; deferred = new ArrayList&lt;&gt;();<a name="line.510"></a>
<span class="sourceLineNo">511</span><a name="line.511"></a>
<span class="sourceLineNo">512</span> for (ServerName sn : toStart) {<a name="line.512"></a>
<span class="sourceLineNo">513</span> try {<a name="line.513"></a>
<span class="sourceLineNo">514</span> if (<a name="line.514"></a>
<span class="sourceLineNo">515</span> !clusterManager.isRunning(ServiceType.HBASE_REGIONSERVER, sn.getHostname(), sn.getPort())<a name="line.515"></a>
<span class="sourceLineNo">516</span> &amp;&amp; master.getPort() != sn.getPort()<a name="line.516"></a>
<span class="sourceLineNo">517</span> ) {<a name="line.517"></a>
<span class="sourceLineNo">518</span> LOG.info("Restoring cluster - starting initial region server: {}", sn.getAddress());<a name="line.518"></a>
<span class="sourceLineNo">519</span> startRegionServer(sn.getHostname(), sn.getPort());<a name="line.519"></a>
<span class="sourceLineNo">520</span> }<a name="line.520"></a>
<span class="sourceLineNo">521</span> } catch (IOException ex) {<a name="line.521"></a>
<span class="sourceLineNo">522</span> deferred.add(ex);<a name="line.522"></a>
<span class="sourceLineNo">523</span> }<a name="line.523"></a>
<span class="sourceLineNo">524</span> }<a name="line.524"></a>
<span class="sourceLineNo">525</span><a name="line.525"></a>
<span class="sourceLineNo">526</span> for (ServerName sn : toKill) {<a name="line.526"></a>
<span class="sourceLineNo">527</span> try {<a name="line.527"></a>
<span class="sourceLineNo">528</span> if (<a name="line.528"></a>
<span class="sourceLineNo">529</span> clusterManager.isRunning(ServiceType.HBASE_REGIONSERVER, sn.getHostname(), sn.getPort())<a name="line.529"></a>
<span class="sourceLineNo">530</span> &amp;&amp; master.getPort() != sn.getPort()<a name="line.530"></a>
<span class="sourceLineNo">531</span> ) {<a name="line.531"></a>
<span class="sourceLineNo">532</span> LOG.info("Restoring cluster - stopping initial region server: {}", sn.getAddress());<a name="line.532"></a>
<span class="sourceLineNo">533</span> stopRegionServer(sn);<a name="line.533"></a>
<span class="sourceLineNo">534</span> }<a name="line.534"></a>
<span class="sourceLineNo">535</span> } catch (IOException ex) {<a name="line.535"></a>
<span class="sourceLineNo">536</span> deferred.add(ex);<a name="line.536"></a>
<span class="sourceLineNo">537</span> }<a name="line.537"></a>
<span class="sourceLineNo">538</span> }<a name="line.538"></a>
<span class="sourceLineNo">539</span> if (!deferred.isEmpty()) {<a name="line.539"></a>
<span class="sourceLineNo">540</span> LOG.warn("Restoring cluster - restoring region servers reported {} errors:", deferred.size());<a name="line.540"></a>
<span class="sourceLineNo">541</span> for (int i = 0; i &lt; deferred.size() &amp;&amp; i &lt; 3; i++) {<a name="line.541"></a>
<span class="sourceLineNo">542</span> LOG.warn(Objects.toString(deferred.get(i)));<a name="line.542"></a>
<span class="sourceLineNo">543</span> }<a name="line.543"></a>
<span class="sourceLineNo">544</span> }<a name="line.544"></a>
<span class="sourceLineNo">545</span><a name="line.545"></a>
<span class="sourceLineNo">546</span> return deferred.isEmpty();<a name="line.546"></a>
<span class="sourceLineNo">547</span> }<a name="line.547"></a>
<span class="sourceLineNo">548</span><a name="line.548"></a>
<span class="sourceLineNo">549</span> protected boolean restoreAdmin() throws IOException {<a name="line.549"></a>
<span class="sourceLineNo">550</span> // While restoring above, if the HBase Master which was initially the Active one, was down<a name="line.550"></a>
<span class="sourceLineNo">551</span> // and the restore put the cluster back to Initial configuration, HAdmin instance will need<a name="line.551"></a>
<span class="sourceLineNo">552</span> // to refresh its connections (otherwise it will return incorrect information) or we can<a name="line.552"></a>
<span class="sourceLineNo">553</span> // point it to new instance.<a name="line.553"></a>
<span class="sourceLineNo">554</span> admin.close();<a name="line.554"></a>
<span class="sourceLineNo">555</span> this.admin = this.connection.getAdmin();<a name="line.555"></a>
<span class="sourceLineNo">556</span> LOG.info("Added new HBaseAdmin");<a name="line.556"></a>
<span class="sourceLineNo">557</span> return true;<a name="line.557"></a>
<span class="sourceLineNo">558</span> }<a name="line.558"></a>
<span class="sourceLineNo">559</span>}<a name="line.559"></a>
</pre>
</div>
</body>
</html>