| <!DOCTYPE HTML> |
| <html lang="en"> |
| <head> |
| <!-- Generated by javadoc (17) --> |
| <title>Source code</title> |
| <meta name="viewport" content="width=device-width, initial-scale=1"> |
| <meta name="description" content="source: package: org.apache.hadoop.hbase, class: DistributedHBaseCluster"> |
| <meta name="generator" content="javadoc/SourceToHTMLConverter"> |
| <link rel="stylesheet" type="text/css" href="../../../../../stylesheet.css" title="Style"> |
| </head> |
| <body class="source-page"> |
| <main role="main"> |
| <div class="source-container"> |
| <pre><span class="source-line-no">001</span><span id="line-1">/*</span> |
| <span class="source-line-no">002</span><span id="line-2"> * Licensed to the Apache Software Foundation (ASF) under one</span> |
| <span class="source-line-no">003</span><span id="line-3"> * or more contributor license agreements. See the NOTICE file</span> |
| <span class="source-line-no">004</span><span id="line-4"> * distributed with this work for additional information</span> |
| <span class="source-line-no">005</span><span id="line-5"> * regarding copyright ownership. The ASF licenses this file</span> |
| <span class="source-line-no">006</span><span id="line-6"> * to you under the Apache License, Version 2.0 (the</span> |
| <span class="source-line-no">007</span><span id="line-7"> * "License"); you may not use this file except in compliance</span> |
| <span class="source-line-no">008</span><span id="line-8"> * with the License. You may obtain a copy of the License at</span> |
| <span class="source-line-no">009</span><span id="line-9"> *</span> |
| <span class="source-line-no">010</span><span id="line-10"> * http://www.apache.org/licenses/LICENSE-2.0</span> |
| <span class="source-line-no">011</span><span id="line-11"> *</span> |
| <span class="source-line-no">012</span><span id="line-12"> * Unless required by applicable law or agreed to in writing, software</span> |
| <span class="source-line-no">013</span><span id="line-13"> * distributed under the License is distributed on an "AS IS" BASIS,</span> |
| <span class="source-line-no">014</span><span id="line-14"> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span> |
| <span class="source-line-no">015</span><span id="line-15"> * See the License for the specific language governing permissions and</span> |
| <span class="source-line-no">016</span><span id="line-16"> * limitations under the License.</span> |
| <span class="source-line-no">017</span><span id="line-17"> */</span> |
| <span class="source-line-no">018</span><span id="line-18">package org.apache.hadoop.hbase;</span> |
| <span class="source-line-no">019</span><span id="line-19"></span> |
| <span class="source-line-no">020</span><span id="line-20">import java.io.IOException;</span> |
| <span class="source-line-no">021</span><span id="line-21">import java.util.ArrayList;</span> |
| <span class="source-line-no">022</span><span id="line-22">import java.util.Comparator;</span> |
| <span class="source-line-no">023</span><span id="line-23">import java.util.EnumSet;</span> |
| <span class="source-line-no">024</span><span id="line-24">import java.util.HashSet;</span> |
| <span class="source-line-no">025</span><span id="line-25">import java.util.List;</span> |
| <span class="source-line-no">026</span><span id="line-26">import java.util.Objects;</span> |
| <span class="source-line-no">027</span><span id="line-27">import java.util.Set;</span> |
| <span class="source-line-no">028</span><span id="line-28">import java.util.TreeSet;</span> |
| <span class="source-line-no">029</span><span id="line-29">import org.apache.hadoop.conf.Configuration;</span> |
| <span class="source-line-no">030</span><span id="line-30">import org.apache.hadoop.hbase.ClusterManager.ServiceType;</span> |
| <span class="source-line-no">031</span><span id="line-31">import org.apache.hadoop.hbase.client.Admin;</span> |
| <span class="source-line-no">032</span><span id="line-32">import org.apache.hadoop.hbase.client.Connection;</span> |
| <span class="source-line-no">033</span><span id="line-33">import org.apache.hadoop.hbase.client.ConnectionFactory;</span> |
| <span class="source-line-no">034</span><span id="line-34">import org.apache.hadoop.hbase.client.RegionInfo;</span> |
| <span class="source-line-no">035</span><span id="line-35">import org.apache.hadoop.hbase.client.RegionLocator;</span> |
| <span class="source-line-no">036</span><span id="line-36">import org.apache.hadoop.hbase.util.Bytes;</span> |
| <span class="source-line-no">037</span><span id="line-37">import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;</span> |
| <span class="source-line-no">038</span><span id="line-38">import org.apache.hadoop.hbase.util.Threads;</span> |
| <span class="source-line-no">039</span><span id="line-39">import org.apache.yetus.audience.InterfaceAudience;</span> |
| <span class="source-line-no">040</span><span id="line-40">import org.slf4j.Logger;</span> |
| <span class="source-line-no">041</span><span id="line-41">import org.slf4j.LoggerFactory;</span> |
| <span class="source-line-no">042</span><span id="line-42"></span> |
| <span class="source-line-no">043</span><span id="line-43">/**</span> |
| <span class="source-line-no">044</span><span id="line-44"> * Manages the interactions with an already deployed distributed cluster (as opposed to a</span> |
| <span class="source-line-no">045</span><span id="line-45"> * pseudo-distributed, or mini/local cluster). This is used by integration and system tests.</span> |
| <span class="source-line-no">046</span><span id="line-46"> */</span> |
| <span class="source-line-no">047</span><span id="line-47">@InterfaceAudience.Private</span> |
| <span class="source-line-no">048</span><span id="line-48">public class DistributedHBaseCluster extends HBaseClusterInterface {</span> |
| <span class="source-line-no">049</span><span id="line-49"></span> |
| <span class="source-line-no">050</span><span id="line-50"> private static final Logger LOG = LoggerFactory.getLogger(DistributedHBaseCluster.class);</span> |
| <span class="source-line-no">051</span><span id="line-51"></span> |
| <span class="source-line-no">052</span><span id="line-52"> private Admin admin;</span> |
| <span class="source-line-no">053</span><span id="line-53"> private final Connection connection;</span> |
| <span class="source-line-no">054</span><span id="line-54"></span> |
| <span class="source-line-no">055</span><span id="line-55"> private ClusterManager clusterManager;</span> |
| <span class="source-line-no">056</span><span id="line-56"> /**</span> |
| <span class="source-line-no">057</span><span id="line-57"> * List of RegionServers killed so far. ServerName also comprises startCode of a server, so any</span> |
| <span class="source-line-no">058</span><span id="line-58"> * restarted instances of the same server will have different ServerName and will not coincide</span> |
| <span class="source-line-no">059</span><span id="line-59"> * with past dead ones. So there's no need to cleanup this list.</span> |
| <span class="source-line-no">060</span><span id="line-60"> */</span> |
| <span class="source-line-no">061</span><span id="line-61"> private final Set<ServerName> killedRegionServers = new HashSet<>();</span> |
| <span class="source-line-no">062</span><span id="line-62"></span> |
| <span class="source-line-no">063</span><span id="line-63"> public DistributedHBaseCluster(Configuration conf, ClusterManager clusterManager)</span> |
| <span class="source-line-no">064</span><span id="line-64"> throws IOException {</span> |
| <span class="source-line-no">065</span><span id="line-65"> super(conf);</span> |
| <span class="source-line-no">066</span><span id="line-66"> this.clusterManager = clusterManager;</span> |
| <span class="source-line-no">067</span><span id="line-67"> this.connection = ConnectionFactory.createConnection(conf);</span> |
| <span class="source-line-no">068</span><span id="line-68"> this.admin = this.connection.getAdmin();</span> |
| <span class="source-line-no">069</span><span id="line-69"> this.initialClusterStatus = getClusterMetrics();</span> |
| <span class="source-line-no">070</span><span id="line-70"> }</span> |
| <span class="source-line-no">071</span><span id="line-71"></span> |
| <span class="source-line-no">072</span><span id="line-72"> public void setClusterManager(ClusterManager clusterManager) {</span> |
| <span class="source-line-no">073</span><span id="line-73"> this.clusterManager = clusterManager;</span> |
| <span class="source-line-no">074</span><span id="line-74"> }</span> |
| <span class="source-line-no">075</span><span id="line-75"></span> |
| <span class="source-line-no">076</span><span id="line-76"> public ClusterManager getClusterManager() {</span> |
| <span class="source-line-no">077</span><span id="line-77"> return clusterManager;</span> |
| <span class="source-line-no">078</span><span id="line-78"> }</span> |
| <span class="source-line-no">079</span><span id="line-79"></span> |
| <span class="source-line-no">080</span><span id="line-80"> /**</span> |
| <span class="source-line-no">081</span><span id="line-81"> * Returns a ClusterStatus for this HBase cluster</span> |
| <span class="source-line-no">082</span><span id="line-82"> */</span> |
| <span class="source-line-no">083</span><span id="line-83"> @Override</span> |
| <span class="source-line-no">084</span><span id="line-84"> public ClusterMetrics getClusterMetrics() throws IOException {</span> |
| <span class="source-line-no">085</span><span id="line-85"> return admin.getClusterMetrics();</span> |
| <span class="source-line-no">086</span><span id="line-86"> }</span> |
| <span class="source-line-no">087</span><span id="line-87"></span> |
| <span class="source-line-no">088</span><span id="line-88"> @Override</span> |
| <span class="source-line-no">089</span><span id="line-89"> public ClusterMetrics getInitialClusterMetrics() throws IOException {</span> |
| <span class="source-line-no">090</span><span id="line-90"> return initialClusterStatus;</span> |
| <span class="source-line-no">091</span><span id="line-91"> }</span> |
| <span class="source-line-no">092</span><span id="line-92"></span> |
| <span class="source-line-no">093</span><span id="line-93"> @Override</span> |
| <span class="source-line-no">094</span><span id="line-94"> public void close() throws IOException {</span> |
| <span class="source-line-no">095</span><span id="line-95"> if (this.admin != null) {</span> |
| <span class="source-line-no">096</span><span id="line-96"> admin.close();</span> |
| <span class="source-line-no">097</span><span id="line-97"> }</span> |
| <span class="source-line-no">098</span><span id="line-98"> if (this.connection != null && !this.connection.isClosed()) {</span> |
| <span class="source-line-no">099</span><span id="line-99"> this.connection.close();</span> |
| <span class="source-line-no">100</span><span id="line-100"> }</span> |
| <span class="source-line-no">101</span><span id="line-101"> }</span> |
| <span class="source-line-no">102</span><span id="line-102"></span> |
| <span class="source-line-no">103</span><span id="line-103"> @Override</span> |
| <span class="source-line-no">104</span><span id="line-104"> public void startRegionServer(String hostname, int port) throws IOException {</span> |
| <span class="source-line-no">105</span><span id="line-105"> LOG.info("Starting RS on: {}", hostname);</span> |
| <span class="source-line-no">106</span><span id="line-106"> clusterManager.start(ServiceType.HBASE_REGIONSERVER, hostname, port);</span> |
| <span class="source-line-no">107</span><span id="line-107"> }</span> |
| <span class="source-line-no">108</span><span id="line-108"></span> |
| <span class="source-line-no">109</span><span id="line-109"> @Override</span> |
| <span class="source-line-no">110</span><span id="line-110"> public void killRegionServer(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">111</span><span id="line-111"> LOG.info("Aborting RS: {}", serverName.getServerName());</span> |
| <span class="source-line-no">112</span><span id="line-112"> killedRegionServers.add(serverName);</span> |
| <span class="source-line-no">113</span><span id="line-113"> clusterManager.kill(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),</span> |
| <span class="source-line-no">114</span><span id="line-114"> serverName.getPort());</span> |
| <span class="source-line-no">115</span><span id="line-115"> }</span> |
| <span class="source-line-no">116</span><span id="line-116"></span> |
| <span class="source-line-no">117</span><span id="line-117"> @Override</span> |
| <span class="source-line-no">118</span><span id="line-118"> public boolean isKilledRS(ServerName serverName) {</span> |
| <span class="source-line-no">119</span><span id="line-119"> return killedRegionServers.contains(serverName);</span> |
| <span class="source-line-no">120</span><span id="line-120"> }</span> |
| <span class="source-line-no">121</span><span id="line-121"></span> |
| <span class="source-line-no">122</span><span id="line-122"> @Override</span> |
| <span class="source-line-no">123</span><span id="line-123"> public void stopRegionServer(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">124</span><span id="line-124"> LOG.info("Stopping RS: {}", serverName.getServerName());</span> |
| <span class="source-line-no">125</span><span id="line-125"> clusterManager.stop(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),</span> |
| <span class="source-line-no">126</span><span id="line-126"> serverName.getPort());</span> |
| <span class="source-line-no">127</span><span id="line-127"> }</span> |
| <span class="source-line-no">128</span><span id="line-128"></span> |
| <span class="source-line-no">129</span><span id="line-129"> @Override</span> |
| <span class="source-line-no">130</span><span id="line-130"> public void waitForRegionServerToStop(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">131</span><span id="line-131"> waitForServiceToStop(ServiceType.HBASE_REGIONSERVER, serverName, timeout);</span> |
| <span class="source-line-no">132</span><span id="line-132"> }</span> |
| <span class="source-line-no">133</span><span id="line-133"></span> |
| <span class="source-line-no">134</span><span id="line-134"> @Override</span> |
| <span class="source-line-no">135</span><span id="line-135"> public void suspendRegionServer(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">136</span><span id="line-136"> LOG.info("Suspend RS: {}", serverName.getServerName());</span> |
| <span class="source-line-no">137</span><span id="line-137"> clusterManager.suspend(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),</span> |
| <span class="source-line-no">138</span><span id="line-138"> serverName.getPort());</span> |
| <span class="source-line-no">139</span><span id="line-139"> }</span> |
| <span class="source-line-no">140</span><span id="line-140"></span> |
| <span class="source-line-no">141</span><span id="line-141"> @Override</span> |
| <span class="source-line-no">142</span><span id="line-142"> public void resumeRegionServer(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">143</span><span id="line-143"> LOG.info("Resume RS: {}", serverName.getServerName());</span> |
| <span class="source-line-no">144</span><span id="line-144"> clusterManager.resume(ServiceType.HBASE_REGIONSERVER, serverName.getHostname(),</span> |
| <span class="source-line-no">145</span><span id="line-145"> serverName.getPort());</span> |
| <span class="source-line-no">146</span><span id="line-146"> }</span> |
| <span class="source-line-no">147</span><span id="line-147"></span> |
| <span class="source-line-no">148</span><span id="line-148"> @Override</span> |
| <span class="source-line-no">149</span><span id="line-149"> public void startZkNode(String hostname, int port) throws IOException {</span> |
| <span class="source-line-no">150</span><span id="line-150"> LOG.info("Starting ZooKeeper node on: {}", hostname);</span> |
| <span class="source-line-no">151</span><span id="line-151"> clusterManager.start(ServiceType.ZOOKEEPER_SERVER, hostname, port);</span> |
| <span class="source-line-no">152</span><span id="line-152"> }</span> |
| <span class="source-line-no">153</span><span id="line-153"></span> |
| <span class="source-line-no">154</span><span id="line-154"> @Override</span> |
| <span class="source-line-no">155</span><span id="line-155"> public void killZkNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">156</span><span id="line-156"> LOG.info("Aborting ZooKeeper node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">157</span><span id="line-157"> clusterManager.kill(ServiceType.ZOOKEEPER_SERVER, serverName.getHostname(),</span> |
| <span class="source-line-no">158</span><span id="line-158"> serverName.getPort());</span> |
| <span class="source-line-no">159</span><span id="line-159"> }</span> |
| <span class="source-line-no">160</span><span id="line-160"></span> |
| <span class="source-line-no">161</span><span id="line-161"> @Override</span> |
| <span class="source-line-no">162</span><span id="line-162"> public void stopZkNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">163</span><span id="line-163"> LOG.info("Stopping ZooKeeper node: {}", serverName.getServerName());</span> |
| <span class="source-line-no">164</span><span id="line-164"> clusterManager.stop(ServiceType.ZOOKEEPER_SERVER, serverName.getHostname(),</span> |
| <span class="source-line-no">165</span><span id="line-165"> serverName.getPort());</span> |
| <span class="source-line-no">166</span><span id="line-166"> }</span> |
| <span class="source-line-no">167</span><span id="line-167"></span> |
| <span class="source-line-no">168</span><span id="line-168"> @Override</span> |
| <span class="source-line-no">169</span><span id="line-169"> public void waitForZkNodeToStart(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">170</span><span id="line-170"> waitForServiceToStart(ServiceType.ZOOKEEPER_SERVER, serverName, timeout);</span> |
| <span class="source-line-no">171</span><span id="line-171"> }</span> |
| <span class="source-line-no">172</span><span id="line-172"></span> |
| <span class="source-line-no">173</span><span id="line-173"> @Override</span> |
| <span class="source-line-no">174</span><span id="line-174"> public void waitForZkNodeToStop(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">175</span><span id="line-175"> waitForServiceToStop(ServiceType.ZOOKEEPER_SERVER, serverName, timeout);</span> |
| <span class="source-line-no">176</span><span id="line-176"> }</span> |
| <span class="source-line-no">177</span><span id="line-177"></span> |
| <span class="source-line-no">178</span><span id="line-178"> @Override</span> |
| <span class="source-line-no">179</span><span id="line-179"> public void startDataNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">180</span><span id="line-180"> LOG.info("Starting data node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">181</span><span id="line-181"> clusterManager.start(ServiceType.HADOOP_DATANODE, serverName.getHostname(),</span> |
| <span class="source-line-no">182</span><span id="line-182"> serverName.getPort());</span> |
| <span class="source-line-no">183</span><span id="line-183"> }</span> |
| <span class="source-line-no">184</span><span id="line-184"></span> |
| <span class="source-line-no">185</span><span id="line-185"> @Override</span> |
| <span class="source-line-no">186</span><span id="line-186"> public void killDataNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">187</span><span id="line-187"> LOG.info("Aborting data node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">188</span><span id="line-188"> clusterManager.kill(ServiceType.HADOOP_DATANODE, serverName.getHostname(),</span> |
| <span class="source-line-no">189</span><span id="line-189"> serverName.getPort());</span> |
| <span class="source-line-no">190</span><span id="line-190"> }</span> |
| <span class="source-line-no">191</span><span id="line-191"></span> |
| <span class="source-line-no">192</span><span id="line-192"> @Override</span> |
| <span class="source-line-no">193</span><span id="line-193"> public void stopDataNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">194</span><span id="line-194"> LOG.info("Stopping data node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">195</span><span id="line-195"> clusterManager.stop(ServiceType.HADOOP_DATANODE, serverName.getHostname(),</span> |
| <span class="source-line-no">196</span><span id="line-196"> serverName.getPort());</span> |
| <span class="source-line-no">197</span><span id="line-197"> }</span> |
| <span class="source-line-no">198</span><span id="line-198"></span> |
| <span class="source-line-no">199</span><span id="line-199"> @Override</span> |
| <span class="source-line-no">200</span><span id="line-200"> public void waitForDataNodeToStart(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">201</span><span id="line-201"> waitForServiceToStart(ServiceType.HADOOP_DATANODE, serverName, timeout);</span> |
| <span class="source-line-no">202</span><span id="line-202"> }</span> |
| <span class="source-line-no">203</span><span id="line-203"></span> |
| <span class="source-line-no">204</span><span id="line-204"> @Override</span> |
| <span class="source-line-no">205</span><span id="line-205"> public void waitForDataNodeToStop(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">206</span><span id="line-206"> waitForServiceToStop(ServiceType.HADOOP_DATANODE, serverName, timeout);</span> |
| <span class="source-line-no">207</span><span id="line-207"> }</span> |
| <span class="source-line-no">208</span><span id="line-208"></span> |
| <span class="source-line-no">209</span><span id="line-209"> @Override</span> |
| <span class="source-line-no">210</span><span id="line-210"> public void startNameNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">211</span><span id="line-211"> LOG.info("Starting name node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">212</span><span id="line-212"> clusterManager.start(ServiceType.HADOOP_NAMENODE, serverName.getHostname(),</span> |
| <span class="source-line-no">213</span><span id="line-213"> serverName.getPort());</span> |
| <span class="source-line-no">214</span><span id="line-214"> }</span> |
| <span class="source-line-no">215</span><span id="line-215"></span> |
| <span class="source-line-no">216</span><span id="line-216"> @Override</span> |
| <span class="source-line-no">217</span><span id="line-217"> public void killNameNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">218</span><span id="line-218"> LOG.info("Aborting name node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">219</span><span id="line-219"> clusterManager.kill(ServiceType.HADOOP_NAMENODE, serverName.getHostname(),</span> |
| <span class="source-line-no">220</span><span id="line-220"> serverName.getPort());</span> |
| <span class="source-line-no">221</span><span id="line-221"> }</span> |
| <span class="source-line-no">222</span><span id="line-222"></span> |
| <span class="source-line-no">223</span><span id="line-223"> @Override</span> |
| <span class="source-line-no">224</span><span id="line-224"> public void stopNameNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">225</span><span id="line-225"> LOG.info("Stopping name node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">226</span><span id="line-226"> clusterManager.stop(ServiceType.HADOOP_NAMENODE, serverName.getHostname(),</span> |
| <span class="source-line-no">227</span><span id="line-227"> serverName.getPort());</span> |
| <span class="source-line-no">228</span><span id="line-228"> }</span> |
| <span class="source-line-no">229</span><span id="line-229"></span> |
| <span class="source-line-no">230</span><span id="line-230"> @Override</span> |
| <span class="source-line-no">231</span><span id="line-231"> public void waitForNameNodeToStart(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">232</span><span id="line-232"> waitForServiceToStart(ServiceType.HADOOP_NAMENODE, serverName, timeout);</span> |
| <span class="source-line-no">233</span><span id="line-233"> }</span> |
| <span class="source-line-no">234</span><span id="line-234"></span> |
| <span class="source-line-no">235</span><span id="line-235"> @Override</span> |
| <span class="source-line-no">236</span><span id="line-236"> public void waitForNameNodeToStop(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">237</span><span id="line-237"> waitForServiceToStop(ServiceType.HADOOP_NAMENODE, serverName, timeout);</span> |
| <span class="source-line-no">238</span><span id="line-238"> }</span> |
| <span class="source-line-no">239</span><span id="line-239"></span> |
| <span class="source-line-no">240</span><span id="line-240"> @Override</span> |
| <span class="source-line-no">241</span><span id="line-241"> public void startJournalNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">242</span><span id="line-242"> LOG.info("Starting journal node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">243</span><span id="line-243"> clusterManager.start(ServiceType.HADOOP_JOURNALNODE, serverName.getHostname(),</span> |
| <span class="source-line-no">244</span><span id="line-244"> serverName.getPort());</span> |
| <span class="source-line-no">245</span><span id="line-245"> }</span> |
| <span class="source-line-no">246</span><span id="line-246"></span> |
| <span class="source-line-no">247</span><span id="line-247"> @Override</span> |
| <span class="source-line-no">248</span><span id="line-248"> public void killJournalNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">249</span><span id="line-249"> LOG.info("Aborting journal node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">250</span><span id="line-250"> clusterManager.kill(ServiceType.HADOOP_JOURNALNODE, serverName.getHostname(),</span> |
| <span class="source-line-no">251</span><span id="line-251"> serverName.getPort());</span> |
| <span class="source-line-no">252</span><span id="line-252"> }</span> |
| <span class="source-line-no">253</span><span id="line-253"></span> |
| <span class="source-line-no">254</span><span id="line-254"> @Override</span> |
| <span class="source-line-no">255</span><span id="line-255"> public void stopJournalNode(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">256</span><span id="line-256"> LOG.info("Stopping journal node on: {}", serverName.getServerName());</span> |
| <span class="source-line-no">257</span><span id="line-257"> clusterManager.stop(ServiceType.HADOOP_JOURNALNODE, serverName.getHostname(),</span> |
| <span class="source-line-no">258</span><span id="line-258"> serverName.getPort());</span> |
| <span class="source-line-no">259</span><span id="line-259"> }</span> |
| <span class="source-line-no">260</span><span id="line-260"></span> |
| <span class="source-line-no">261</span><span id="line-261"> @Override</span> |
| <span class="source-line-no">262</span><span id="line-262"> public void waitForJournalNodeToStart(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">263</span><span id="line-263"> waitForServiceToStart(ServiceType.HADOOP_JOURNALNODE, serverName, timeout);</span> |
| <span class="source-line-no">264</span><span id="line-264"> }</span> |
| <span class="source-line-no">265</span><span id="line-265"></span> |
| <span class="source-line-no">266</span><span id="line-266"> @Override</span> |
| <span class="source-line-no">267</span><span id="line-267"> public void waitForJournalNodeToStop(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">268</span><span id="line-268"> waitForServiceToStop(ServiceType.HADOOP_JOURNALNODE, serverName, timeout);</span> |
| <span class="source-line-no">269</span><span id="line-269"> }</span> |
| <span class="source-line-no">270</span><span id="line-270"></span> |
| <span class="source-line-no">271</span><span id="line-271"> private void waitForServiceToStop(ServiceType service, ServerName serverName, long timeout)</span> |
| <span class="source-line-no">272</span><span id="line-272"> throws IOException {</span> |
| <span class="source-line-no">273</span><span id="line-273"> LOG.info("Waiting for service: {} to stop: {}", service, serverName.getServerName());</span> |
| <span class="source-line-no">274</span><span id="line-274"> long start = EnvironmentEdgeManager.currentTime();</span> |
| <span class="source-line-no">275</span><span id="line-275"></span> |
| <span class="source-line-no">276</span><span id="line-276"> while ((EnvironmentEdgeManager.currentTime() - start) < timeout) {</span> |
| <span class="source-line-no">277</span><span id="line-277"> if (!clusterManager.isRunning(service, serverName.getHostname(), serverName.getPort())) {</span> |
| <span class="source-line-no">278</span><span id="line-278"> return;</span> |
| <span class="source-line-no">279</span><span id="line-279"> }</span> |
| <span class="source-line-no">280</span><span id="line-280"> Threads.sleep(100);</span> |
| <span class="source-line-no">281</span><span id="line-281"> }</span> |
| <span class="source-line-no">282</span><span id="line-282"> throw new IOException("Timed-out waiting for service to stop: " + serverName);</span> |
| <span class="source-line-no">283</span><span id="line-283"> }</span> |
| <span class="source-line-no">284</span><span id="line-284"></span> |
| <span class="source-line-no">285</span><span id="line-285"> private void waitForServiceToStart(ServiceType service, ServerName serverName, long timeout)</span> |
| <span class="source-line-no">286</span><span id="line-286"> throws IOException {</span> |
| <span class="source-line-no">287</span><span id="line-287"> LOG.info("Waiting for service: {} to start: {}", service, serverName.getServerName());</span> |
| <span class="source-line-no">288</span><span id="line-288"> long start = EnvironmentEdgeManager.currentTime();</span> |
| <span class="source-line-no">289</span><span id="line-289"></span> |
| <span class="source-line-no">290</span><span id="line-290"> while ((EnvironmentEdgeManager.currentTime() - start) < timeout) {</span> |
| <span class="source-line-no">291</span><span id="line-291"> if (clusterManager.isRunning(service, serverName.getHostname(), serverName.getPort())) {</span> |
| <span class="source-line-no">292</span><span id="line-292"> return;</span> |
| <span class="source-line-no">293</span><span id="line-293"> }</span> |
| <span class="source-line-no">294</span><span id="line-294"> Threads.sleep(100);</span> |
| <span class="source-line-no">295</span><span id="line-295"> }</span> |
| <span class="source-line-no">296</span><span id="line-296"> throw new IOException("Timed-out waiting for service to start: " + serverName);</span> |
| <span class="source-line-no">297</span><span id="line-297"> }</span> |
| <span class="source-line-no">298</span><span id="line-298"></span> |
| <span class="source-line-no">299</span><span id="line-299"> @Override</span> |
| <span class="source-line-no">300</span><span id="line-300"> public void startMaster(String hostname, int port) throws IOException {</span> |
| <span class="source-line-no">301</span><span id="line-301"> LOG.info("Starting Master on: {}:{}", hostname, port);</span> |
| <span class="source-line-no">302</span><span id="line-302"> clusterManager.start(ServiceType.HBASE_MASTER, hostname, port);</span> |
| <span class="source-line-no">303</span><span id="line-303"> }</span> |
| <span class="source-line-no">304</span><span id="line-304"></span> |
| <span class="source-line-no">305</span><span id="line-305"> @Override</span> |
| <span class="source-line-no">306</span><span id="line-306"> public void killMaster(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">307</span><span id="line-307"> LOG.info("Aborting Master: {}", serverName.getServerName());</span> |
| <span class="source-line-no">308</span><span id="line-308"> clusterManager.kill(ServiceType.HBASE_MASTER, serverName.getHostname(), serverName.getPort());</span> |
| <span class="source-line-no">309</span><span id="line-309"> }</span> |
| <span class="source-line-no">310</span><span id="line-310"></span> |
| <span class="source-line-no">311</span><span id="line-311"> @Override</span> |
| <span class="source-line-no">312</span><span id="line-312"> public void stopMaster(ServerName serverName) throws IOException {</span> |
| <span class="source-line-no">313</span><span id="line-313"> LOG.info("Stopping Master: {}", serverName.getServerName());</span> |
| <span class="source-line-no">314</span><span id="line-314"> clusterManager.stop(ServiceType.HBASE_MASTER, serverName.getHostname(), serverName.getPort());</span> |
| <span class="source-line-no">315</span><span id="line-315"> }</span> |
| <span class="source-line-no">316</span><span id="line-316"></span> |
| <span class="source-line-no">317</span><span id="line-317"> @Override</span> |
| <span class="source-line-no">318</span><span id="line-318"> public void waitForMasterToStop(ServerName serverName, long timeout) throws IOException {</span> |
| <span class="source-line-no">319</span><span id="line-319"> waitForServiceToStop(ServiceType.HBASE_MASTER, serverName, timeout);</span> |
| <span class="source-line-no">320</span><span id="line-320"> }</span> |
| <span class="source-line-no">321</span><span id="line-321"></span> |
| <span class="source-line-no">322</span><span id="line-322"> @Override</span> |
| <span class="source-line-no">323</span><span id="line-323"> public boolean waitForActiveAndReadyMaster(long timeout) throws IOException {</span> |
| <span class="source-line-no">324</span><span id="line-324"> long start = EnvironmentEdgeManager.currentTime();</span> |
| <span class="source-line-no">325</span><span id="line-325"> while (EnvironmentEdgeManager.currentTime() - start < timeout) {</span> |
| <span class="source-line-no">326</span><span id="line-326"> try {</span> |
| <span class="source-line-no">327</span><span id="line-327"> connection.getAdmin().getClusterMetrics(EnumSet.of(ClusterMetrics.Option.HBASE_VERSION));</span> |
| <span class="source-line-no">328</span><span id="line-328"> return true;</span> |
| <span class="source-line-no">329</span><span id="line-329"> } catch (MasterNotRunningException m) {</span> |
| <span class="source-line-no">330</span><span id="line-330"> LOG.warn("Master not started yet " + m);</span> |
| <span class="source-line-no">331</span><span id="line-331"> } catch (ZooKeeperConnectionException e) {</span> |
| <span class="source-line-no">332</span><span id="line-332"> LOG.warn("Failed to connect to ZK " + e);</span> |
| <span class="source-line-no">333</span><span id="line-333"> }</span> |
| <span class="source-line-no">334</span><span id="line-334"> Threads.sleep(1000);</span> |
| <span class="source-line-no">335</span><span id="line-335"> }</span> |
| <span class="source-line-no">336</span><span id="line-336"> return false;</span> |
| <span class="source-line-no">337</span><span id="line-337"> }</span> |
| <span class="source-line-no">338</span><span id="line-338"></span> |
| <span class="source-line-no">339</span><span id="line-339"> @Override</span> |
| <span class="source-line-no">340</span><span id="line-340"> public ServerName getServerHoldingRegion(TableName tn, byte[] regionName) throws IOException {</span> |
| <span class="source-line-no">341</span><span id="line-341"> byte[] startKey = RegionInfo.getStartKey(regionName);</span> |
| <span class="source-line-no">342</span><span id="line-342"> HRegionLocation regionLoc = null;</span> |
| <span class="source-line-no">343</span><span id="line-343"> try (RegionLocator locator = connection.getRegionLocator(tn)) {</span> |
| <span class="source-line-no">344</span><span id="line-344"> regionLoc = locator.getRegionLocation(startKey, true);</span> |
| <span class="source-line-no">345</span><span id="line-345"> }</span> |
| <span class="source-line-no">346</span><span id="line-346"> if (regionLoc == null) {</span> |
| <span class="source-line-no">347</span><span id="line-347"> LOG.warn("Cannot find region server holding region {}", Bytes.toStringBinary(regionName));</span> |
| <span class="source-line-no">348</span><span id="line-348"> return null;</span> |
| <span class="source-line-no">349</span><span id="line-349"> }</span> |
| <span class="source-line-no">350</span><span id="line-350"> return regionLoc.getServerName();</span> |
| <span class="source-line-no">351</span><span id="line-351"> }</span> |
| <span class="source-line-no">352</span><span id="line-352"></span> |
| <span class="source-line-no">353</span><span id="line-353"> @Override</span> |
| <span class="source-line-no">354</span><span id="line-354"> public void waitUntilShutDown() {</span> |
| <span class="source-line-no">355</span><span id="line-355"> // Simply wait for a few seconds for now (after issuing serverManager.kill</span> |
| <span class="source-line-no">356</span><span id="line-356"> throw new RuntimeException(HConstants.NOT_IMPLEMENTED);</span> |
| <span class="source-line-no">357</span><span id="line-357"> }</span> |
| <span class="source-line-no">358</span><span id="line-358"></span> |
| <span class="source-line-no">359</span><span id="line-359"> @Override</span> |
| <span class="source-line-no">360</span><span id="line-360"> public void shutdown() throws IOException {</span> |
| <span class="source-line-no">361</span><span id="line-361"> // not sure we want this</span> |
| <span class="source-line-no">362</span><span id="line-362"> throw new RuntimeException(HConstants.NOT_IMPLEMENTED);</span> |
| <span class="source-line-no">363</span><span id="line-363"> }</span> |
| <span class="source-line-no">364</span><span id="line-364"></span> |
| <span class="source-line-no">365</span><span id="line-365"> @Override</span> |
| <span class="source-line-no">366</span><span id="line-366"> public boolean isDistributedCluster() {</span> |
| <span class="source-line-no">367</span><span id="line-367"> return true;</span> |
| <span class="source-line-no">368</span><span id="line-368"> }</span> |
| <span class="source-line-no">369</span><span id="line-369"></span> |
| <span class="source-line-no">370</span><span id="line-370"> @Override</span> |
| <span class="source-line-no">371</span><span id="line-371"> public boolean restoreClusterMetrics(ClusterMetrics initial) throws IOException {</span> |
| <span class="source-line-no">372</span><span id="line-372"> ClusterMetrics current = getClusterMetrics();</span> |
| <span class="source-line-no">373</span><span id="line-373"></span> |
| <span class="source-line-no">374</span><span id="line-374"> LOG.info("Restoring cluster - started");</span> |
| <span class="source-line-no">375</span><span id="line-375"></span> |
| <span class="source-line-no">376</span><span id="line-376"> // do a best effort restore</span> |
| <span class="source-line-no">377</span><span id="line-377"> boolean success = restoreMasters(initial, current);</span> |
| <span class="source-line-no">378</span><span id="line-378"> success = restoreRegionServers(initial, current) && success;</span> |
| <span class="source-line-no">379</span><span id="line-379"> success = restoreAdmin() && success;</span> |
| <span class="source-line-no">380</span><span id="line-380"></span> |
| <span class="source-line-no">381</span><span id="line-381"> LOG.info("Restoring cluster - done");</span> |
| <span class="source-line-no">382</span><span id="line-382"> return success;</span> |
| <span class="source-line-no">383</span><span id="line-383"> }</span> |
| <span class="source-line-no">384</span><span id="line-384"></span> |
| <span class="source-line-no">385</span><span id="line-385"> protected boolean restoreMasters(ClusterMetrics initial, ClusterMetrics current) {</span> |
| <span class="source-line-no">386</span><span id="line-386"> List<IOException> deferred = new ArrayList<>();</span> |
| <span class="source-line-no">387</span><span id="line-387"> // check whether current master has changed</span> |
| <span class="source-line-no">388</span><span id="line-388"> final ServerName initMaster = initial.getMasterName();</span> |
| <span class="source-line-no">389</span><span id="line-389"> if (!ServerName.isSameAddress(initMaster, current.getMasterName())) {</span> |
| <span class="source-line-no">390</span><span id="line-390"> LOG.info("Restoring cluster - Initial active master : {} has changed to : {}",</span> |
| <span class="source-line-no">391</span><span id="line-391"> initMaster.getAddress(), current.getMasterName().getAddress());</span> |
| <span class="source-line-no">392</span><span id="line-392"> // If initial master is stopped, start it, before restoring the state.</span> |
| <span class="source-line-no">393</span><span id="line-393"> // It will come up as a backup master, if there is already an active master.</span> |
| <span class="source-line-no">394</span><span id="line-394"> try {</span> |
| <span class="source-line-no">395</span><span id="line-395"> if (</span> |
| <span class="source-line-no">396</span><span id="line-396"> !clusterManager.isRunning(ServiceType.HBASE_MASTER, initMaster.getHostname(),</span> |
| <span class="source-line-no">397</span><span id="line-397"> initMaster.getPort())</span> |
| <span class="source-line-no">398</span><span id="line-398"> ) {</span> |
| <span class="source-line-no">399</span><span id="line-399"> LOG.info("Restoring cluster - starting initial active master at:{}",</span> |
| <span class="source-line-no">400</span><span id="line-400"> initMaster.getAddress());</span> |
| <span class="source-line-no">401</span><span id="line-401"> startMaster(initMaster.getHostname(), initMaster.getPort());</span> |
| <span class="source-line-no">402</span><span id="line-402"> }</span> |
| <span class="source-line-no">403</span><span id="line-403"></span> |
| <span class="source-line-no">404</span><span id="line-404"> // master has changed, we would like to undo this.</span> |
| <span class="source-line-no">405</span><span id="line-405"> // 1. Kill the current backups</span> |
| <span class="source-line-no">406</span><span id="line-406"> // 2. Stop current master</span> |
| <span class="source-line-no">407</span><span id="line-407"> // 3. Start backup masters</span> |
| <span class="source-line-no">408</span><span id="line-408"> for (ServerName currentBackup : current.getBackupMasterNames()) {</span> |
| <span class="source-line-no">409</span><span id="line-409"> if (!ServerName.isSameAddress(currentBackup, initMaster)) {</span> |
| <span class="source-line-no">410</span><span id="line-410"> LOG.info("Restoring cluster - stopping backup master: {}", currentBackup);</span> |
| <span class="source-line-no">411</span><span id="line-411"> stopMaster(currentBackup);</span> |
| <span class="source-line-no">412</span><span id="line-412"> }</span> |
| <span class="source-line-no">413</span><span id="line-413"> }</span> |
| <span class="source-line-no">414</span><span id="line-414"> LOG.info("Restoring cluster - stopping active master: {}", current.getMasterName());</span> |
| <span class="source-line-no">415</span><span id="line-415"> stopMaster(current.getMasterName());</span> |
| <span class="source-line-no">416</span><span id="line-416"> waitForActiveAndReadyMaster(); // wait so that active master takes over</span> |
| <span class="source-line-no">417</span><span id="line-417"> } catch (IOException ex) {</span> |
| <span class="source-line-no">418</span><span id="line-418"> // if we fail to start the initial active master, we do not want to continue stopping</span> |
| <span class="source-line-no">419</span><span id="line-419"> // backup masters. Just keep what we have now</span> |
| <span class="source-line-no">420</span><span id="line-420"> deferred.add(ex);</span> |
| <span class="source-line-no">421</span><span id="line-421"> }</span> |
| <span class="source-line-no">422</span><span id="line-422"></span> |
| <span class="source-line-no">423</span><span id="line-423"> // start backup masters</span> |
| <span class="source-line-no">424</span><span id="line-424"> for (ServerName backup : initial.getBackupMasterNames()) {</span> |
| <span class="source-line-no">425</span><span id="line-425"> try {</span> |
| <span class="source-line-no">426</span><span id="line-426"> // these are not started in backup mode, but we should already have an active master</span> |
| <span class="source-line-no">427</span><span id="line-427"> if (</span> |
| <span class="source-line-no">428</span><span id="line-428"> !clusterManager.isRunning(ServiceType.HBASE_MASTER, backup.getHostname(),</span> |
| <span class="source-line-no">429</span><span id="line-429"> backup.getPort())</span> |
| <span class="source-line-no">430</span><span id="line-430"> ) {</span> |
| <span class="source-line-no">431</span><span id="line-431"> LOG.info("Restoring cluster - starting initial backup master: {}", backup.getAddress());</span> |
| <span class="source-line-no">432</span><span id="line-432"> startMaster(backup.getHostname(), backup.getPort());</span> |
| <span class="source-line-no">433</span><span id="line-433"> }</span> |
| <span class="source-line-no">434</span><span id="line-434"> } catch (IOException ex) {</span> |
| <span class="source-line-no">435</span><span id="line-435"> deferred.add(ex);</span> |
| <span class="source-line-no">436</span><span id="line-436"> }</span> |
| <span class="source-line-no">437</span><span id="line-437"> }</span> |
| <span class="source-line-no">438</span><span id="line-438"> } else {</span> |
| <span class="source-line-no">439</span><span id="line-439"> // current master has not changed, match up backup masters</span> |
| <span class="source-line-no">440</span><span id="line-440"> Set<ServerName> toStart = new TreeSet<>(new ServerNameIgnoreStartCodeComparator());</span> |
| <span class="source-line-no">441</span><span id="line-441"> Set<ServerName> toKill = new TreeSet<>(new ServerNameIgnoreStartCodeComparator());</span> |
| <span class="source-line-no">442</span><span id="line-442"> toStart.addAll(initial.getBackupMasterNames());</span> |
| <span class="source-line-no">443</span><span id="line-443"> toKill.addAll(current.getBackupMasterNames());</span> |
| <span class="source-line-no">444</span><span id="line-444"></span> |
| <span class="source-line-no">445</span><span id="line-445"> for (ServerName server : current.getBackupMasterNames()) {</span> |
| <span class="source-line-no">446</span><span id="line-446"> toStart.remove(server);</span> |
| <span class="source-line-no">447</span><span id="line-447"> }</span> |
| <span class="source-line-no">448</span><span id="line-448"> for (ServerName server : initial.getBackupMasterNames()) {</span> |
| <span class="source-line-no">449</span><span id="line-449"> toKill.remove(server);</span> |
| <span class="source-line-no">450</span><span id="line-450"> }</span> |
| <span class="source-line-no">451</span><span id="line-451"></span> |
| <span class="source-line-no">452</span><span id="line-452"> for (ServerName sn : toStart) {</span> |
| <span class="source-line-no">453</span><span id="line-453"> try {</span> |
| <span class="source-line-no">454</span><span id="line-454"> if (!clusterManager.isRunning(ServiceType.HBASE_MASTER, sn.getHostname(), sn.getPort())) {</span> |
| <span class="source-line-no">455</span><span id="line-455"> LOG.info("Restoring cluster - starting initial backup master: {}", sn.getAddress());</span> |
| <span class="source-line-no">456</span><span id="line-456"> startMaster(sn.getHostname(), sn.getPort());</span> |
| <span class="source-line-no">457</span><span id="line-457"> }</span> |
| <span class="source-line-no">458</span><span id="line-458"> } catch (IOException ex) {</span> |
| <span class="source-line-no">459</span><span id="line-459"> deferred.add(ex);</span> |
| <span class="source-line-no">460</span><span id="line-460"> }</span> |
| <span class="source-line-no">461</span><span id="line-461"> }</span> |
| <span class="source-line-no">462</span><span id="line-462"></span> |
| <span class="source-line-no">463</span><span id="line-463"> for (ServerName sn : toKill) {</span> |
| <span class="source-line-no">464</span><span id="line-464"> try {</span> |
| <span class="source-line-no">465</span><span id="line-465"> if (clusterManager.isRunning(ServiceType.HBASE_MASTER, sn.getHostname(), sn.getPort())) {</span> |
| <span class="source-line-no">466</span><span id="line-466"> LOG.info("Restoring cluster - stopping backup master: {}", sn.getAddress());</span> |
| <span class="source-line-no">467</span><span id="line-467"> stopMaster(sn);</span> |
| <span class="source-line-no">468</span><span id="line-468"> }</span> |
| <span class="source-line-no">469</span><span id="line-469"> } catch (IOException ex) {</span> |
| <span class="source-line-no">470</span><span id="line-470"> deferred.add(ex);</span> |
| <span class="source-line-no">471</span><span id="line-471"> }</span> |
| <span class="source-line-no">472</span><span id="line-472"> }</span> |
| <span class="source-line-no">473</span><span id="line-473"> }</span> |
| <span class="source-line-no">474</span><span id="line-474"> if (!deferred.isEmpty()) {</span> |
| <span class="source-line-no">475</span><span id="line-475"> LOG.warn("Restoring cluster - restoring region servers reported {} errors:", deferred.size());</span> |
| <span class="source-line-no">476</span><span id="line-476"> for (int i = 0; i < deferred.size() && i < 3; i++) {</span> |
| <span class="source-line-no">477</span><span id="line-477"> LOG.warn(Objects.toString(deferred.get(i)));</span> |
| <span class="source-line-no">478</span><span id="line-478"> }</span> |
| <span class="source-line-no">479</span><span id="line-479"> }</span> |
| <span class="source-line-no">480</span><span id="line-480"></span> |
| <span class="source-line-no">481</span><span id="line-481"> return deferred.isEmpty();</span> |
| <span class="source-line-no">482</span><span id="line-482"> }</span> |
| <span class="source-line-no">483</span><span id="line-483"></span> |
| <span class="source-line-no">484</span><span id="line-484"> private static class ServerNameIgnoreStartCodeComparator implements Comparator<ServerName> {</span> |
| <span class="source-line-no">485</span><span id="line-485"> @Override</span> |
| <span class="source-line-no">486</span><span id="line-486"> public int compare(ServerName o1, ServerName o2) {</span> |
| <span class="source-line-no">487</span><span id="line-487"> int compare = o1.getHostname().compareToIgnoreCase(o2.getHostname());</span> |
| <span class="source-line-no">488</span><span id="line-488"> if (compare != 0) return compare;</span> |
| <span class="source-line-no">489</span><span id="line-489"> compare = o1.getPort() - o2.getPort();</span> |
| <span class="source-line-no">490</span><span id="line-490"> if (compare != 0) return compare;</span> |
| <span class="source-line-no">491</span><span id="line-491"> return 0;</span> |
| <span class="source-line-no">492</span><span id="line-492"> }</span> |
| <span class="source-line-no">493</span><span id="line-493"> }</span> |
| <span class="source-line-no">494</span><span id="line-494"></span> |
| <span class="source-line-no">495</span><span id="line-495"> protected boolean restoreRegionServers(ClusterMetrics initial, ClusterMetrics current) {</span> |
| <span class="source-line-no">496</span><span id="line-496"> Set<ServerName> toStart = new TreeSet<>(new ServerNameIgnoreStartCodeComparator());</span> |
| <span class="source-line-no">497</span><span id="line-497"> Set<ServerName> toKill = new TreeSet<>(new ServerNameIgnoreStartCodeComparator());</span> |
| <span class="source-line-no">498</span><span id="line-498"> toStart.addAll(initial.getLiveServerMetrics().keySet());</span> |
| <span class="source-line-no">499</span><span id="line-499"> toKill.addAll(current.getLiveServerMetrics().keySet());</span> |
| <span class="source-line-no">500</span><span id="line-500"></span> |
| <span class="source-line-no">501</span><span id="line-501"> ServerName master = initial.getMasterName();</span> |
| <span class="source-line-no">502</span><span id="line-502"></span> |
| <span class="source-line-no">503</span><span id="line-503"> for (ServerName server : current.getLiveServerMetrics().keySet()) {</span> |
| <span class="source-line-no">504</span><span id="line-504"> toStart.remove(server);</span> |
| <span class="source-line-no">505</span><span id="line-505"> }</span> |
| <span class="source-line-no">506</span><span id="line-506"> for (ServerName server : initial.getLiveServerMetrics().keySet()) {</span> |
| <span class="source-line-no">507</span><span id="line-507"> toKill.remove(server);</span> |
| <span class="source-line-no">508</span><span id="line-508"> }</span> |
| <span class="source-line-no">509</span><span id="line-509"></span> |
| <span class="source-line-no">510</span><span id="line-510"> List<IOException> deferred = new ArrayList<>();</span> |
| <span class="source-line-no">511</span><span id="line-511"></span> |
| <span class="source-line-no">512</span><span id="line-512"> for (ServerName sn : toStart) {</span> |
| <span class="source-line-no">513</span><span id="line-513"> try {</span> |
| <span class="source-line-no">514</span><span id="line-514"> if (</span> |
| <span class="source-line-no">515</span><span id="line-515"> !clusterManager.isRunning(ServiceType.HBASE_REGIONSERVER, sn.getHostname(), sn.getPort())</span> |
| <span class="source-line-no">516</span><span id="line-516"> && master.getPort() != sn.getPort()</span> |
| <span class="source-line-no">517</span><span id="line-517"> ) {</span> |
| <span class="source-line-no">518</span><span id="line-518"> LOG.info("Restoring cluster - starting initial region server: {}", sn.getAddress());</span> |
| <span class="source-line-no">519</span><span id="line-519"> startRegionServer(sn.getHostname(), sn.getPort());</span> |
| <span class="source-line-no">520</span><span id="line-520"> }</span> |
| <span class="source-line-no">521</span><span id="line-521"> } catch (IOException ex) {</span> |
| <span class="source-line-no">522</span><span id="line-522"> deferred.add(ex);</span> |
| <span class="source-line-no">523</span><span id="line-523"> }</span> |
| <span class="source-line-no">524</span><span id="line-524"> }</span> |
| <span class="source-line-no">525</span><span id="line-525"></span> |
| <span class="source-line-no">526</span><span id="line-526"> for (ServerName sn : toKill) {</span> |
| <span class="source-line-no">527</span><span id="line-527"> try {</span> |
| <span class="source-line-no">528</span><span id="line-528"> if (</span> |
| <span class="source-line-no">529</span><span id="line-529"> clusterManager.isRunning(ServiceType.HBASE_REGIONSERVER, sn.getHostname(), sn.getPort())</span> |
| <span class="source-line-no">530</span><span id="line-530"> && master.getPort() != sn.getPort()</span> |
| <span class="source-line-no">531</span><span id="line-531"> ) {</span> |
| <span class="source-line-no">532</span><span id="line-532"> LOG.info("Restoring cluster - stopping initial region server: {}", sn.getAddress());</span> |
| <span class="source-line-no">533</span><span id="line-533"> stopRegionServer(sn);</span> |
| <span class="source-line-no">534</span><span id="line-534"> }</span> |
| <span class="source-line-no">535</span><span id="line-535"> } catch (IOException ex) {</span> |
| <span class="source-line-no">536</span><span id="line-536"> deferred.add(ex);</span> |
| <span class="source-line-no">537</span><span id="line-537"> }</span> |
| <span class="source-line-no">538</span><span id="line-538"> }</span> |
| <span class="source-line-no">539</span><span id="line-539"> if (!deferred.isEmpty()) {</span> |
| <span class="source-line-no">540</span><span id="line-540"> LOG.warn("Restoring cluster - restoring region servers reported {} errors:", deferred.size());</span> |
| <span class="source-line-no">541</span><span id="line-541"> for (int i = 0; i < deferred.size() && i < 3; i++) {</span> |
| <span class="source-line-no">542</span><span id="line-542"> LOG.warn(Objects.toString(deferred.get(i)));</span> |
| <span class="source-line-no">543</span><span id="line-543"> }</span> |
| <span class="source-line-no">544</span><span id="line-544"> }</span> |
| <span class="source-line-no">545</span><span id="line-545"></span> |
| <span class="source-line-no">546</span><span id="line-546"> return deferred.isEmpty();</span> |
| <span class="source-line-no">547</span><span id="line-547"> }</span> |
| <span class="source-line-no">548</span><span id="line-548"></span> |
| <span class="source-line-no">549</span><span id="line-549"> protected boolean restoreAdmin() throws IOException {</span> |
| <span class="source-line-no">550</span><span id="line-550"> // While restoring above, if the HBase Master which was initially the Active one, was down</span> |
| <span class="source-line-no">551</span><span id="line-551"> // and the restore put the cluster back to Initial configuration, HAdmin instance will need</span> |
| <span class="source-line-no">552</span><span id="line-552"> // to refresh its connections (otherwise it will return incorrect information) or we can</span> |
| <span class="source-line-no">553</span><span id="line-553"> // point it to new instance.</span> |
| <span class="source-line-no">554</span><span id="line-554"> admin.close();</span> |
| <span class="source-line-no">555</span><span id="line-555"> this.admin = this.connection.getAdmin();</span> |
| <span class="source-line-no">556</span><span id="line-556"> LOG.info("Added new HBaseAdmin");</span> |
| <span class="source-line-no">557</span><span id="line-557"> return true;</span> |
| <span class="source-line-no">558</span><span id="line-558"> }</span> |
| <span class="source-line-no">559</span><span id="line-559">}</span> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </main> |
| </body> |
| </html> |
