blob: 3e0d8f158b38ea8c49ccd040adb7e3196343b08f [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.rsgroup;<a name="line.18"></a>
<span class="sourceLineNo">019</span><a name="line.19"></a>
<span class="sourceLineNo">020</span>import static org.junit.Assert.assertEquals;<a name="line.20"></a>
<span class="sourceLineNo">021</span>import static org.junit.Assert.assertFalse;<a name="line.21"></a>
<span class="sourceLineNo">022</span><a name="line.22"></a>
<span class="sourceLineNo">023</span>import org.apache.hadoop.hbase.HBaseClassTestRule;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import org.apache.hadoop.hbase.HBaseClusterInterface;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import org.apache.hadoop.hbase.HBaseTestingUtil;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.SingleProcessHBaseCluster;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.StartTestingClusterOption;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.TableName;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.Waiter;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.client.Admin;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.master.HMaster;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.master.ServerManager;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.testclassification.MediumTests;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.testclassification.RSGroupTests;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.junit.AfterClass;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.junit.BeforeClass;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.junit.ClassRule;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.junit.Rule;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.junit.Test;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.junit.experimental.categories.Category;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.junit.rules.TestName;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.slf4j.Logger;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.slf4j.LoggerFactory;<a name="line.46"></a>
<span class="sourceLineNo">047</span><a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.apache.hbase.thirdparty.com.google.common.collect.Sets;<a name="line.48"></a>
<span class="sourceLineNo">049</span><a name="line.49"></a>
<span class="sourceLineNo">050</span>/**<a name="line.50"></a>
<span class="sourceLineNo">051</span> * This tests that GroupBasedBalancer will use data in zk to do balancing during master startup.<a name="line.51"></a>
<span class="sourceLineNo">052</span> * This does not test retain assignment.<a name="line.52"></a>
<span class="sourceLineNo">053</span> * &lt;p/&gt;<a name="line.53"></a>
<span class="sourceLineNo">054</span> * The tests brings up 3 RS, creates a new RS group 'my_group', moves 1 RS to 'my_group', assigns<a name="line.54"></a>
<span class="sourceLineNo">055</span> * 'hbase:rsgroup' to 'my_group', and kill the only server in that group so that 'hbase:rsgroup'<a name="line.55"></a>
<span class="sourceLineNo">056</span> * table isn't available. It then kills the active master and waits for backup master to come<a name="line.56"></a>
<span class="sourceLineNo">057</span> * online. In new master, RSGroupInfoManagerImpl gets the data from zk and waits for the expected<a name="line.57"></a>
<span class="sourceLineNo">058</span> * assignment with a timeout.<a name="line.58"></a>
<span class="sourceLineNo">059</span> */<a name="line.59"></a>
<span class="sourceLineNo">060</span>@Category({ RSGroupTests.class, MediumTests.class })<a name="line.60"></a>
<span class="sourceLineNo">061</span>public class TestRSGroupsOfflineMode extends TestRSGroupsBase {<a name="line.61"></a>
<span class="sourceLineNo">062</span><a name="line.62"></a>
<span class="sourceLineNo">063</span> @ClassRule<a name="line.63"></a>
<span class="sourceLineNo">064</span> public static final HBaseClassTestRule CLASS_RULE =<a name="line.64"></a>
<span class="sourceLineNo">065</span> HBaseClassTestRule.forClass(TestRSGroupsOfflineMode.class);<a name="line.65"></a>
<span class="sourceLineNo">066</span><a name="line.66"></a>
<span class="sourceLineNo">067</span> private static final Logger LOG = LoggerFactory.getLogger(TestRSGroupsOfflineMode.class);<a name="line.67"></a>
<span class="sourceLineNo">068</span> private static HMaster master;<a name="line.68"></a>
<span class="sourceLineNo">069</span> private static Admin hbaseAdmin;<a name="line.69"></a>
<span class="sourceLineNo">070</span> private static HBaseTestingUtil TEST_UTIL;<a name="line.70"></a>
<span class="sourceLineNo">071</span> private static HBaseClusterInterface cluster;<a name="line.71"></a>
<span class="sourceLineNo">072</span> private final static long WAIT_TIMEOUT = 60000 * 5;<a name="line.72"></a>
<span class="sourceLineNo">073</span><a name="line.73"></a>
<span class="sourceLineNo">074</span> @Rule<a name="line.74"></a>
<span class="sourceLineNo">075</span> public TestName name = new TestName();<a name="line.75"></a>
<span class="sourceLineNo">076</span><a name="line.76"></a>
<span class="sourceLineNo">077</span> @BeforeClass<a name="line.77"></a>
<span class="sourceLineNo">078</span> public static void setUp() throws Exception {<a name="line.78"></a>
<span class="sourceLineNo">079</span> TEST_UTIL = new HBaseTestingUtil();<a name="line.79"></a>
<span class="sourceLineNo">080</span> RSGroupUtil.enableRSGroup(TEST_UTIL.getConfiguration());<a name="line.80"></a>
<span class="sourceLineNo">081</span> TEST_UTIL.getConfiguration().set(ServerManager.WAIT_ON_REGIONSERVERS_MINTOSTART, "1");<a name="line.81"></a>
<span class="sourceLineNo">082</span> StartTestingClusterOption option =<a name="line.82"></a>
<span class="sourceLineNo">083</span> StartTestingClusterOption.builder().numMasters(2).numRegionServers(3).numDataNodes(3).build();<a name="line.83"></a>
<span class="sourceLineNo">084</span> TEST_UTIL.startMiniCluster(option);<a name="line.84"></a>
<span class="sourceLineNo">085</span> cluster = TEST_UTIL.getHBaseCluster();<a name="line.85"></a>
<span class="sourceLineNo">086</span> master = ((SingleProcessHBaseCluster) cluster).getMaster();<a name="line.86"></a>
<span class="sourceLineNo">087</span> master.balanceSwitch(false);<a name="line.87"></a>
<span class="sourceLineNo">088</span> hbaseAdmin = TEST_UTIL.getAdmin();<a name="line.88"></a>
<span class="sourceLineNo">089</span> // wait till the balancer is in online mode<a name="line.89"></a>
<span class="sourceLineNo">090</span> TEST_UTIL.waitFor(WAIT_TIMEOUT, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.90"></a>
<span class="sourceLineNo">091</span> @Override<a name="line.91"></a>
<span class="sourceLineNo">092</span> public boolean evaluate() throws Exception {<a name="line.92"></a>
<span class="sourceLineNo">093</span> return master.isInitialized() &amp;&amp;<a name="line.93"></a>
<span class="sourceLineNo">094</span> ((RSGroupBasedLoadBalancer) master.getLoadBalancer()).isOnline() &amp;&amp;<a name="line.94"></a>
<span class="sourceLineNo">095</span> master.getServerManager().getOnlineServersList().size() &gt;= 3;<a name="line.95"></a>
<span class="sourceLineNo">096</span> }<a name="line.96"></a>
<span class="sourceLineNo">097</span> });<a name="line.97"></a>
<span class="sourceLineNo">098</span> }<a name="line.98"></a>
<span class="sourceLineNo">099</span><a name="line.99"></a>
<span class="sourceLineNo">100</span> @AfterClass<a name="line.100"></a>
<span class="sourceLineNo">101</span> public static void tearDown() throws Exception {<a name="line.101"></a>
<span class="sourceLineNo">102</span> TEST_UTIL.shutdownMiniCluster();<a name="line.102"></a>
<span class="sourceLineNo">103</span> }<a name="line.103"></a>
<span class="sourceLineNo">104</span><a name="line.104"></a>
<span class="sourceLineNo">105</span> @Test<a name="line.105"></a>
<span class="sourceLineNo">106</span> public void testOffline() throws Exception, InterruptedException {<a name="line.106"></a>
<span class="sourceLineNo">107</span> // Table should be after group table name so it gets assigned later.<a name="line.107"></a>
<span class="sourceLineNo">108</span> final TableName failoverTable = TableName.valueOf(getNameWithoutIndex(name.getMethodName()));<a name="line.108"></a>
<span class="sourceLineNo">109</span> TEST_UTIL.createTable(failoverTable, Bytes.toBytes("f"));<a name="line.109"></a>
<span class="sourceLineNo">110</span> final HRegionServer killRS = ((SingleProcessHBaseCluster) cluster).getRegionServer(0);<a name="line.110"></a>
<span class="sourceLineNo">111</span> final HRegionServer groupRS = ((SingleProcessHBaseCluster) cluster).getRegionServer(1);<a name="line.111"></a>
<span class="sourceLineNo">112</span> final HRegionServer failoverRS = ((SingleProcessHBaseCluster) cluster).getRegionServer(2);<a name="line.112"></a>
<span class="sourceLineNo">113</span> String newGroup = "my_group";<a name="line.113"></a>
<span class="sourceLineNo">114</span> Admin admin = TEST_UTIL.getAdmin();<a name="line.114"></a>
<span class="sourceLineNo">115</span> admin.addRSGroup(newGroup);<a name="line.115"></a>
<span class="sourceLineNo">116</span> if (master.getAssignmentManager().getRegionStates().getRegionAssignments()<a name="line.116"></a>
<span class="sourceLineNo">117</span> .containsValue(failoverRS.getServerName())) {<a name="line.117"></a>
<span class="sourceLineNo">118</span> for (RegionInfo regionInfo : hbaseAdmin.getRegions(failoverRS.getServerName())) {<a name="line.118"></a>
<span class="sourceLineNo">119</span> hbaseAdmin.move(regionInfo.getEncodedNameAsBytes(), failoverRS.getServerName());<a name="line.119"></a>
<span class="sourceLineNo">120</span> }<a name="line.120"></a>
<span class="sourceLineNo">121</span> LOG.info("Waiting for region unassignments on failover RS...");<a name="line.121"></a>
<span class="sourceLineNo">122</span> TEST_UTIL.waitFor(WAIT_TIMEOUT, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.122"></a>
<span class="sourceLineNo">123</span> @Override<a name="line.123"></a>
<span class="sourceLineNo">124</span> public boolean evaluate() throws Exception {<a name="line.124"></a>
<span class="sourceLineNo">125</span> return !master.getServerManager().getLoad(failoverRS.getServerName()).getRegionMetrics()<a name="line.125"></a>
<span class="sourceLineNo">126</span> .isEmpty();<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> }<a name="line.129"></a>
<span class="sourceLineNo">130</span><a name="line.130"></a>
<span class="sourceLineNo">131</span> // Move server to group and make sure all tables are assigned.<a name="line.131"></a>
<span class="sourceLineNo">132</span> admin.moveServersToRSGroup(Sets.newHashSet(groupRS.getServerName().getAddress()), newGroup);<a name="line.132"></a>
<span class="sourceLineNo">133</span> TEST_UTIL.waitFor(WAIT_TIMEOUT, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.133"></a>
<span class="sourceLineNo">134</span> @Override<a name="line.134"></a>
<span class="sourceLineNo">135</span> public boolean evaluate() throws Exception {<a name="line.135"></a>
<span class="sourceLineNo">136</span> return groupRS.getNumberOfOnlineRegions() &lt; 1 &amp;&amp;<a name="line.136"></a>
<span class="sourceLineNo">137</span> master.getAssignmentManager().getRegionStates().getRegionsInTransitionCount() &lt; 1;<a name="line.137"></a>
<span class="sourceLineNo">138</span> }<a name="line.138"></a>
<span class="sourceLineNo">139</span> });<a name="line.139"></a>
<span class="sourceLineNo">140</span> // Move table to group and wait.<a name="line.140"></a>
<span class="sourceLineNo">141</span> admin.setRSGroup(Sets.newHashSet(RSGroupInfoManagerImpl.RSGROUP_TABLE_NAME), newGroup);<a name="line.141"></a>
<span class="sourceLineNo">142</span> LOG.info("Waiting for move table...");<a name="line.142"></a>
<span class="sourceLineNo">143</span> TEST_UTIL.waitFor(WAIT_TIMEOUT, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.143"></a>
<span class="sourceLineNo">144</span> @Override<a name="line.144"></a>
<span class="sourceLineNo">145</span> public boolean evaluate() throws Exception {<a name="line.145"></a>
<span class="sourceLineNo">146</span> return groupRS.getNumberOfOnlineRegions() == 1;<a name="line.146"></a>
<span class="sourceLineNo">147</span> }<a name="line.147"></a>
<span class="sourceLineNo">148</span> });<a name="line.148"></a>
<span class="sourceLineNo">149</span><a name="line.149"></a>
<span class="sourceLineNo">150</span> groupRS.stop("die");<a name="line.150"></a>
<span class="sourceLineNo">151</span> // Race condition here.<a name="line.151"></a>
<span class="sourceLineNo">152</span> TEST_UTIL.getHBaseCluster().getMaster().stopMaster();<a name="line.152"></a>
<span class="sourceLineNo">153</span> LOG.info("Waiting for offline mode...");<a name="line.153"></a>
<span class="sourceLineNo">154</span> TEST_UTIL.waitFor(WAIT_TIMEOUT, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.154"></a>
<span class="sourceLineNo">155</span> @Override<a name="line.155"></a>
<span class="sourceLineNo">156</span> public boolean evaluate() throws Exception {<a name="line.156"></a>
<span class="sourceLineNo">157</span> return TEST_UTIL.getHBaseCluster().getMaster() != null &amp;&amp;<a name="line.157"></a>
<span class="sourceLineNo">158</span> TEST_UTIL.getHBaseCluster().getMaster().isActiveMaster() &amp;&amp;<a name="line.158"></a>
<span class="sourceLineNo">159</span> TEST_UTIL.getHBaseCluster().getMaster().isInitialized() &amp;&amp;<a name="line.159"></a>
<span class="sourceLineNo">160</span> TEST_UTIL.getHBaseCluster().getMaster().getServerManager().getOnlineServers().size() &lt;= 3;<a name="line.160"></a>
<span class="sourceLineNo">161</span> }<a name="line.161"></a>
<span class="sourceLineNo">162</span> });<a name="line.162"></a>
<span class="sourceLineNo">163</span><a name="line.163"></a>
<span class="sourceLineNo">164</span> // Get groupInfoManager from the new active master.<a name="line.164"></a>
<span class="sourceLineNo">165</span> RSGroupInfoManager groupMgr =<a name="line.165"></a>
<span class="sourceLineNo">166</span> ((SingleProcessHBaseCluster) cluster).getMaster().getRSGroupInfoManager();<a name="line.166"></a>
<span class="sourceLineNo">167</span> // Make sure balancer is in offline mode, since this is what we're testing.<a name="line.167"></a>
<span class="sourceLineNo">168</span> assertFalse(groupMgr.isOnline());<a name="line.168"></a>
<span class="sourceLineNo">169</span> // Kill final regionserver to see the failover happens for all tables except GROUP table since<a name="line.169"></a>
<span class="sourceLineNo">170</span> // it's group does not have any online RS.<a name="line.170"></a>
<span class="sourceLineNo">171</span> killRS.stop("die");<a name="line.171"></a>
<span class="sourceLineNo">172</span> master = TEST_UTIL.getHBaseCluster().getMaster();<a name="line.172"></a>
<span class="sourceLineNo">173</span> LOG.info("Waiting for new table assignment...");<a name="line.173"></a>
<span class="sourceLineNo">174</span> TEST_UTIL.waitFor(WAIT_TIMEOUT, new Waiter.Predicate&lt;Exception&gt;() {<a name="line.174"></a>
<span class="sourceLineNo">175</span> @Override<a name="line.175"></a>
<span class="sourceLineNo">176</span> public boolean evaluate() throws Exception {<a name="line.176"></a>
<span class="sourceLineNo">177</span> return failoverRS.getRegions(failoverTable).size() &gt;= 1;<a name="line.177"></a>
<span class="sourceLineNo">178</span> }<a name="line.178"></a>
<span class="sourceLineNo">179</span> });<a name="line.179"></a>
<span class="sourceLineNo">180</span> assertEquals(0, failoverRS.getRegions(RSGroupInfoManagerImpl.RSGROUP_TABLE_NAME).size());<a name="line.180"></a>
<span class="sourceLineNo">181</span><a name="line.181"></a>
<span class="sourceLineNo">182</span> // Need this for minicluster to shutdown cleanly.<a name="line.182"></a>
<span class="sourceLineNo">183</span> master.stopMaster();<a name="line.183"></a>
<span class="sourceLineNo">184</span> }<a name="line.184"></a>
<span class="sourceLineNo">185</span>}<a name="line.185"></a>
</pre>
</div>
</body>
</html>