| <!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.rsgroup, class: RSGroupInfoManagerImpl, class: RSGroupStartupWorker"> |
| <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.rsgroup;</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.ByteArrayInputStream;</span> |
| <span class="source-line-no">021</span><span id="line-21">import java.io.IOException;</span> |
| <span class="source-line-no">022</span><span id="line-22">import java.util.ArrayList;</span> |
| <span class="source-line-no">023</span><span id="line-23">import java.util.Collection;</span> |
| <span class="source-line-no">024</span><span id="line-24">import java.util.Collections;</span> |
| <span class="source-line-no">025</span><span id="line-25">import java.util.HashMap;</span> |
| <span class="source-line-no">026</span><span id="line-26">import java.util.HashSet;</span> |
| <span class="source-line-no">027</span><span id="line-27">import java.util.LinkedList;</span> |
| <span class="source-line-no">028</span><span id="line-28">import java.util.List;</span> |
| <span class="source-line-no">029</span><span id="line-29">import java.util.Map;</span> |
| <span class="source-line-no">030</span><span id="line-30">import java.util.Optional;</span> |
| <span class="source-line-no">031</span><span id="line-31">import java.util.OptionalLong;</span> |
| <span class="source-line-no">032</span><span id="line-32">import java.util.Set;</span> |
| <span class="source-line-no">033</span><span id="line-33">import java.util.SortedSet;</span> |
| <span class="source-line-no">034</span><span id="line-34">import java.util.TreeSet;</span> |
| <span class="source-line-no">035</span><span id="line-35">import java.util.concurrent.Future;</span> |
| <span class="source-line-no">036</span><span id="line-36">import java.util.function.Function;</span> |
| <span class="source-line-no">037</span><span id="line-37">import java.util.stream.Collectors;</span> |
| <span class="source-line-no">038</span><span id="line-38">import org.apache.commons.lang3.StringUtils;</span> |
| <span class="source-line-no">039</span><span id="line-39">import org.apache.hadoop.conf.Configuration;</span> |
| <span class="source-line-no">040</span><span id="line-40">import org.apache.hadoop.hbase.Coprocessor;</span> |
| <span class="source-line-no">041</span><span id="line-41">import org.apache.hadoop.hbase.DoNotRetryIOException;</span> |
| <span class="source-line-no">042</span><span id="line-42">import org.apache.hadoop.hbase.HConstants;</span> |
| <span class="source-line-no">043</span><span id="line-43">import org.apache.hadoop.hbase.NamespaceDescriptor;</span> |
| <span class="source-line-no">044</span><span id="line-44">import org.apache.hadoop.hbase.ServerName;</span> |
| <span class="source-line-no">045</span><span id="line-45">import org.apache.hadoop.hbase.TableDescriptors;</span> |
| <span class="source-line-no">046</span><span id="line-46">import org.apache.hadoop.hbase.TableName;</span> |
| <span class="source-line-no">047</span><span id="line-47">import org.apache.hadoop.hbase.client.AsyncClusterConnection;</span> |
| <span class="source-line-no">048</span><span id="line-48">import org.apache.hadoop.hbase.client.AsyncTable;</span> |
| <span class="source-line-no">049</span><span id="line-49">import org.apache.hadoop.hbase.client.BalanceRequest;</span> |
| <span class="source-line-no">050</span><span id="line-50">import org.apache.hadoop.hbase.client.BalanceResponse;</span> |
| <span class="source-line-no">051</span><span id="line-51">import org.apache.hadoop.hbase.client.ColumnFamilyDescriptorBuilder;</span> |
| <span class="source-line-no">052</span><span id="line-52">import org.apache.hadoop.hbase.client.CoprocessorDescriptorBuilder;</span> |
| <span class="source-line-no">053</span><span id="line-53">import org.apache.hadoop.hbase.client.Delete;</span> |
| <span class="source-line-no">054</span><span id="line-54">import org.apache.hadoop.hbase.client.Get;</span> |
| <span class="source-line-no">055</span><span id="line-55">import org.apache.hadoop.hbase.client.Mutation;</span> |
| <span class="source-line-no">056</span><span id="line-56">import org.apache.hadoop.hbase.client.Put;</span> |
| <span class="source-line-no">057</span><span id="line-57">import org.apache.hadoop.hbase.client.RegionInfo;</span> |
| <span class="source-line-no">058</span><span id="line-58">import org.apache.hadoop.hbase.client.Result;</span> |
| <span class="source-line-no">059</span><span id="line-59">import org.apache.hadoop.hbase.client.ResultScanner;</span> |
| <span class="source-line-no">060</span><span id="line-60">import org.apache.hadoop.hbase.client.TableDescriptor;</span> |
| <span class="source-line-no">061</span><span id="line-61">import org.apache.hadoop.hbase.client.TableDescriptorBuilder;</span> |
| <span class="source-line-no">062</span><span id="line-62">import org.apache.hadoop.hbase.client.TableState;</span> |
| <span class="source-line-no">063</span><span id="line-63">import org.apache.hadoop.hbase.constraint.ConstraintException;</span> |
| <span class="source-line-no">064</span><span id="line-64">import org.apache.hadoop.hbase.coprocessor.MultiRowMutationEndpoint;</span> |
| <span class="source-line-no">065</span><span id="line-65">import org.apache.hadoop.hbase.exceptions.DeserializationException;</span> |
| <span class="source-line-no">066</span><span id="line-66">import org.apache.hadoop.hbase.master.LoadBalancer;</span> |
| <span class="source-line-no">067</span><span id="line-67">import org.apache.hadoop.hbase.master.MasterServices;</span> |
| <span class="source-line-no">068</span><span id="line-68">import org.apache.hadoop.hbase.master.RegionPlan;</span> |
| <span class="source-line-no">069</span><span id="line-69">import org.apache.hadoop.hbase.master.RegionState;</span> |
| <span class="source-line-no">070</span><span id="line-70">import org.apache.hadoop.hbase.master.ServerListener;</span> |
| <span class="source-line-no">071</span><span id="line-71">import org.apache.hadoop.hbase.master.ServerManager;</span> |
| <span class="source-line-no">072</span><span id="line-72">import org.apache.hadoop.hbase.master.TableStateManager;</span> |
| <span class="source-line-no">073</span><span id="line-73">import org.apache.hadoop.hbase.master.assignment.RegionStateNode;</span> |
| <span class="source-line-no">074</span><span id="line-74">import org.apache.hadoop.hbase.master.procedure.CreateTableProcedure;</span> |
| <span class="source-line-no">075</span><span id="line-75">import org.apache.hadoop.hbase.master.procedure.MasterProcedureEnv;</span> |
| <span class="source-line-no">076</span><span id="line-76">import org.apache.hadoop.hbase.master.procedure.MasterProcedureUtil;</span> |
| <span class="source-line-no">077</span><span id="line-77">import org.apache.hadoop.hbase.master.procedure.ProcedureSyncWait;</span> |
| <span class="source-line-no">078</span><span id="line-78">import org.apache.hadoop.hbase.net.Address;</span> |
| <span class="source-line-no">079</span><span id="line-79">import org.apache.hadoop.hbase.procedure2.Procedure;</span> |
| <span class="source-line-no">080</span><span id="line-80">import org.apache.hadoop.hbase.procedure2.ProcedureExecutor;</span> |
| <span class="source-line-no">081</span><span id="line-81">import org.apache.hadoop.hbase.protobuf.ProtobufMagic;</span> |
| <span class="source-line-no">082</span><span id="line-82">import org.apache.hadoop.hbase.regionserver.DisabledRegionSplitPolicy;</span> |
| <span class="source-line-no">083</span><span id="line-83">import org.apache.hadoop.hbase.util.Bytes;</span> |
| <span class="source-line-no">084</span><span id="line-84">import org.apache.hadoop.hbase.util.FutureUtils;</span> |
| <span class="source-line-no">085</span><span id="line-85">import org.apache.hadoop.hbase.util.Pair;</span> |
| <span class="source-line-no">086</span><span id="line-86">import org.apache.hadoop.hbase.util.Threads;</span> |
| <span class="source-line-no">087</span><span id="line-87">import org.apache.hadoop.hbase.zookeeper.ZKUtil;</span> |
| <span class="source-line-no">088</span><span id="line-88">import org.apache.hadoop.hbase.zookeeper.ZKWatcher;</span> |
| <span class="source-line-no">089</span><span id="line-89">import org.apache.hadoop.hbase.zookeeper.ZNodePaths;</span> |
| <span class="source-line-no">090</span><span id="line-90">import org.apache.hadoop.util.Shell;</span> |
| <span class="source-line-no">091</span><span id="line-91">import org.apache.yetus.audience.InterfaceAudience;</span> |
| <span class="source-line-no">092</span><span id="line-92">import org.apache.zookeeper.KeeperException;</span> |
| <span class="source-line-no">093</span><span id="line-93">import org.slf4j.Logger;</span> |
| <span class="source-line-no">094</span><span id="line-94">import org.slf4j.LoggerFactory;</span> |
| <span class="source-line-no">095</span><span id="line-95"></span> |
| <span class="source-line-no">096</span><span id="line-96">import org.apache.hbase.thirdparty.com.google.common.collect.ImmutableMap;</span> |
| <span class="source-line-no">097</span><span id="line-97">import org.apache.hbase.thirdparty.com.google.common.collect.Lists;</span> |
| <span class="source-line-no">098</span><span id="line-98">import org.apache.hbase.thirdparty.com.google.common.collect.Maps;</span> |
| <span class="source-line-no">099</span><span id="line-99">import org.apache.hbase.thirdparty.com.google.common.collect.Sets;</span> |
| <span class="source-line-no">100</span><span id="line-100"></span> |
| <span class="source-line-no">101</span><span id="line-101">import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;</span> |
| <span class="source-line-no">102</span><span id="line-102">import org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos.MutationProto;</span> |
| <span class="source-line-no">103</span><span id="line-103">import org.apache.hadoop.hbase.shaded.protobuf.generated.MultiRowMutationProtos.MultiRowMutationService;</span> |
| <span class="source-line-no">104</span><span id="line-104">import org.apache.hadoop.hbase.shaded.protobuf.generated.MultiRowMutationProtos.MutateRowsRequest;</span> |
| <span class="source-line-no">105</span><span id="line-105">import org.apache.hadoop.hbase.shaded.protobuf.generated.MultiRowMutationProtos.MutateRowsResponse;</span> |
| <span class="source-line-no">106</span><span id="line-106">import org.apache.hadoop.hbase.shaded.protobuf.generated.RSGroupProtos;</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"> * This is an implementation of {@link RSGroupInfoManager} which makes use of an HBase table as the</span> |
| <span class="source-line-no">110</span><span id="line-110"> * persistence store for the group information. It also makes use of zookeeper to store group</span> |
| <span class="source-line-no">111</span><span id="line-111"> * information needed for bootstrapping during offline mode.</span> |
| <span class="source-line-no">112</span><span id="line-112"> * <h2>Concurrency</h2> RSGroup state is kept locally in Maps. There is a rsgroup name to cached</span> |
| <span class="source-line-no">113</span><span id="line-113"> * RSGroupInfo Map at {@link RSGroupInfoHolder#groupName2Group}. These Maps are persisted to the</span> |
| <span class="source-line-no">114</span><span id="line-114"> * hbase:rsgroup table (and cached in zk) on each modification.</span> |
| <span class="source-line-no">115</span><span id="line-115"> * <p/></span> |
| <span class="source-line-no">116</span><span id="line-116"> * Mutations on state are synchronized but reads can continue without having to wait on an instance</span> |
| <span class="source-line-no">117</span><span id="line-117"> * monitor, mutations do wholesale replace of the Maps on update -- Copy-On-Write; the local Maps of</span> |
| <span class="source-line-no">118</span><span id="line-118"> * state are read-only, just-in-case (see flushConfig).</span> |
| <span class="source-line-no">119</span><span id="line-119"> * <p/></span> |
| <span class="source-line-no">120</span><span id="line-120"> * Reads must not block else there is a danger we'll deadlock.</span> |
| <span class="source-line-no">121</span><span id="line-121"> * <p/></span> |
| <span class="source-line-no">122</span><span id="line-122"> * Clients of this class, the {@link RSGroupAdminEndpoint} for example, want to query and then act</span> |
| <span class="source-line-no">123</span><span id="line-123"> * on the results of the query modifying cache in zookeeper without another thread making</span> |
| <span class="source-line-no">124</span><span id="line-124"> * intermediate modifications. These clients synchronize on the 'this' instance so no other has</span> |
| <span class="source-line-no">125</span><span id="line-125"> * access concurrently. Reads must be able to continue concurrently.</span> |
| <span class="source-line-no">126</span><span id="line-126"> */</span> |
| <span class="source-line-no">127</span><span id="line-127">@InterfaceAudience.Private</span> |
| <span class="source-line-no">128</span><span id="line-128">final class RSGroupInfoManagerImpl implements RSGroupInfoManager {</span> |
| <span class="source-line-no">129</span><span id="line-129"> private static final Logger LOG = LoggerFactory.getLogger(RSGroupInfoManagerImpl.class);</span> |
| <span class="source-line-no">130</span><span id="line-130"></span> |
| <span class="source-line-no">131</span><span id="line-131"> // Assigned before user tables</span> |
| <span class="source-line-no">132</span><span id="line-132"> static final TableName RSGROUP_TABLE_NAME =</span> |
| <span class="source-line-no">133</span><span id="line-133"> TableName.valueOf(NamespaceDescriptor.SYSTEM_NAMESPACE_NAME_STR, "rsgroup");</span> |
| <span class="source-line-no">134</span><span id="line-134"></span> |
| <span class="source-line-no">135</span><span id="line-135"> static final String KEEP_ONE_SERVER_IN_DEFAULT_ERROR_MESSAGE =</span> |
| <span class="source-line-no">136</span><span id="line-136"> "should keep at least " + "one server in 'default' RSGroup.";</span> |
| <span class="source-line-no">137</span><span id="line-137"></span> |
| <span class="source-line-no">138</span><span id="line-138"> /** Define the config key of retries threshold when movements failed */</span> |
| <span class="source-line-no">139</span><span id="line-139"> static final String FAILED_MOVE_MAX_RETRY = "hbase.rsgroup.move.max.retry";</span> |
| <span class="source-line-no">140</span><span id="line-140"></span> |
| <span class="source-line-no">141</span><span id="line-141"> /** Define the default number of retries */</span> |
| <span class="source-line-no">142</span><span id="line-142"> static final int DEFAULT_MAX_RETRY_VALUE = 50;</span> |
| <span class="source-line-no">143</span><span id="line-143"></span> |
| <span class="source-line-no">144</span><span id="line-144"> private static final String RS_GROUP_ZNODE = "rsgroup";</span> |
| <span class="source-line-no">145</span><span id="line-145"></span> |
| <span class="source-line-no">146</span><span id="line-146"> static final byte[] META_FAMILY_BYTES = Bytes.toBytes("m");</span> |
| <span class="source-line-no">147</span><span id="line-147"></span> |
| <span class="source-line-no">148</span><span id="line-148"> static final byte[] META_QUALIFIER_BYTES = Bytes.toBytes("i");</span> |
| <span class="source-line-no">149</span><span id="line-149"></span> |
| <span class="source-line-no">150</span><span id="line-150"> static final String MIGRATE_THREAD_NAME = "Migrate-RSGroup-Tables";</span> |
| <span class="source-line-no">151</span><span id="line-151"></span> |
| <span class="source-line-no">152</span><span id="line-152"> private static final byte[] ROW_KEY = { 0 };</span> |
| <span class="source-line-no">153</span><span id="line-153"></span> |
| <span class="source-line-no">154</span><span id="line-154"> /** Table descriptor for <code>hbase:rsgroup</code> catalog table */</span> |
| <span class="source-line-no">155</span><span id="line-155"> private static final TableDescriptor RSGROUP_TABLE_DESC;</span> |
| <span class="source-line-no">156</span><span id="line-156"> static {</span> |
| <span class="source-line-no">157</span><span id="line-157"> TableDescriptorBuilder builder = TableDescriptorBuilder.newBuilder(RSGROUP_TABLE_NAME)</span> |
| <span class="source-line-no">158</span><span id="line-158"> .setColumnFamily(ColumnFamilyDescriptorBuilder.of(META_FAMILY_BYTES))</span> |
| <span class="source-line-no">159</span><span id="line-159"> .setRegionSplitPolicyClassName(DisabledRegionSplitPolicy.class.getName());</span> |
| <span class="source-line-no">160</span><span id="line-160"> try {</span> |
| <span class="source-line-no">161</span><span id="line-161"> builder.setCoprocessor(</span> |
| <span class="source-line-no">162</span><span id="line-162"> CoprocessorDescriptorBuilder.newBuilder(MultiRowMutationEndpoint.class.getName())</span> |
| <span class="source-line-no">163</span><span id="line-163"> .setPriority(Coprocessor.PRIORITY_SYSTEM).build());</span> |
| <span class="source-line-no">164</span><span id="line-164"> } catch (IOException ex) {</span> |
| <span class="source-line-no">165</span><span id="line-165"> throw new Error(ex);</span> |
| <span class="source-line-no">166</span><span id="line-166"> }</span> |
| <span class="source-line-no">167</span><span id="line-167"> RSGROUP_TABLE_DESC = builder.build();</span> |
| <span class="source-line-no">168</span><span id="line-168"> }</span> |
| <span class="source-line-no">169</span><span id="line-169"></span> |
| <span class="source-line-no">170</span><span id="line-170"> // There two Maps are immutable and wholesale replaced on each modification</span> |
| <span class="source-line-no">171</span><span id="line-171"> // so are safe to access concurrently. See class comment.</span> |
| <span class="source-line-no">172</span><span id="line-172"> private static final class RSGroupInfoHolder {</span> |
| <span class="source-line-no">173</span><span id="line-173"> final ImmutableMap<String, RSGroupInfo> groupName2Group;</span> |
| <span class="source-line-no">174</span><span id="line-174"> final ImmutableMap<TableName, RSGroupInfo> tableName2Group;</span> |
| <span class="source-line-no">175</span><span id="line-175"></span> |
| <span class="source-line-no">176</span><span id="line-176"> RSGroupInfoHolder() {</span> |
| <span class="source-line-no">177</span><span id="line-177"> this(Collections.emptyMap());</span> |
| <span class="source-line-no">178</span><span id="line-178"> }</span> |
| <span class="source-line-no">179</span><span id="line-179"></span> |
| <span class="source-line-no">180</span><span id="line-180"> RSGroupInfoHolder(Map<String, RSGroupInfo> rsGroupMap) {</span> |
| <span class="source-line-no">181</span><span id="line-181"> ImmutableMap.Builder<String, RSGroupInfo> group2Name2GroupBuilder = ImmutableMap.builder();</span> |
| <span class="source-line-no">182</span><span id="line-182"> ImmutableMap.Builder<TableName, RSGroupInfo> tableName2GroupBuilder = ImmutableMap.builder();</span> |
| <span class="source-line-no">183</span><span id="line-183"> rsGroupMap.forEach((groupName, rsGroupInfo) -> {</span> |
| <span class="source-line-no">184</span><span id="line-184"> group2Name2GroupBuilder.put(groupName, rsGroupInfo);</span> |
| <span class="source-line-no">185</span><span id="line-185"> if (!groupName.equals(RSGroupInfo.DEFAULT_GROUP)) {</span> |
| <span class="source-line-no">186</span><span id="line-186"> rsGroupInfo.getTables()</span> |
| <span class="source-line-no">187</span><span id="line-187"> .forEach(tableName -> tableName2GroupBuilder.put(tableName, rsGroupInfo));</span> |
| <span class="source-line-no">188</span><span id="line-188"> }</span> |
| <span class="source-line-no">189</span><span id="line-189"> });</span> |
| <span class="source-line-no">190</span><span id="line-190"> this.groupName2Group = group2Name2GroupBuilder.build();</span> |
| <span class="source-line-no">191</span><span id="line-191"> this.tableName2Group = tableName2GroupBuilder.build();</span> |
| <span class="source-line-no">192</span><span id="line-192"> }</span> |
| <span class="source-line-no">193</span><span id="line-193"> }</span> |
| <span class="source-line-no">194</span><span id="line-194"></span> |
| <span class="source-line-no">195</span><span id="line-195"> private volatile RSGroupInfoHolder holder = new RSGroupInfoHolder();</span> |
| <span class="source-line-no">196</span><span id="line-196"></span> |
| <span class="source-line-no">197</span><span id="line-197"> private final MasterServices masterServices;</span> |
| <span class="source-line-no">198</span><span id="line-198"> private final AsyncClusterConnection conn;</span> |
| <span class="source-line-no">199</span><span id="line-199"> private final ZKWatcher watcher;</span> |
| <span class="source-line-no">200</span><span id="line-200"> private final RSGroupStartupWorker rsGroupStartupWorker;</span> |
| <span class="source-line-no">201</span><span id="line-201"> // contains list of groups that were last flushed to persistent store</span> |
| <span class="source-line-no">202</span><span id="line-202"> private Set<String> prevRSGroups = new HashSet<>();</span> |
| <span class="source-line-no">203</span><span id="line-203"></span> |
| <span class="source-line-no">204</span><span id="line-204"> // Package visibility for testing</span> |
| <span class="source-line-no">205</span><span id="line-205"> static class RSGroupMappingScript {</span> |
| <span class="source-line-no">206</span><span id="line-206"> static final String RS_GROUP_MAPPING_SCRIPT = "hbase.rsgroup.table.mapping.script";</span> |
| <span class="source-line-no">207</span><span id="line-207"> static final String RS_GROUP_MAPPING_SCRIPT_TIMEOUT =</span> |
| <span class="source-line-no">208</span><span id="line-208"> "hbase.rsgroup.table.mapping.script.timeout";</span> |
| <span class="source-line-no">209</span><span id="line-209"></span> |
| <span class="source-line-no">210</span><span id="line-210"> private final String script;</span> |
| <span class="source-line-no">211</span><span id="line-211"> private final long scriptTimeout;</span> |
| <span class="source-line-no">212</span><span id="line-212"></span> |
| <span class="source-line-no">213</span><span id="line-213"> RSGroupMappingScript(Configuration conf) {</span> |
| <span class="source-line-no">214</span><span id="line-214"> script = conf.get(RS_GROUP_MAPPING_SCRIPT);</span> |
| <span class="source-line-no">215</span><span id="line-215"> scriptTimeout = conf.getLong(RS_GROUP_MAPPING_SCRIPT_TIMEOUT, 5000); // 5 seconds</span> |
| <span class="source-line-no">216</span><span id="line-216"> }</span> |
| <span class="source-line-no">217</span><span id="line-217"></span> |
| <span class="source-line-no">218</span><span id="line-218"> String getRSGroup(String namespace, String tablename) {</span> |
| <span class="source-line-no">219</span><span id="line-219"> if (script == null || script.isEmpty()) {</span> |
| <span class="source-line-no">220</span><span id="line-220"> return null;</span> |
| <span class="source-line-no">221</span><span id="line-221"> }</span> |
| <span class="source-line-no">222</span><span id="line-222"> Shell.ShellCommandExecutor rsgroupMappingScript =</span> |
| <span class="source-line-no">223</span><span id="line-223"> new Shell.ShellCommandExecutor(new String[] { script, "", "" }, null, null, scriptTimeout);</span> |
| <span class="source-line-no">224</span><span id="line-224"></span> |
| <span class="source-line-no">225</span><span id="line-225"> String[] exec = rsgroupMappingScript.getExecString();</span> |
| <span class="source-line-no">226</span><span id="line-226"> exec[1] = namespace;</span> |
| <span class="source-line-no">227</span><span id="line-227"> exec[2] = tablename;</span> |
| <span class="source-line-no">228</span><span id="line-228"> try {</span> |
| <span class="source-line-no">229</span><span id="line-229"> rsgroupMappingScript.execute();</span> |
| <span class="source-line-no">230</span><span id="line-230"> } catch (IOException e) {</span> |
| <span class="source-line-no">231</span><span id="line-231"> // This exception may happen, like process doesn't have permission to run this script.</span> |
| <span class="source-line-no">232</span><span id="line-232"> LOG.error("{}, placing {} back to default rsgroup", e.getMessage(),</span> |
| <span class="source-line-no">233</span><span id="line-233"> TableName.valueOf(namespace, tablename));</span> |
| <span class="source-line-no">234</span><span id="line-234"> return RSGroupInfo.DEFAULT_GROUP;</span> |
| <span class="source-line-no">235</span><span id="line-235"> }</span> |
| <span class="source-line-no">236</span><span id="line-236"> return rsgroupMappingScript.getOutput().trim();</span> |
| <span class="source-line-no">237</span><span id="line-237"> }</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"> private RSGroupMappingScript script;</span> |
| <span class="source-line-no">241</span><span id="line-241"></span> |
| <span class="source-line-no">242</span><span id="line-242"> private RSGroupInfoManagerImpl(MasterServices masterServices) {</span> |
| <span class="source-line-no">243</span><span id="line-243"> this.masterServices = masterServices;</span> |
| <span class="source-line-no">244</span><span id="line-244"> this.watcher = masterServices.getZooKeeper();</span> |
| <span class="source-line-no">245</span><span id="line-245"> this.conn = masterServices.getAsyncClusterConnection();</span> |
| <span class="source-line-no">246</span><span id="line-246"> this.rsGroupStartupWorker = new RSGroupStartupWorker();</span> |
| <span class="source-line-no">247</span><span id="line-247"> this.script = new RSGroupMappingScript(masterServices.getConfiguration());</span> |
| <span class="source-line-no">248</span><span id="line-248"> }</span> |
| <span class="source-line-no">249</span><span id="line-249"></span> |
| <span class="source-line-no">250</span><span id="line-250"> private synchronized void updateDefaultServers() {</span> |
| <span class="source-line-no">251</span><span id="line-251"> LOG.info("Updating default servers.");</span> |
| <span class="source-line-no">252</span><span id="line-252"> Map<String, RSGroupInfo> newGroupMap = Maps.newHashMap(holder.groupName2Group);</span> |
| <span class="source-line-no">253</span><span id="line-253"> RSGroupInfo oldDefaultGroupInfo = getRSGroup(RSGroupInfo.DEFAULT_GROUP);</span> |
| <span class="source-line-no">254</span><span id="line-254"> assert oldDefaultGroupInfo != null;</span> |
| <span class="source-line-no">255</span><span id="line-255"> RSGroupInfo newDefaultGroupInfo =</span> |
| <span class="source-line-no">256</span><span id="line-256"> new RSGroupInfo(RSGroupInfo.DEFAULT_GROUP, getDefaultServers());</span> |
| <span class="source-line-no">257</span><span id="line-257"> newDefaultGroupInfo.addAllTables(oldDefaultGroupInfo.getTables());</span> |
| <span class="source-line-no">258</span><span id="line-258"> newGroupMap.put(RSGroupInfo.DEFAULT_GROUP, newDefaultGroupInfo);</span> |
| <span class="source-line-no">259</span><span id="line-259"> // do not need to persist, as we do not persist default group.</span> |
| <span class="source-line-no">260</span><span id="line-260"> resetRSGroupMap(newGroupMap);</span> |
| <span class="source-line-no">261</span><span id="line-261"> LOG.info("Updated default servers, {} servers", newDefaultGroupInfo.getServers().size());</span> |
| <span class="source-line-no">262</span><span id="line-262"> if (LOG.isDebugEnabled()) {</span> |
| <span class="source-line-no">263</span><span id="line-263"> LOG.debug("New default servers list: {}", newDefaultGroupInfo.getServers());</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"></span> |
| <span class="source-line-no">267</span><span id="line-267"> private synchronized void init() throws IOException {</span> |
| <span class="source-line-no">268</span><span id="line-268"> refresh(false);</span> |
| <span class="source-line-no">269</span><span id="line-269"> masterServices.getServerManager().registerListener(new ServerListener() {</span> |
| <span class="source-line-no">270</span><span id="line-270"></span> |
| <span class="source-line-no">271</span><span id="line-271"> @Override</span> |
| <span class="source-line-no">272</span><span id="line-272"> public void serverAdded(ServerName serverName) {</span> |
| <span class="source-line-no">273</span><span id="line-273"> updateDefaultServers();</span> |
| <span class="source-line-no">274</span><span id="line-274"> }</span> |
| <span class="source-line-no">275</span><span id="line-275"></span> |
| <span class="source-line-no">276</span><span id="line-276"> @Override</span> |
| <span class="source-line-no">277</span><span id="line-277"> public void serverRemoved(ServerName serverName) {</span> |
| <span class="source-line-no">278</span><span id="line-278"> updateDefaultServers();</span> |
| <span class="source-line-no">279</span><span id="line-279"> }</span> |
| <span class="source-line-no">280</span><span id="line-280"> });</span> |
| <span class="source-line-no">281</span><span id="line-281"> }</span> |
| <span class="source-line-no">282</span><span id="line-282"></span> |
| <span class="source-line-no">283</span><span id="line-283"> static RSGroupInfoManager getInstance(MasterServices masterServices) throws IOException {</span> |
| <span class="source-line-no">284</span><span id="line-284"> RSGroupInfoManagerImpl instance = new RSGroupInfoManagerImpl(masterServices);</span> |
| <span class="source-line-no">285</span><span id="line-285"> instance.init();</span> |
| <span class="source-line-no">286</span><span id="line-286"> return instance;</span> |
| <span class="source-line-no">287</span><span id="line-287"> }</span> |
| <span class="source-line-no">288</span><span id="line-288"></span> |
| <span class="source-line-no">289</span><span id="line-289"> public void start() {</span> |
| <span class="source-line-no">290</span><span id="line-290"> // create system table of rsgroup</span> |
| <span class="source-line-no">291</span><span id="line-291"> rsGroupStartupWorker.start();</span> |
| <span class="source-line-no">292</span><span id="line-292"> }</span> |
| <span class="source-line-no">293</span><span id="line-293"></span> |
| <span class="source-line-no">294</span><span id="line-294"> @Override</span> |
| <span class="source-line-no">295</span><span id="line-295"> public synchronized void addRSGroup(RSGroupInfo rsGroupInfo) throws IOException {</span> |
| <span class="source-line-no">296</span><span id="line-296"> checkGroupName(rsGroupInfo.getName());</span> |
| <span class="source-line-no">297</span><span id="line-297"> Map<String, RSGroupInfo> rsGroupMap = holder.groupName2Group;</span> |
| <span class="source-line-no">298</span><span id="line-298"> if (</span> |
| <span class="source-line-no">299</span><span id="line-299"> rsGroupMap.get(rsGroupInfo.getName()) != null</span> |
| <span class="source-line-no">300</span><span id="line-300"> || rsGroupInfo.getName().equals(RSGroupInfo.DEFAULT_GROUP)</span> |
| <span class="source-line-no">301</span><span id="line-301"> ) {</span> |
| <span class="source-line-no">302</span><span id="line-302"> throw new ConstraintException("Group already exists: " + rsGroupInfo.getName());</span> |
| <span class="source-line-no">303</span><span id="line-303"> }</span> |
| <span class="source-line-no">304</span><span id="line-304"> Map<String, RSGroupInfo> newGroupMap = Maps.newHashMap(rsGroupMap);</span> |
| <span class="source-line-no">305</span><span id="line-305"> newGroupMap.put(rsGroupInfo.getName(), rsGroupInfo);</span> |
| <span class="source-line-no">306</span><span id="line-306"> flushConfig(newGroupMap);</span> |
| <span class="source-line-no">307</span><span id="line-307"> LOG.info("Add group {} done.", rsGroupInfo.getName());</span> |
| <span class="source-line-no">308</span><span id="line-308"> }</span> |
| <span class="source-line-no">309</span><span id="line-309"></span> |
| <span class="source-line-no">310</span><span id="line-310"> private RSGroupInfo getRSGroupInfo(final String groupName) throws ConstraintException {</span> |
| <span class="source-line-no">311</span><span id="line-311"> RSGroupInfo rsGroupInfo = holder.groupName2Group.get(groupName);</span> |
| <span class="source-line-no">312</span><span id="line-312"> if (rsGroupInfo == null) {</span> |
| <span class="source-line-no">313</span><span id="line-313"> throw new ConstraintException("RSGroup " + groupName + " does not exist");</span> |
| <span class="source-line-no">314</span><span id="line-314"> }</span> |
| <span class="source-line-no">315</span><span id="line-315"> return rsGroupInfo;</span> |
| <span class="source-line-no">316</span><span id="line-316"> }</span> |
| <span class="source-line-no">317</span><span id="line-317"></span> |
| <span class="source-line-no">318</span><span id="line-318"> /** Returns Set of online Servers named for their hostname and port (not ServerName). */</span> |
| <span class="source-line-no">319</span><span id="line-319"> private Set<Address> getOnlineServers() {</span> |
| <span class="source-line-no">320</span><span id="line-320"> return masterServices.getServerManager().getOnlineServers().keySet().stream()</span> |
| <span class="source-line-no">321</span><span id="line-321"> .map(ServerName::getAddress).collect(Collectors.toSet());</span> |
| <span class="source-line-no">322</span><span id="line-322"> }</span> |
| <span class="source-line-no">323</span><span id="line-323"></span> |
| <span class="source-line-no">324</span><span id="line-324"> public synchronized Set<Address> moveServers(Set<Address> servers, String srcGroup,</span> |
| <span class="source-line-no">325</span><span id="line-325"> String dstGroup) throws IOException {</span> |
| <span class="source-line-no">326</span><span id="line-326"> RSGroupInfo src = getRSGroupInfo(srcGroup);</span> |
| <span class="source-line-no">327</span><span id="line-327"> RSGroupInfo dst = getRSGroupInfo(dstGroup);</span> |
| <span class="source-line-no">328</span><span id="line-328"> Set<Address> movedServers = new HashSet<>();</span> |
| <span class="source-line-no">329</span><span id="line-329"> // If destination is 'default' rsgroup, only add servers that are online. If not online, drop</span> |
| <span class="source-line-no">330</span><span id="line-330"> // it. If not 'default' group, add server to 'dst' rsgroup EVEN IF IT IS NOT online (could be a</span> |
| <span class="source-line-no">331</span><span id="line-331"> // rsgroup of dead servers that are to come back later).</span> |
| <span class="source-line-no">332</span><span id="line-332"> Set<Address> onlineServers =</span> |
| <span class="source-line-no">333</span><span id="line-333"> dst.getName().equals(RSGroupInfo.DEFAULT_GROUP) ? getOnlineServers() : null;</span> |
| <span class="source-line-no">334</span><span id="line-334"> for (Address el : servers) {</span> |
| <span class="source-line-no">335</span><span id="line-335"> src.removeServer(el);</span> |
| <span class="source-line-no">336</span><span id="line-336"> if (onlineServers != null) {</span> |
| <span class="source-line-no">337</span><span id="line-337"> if (!onlineServers.contains(el)) {</span> |
| <span class="source-line-no">338</span><span id="line-338"> if (LOG.isDebugEnabled()) {</span> |
| <span class="source-line-no">339</span><span id="line-339"> LOG.debug("Dropping " + el + " during move-to-default RSGroup because not online");</span> |
| <span class="source-line-no">340</span><span id="line-340"> }</span> |
| <span class="source-line-no">341</span><span id="line-341"> continue;</span> |
| <span class="source-line-no">342</span><span id="line-342"> }</span> |
| <span class="source-line-no">343</span><span id="line-343"> }</span> |
| <span class="source-line-no">344</span><span id="line-344"> dst.addServer(el);</span> |
| <span class="source-line-no">345</span><span id="line-345"> movedServers.add(el);</span> |
| <span class="source-line-no">346</span><span id="line-346"> }</span> |
| <span class="source-line-no">347</span><span id="line-347"> Map<String, RSGroupInfo> newGroupMap = Maps.newHashMap(holder.groupName2Group);</span> |
| <span class="source-line-no">348</span><span id="line-348"> newGroupMap.put(src.getName(), src);</span> |
| <span class="source-line-no">349</span><span id="line-349"> newGroupMap.put(dst.getName(), dst);</span> |
| <span class="source-line-no">350</span><span id="line-350"> flushConfig(newGroupMap);</span> |
| <span class="source-line-no">351</span><span id="line-351"> return movedServers;</span> |
| <span class="source-line-no">352</span><span id="line-352"> }</span> |
| <span class="source-line-no">353</span><span id="line-353"></span> |
| <span class="source-line-no">354</span><span id="line-354"> @Override</span> |
| <span class="source-line-no">355</span><span id="line-355"> public RSGroupInfo getRSGroupOfServer(Address serverHostPort) {</span> |
| <span class="source-line-no">356</span><span id="line-356"> for (RSGroupInfo info : holder.groupName2Group.values()) {</span> |
| <span class="source-line-no">357</span><span id="line-357"> if (info.containsServer(serverHostPort)) {</span> |
| <span class="source-line-no">358</span><span id="line-358"> return info;</span> |
| <span class="source-line-no">359</span><span id="line-359"> }</span> |
| <span class="source-line-no">360</span><span id="line-360"> }</span> |
| <span class="source-line-no">361</span><span id="line-361"> return null;</span> |
| <span class="source-line-no">362</span><span id="line-362"> }</span> |
| <span class="source-line-no">363</span><span id="line-363"></span> |
| <span class="source-line-no">364</span><span id="line-364"> @Override</span> |
| <span class="source-line-no">365</span><span id="line-365"> public RSGroupInfo getRSGroup(String groupName) {</span> |
| <span class="source-line-no">366</span><span id="line-366"> return holder.groupName2Group.get(groupName);</span> |
| <span class="source-line-no">367</span><span id="line-367"> }</span> |
| <span class="source-line-no">368</span><span id="line-368"></span> |
| <span class="source-line-no">369</span><span id="line-369"> @Override</span> |
| <span class="source-line-no">370</span><span id="line-370"> public synchronized void removeRSGroup(String groupName) throws IOException {</span> |
| <span class="source-line-no">371</span><span id="line-371"> RSGroupInfo rsGroupInfo = getRSGroupInfo(groupName);</span> |
| <span class="source-line-no">372</span><span id="line-372"> int serverCount = rsGroupInfo.getServers().size();</span> |
| <span class="source-line-no">373</span><span id="line-373"> if (serverCount > 0) {</span> |
| <span class="source-line-no">374</span><span id="line-374"> throw new ConstraintException("RSGroup " + groupName + " has " + serverCount</span> |
| <span class="source-line-no">375</span><span id="line-375"> + " servers; you must remove these servers from the RSGroup before"</span> |
| <span class="source-line-no">376</span><span id="line-376"> + " the RSGroup can be removed.");</span> |
| <span class="source-line-no">377</span><span id="line-377"> }</span> |
| <span class="source-line-no">378</span><span id="line-378"> for (TableDescriptor td : masterServices.getTableDescriptors().getAll().values()) {</span> |
| <span class="source-line-no">379</span><span id="line-379"> if (td.getRegionServerGroup().map(groupName::equals).orElse(false)) {</span> |
| <span class="source-line-no">380</span><span id="line-380"> throw new ConstraintException("RSGroup " + groupName + " is already referenced by "</span> |
| <span class="source-line-no">381</span><span id="line-381"> + td.getTableName() + "; you must remove all the tables from the RSGroup before "</span> |
| <span class="source-line-no">382</span><span id="line-382"> + "the RSGroup can be removed.");</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"> for (NamespaceDescriptor ns : masterServices.getClusterSchema().getNamespaces()) {</span> |
| <span class="source-line-no">386</span><span id="line-386"> String nsGroup = ns.getConfigurationValue(RSGroupInfo.NAMESPACE_DESC_PROP_GROUP);</span> |
| <span class="source-line-no">387</span><span id="line-387"> if (nsGroup != null && nsGroup.equals(groupName)) {</span> |
| <span class="source-line-no">388</span><span id="line-388"> throw new ConstraintException(</span> |
| <span class="source-line-no">389</span><span id="line-389"> "RSGroup " + groupName + " is referenced by namespace: " + ns.getName());</span> |
| <span class="source-line-no">390</span><span id="line-390"> }</span> |
| <span class="source-line-no">391</span><span id="line-391"> }</span> |
| <span class="source-line-no">392</span><span id="line-392"> Map<String, RSGroupInfo> rsGroupMap = holder.groupName2Group;</span> |
| <span class="source-line-no">393</span><span id="line-393"> if (!rsGroupMap.containsKey(groupName) || groupName.equals(RSGroupInfo.DEFAULT_GROUP)) {</span> |
| <span class="source-line-no">394</span><span id="line-394"> throw new ConstraintException(</span> |
| <span class="source-line-no">395</span><span id="line-395"> "Group " + groupName + " does not exist or is a reserved " + "group");</span> |
| <span class="source-line-no">396</span><span id="line-396"> }</span> |
| <span class="source-line-no">397</span><span id="line-397"> Map<String, RSGroupInfo> newGroupMap = Maps.newHashMap(rsGroupMap);</span> |
| <span class="source-line-no">398</span><span id="line-398"> newGroupMap.remove(groupName);</span> |
| <span class="source-line-no">399</span><span id="line-399"> flushConfig(newGroupMap);</span> |
| <span class="source-line-no">400</span><span id="line-400"> LOG.info("Remove group {} done", groupName);</span> |
| <span class="source-line-no">401</span><span id="line-401"> }</span> |
| <span class="source-line-no">402</span><span id="line-402"></span> |
| <span class="source-line-no">403</span><span id="line-403"> @Override</span> |
| <span class="source-line-no">404</span><span id="line-404"> public List<RSGroupInfo> listRSGroups() {</span> |
| <span class="source-line-no">405</span><span id="line-405"> return Lists.newArrayList(holder.groupName2Group.values());</span> |
| <span class="source-line-no">406</span><span id="line-406"> }</span> |
| <span class="source-line-no">407</span><span id="line-407"></span> |
| <span class="source-line-no">408</span><span id="line-408"> @Override</span> |
| <span class="source-line-no">409</span><span id="line-409"> public boolean isOnline() {</span> |
| <span class="source-line-no">410</span><span id="line-410"> return rsGroupStartupWorker.isOnline();</span> |
| <span class="source-line-no">411</span><span id="line-411"> }</span> |
| <span class="source-line-no">412</span><span id="line-412"></span> |
| <span class="source-line-no">413</span><span id="line-413"> @Override</span> |
| <span class="source-line-no">414</span><span id="line-414"> public synchronized void removeServers(Set<Address> servers) throws IOException {</span> |
| <span class="source-line-no">415</span><span id="line-415"> if (servers == null || servers.isEmpty()) {</span> |
| <span class="source-line-no">416</span><span id="line-416"> throw new ConstraintException("The set of servers to remove cannot be null or empty.");</span> |
| <span class="source-line-no">417</span><span id="line-417"> }</span> |
| <span class="source-line-no">418</span><span id="line-418"></span> |
| <span class="source-line-no">419</span><span id="line-419"> // check the set of servers</span> |
| <span class="source-line-no">420</span><span id="line-420"> checkForDeadOrOnlineServers(servers);</span> |
| <span class="source-line-no">421</span><span id="line-421"></span> |
| <span class="source-line-no">422</span><span id="line-422"> Map<String, RSGroupInfo> rsGroupInfos = new HashMap<String, RSGroupInfo>();</span> |
| <span class="source-line-no">423</span><span id="line-423"> for (Address el : servers) {</span> |
| <span class="source-line-no">424</span><span id="line-424"> RSGroupInfo rsGroupInfo = getRSGroupOfServer(el);</span> |
| <span class="source-line-no">425</span><span id="line-425"> if (rsGroupInfo != null) {</span> |
| <span class="source-line-no">426</span><span id="line-426"> RSGroupInfo newRsGroupInfo = rsGroupInfos.get(rsGroupInfo.getName());</span> |
| <span class="source-line-no">427</span><span id="line-427"> if (newRsGroupInfo == null) {</span> |
| <span class="source-line-no">428</span><span id="line-428"> rsGroupInfo.removeServer(el);</span> |
| <span class="source-line-no">429</span><span id="line-429"> rsGroupInfos.put(rsGroupInfo.getName(), rsGroupInfo);</span> |
| <span class="source-line-no">430</span><span id="line-430"> } else {</span> |
| <span class="source-line-no">431</span><span id="line-431"> newRsGroupInfo.removeServer(el);</span> |
| <span class="source-line-no">432</span><span id="line-432"> rsGroupInfos.put(newRsGroupInfo.getName(), newRsGroupInfo);</span> |
| <span class="source-line-no">433</span><span id="line-433"> }</span> |
| <span class="source-line-no">434</span><span id="line-434"> } else {</span> |
| <span class="source-line-no">435</span><span id="line-435"> LOG.warn("Server " + el + " does not belong to any rsgroup.");</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"></span> |
| <span class="source-line-no">439</span><span id="line-439"> if (rsGroupInfos.size() > 0) {</span> |
| <span class="source-line-no">440</span><span id="line-440"> Map<String, RSGroupInfo> newGroupMap = Maps.newHashMap(holder.groupName2Group);</span> |
| <span class="source-line-no">441</span><span id="line-441"> newGroupMap.putAll(rsGroupInfos);</span> |
| <span class="source-line-no">442</span><span id="line-442"> flushConfig(newGroupMap);</span> |
| <span class="source-line-no">443</span><span id="line-443"> }</span> |
| <span class="source-line-no">444</span><span id="line-444"> LOG.info("Remove decommissioned servers {} from RSGroup done", servers);</span> |
| <span class="source-line-no">445</span><span id="line-445"> }</span> |
| <span class="source-line-no">446</span><span id="line-446"></span> |
| <span class="source-line-no">447</span><span id="line-447"> private List<RSGroupInfo> retrieveGroupListFromGroupTable() throws IOException {</span> |
| <span class="source-line-no">448</span><span id="line-448"> List<RSGroupInfo> rsGroupInfoList = Lists.newArrayList();</span> |
| <span class="source-line-no">449</span><span id="line-449"> AsyncTable<?> table = conn.getTable(RSGROUP_TABLE_NAME);</span> |
| <span class="source-line-no">450</span><span id="line-450"> try (ResultScanner scanner = table.getScanner(META_FAMILY_BYTES, META_QUALIFIER_BYTES)) {</span> |
| <span class="source-line-no">451</span><span id="line-451"> for (Result result;;) {</span> |
| <span class="source-line-no">452</span><span id="line-452"> result = scanner.next();</span> |
| <span class="source-line-no">453</span><span id="line-453"> if (result == null) {</span> |
| <span class="source-line-no">454</span><span id="line-454"> break;</span> |
| <span class="source-line-no">455</span><span id="line-455"> }</span> |
| <span class="source-line-no">456</span><span id="line-456"> RSGroupProtos.RSGroupInfo proto = RSGroupProtos.RSGroupInfo</span> |
| <span class="source-line-no">457</span><span id="line-457"> .parseFrom(result.getValue(META_FAMILY_BYTES, META_QUALIFIER_BYTES));</span> |
| <span class="source-line-no">458</span><span id="line-458"> rsGroupInfoList.add(ProtobufUtil.toGroupInfo(proto));</span> |
| <span class="source-line-no">459</span><span id="line-459"> }</span> |
| <span class="source-line-no">460</span><span id="line-460"> }</span> |
| <span class="source-line-no">461</span><span id="line-461"> return rsGroupInfoList;</span> |
| <span class="source-line-no">462</span><span id="line-462"> }</span> |
| <span class="source-line-no">463</span><span id="line-463"></span> |
| <span class="source-line-no">464</span><span id="line-464"> private List<RSGroupInfo> retrieveGroupListFromZookeeper() throws IOException {</span> |
| <span class="source-line-no">465</span><span id="line-465"> String groupBasePath = ZNodePaths.joinZNode(watcher.getZNodePaths().baseZNode, RS_GROUP_ZNODE);</span> |
| <span class="source-line-no">466</span><span id="line-466"> List<RSGroupInfo> RSGroupInfoList = Lists.newArrayList();</span> |
| <span class="source-line-no">467</span><span id="line-467"> // Overwrite any info stored by table, this takes precedence</span> |
| <span class="source-line-no">468</span><span id="line-468"> try {</span> |
| <span class="source-line-no">469</span><span id="line-469"> if (ZKUtil.checkExists(watcher, groupBasePath) != -1) {</span> |
| <span class="source-line-no">470</span><span id="line-470"> List<String> children = ZKUtil.listChildrenAndWatchForNewChildren(watcher, groupBasePath);</span> |
| <span class="source-line-no">471</span><span id="line-471"> if (children == null) {</span> |
| <span class="source-line-no">472</span><span id="line-472"> return RSGroupInfoList;</span> |
| <span class="source-line-no">473</span><span id="line-473"> }</span> |
| <span class="source-line-no">474</span><span id="line-474"> for (String znode : children) {</span> |
| <span class="source-line-no">475</span><span id="line-475"> byte[] data = ZKUtil.getData(watcher, ZNodePaths.joinZNode(groupBasePath, znode));</span> |
| <span class="source-line-no">476</span><span id="line-476"> if (data != null && data.length > 0) {</span> |
| <span class="source-line-no">477</span><span id="line-477"> ProtobufUtil.expectPBMagicPrefix(data);</span> |
| <span class="source-line-no">478</span><span id="line-478"> ByteArrayInputStream bis =</span> |
| <span class="source-line-no">479</span><span id="line-479"> new ByteArrayInputStream(data, ProtobufUtil.lengthOfPBMagic(), data.length);</span> |
| <span class="source-line-no">480</span><span id="line-480"> RSGroupInfoList.add(ProtobufUtil.toGroupInfo(RSGroupProtos.RSGroupInfo.parseFrom(bis)));</span> |
| <span class="source-line-no">481</span><span id="line-481"> }</span> |
| <span class="source-line-no">482</span><span id="line-482"> }</span> |
| <span class="source-line-no">483</span><span id="line-483"> LOG.debug("Read ZK GroupInfo count:" + RSGroupInfoList.size());</span> |
| <span class="source-line-no">484</span><span id="line-484"> }</span> |
| <span class="source-line-no">485</span><span id="line-485"> } catch (KeeperException | DeserializationException | InterruptedException e) {</span> |
| <span class="source-line-no">486</span><span id="line-486"> throw new IOException("Failed to read rsGroupZNode", e);</span> |
| <span class="source-line-no">487</span><span id="line-487"> }</span> |
| <span class="source-line-no">488</span><span id="line-488"> return RSGroupInfoList;</span> |
| <span class="source-line-no">489</span><span id="line-489"> }</span> |
| <span class="source-line-no">490</span><span id="line-490"></span> |
| <span class="source-line-no">491</span><span id="line-491"> private void migrate(Collection<RSGroupInfo> groupList) {</span> |
| <span class="source-line-no">492</span><span id="line-492"> TableDescriptors tds = masterServices.getTableDescriptors();</span> |
| <span class="source-line-no">493</span><span id="line-493"> ProcedureExecutor<MasterProcedureEnv> procExec = masterServices.getMasterProcedureExecutor();</span> |
| <span class="source-line-no">494</span><span id="line-494"> for (RSGroupInfo groupInfo : groupList) {</span> |
| <span class="source-line-no">495</span><span id="line-495"> if (groupInfo.getName().equals(RSGroupInfo.DEFAULT_GROUP)) {</span> |
| <span class="source-line-no">496</span><span id="line-496"> continue;</span> |
| <span class="source-line-no">497</span><span id="line-497"> }</span> |
| <span class="source-line-no">498</span><span id="line-498"> SortedSet<TableName> failedTables = new TreeSet<>();</span> |
| <span class="source-line-no">499</span><span id="line-499"> List<MigrateRSGroupProcedure> procs = new ArrayList<>();</span> |
| <span class="source-line-no">500</span><span id="line-500"> for (TableName tableName : groupInfo.getTables()) {</span> |
| <span class="source-line-no">501</span><span id="line-501"> LOG.debug("Migrating {} in group {}", tableName, groupInfo.getName());</span> |
| <span class="source-line-no">502</span><span id="line-502"> TableDescriptor oldTd;</span> |
| <span class="source-line-no">503</span><span id="line-503"> try {</span> |
| <span class="source-line-no">504</span><span id="line-504"> oldTd = tds.get(tableName);</span> |
| <span class="source-line-no">505</span><span id="line-505"> } catch (IOException e) {</span> |
| <span class="source-line-no">506</span><span id="line-506"> LOG.warn("Failed to migrate {} in group {}", tableName, groupInfo.getName(), e);</span> |
| <span class="source-line-no">507</span><span id="line-507"> failedTables.add(tableName);</span> |
| <span class="source-line-no">508</span><span id="line-508"> continue;</span> |
| <span class="source-line-no">509</span><span id="line-509"> }</span> |
| <span class="source-line-no">510</span><span id="line-510"> if (oldTd == null) {</span> |
| <span class="source-line-no">511</span><span id="line-511"> continue;</span> |
| <span class="source-line-no">512</span><span id="line-512"> }</span> |
| <span class="source-line-no">513</span><span id="line-513"> if (oldTd.getRegionServerGroup().isPresent()) {</span> |
| <span class="source-line-no">514</span><span id="line-514"> // either we have already migrated it or that user has set the rs group using the new</span> |
| <span class="source-line-no">515</span><span id="line-515"> // code which will set the group directly on table descriptor, skip.</span> |
| <span class="source-line-no">516</span><span id="line-516"> LOG.debug("Skip migrating {} since it is already in group {}", tableName,</span> |
| <span class="source-line-no">517</span><span id="line-517"> oldTd.getRegionServerGroup().get());</span> |
| <span class="source-line-no">518</span><span id="line-518"> continue;</span> |
| <span class="source-line-no">519</span><span id="line-519"> }</span> |
| <span class="source-line-no">520</span><span id="line-520"> // This is a bit tricky. Since we know that the region server group config in</span> |
| <span class="source-line-no">521</span><span id="line-521"> // TableDescriptor will only be used at master side, it is fine to just update the table</span> |
| <span class="source-line-no">522</span><span id="line-522"> // descriptor on file system and also the cache, without reopening all the regions. This</span> |
| <span class="source-line-no">523</span><span id="line-523"> // will be much faster than the normal modifyTable. And when upgrading, we will update</span> |
| <span class="source-line-no">524</span><span id="line-524"> // master first and then region server, so after all the region servers has been reopened,</span> |
| <span class="source-line-no">525</span><span id="line-525"> // the new TableDescriptor will be loaded.</span> |
| <span class="source-line-no">526</span><span id="line-526"> MigrateRSGroupProcedure proc =</span> |
| <span class="source-line-no">527</span><span id="line-527"> new MigrateRSGroupProcedure(procExec.getEnvironment(), tableName);</span> |
| <span class="source-line-no">528</span><span id="line-528"> procExec.submitProcedure(proc);</span> |
| <span class="source-line-no">529</span><span id="line-529"> procs.add(proc);</span> |
| <span class="source-line-no">530</span><span id="line-530"> }</span> |
| <span class="source-line-no">531</span><span id="line-531"> for (MigrateRSGroupProcedure proc : procs) {</span> |
| <span class="source-line-no">532</span><span id="line-532"> try {</span> |
| <span class="source-line-no">533</span><span id="line-533"> ProcedureSyncWait.waitForProcedureToComplete(procExec, proc, 60000);</span> |
| <span class="source-line-no">534</span><span id="line-534"> } catch (IOException e) {</span> |
| <span class="source-line-no">535</span><span id="line-535"> LOG.warn("Failed to migrate rs group {} for table {}", groupInfo.getName(),</span> |
| <span class="source-line-no">536</span><span id="line-536"> proc.getTableName());</span> |
| <span class="source-line-no">537</span><span id="line-537"> failedTables.add(proc.getTableName());</span> |
| <span class="source-line-no">538</span><span id="line-538"> }</span> |
| <span class="source-line-no">539</span><span id="line-539"> }</span> |
| <span class="source-line-no">540</span><span id="line-540"> LOG.debug("Done migrating {}, failed tables {}", groupInfo.getName(), failedTables);</span> |
| <span class="source-line-no">541</span><span id="line-541"> synchronized (RSGroupInfoManagerImpl.this) {</span> |
| <span class="source-line-no">542</span><span id="line-542"> Map<String, RSGroupInfo> rsGroupMap = holder.groupName2Group;</span> |
| <span class="source-line-no">543</span><span id="line-543"> RSGroupInfo currentInfo = rsGroupMap.get(groupInfo.getName());</span> |
| <span class="source-line-no">544</span><span id="line-544"> if (currentInfo != null) {</span> |
| <span class="source-line-no">545</span><span id="line-545"> RSGroupInfo newInfo =</span> |
| <span class="source-line-no">546</span><span id="line-546"> new RSGroupInfo(currentInfo.getName(), currentInfo.getServers(), failedTables);</span> |
| <span class="source-line-no">547</span><span id="line-547"> Map<String, RSGroupInfo> newGroupMap = new HashMap<>(rsGroupMap);</span> |
| <span class="source-line-no">548</span><span id="line-548"> newGroupMap.put(groupInfo.getName(), newInfo);</span> |
| <span class="source-line-no">549</span><span id="line-549"> try {</span> |
| <span class="source-line-no">550</span><span id="line-550"> flushConfig(newGroupMap);</span> |
| <span class="source-line-no">551</span><span id="line-551"> } catch (IOException e) {</span> |
| <span class="source-line-no">552</span><span id="line-552"> LOG.warn("Failed to persist rs group {}", newInfo.getName(), e);</span> |
| <span class="source-line-no">553</span><span id="line-553"> }</span> |
| <span class="source-line-no">554</span><span id="line-554"> }</span> |
| <span class="source-line-no">555</span><span id="line-555"> }</span> |
| <span class="source-line-no">556</span><span id="line-556"> }</span> |
| <span class="source-line-no">557</span><span id="line-557"> }</span> |
| <span class="source-line-no">558</span><span id="line-558"></span> |
| <span class="source-line-no">559</span><span id="line-559"> // Migrate the table rs group info from RSGroupInfo into the table descriptor</span> |
| <span class="source-line-no">560</span><span id="line-560"> // Notice that we do not want to block the initialize so this will be done in background, and</span> |
| <span class="source-line-no">561</span><span id="line-561"> // during the migrating, the rs group info maybe incomplete and cause region to be misplaced.</span> |
| <span class="source-line-no">562</span><span id="line-562"> private void migrate() {</span> |
| <span class="source-line-no">563</span><span id="line-563"> Thread migrateThread = new Thread(MIGRATE_THREAD_NAME) {</span> |
| <span class="source-line-no">564</span><span id="line-564"></span> |
| <span class="source-line-no">565</span><span id="line-565"> @Override</span> |
| <span class="source-line-no">566</span><span id="line-566"> public void run() {</span> |
| <span class="source-line-no">567</span><span id="line-567"> LOG.info("Start migrating table rs group config");</span> |
| <span class="source-line-no">568</span><span id="line-568"> while (!masterServices.isStopped()) {</span> |
| <span class="source-line-no">569</span><span id="line-569"> Collection<RSGroupInfo> groups = holder.groupName2Group.values();</span> |
| <span class="source-line-no">570</span><span id="line-570"> boolean hasTables = groups.stream().anyMatch(r -> !r.getTables().isEmpty());</span> |
| <span class="source-line-no">571</span><span id="line-571"> if (!hasTables) {</span> |
| <span class="source-line-no">572</span><span id="line-572"> break;</span> |
| <span class="source-line-no">573</span><span id="line-573"> }</span> |
| <span class="source-line-no">574</span><span id="line-574"> migrate(groups);</span> |
| <span class="source-line-no">575</span><span id="line-575"> }</span> |
| <span class="source-line-no">576</span><span id="line-576"> LOG.info("Done migrating table rs group info");</span> |
| <span class="source-line-no">577</span><span id="line-577"> }</span> |
| <span class="source-line-no">578</span><span id="line-578"> };</span> |
| <span class="source-line-no">579</span><span id="line-579"> migrateThread.setDaemon(true);</span> |
| <span class="source-line-no">580</span><span id="line-580"> migrateThread.start();</span> |
| <span class="source-line-no">581</span><span id="line-581"> }</span> |
| <span class="source-line-no">582</span><span id="line-582"></span> |
| <span class="source-line-no">583</span><span id="line-583"> /**</span> |
| <span class="source-line-no">584</span><span id="line-584"> * Read rsgroup info from the source of truth, the hbase:rsgroup table. Update zk cache. Called on</span> |
| <span class="source-line-no">585</span><span id="line-585"> * startup of the manager.</span> |
| <span class="source-line-no">586</span><span id="line-586"> */</span> |
| <span class="source-line-no">587</span><span id="line-587"> private synchronized void refresh(boolean forceOnline) throws IOException {</span> |
| <span class="source-line-no">588</span><span id="line-588"> List<RSGroupInfo> groupList = new ArrayList<>();</span> |
| <span class="source-line-no">589</span><span id="line-589"></span> |
| <span class="source-line-no">590</span><span id="line-590"> // Overwrite anything read from zk, group table is source of truth</span> |
| <span class="source-line-no">591</span><span id="line-591"> // if online read from GROUP table</span> |
| <span class="source-line-no">592</span><span id="line-592"> if (forceOnline || isOnline()) {</span> |
| <span class="source-line-no">593</span><span id="line-593"> LOG.debug("Refreshing in Online mode.");</span> |
| <span class="source-line-no">594</span><span id="line-594"> groupList.addAll(retrieveGroupListFromGroupTable());</span> |
| <span class="source-line-no">595</span><span id="line-595"> } else {</span> |
| <span class="source-line-no">596</span><span id="line-596"> LOG.debug("Refreshing in Offline mode.");</span> |
| <span class="source-line-no">597</span><span id="line-597"> groupList.addAll(retrieveGroupListFromZookeeper());</span> |
| <span class="source-line-no">598</span><span id="line-598"> }</span> |
| <span class="source-line-no">599</span><span id="line-599"></span> |
| <span class="source-line-no">600</span><span id="line-600"> // This is added to the last of the list so it overwrites the 'default' rsgroup loaded</span> |
| <span class="source-line-no">601</span><span id="line-601"> // from region group table or zk</span> |
| <span class="source-line-no">602</span><span id="line-602"> groupList.add(new RSGroupInfo(RSGroupInfo.DEFAULT_GROUP, getDefaultServers(groupList)));</span> |
| <span class="source-line-no">603</span><span id="line-603"></span> |
| <span class="source-line-no">604</span><span id="line-604"> // populate the data</span> |
| <span class="source-line-no">605</span><span id="line-605"> HashMap<String, RSGroupInfo> newGroupMap = Maps.newHashMap();</span> |
| <span class="source-line-no">606</span><span id="line-606"> for (RSGroupInfo group : groupList) {</span> |
| <span class="source-line-no">607</span><span id="line-607"> newGroupMap.put(group.getName(), group);</span> |
| <span class="source-line-no">608</span><span id="line-608"> }</span> |
| <span class="source-line-no">609</span><span id="line-609"> resetRSGroupMap(newGroupMap);</span> |
| <span class="source-line-no">610</span><span id="line-610"> updateCacheOfRSGroups(newGroupMap.keySet());</span> |
| <span class="source-line-no">611</span><span id="line-611"> }</span> |
| <span class="source-line-no">612</span><span id="line-612"></span> |
| <span class="source-line-no">613</span><span id="line-613"> private void flushConfigTable(Map<String, RSGroupInfo> groupMap) throws IOException {</span> |
| <span class="source-line-no">614</span><span id="line-614"> List<Mutation> mutations = Lists.newArrayList();</span> |
| <span class="source-line-no">615</span><span id="line-615"></span> |
| <span class="source-line-no">616</span><span id="line-616"> // populate deletes</span> |
| <span class="source-line-no">617</span><span id="line-617"> for (String groupName : prevRSGroups) {</span> |
| <span class="source-line-no">618</span><span id="line-618"> if (!groupMap.containsKey(groupName)) {</span> |
| <span class="source-line-no">619</span><span id="line-619"> Delete d = new Delete(Bytes.toBytes(groupName));</span> |
| <span class="source-line-no">620</span><span id="line-620"> mutations.add(d);</span> |
| <span class="source-line-no">621</span><span id="line-621"> }</span> |
| <span class="source-line-no">622</span><span id="line-622"> }</span> |
| <span class="source-line-no">623</span><span id="line-623"></span> |
| <span class="source-line-no">624</span><span id="line-624"> // populate puts</span> |
| <span class="source-line-no">625</span><span id="line-625"> for (RSGroupInfo gi : groupMap.values()) {</span> |
| <span class="source-line-no">626</span><span id="line-626"> if (!gi.getName().equals(RSGroupInfo.DEFAULT_GROUP)) {</span> |
| <span class="source-line-no">627</span><span id="line-627"> RSGroupProtos.RSGroupInfo proto = ProtobufUtil.toProtoGroupInfo(gi);</span> |
| <span class="source-line-no">628</span><span id="line-628"> Put p = new Put(Bytes.toBytes(gi.getName()));</span> |
| <span class="source-line-no">629</span><span id="line-629"> p.addColumn(META_FAMILY_BYTES, META_QUALIFIER_BYTES, proto.toByteArray());</span> |
| <span class="source-line-no">630</span><span id="line-630"> mutations.add(p);</span> |
| <span class="source-line-no">631</span><span id="line-631"> }</span> |
| <span class="source-line-no">632</span><span id="line-632"> }</span> |
| <span class="source-line-no">633</span><span id="line-633"></span> |
| <span class="source-line-no">634</span><span id="line-634"> if (mutations.size() > 0) {</span> |
| <span class="source-line-no">635</span><span id="line-635"> multiMutate(mutations);</span> |
| <span class="source-line-no">636</span><span id="line-636"> }</span> |
| <span class="source-line-no">637</span><span id="line-637"> }</span> |
| <span class="source-line-no">638</span><span id="line-638"></span> |
| <span class="source-line-no">639</span><span id="line-639"> private synchronized void flushConfig() throws IOException {</span> |
| <span class="source-line-no">640</span><span id="line-640"> flushConfig(holder.groupName2Group);</span> |
| <span class="source-line-no">641</span><span id="line-641"> }</span> |
| <span class="source-line-no">642</span><span id="line-642"></span> |
| <span class="source-line-no">643</span><span id="line-643"> private synchronized void flushConfig(Map<String, RSGroupInfo> newGroupMap) throws IOException {</span> |
| <span class="source-line-no">644</span><span id="line-644"> // For offline mode persistence is still unavailable</span> |
| <span class="source-line-no">645</span><span id="line-645"> // We're refreshing in-memory state but only for servers in default group</span> |
| <span class="source-line-no">646</span><span id="line-646"> if (!isOnline()) {</span> |
| <span class="source-line-no">647</span><span id="line-647"> if (newGroupMap == holder.groupName2Group) {</span> |
| <span class="source-line-no">648</span><span id="line-648"> // When newGroupMap is this.rsGroupMap itself,</span> |
| <span class="source-line-no">649</span><span id="line-649"> // do not need to check default group and other groups as followed</span> |
| <span class="source-line-no">650</span><span id="line-650"> return;</span> |
| <span class="source-line-no">651</span><span id="line-651"> }</span> |
| <span class="source-line-no">652</span><span id="line-652"></span> |
| <span class="source-line-no">653</span><span id="line-653"> LOG.debug("Offline mode, cannot persist to {}", RSGROUP_TABLE_NAME);</span> |
| <span class="source-line-no">654</span><span id="line-654"></span> |
| <span class="source-line-no">655</span><span id="line-655"> Map<String, RSGroupInfo> oldGroupMap = Maps.newHashMap(holder.groupName2Group);</span> |
| <span class="source-line-no">656</span><span id="line-656"> RSGroupInfo oldDefaultGroup = oldGroupMap.remove(RSGroupInfo.DEFAULT_GROUP);</span> |
| <span class="source-line-no">657</span><span id="line-657"> RSGroupInfo newDefaultGroup = newGroupMap.remove(RSGroupInfo.DEFAULT_GROUP);</span> |
| <span class="source-line-no">658</span><span id="line-658"> if (</span> |
| <span class="source-line-no">659</span><span id="line-659"> !oldGroupMap.equals(newGroupMap)</span> |
| <span class="source-line-no">660</span><span id="line-660"> /* compare both tables and servers in other groups */ || !oldDefaultGroup.getTables()</span> |
| <span class="source-line-no">661</span><span id="line-661"> .equals(newDefaultGroup.getTables())</span> |
| <span class="source-line-no">662</span><span id="line-662"> /* compare tables in default group */</span> |
| <span class="source-line-no">663</span><span id="line-663"> ) {</span> |
| <span class="source-line-no">664</span><span id="line-664"> throw new IOException("Only servers in default group can be updated during offline mode");</span> |
| <span class="source-line-no">665</span><span id="line-665"> }</span> |
| <span class="source-line-no">666</span><span id="line-666"></span> |
| <span class="source-line-no">667</span><span id="line-667"> // Restore newGroupMap by putting its default group back</span> |
| <span class="source-line-no">668</span><span id="line-668"> newGroupMap.put(RSGroupInfo.DEFAULT_GROUP, newDefaultGroup);</span> |
| <span class="source-line-no">669</span><span id="line-669"></span> |
| <span class="source-line-no">670</span><span id="line-670"> // Refresh rsGroupMap</span> |
| <span class="source-line-no">671</span><span id="line-671"> // according to the inputted newGroupMap (an updated copy of rsGroupMap)</span> |
| <span class="source-line-no">672</span><span id="line-672"> this.holder = new RSGroupInfoHolder(newGroupMap);</span> |
| <span class="source-line-no">673</span><span id="line-673"></span> |
| <span class="source-line-no">674</span><span id="line-674"> LOG.debug("New RSGroup map: {}", newGroupMap);</span> |
| <span class="source-line-no">675</span><span id="line-675"></span> |
| <span class="source-line-no">676</span><span id="line-676"> // Do not need to update tableMap</span> |
| <span class="source-line-no">677</span><span id="line-677"> // because only the update on servers in default group is allowed above,</span> |
| <span class="source-line-no">678</span><span id="line-678"> // or IOException will be thrown</span> |
| <span class="source-line-no">679</span><span id="line-679"> return;</span> |
| <span class="source-line-no">680</span><span id="line-680"> }</span> |
| <span class="source-line-no">681</span><span id="line-681"></span> |
| <span class="source-line-no">682</span><span id="line-682"> /* For online mode, persist to hbase:rsgroup and Zookeeper */</span> |
| <span class="source-line-no">683</span><span id="line-683"> LOG.debug("Online mode, persisting to {} and ZK", RSGROUP_TABLE_NAME);</span> |
| <span class="source-line-no">684</span><span id="line-684"> flushConfigTable(newGroupMap);</span> |
| <span class="source-line-no">685</span><span id="line-685"></span> |
| <span class="source-line-no">686</span><span id="line-686"> // Make changes visible after having been persisted to the source of truth</span> |
| <span class="source-line-no">687</span><span id="line-687"> resetRSGroupMap(newGroupMap);</span> |
| <span class="source-line-no">688</span><span id="line-688"> saveRSGroupMapToZK(newGroupMap);</span> |
| <span class="source-line-no">689</span><span id="line-689"> updateCacheOfRSGroups(newGroupMap.keySet());</span> |
| <span class="source-line-no">690</span><span id="line-690"> LOG.info("Flush config done, new RSGroup map: {}", newGroupMap);</span> |
| <span class="source-line-no">691</span><span id="line-691"> }</span> |
| <span class="source-line-no">692</span><span id="line-692"></span> |
| <span class="source-line-no">693</span><span id="line-693"> private void saveRSGroupMapToZK(Map<String, RSGroupInfo> newGroupMap) throws IOException {</span> |
| <span class="source-line-no">694</span><span id="line-694"> LOG.debug("Saving RSGroup info to ZK");</span> |
| <span class="source-line-no">695</span><span id="line-695"> try {</span> |
| <span class="source-line-no">696</span><span id="line-696"> String groupBasePath =</span> |
| <span class="source-line-no">697</span><span id="line-697"> ZNodePaths.joinZNode(watcher.getZNodePaths().baseZNode, RS_GROUP_ZNODE);</span> |
| <span class="source-line-no">698</span><span id="line-698"> ZKUtil.createAndFailSilent(watcher, groupBasePath, ProtobufMagic.PB_MAGIC);</span> |
| <span class="source-line-no">699</span><span id="line-699"></span> |
| <span class="source-line-no">700</span><span id="line-700"> List<ZKUtil.ZKUtilOp> zkOps = new ArrayList<>(newGroupMap.size());</span> |
| <span class="source-line-no">701</span><span id="line-701"> for (String groupName : prevRSGroups) {</span> |
| <span class="source-line-no">702</span><span id="line-702"> if (!newGroupMap.containsKey(groupName)) {</span> |
| <span class="source-line-no">703</span><span id="line-703"> String znode = ZNodePaths.joinZNode(groupBasePath, groupName);</span> |
| <span class="source-line-no">704</span><span id="line-704"> zkOps.add(ZKUtil.ZKUtilOp.deleteNodeFailSilent(znode));</span> |
| <span class="source-line-no">705</span><span id="line-705"> }</span> |
| <span class="source-line-no">706</span><span id="line-706"> }</span> |
| <span class="source-line-no">707</span><span id="line-707"></span> |
| <span class="source-line-no">708</span><span id="line-708"> for (RSGroupInfo gi : newGroupMap.values()) {</span> |
| <span class="source-line-no">709</span><span id="line-709"> if (!gi.getName().equals(RSGroupInfo.DEFAULT_GROUP)) {</span> |
| <span class="source-line-no">710</span><span id="line-710"> String znode = ZNodePaths.joinZNode(groupBasePath, gi.getName());</span> |
| <span class="source-line-no">711</span><span id="line-711"> RSGroupProtos.RSGroupInfo proto = ProtobufUtil.toProtoGroupInfo(gi);</span> |
| <span class="source-line-no">712</span><span id="line-712"> LOG.debug("Updating znode: " + znode);</span> |
| <span class="source-line-no">713</span><span id="line-713"> ZKUtil.createAndFailSilent(watcher, znode);</span> |
| <span class="source-line-no">714</span><span id="line-714"> zkOps.add(ZKUtil.ZKUtilOp.deleteNodeFailSilent(znode));</span> |
| <span class="source-line-no">715</span><span id="line-715"> zkOps.add(ZKUtil.ZKUtilOp.createAndFailSilent(znode,</span> |
| <span class="source-line-no">716</span><span id="line-716"> ProtobufUtil.prependPBMagic(proto.toByteArray())));</span> |
| <span class="source-line-no">717</span><span id="line-717"> }</span> |
| <span class="source-line-no">718</span><span id="line-718"> }</span> |
| <span class="source-line-no">719</span><span id="line-719"> LOG.debug("Writing ZK GroupInfo count: " + zkOps.size());</span> |
| <span class="source-line-no">720</span><span id="line-720"></span> |
| <span class="source-line-no">721</span><span id="line-721"> ZKUtil.multiOrSequential(watcher, zkOps, false);</span> |
| <span class="source-line-no">722</span><span id="line-722"> } catch (KeeperException e) {</span> |
| <span class="source-line-no">723</span><span id="line-723"> LOG.error("Failed to write to rsGroupZNode", e);</span> |
| <span class="source-line-no">724</span><span id="line-724"> masterServices.abort("Failed to write to rsGroupZNode", e);</span> |
| <span class="source-line-no">725</span><span id="line-725"> throw new IOException("Failed to write to rsGroupZNode", e);</span> |
| <span class="source-line-no">726</span><span id="line-726"> }</span> |
| <span class="source-line-no">727</span><span id="line-727"> }</span> |
| <span class="source-line-no">728</span><span id="line-728"></span> |
| <span class="source-line-no">729</span><span id="line-729"> /**</span> |
| <span class="source-line-no">730</span><span id="line-730"> * Make changes visible. Caller must be synchronized on 'this'.</span> |
| <span class="source-line-no">731</span><span id="line-731"> */</span> |
| <span class="source-line-no">732</span><span id="line-732"> private void resetRSGroupMap(Map<String, RSGroupInfo> newRSGroupMap) {</span> |
| <span class="source-line-no">733</span><span id="line-733"> this.holder = new RSGroupInfoHolder(newRSGroupMap);</span> |
| <span class="source-line-no">734</span><span id="line-734"> }</span> |
| <span class="source-line-no">735</span><span id="line-735"></span> |
| <span class="source-line-no">736</span><span id="line-736"> /**</span> |
| <span class="source-line-no">737</span><span id="line-737"> * Update cache of rsgroups. Caller must be synchronized on 'this'.</span> |
| <span class="source-line-no">738</span><span id="line-738"> * @param currentGroups Current list of Groups.</span> |
| <span class="source-line-no">739</span><span id="line-739"> */</span> |
| <span class="source-line-no">740</span><span id="line-740"> private void updateCacheOfRSGroups(final Set<String> currentGroups) {</span> |
| <span class="source-line-no">741</span><span id="line-741"> this.prevRSGroups.clear();</span> |
| <span class="source-line-no">742</span><span id="line-742"> this.prevRSGroups.addAll(currentGroups);</span> |
| <span class="source-line-no">743</span><span id="line-743"> }</span> |
| <span class="source-line-no">744</span><span id="line-744"></span> |
| <span class="source-line-no">745</span><span id="line-745"> // Called by ServerEventsListenerThread. Presume it has lock on this manager when it runs.</span> |
| <span class="source-line-no">746</span><span id="line-746"> private SortedSet<Address> getDefaultServers() {</span> |
| <span class="source-line-no">747</span><span id="line-747"> return getDefaultServers(listRSGroups()/* get from rsGroupMap */);</span> |
| <span class="source-line-no">748</span><span id="line-748"> }</span> |
| <span class="source-line-no">749</span><span id="line-749"></span> |
| <span class="source-line-no">750</span><span id="line-750"> // Called by ServerEventsListenerThread. Presume it has lock on this manager when it runs.</span> |
| <span class="source-line-no">751</span><span id="line-751"> private SortedSet<Address> getDefaultServers(List<RSGroupInfo> rsGroupInfoList) {</span> |
| <span class="source-line-no">752</span><span id="line-752"> // Build a list of servers in other groups than default group, from rsGroupMap</span> |
| <span class="source-line-no">753</span><span id="line-753"> Set<Address> serversInOtherGroup = new HashSet<>();</span> |
| <span class="source-line-no">754</span><span id="line-754"> for (RSGroupInfo group : rsGroupInfoList) {</span> |
| <span class="source-line-no">755</span><span id="line-755"> if (!RSGroupInfo.DEFAULT_GROUP.equals(group.getName())) { // not default group</span> |
| <span class="source-line-no">756</span><span id="line-756"> serversInOtherGroup.addAll(group.getServers());</span> |
| <span class="source-line-no">757</span><span id="line-757"> }</span> |
| <span class="source-line-no">758</span><span id="line-758"> }</span> |
| <span class="source-line-no">759</span><span id="line-759"></span> |
| <span class="source-line-no">760</span><span id="line-760"> // Get all online servers from Zookeeper and find out servers in default group</span> |
| <span class="source-line-no">761</span><span id="line-761"> SortedSet<Address> defaultServers = Sets.newTreeSet();</span> |
| <span class="source-line-no">762</span><span id="line-762"> for (ServerName serverName : masterServices.getServerManager().getOnlineServers().keySet()) {</span> |
| <span class="source-line-no">763</span><span id="line-763"> Address server = Address.fromParts(serverName.getHostname(), serverName.getPort());</span> |
| <span class="source-line-no">764</span><span id="line-764"> if (!serversInOtherGroup.contains(server)) { // not in other groups</span> |
| <span class="source-line-no">765</span><span id="line-765"> defaultServers.add(server);</span> |
| <span class="source-line-no">766</span><span id="line-766"> }</span> |
| <span class="source-line-no">767</span><span id="line-767"> }</span> |
| <span class="source-line-no">768</span><span id="line-768"> return defaultServers;</span> |
| <span class="source-line-no">769</span><span id="line-769"> }</span> |
| <span class="source-line-no">770</span><span id="line-770"></span> |
| <span class="source-line-no">771</span><span id="line-771"> private class RSGroupStartupWorker extends Thread {</span> |
| <span class="source-line-no">772</span><span id="line-772"> private final Logger LOG = LoggerFactory.getLogger(RSGroupStartupWorker.class);</span> |
| <span class="source-line-no">773</span><span id="line-773"> private volatile boolean online = false;</span> |
| <span class="source-line-no">774</span><span id="line-774"></span> |
| <span class="source-line-no">775</span><span id="line-775"> RSGroupStartupWorker() {</span> |
| <span class="source-line-no">776</span><span id="line-776"> super(RSGroupStartupWorker.class.getName() + "-" + masterServices.getServerName());</span> |
| <span class="source-line-no">777</span><span id="line-777"> setDaemon(true);</span> |
| <span class="source-line-no">778</span><span id="line-778"> }</span> |
| <span class="source-line-no">779</span><span id="line-779"></span> |
| <span class="source-line-no">780</span><span id="line-780"> @Override</span> |
| <span class="source-line-no">781</span><span id="line-781"> public void run() {</span> |
| <span class="source-line-no">782</span><span id="line-782"> if (waitForGroupTableOnline()) {</span> |
| <span class="source-line-no">783</span><span id="line-783"> LOG.info("GroupBasedLoadBalancer is now online");</span> |
| <span class="source-line-no">784</span><span id="line-784"> } else {</span> |
| <span class="source-line-no">785</span><span id="line-785"> LOG.warn("Quit without making region group table online");</span> |
| <span class="source-line-no">786</span><span id="line-786"> }</span> |
| <span class="source-line-no">787</span><span id="line-787"> }</span> |
| <span class="source-line-no">788</span><span id="line-788"></span> |
| <span class="source-line-no">789</span><span id="line-789"> private boolean waitForGroupTableOnline() {</span> |
| <span class="source-line-no">790</span><span id="line-790"> while (isMasterRunning(masterServices)) {</span> |
| <span class="source-line-no">791</span><span id="line-791"> try {</span> |
| <span class="source-line-no">792</span><span id="line-792"> TableStateManager tsm = masterServices.getTableStateManager();</span> |
| <span class="source-line-no">793</span><span id="line-793"> if (!tsm.isTablePresent(RSGROUP_TABLE_NAME)) {</span> |
| <span class="source-line-no">794</span><span id="line-794"> createRSGroupTable();</span> |
| <span class="source-line-no">795</span><span id="line-795"> }</span> |
| <span class="source-line-no">796</span><span id="line-796"> // try reading from the table</span> |
| <span class="source-line-no">797</span><span id="line-797"> FutureUtils.get(conn.getTable(RSGROUP_TABLE_NAME).get(new Get(ROW_KEY)));</span> |
| <span class="source-line-no">798</span><span id="line-798"> LOG.info("RSGroup table={} is online, refreshing cached information", RSGROUP_TABLE_NAME);</span> |
| <span class="source-line-no">799</span><span id="line-799"> RSGroupInfoManagerImpl.this.refresh(true);</span> |
| <span class="source-line-no">800</span><span id="line-800"> online = true;</span> |
| <span class="source-line-no">801</span><span id="line-801"> // flush any inconsistencies between ZK and HTable</span> |
| <span class="source-line-no">802</span><span id="line-802"> RSGroupInfoManagerImpl.this.flushConfig();</span> |
| <span class="source-line-no">803</span><span id="line-803"> // migrate after we are online.</span> |
| <span class="source-line-no">804</span><span id="line-804"> migrate();</span> |
| <span class="source-line-no">805</span><span id="line-805"> return true;</span> |
| <span class="source-line-no">806</span><span id="line-806"> } catch (Exception e) {</span> |
| <span class="source-line-no">807</span><span id="line-807"> LOG.warn("Failed to perform check", e);</span> |
| <span class="source-line-no">808</span><span id="line-808"> // 100ms is short so let's just ignore the interrupt</span> |
| <span class="source-line-no">809</span><span id="line-809"> Threads.sleepWithoutInterrupt(100);</span> |
| <span class="source-line-no">810</span><span id="line-810"> }</span> |
| <span class="source-line-no">811</span><span id="line-811"> }</span> |
| <span class="source-line-no">812</span><span id="line-812"> return false;</span> |
| <span class="source-line-no">813</span><span id="line-813"> }</span> |
| <span class="source-line-no">814</span><span id="line-814"></span> |
| <span class="source-line-no">815</span><span id="line-815"> private void createRSGroupTable() throws IOException {</span> |
| <span class="source-line-no">816</span><span id="line-816"> OptionalLong optProcId = masterServices.getProcedures().stream()</span> |
| <span class="source-line-no">817</span><span id="line-817"> .filter(p -> p instanceof CreateTableProcedure).map(p -> (CreateTableProcedure) p)</span> |
| <span class="source-line-no">818</span><span id="line-818"> .filter(p -> p.getTableName().equals(RSGROUP_TABLE_NAME)).mapToLong(Procedure::getProcId)</span> |
| <span class="source-line-no">819</span><span id="line-819"> .findFirst();</span> |
| <span class="source-line-no">820</span><span id="line-820"> long procId;</span> |
| <span class="source-line-no">821</span><span id="line-821"> if (optProcId.isPresent()) {</span> |
| <span class="source-line-no">822</span><span id="line-822"> procId = optProcId.getAsLong();</span> |
| <span class="source-line-no">823</span><span id="line-823"> } else {</span> |
| <span class="source-line-no">824</span><span id="line-824"> LOG.debug("Creating group table {}", RSGROUP_TABLE_NAME);</span> |
| <span class="source-line-no">825</span><span id="line-825"> procId = masterServices.createSystemTable(RSGROUP_TABLE_DESC);</span> |
| <span class="source-line-no">826</span><span id="line-826"> }</span> |
| <span class="source-line-no">827</span><span id="line-827"> // wait for region to be online</span> |
| <span class="source-line-no">828</span><span id="line-828"> int tries = 600;</span> |
| <span class="source-line-no">829</span><span id="line-829"> while (</span> |
| <span class="source-line-no">830</span><span id="line-830"> !(masterServices.getMasterProcedureExecutor().isFinished(procId))</span> |
| <span class="source-line-no">831</span><span id="line-831"> && masterServices.getMasterProcedureExecutor().isRunning() && tries > 0</span> |
| <span class="source-line-no">832</span><span id="line-832"> ) {</span> |
| <span class="source-line-no">833</span><span id="line-833"> try {</span> |
| <span class="source-line-no">834</span><span id="line-834"> Thread.sleep(100);</span> |
| <span class="source-line-no">835</span><span id="line-835"> } catch (InterruptedException e) {</span> |
| <span class="source-line-no">836</span><span id="line-836"> throw new IOException("Wait interrupted ", e);</span> |
| <span class="source-line-no">837</span><span id="line-837"> }</span> |
| <span class="source-line-no">838</span><span id="line-838"> tries--;</span> |
| <span class="source-line-no">839</span><span id="line-839"> }</span> |
| <span class="source-line-no">840</span><span id="line-840"> if (tries <= 0) {</span> |
| <span class="source-line-no">841</span><span id="line-841"> throw new IOException("Failed to create group table in a given time.");</span> |
| <span class="source-line-no">842</span><span id="line-842"> } else {</span> |
| <span class="source-line-no">843</span><span id="line-843"> Procedure<?> result = masterServices.getMasterProcedureExecutor().getResult(procId);</span> |
| <span class="source-line-no">844</span><span id="line-844"> if (result != null && result.isFailed()) {</span> |
| <span class="source-line-no">845</span><span id="line-845"> throw new IOException(</span> |
| <span class="source-line-no">846</span><span id="line-846"> "Failed to create group table. " + MasterProcedureUtil.unwrapRemoteIOException(result));</span> |
| <span class="source-line-no">847</span><span id="line-847"> }</span> |
| <span class="source-line-no">848</span><span id="line-848"> }</span> |
| <span class="source-line-no">849</span><span id="line-849"> }</span> |
| <span class="source-line-no">850</span><span id="line-850"></span> |
| <span class="source-line-no">851</span><span id="line-851"> public boolean isOnline() {</span> |
| <span class="source-line-no">852</span><span id="line-852"> return online;</span> |
| <span class="source-line-no">853</span><span id="line-853"> }</span> |
| <span class="source-line-no">854</span><span id="line-854"> }</span> |
| <span class="source-line-no">855</span><span id="line-855"></span> |
| <span class="source-line-no">856</span><span id="line-856"> private static boolean isMasterRunning(MasterServices masterServices) {</span> |
| <span class="source-line-no">857</span><span id="line-857"> return !masterServices.isAborted() && !masterServices.isStopped();</span> |
| <span class="source-line-no">858</span><span id="line-858"> }</span> |
| <span class="source-line-no">859</span><span id="line-859"></span> |
| <span class="source-line-no">860</span><span id="line-860"> private void multiMutate(List<Mutation> mutations) throws IOException {</span> |
| <span class="source-line-no">861</span><span id="line-861"> MutateRowsRequest.Builder builder = MutateRowsRequest.newBuilder();</span> |
| <span class="source-line-no">862</span><span id="line-862"> for (Mutation mutation : mutations) {</span> |
| <span class="source-line-no">863</span><span id="line-863"> if (mutation instanceof Put) {</span> |
| <span class="source-line-no">864</span><span id="line-864"> builder</span> |
| <span class="source-line-no">865</span><span id="line-865"> .addMutationRequest(ProtobufUtil.toMutation(MutationProto.MutationType.PUT, mutation));</span> |
| <span class="source-line-no">866</span><span id="line-866"> } else if (mutation instanceof Delete) {</span> |
| <span class="source-line-no">867</span><span id="line-867"> builder</span> |
| <span class="source-line-no">868</span><span id="line-868"> .addMutationRequest(ProtobufUtil.toMutation(MutationProto.MutationType.DELETE, mutation));</span> |
| <span class="source-line-no">869</span><span id="line-869"> } else {</span> |
| <span class="source-line-no">870</span><span id="line-870"> throw new DoNotRetryIOException(</span> |
| <span class="source-line-no">871</span><span id="line-871"> "multiMutate doesn't support " + mutation.getClass().getName());</span> |
| <span class="source-line-no">872</span><span id="line-872"> }</span> |
| <span class="source-line-no">873</span><span id="line-873"> }</span> |
| <span class="source-line-no">874</span><span id="line-874"> MutateRowsRequest request = builder.build();</span> |
| <span class="source-line-no">875</span><span id="line-875"> AsyncTable<?> table = conn.getTable(RSGROUP_TABLE_NAME);</span> |
| <span class="source-line-no">876</span><span id="line-876"> LOG.debug("Multimutating {} with {} mutations", RSGROUP_TABLE_NAME, mutations.size());</span> |
| <span class="source-line-no">877</span><span id="line-877"> FutureUtils.get(table.<MultiRowMutationService, MutateRowsResponse> coprocessorService(</span> |
| <span class="source-line-no">878</span><span id="line-878"> MultiRowMutationService::newStub,</span> |
| <span class="source-line-no">879</span><span id="line-879"> (stub, controller, done) -> stub.mutateRows(controller, request, done), ROW_KEY));</span> |
| <span class="source-line-no">880</span><span id="line-880"> LOG.info("Multimutating {} with {} mutations done", RSGROUP_TABLE_NAME, mutations.size());</span> |
| <span class="source-line-no">881</span><span id="line-881"> }</span> |
| <span class="source-line-no">882</span><span id="line-882"></span> |
| <span class="source-line-no">883</span><span id="line-883"> private void checkGroupName(String groupName) throws ConstraintException {</span> |
| <span class="source-line-no">884</span><span id="line-884"> if (!groupName.matches("[a-zA-Z0-9_]+")) {</span> |
| <span class="source-line-no">885</span><span id="line-885"> throw new ConstraintException("RSGroup name should only contain alphanumeric characters");</span> |
| <span class="source-line-no">886</span><span id="line-886"> }</span> |
| <span class="source-line-no">887</span><span id="line-887"> }</span> |
| <span class="source-line-no">888</span><span id="line-888"></span> |
| <span class="source-line-no">889</span><span id="line-889"> @Override</span> |
| <span class="source-line-no">890</span><span id="line-890"> public RSGroupInfo getRSGroupForTable(TableName tableName) throws IOException {</span> |
| <span class="source-line-no">891</span><span id="line-891"> return holder.tableName2Group.get(tableName);</span> |
| <span class="source-line-no">892</span><span id="line-892"> }</span> |
| <span class="source-line-no">893</span><span id="line-893"></span> |
| <span class="source-line-no">894</span><span id="line-894"> /**</span> |
| <span class="source-line-no">895</span><span id="line-895"> * Check if the set of servers are belong to dead servers list or online servers list.</span> |
| <span class="source-line-no">896</span><span id="line-896"> * @param servers servers to remove</span> |
| <span class="source-line-no">897</span><span id="line-897"> */</span> |
| <span class="source-line-no">898</span><span id="line-898"> private void checkForDeadOrOnlineServers(Set<Address> servers) throws IOException {</span> |
| <span class="source-line-no">899</span><span id="line-899"> // This ugliness is because we only have Address, not ServerName.</span> |
| <span class="source-line-no">900</span><span id="line-900"> Set<Address> onlineServers = new HashSet<>();</span> |
| <span class="source-line-no">901</span><span id="line-901"> List<ServerName> drainingServers = masterServices.getServerManager().getDrainingServersList();</span> |
| <span class="source-line-no">902</span><span id="line-902"> for (ServerName server : masterServices.getServerManager().getOnlineServers().keySet()) {</span> |
| <span class="source-line-no">903</span><span id="line-903"> // Only online but not decommissioned servers are really online</span> |
| <span class="source-line-no">904</span><span id="line-904"> if (!drainingServers.contains(server)) {</span> |
| <span class="source-line-no">905</span><span id="line-905"> onlineServers.add(server.getAddress());</span> |
| <span class="source-line-no">906</span><span id="line-906"> }</span> |
| <span class="source-line-no">907</span><span id="line-907"> }</span> |
| <span class="source-line-no">908</span><span id="line-908"></span> |
| <span class="source-line-no">909</span><span id="line-909"> Set<Address> deadServers = new HashSet<>();</span> |
| <span class="source-line-no">910</span><span id="line-910"> for (ServerName server : masterServices.getServerManager().getDeadServers().copyServerNames()) {</span> |
| <span class="source-line-no">911</span><span id="line-911"> deadServers.add(server.getAddress());</span> |
| <span class="source-line-no">912</span><span id="line-912"> }</span> |
| <span class="source-line-no">913</span><span id="line-913"></span> |
| <span class="source-line-no">914</span><span id="line-914"> for (Address address : servers) {</span> |
| <span class="source-line-no">915</span><span id="line-915"> if (onlineServers.contains(address)) {</span> |
| <span class="source-line-no">916</span><span id="line-916"> throw new DoNotRetryIOException(</span> |
| <span class="source-line-no">917</span><span id="line-917"> "Server " + address + " is an online server, not allowed to remove.");</span> |
| <span class="source-line-no">918</span><span id="line-918"> }</span> |
| <span class="source-line-no">919</span><span id="line-919"> if (deadServers.contains(address)) {</span> |
| <span class="source-line-no">920</span><span id="line-920"> throw new DoNotRetryIOException("Server " + address + " is on the dead servers list,"</span> |
| <span class="source-line-no">921</span><span id="line-921"> + " Maybe it will come back again, not allowed to remove.");</span> |
| <span class="source-line-no">922</span><span id="line-922"> }</span> |
| <span class="source-line-no">923</span><span id="line-923"> }</span> |
| <span class="source-line-no">924</span><span id="line-924"> }</span> |
| <span class="source-line-no">925</span><span id="line-925"></span> |
| <span class="source-line-no">926</span><span id="line-926"> private void checkOnlineServersOnly(Set<Address> servers) throws IOException {</span> |
| <span class="source-line-no">927</span><span id="line-927"> // This uglyness is because we only have Address, not ServerName.</span> |
| <span class="source-line-no">928</span><span id="line-928"> // Online servers are keyed by ServerName.</span> |
| <span class="source-line-no">929</span><span id="line-929"> Set<Address> onlineServers = new HashSet<>();</span> |
| <span class="source-line-no">930</span><span id="line-930"> for (ServerName server : masterServices.getServerManager().getOnlineServers().keySet()) {</span> |
| <span class="source-line-no">931</span><span id="line-931"> onlineServers.add(server.getAddress());</span> |
| <span class="source-line-no">932</span><span id="line-932"> }</span> |
| <span class="source-line-no">933</span><span id="line-933"> for (Address address : servers) {</span> |
| <span class="source-line-no">934</span><span id="line-934"> if (!onlineServers.contains(address)) {</span> |
| <span class="source-line-no">935</span><span id="line-935"> throw new DoNotRetryIOException(</span> |
| <span class="source-line-no">936</span><span id="line-936"> "Server " + address + " is not an online server in 'default' RSGroup.");</span> |
| <span class="source-line-no">937</span><span id="line-937"> }</span> |
| <span class="source-line-no">938</span><span id="line-938"> }</span> |
| <span class="source-line-no">939</span><span id="line-939"> }</span> |
| <span class="source-line-no">940</span><span id="line-940"></span> |
| <span class="source-line-no">941</span><span id="line-941"> /** Returns List of Regions associated with this <code>server</code>. */</span> |
| <span class="source-line-no">942</span><span id="line-942"> private List<RegionInfo> getRegions(final Address server) {</span> |
| <span class="source-line-no">943</span><span id="line-943"> LinkedList<RegionInfo> regions = new LinkedList<>();</span> |
| <span class="source-line-no">944</span><span id="line-944"> for (Map.Entry<RegionInfo, ServerName> el : masterServices.getAssignmentManager()</span> |
| <span class="source-line-no">945</span><span id="line-945"> .getRegionStates().getRegionAssignments().entrySet()) {</span> |
| <span class="source-line-no">946</span><span id="line-946"> if (el.getValue() == null) {</span> |
| <span class="source-line-no">947</span><span id="line-947"> continue;</span> |
| <span class="source-line-no">948</span><span id="line-948"> }</span> |
| <span class="source-line-no">949</span><span id="line-949"></span> |
| <span class="source-line-no">950</span><span id="line-950"> if (el.getValue().getAddress().equals(server)) {</span> |
| <span class="source-line-no">951</span><span id="line-951"> addRegion(regions, el.getKey());</span> |
| <span class="source-line-no">952</span><span id="line-952"> }</span> |
| <span class="source-line-no">953</span><span id="line-953"> }</span> |
| <span class="source-line-no">954</span><span id="line-954"> for (RegionStateNode state : masterServices.getAssignmentManager().getRegionsInTransition()) {</span> |
| <span class="source-line-no">955</span><span id="line-955"> if (</span> |
| <span class="source-line-no">956</span><span id="line-956"> state.getRegionLocation() != null && state.getRegionLocation().getAddress().equals(server)</span> |
| <span class="source-line-no">957</span><span id="line-957"> ) {</span> |
| <span class="source-line-no">958</span><span id="line-958"> addRegion(regions, state.getRegionInfo());</span> |
| <span class="source-line-no">959</span><span id="line-959"> }</span> |
| <span class="source-line-no">960</span><span id="line-960"> }</span> |
| <span class="source-line-no">961</span><span id="line-961"> return regions;</span> |
| <span class="source-line-no">962</span><span id="line-962"> }</span> |
| <span class="source-line-no">963</span><span id="line-963"></span> |
| <span class="source-line-no">964</span><span id="line-964"> private void addRegion(final LinkedList<RegionInfo> regions, RegionInfo hri) {</span> |
| <span class="source-line-no">965</span><span id="line-965"> // If meta, move it last otherwise other unassigns fail because meta is not</span> |
| <span class="source-line-no">966</span><span id="line-966"> // online for them to update state in. This is dodgy. Needs to be made more</span> |
| <span class="source-line-no">967</span><span id="line-967"> // robust. See TODO below.</span> |
| <span class="source-line-no">968</span><span id="line-968"> if (hri.isMetaRegion()) {</span> |
| <span class="source-line-no">969</span><span id="line-969"> regions.addLast(hri);</span> |
| <span class="source-line-no">970</span><span id="line-970"> } else {</span> |
| <span class="source-line-no">971</span><span id="line-971"> regions.addFirst(hri);</span> |
| <span class="source-line-no">972</span><span id="line-972"> }</span> |
| <span class="source-line-no">973</span><span id="line-973"> }</span> |
| <span class="source-line-no">974</span><span id="line-974"></span> |
| <span class="source-line-no">975</span><span id="line-975"> /**</span> |
| <span class="source-line-no">976</span><span id="line-976"> * Move every region from servers which are currently located on these servers, but should not be</span> |
| <span class="source-line-no">977</span><span id="line-977"> * located there.</span> |
| <span class="source-line-no">978</span><span id="line-978"> * @param movedServers the servers that are moved to new group</span> |
| <span class="source-line-no">979</span><span id="line-979"> * @param srcGrpServers all servers in the source group, excluding the movedServers</span> |
| <span class="source-line-no">980</span><span id="line-980"> * @param targetGroupName the target group</span> |
| <span class="source-line-no">981</span><span id="line-981"> * @param sourceGroupName the source group</span> |
| <span class="source-line-no">982</span><span id="line-982"> * @throws IOException if moving the server and tables fail</span> |
| <span class="source-line-no">983</span><span id="line-983"> */</span> |
| <span class="source-line-no">984</span><span id="line-984"> private void moveServerRegionsFromGroup(Set<Address> movedServers, Set<Address> srcGrpServers,</span> |
| <span class="source-line-no">985</span><span id="line-985"> String targetGroupName, String sourceGroupName) throws IOException {</span> |
| <span class="source-line-no">986</span><span id="line-986"> moveRegionsBetweenGroups(movedServers, srcGrpServers, targetGroupName, sourceGroupName,</span> |
| <span class="source-line-no">987</span><span id="line-987"> rs -> getRegions(rs), info -> {</span> |
| <span class="source-line-no">988</span><span id="line-988"> try {</span> |
| <span class="source-line-no">989</span><span id="line-989"> String groupName = RSGroupUtil.getRSGroupInfo(masterServices, this, info.getTable())</span> |
| <span class="source-line-no">990</span><span id="line-990"> .map(RSGroupInfo::getName).orElse(RSGroupInfo.DEFAULT_GROUP);</span> |
| <span class="source-line-no">991</span><span id="line-991"> return groupName.equals(targetGroupName);</span> |
| <span class="source-line-no">992</span><span id="line-992"> } catch (IOException e) {</span> |
| <span class="source-line-no">993</span><span id="line-993"> LOG.warn("Failed to test group for region {} and target group {}", info, targetGroupName);</span> |
| <span class="source-line-no">994</span><span id="line-994"> return false;</span> |
| <span class="source-line-no">995</span><span id="line-995"> }</span> |
| <span class="source-line-no">996</span><span id="line-996"> });</span> |
| <span class="source-line-no">997</span><span id="line-997"> }</span> |
| <span class="source-line-no">998</span><span id="line-998"></span> |
| <span class="source-line-no">999</span><span id="line-999"> private <T> void moveRegionsBetweenGroups(Set<T> regionsOwners, Set<Address> newRegionsOwners,</span> |
| <span class="source-line-no">1000</span><span id="line-1000"> String targetGroupName, String sourceGroupName, Function<T, List<RegionInfo>> getRegionsInfo,</span> |
| <span class="source-line-no">1001</span><span id="line-1001"> Function<RegionInfo, Boolean> validation) throws IOException {</span> |
| <span class="source-line-no">1002</span><span id="line-1002"> // Get server names corresponding to given Addresses</span> |
| <span class="source-line-no">1003</span><span id="line-1003"> List<ServerName> movedServerNames = new ArrayList<>(regionsOwners.size());</span> |
| <span class="source-line-no">1004</span><span id="line-1004"> List<ServerName> srcGrpServerNames = new ArrayList<>(newRegionsOwners.size());</span> |
| <span class="source-line-no">1005</span><span id="line-1005"> for (ServerName serverName : masterServices.getServerManager().getOnlineServers().keySet()) {</span> |
| <span class="source-line-no">1006</span><span id="line-1006"> // In case region move failed in previous attempt, regionsOwners and newRegionsOwners</span> |
| <span class="source-line-no">1007</span><span id="line-1007"> // can have the same servers. So for all servers below both conditions to be checked</span> |
| <span class="source-line-no">1008</span><span id="line-1008"> if (newRegionsOwners.contains(serverName.getAddress())) {</span> |
| <span class="source-line-no">1009</span><span id="line-1009"> srcGrpServerNames.add(serverName);</span> |
| <span class="source-line-no">1010</span><span id="line-1010"> }</span> |
| <span class="source-line-no">1011</span><span id="line-1011"> if (regionsOwners.contains(serverName.getAddress())) {</span> |
| <span class="source-line-no">1012</span><span id="line-1012"> movedServerNames.add(serverName);</span> |
| <span class="source-line-no">1013</span><span id="line-1013"> }</span> |
| <span class="source-line-no">1014</span><span id="line-1014"> }</span> |
| <span class="source-line-no">1015</span><span id="line-1015"> List<Pair<RegionInfo, Future<byte[]>>> assignmentFutures = new ArrayList<>();</span> |
| <span class="source-line-no">1016</span><span id="line-1016"> int retry = 0;</span> |
| <span class="source-line-no">1017</span><span id="line-1017"> Set<String> failedRegions = new HashSet<>();</span> |
| <span class="source-line-no">1018</span><span id="line-1018"> IOException toThrow = null;</span> |
| <span class="source-line-no">1019</span><span id="line-1019"> do {</span> |
| <span class="source-line-no">1020</span><span id="line-1020"> assignmentFutures.clear();</span> |
| <span class="source-line-no">1021</span><span id="line-1021"> failedRegions.clear();</span> |
| <span class="source-line-no">1022</span><span id="line-1022"> for (ServerName owner : movedServerNames) {</span> |
| <span class="source-line-no">1023</span><span id="line-1023"> // Get regions that are associated with this server and filter regions by group tables.</span> |
| <span class="source-line-no">1024</span><span id="line-1024"> for (RegionInfo region : getRegionsInfo.apply((T) owner.getAddress())) {</span> |
| <span class="source-line-no">1025</span><span id="line-1025"> if (!validation.apply(region)) {</span> |
| <span class="source-line-no">1026</span><span id="line-1026"> LOG.info("Moving region {}, which does not belong to RSGroup {}",</span> |
| <span class="source-line-no">1027</span><span id="line-1027"> region.getShortNameToLog(), targetGroupName);</span> |
| <span class="source-line-no">1028</span><span id="line-1028"> // Move region back to source RSGroup servers</span> |
| <span class="source-line-no">1029</span><span id="line-1029"> ServerName dest =</span> |
| <span class="source-line-no">1030</span><span id="line-1030"> masterServices.getLoadBalancer().randomAssignment(region, srcGrpServerNames);</span> |
| <span class="source-line-no">1031</span><span id="line-1031"> if (dest == null) {</span> |
| <span class="source-line-no">1032</span><span id="line-1032"> failedRegions.add(region.getRegionNameAsString());</span> |
| <span class="source-line-no">1033</span><span id="line-1033"> continue;</span> |
| <span class="source-line-no">1034</span><span id="line-1034"> }</span> |
| <span class="source-line-no">1035</span><span id="line-1035"> RegionPlan rp = new RegionPlan(region, owner, dest);</span> |
| <span class="source-line-no">1036</span><span id="line-1036"> try {</span> |
| <span class="source-line-no">1037</span><span id="line-1037"> Future<byte[]> future = masterServices.getAssignmentManager().moveAsync(rp);</span> |
| <span class="source-line-no">1038</span><span id="line-1038"> assignmentFutures.add(Pair.newPair(region, future));</span> |
| <span class="source-line-no">1039</span><span id="line-1039"> } catch (IOException ioe) {</span> |
| <span class="source-line-no">1040</span><span id="line-1040"> failedRegions.add(region.getRegionNameAsString());</span> |
| <span class="source-line-no">1041</span><span id="line-1041"> LOG.debug("Move region {} failed, will retry, current retry time is {}",</span> |
| <span class="source-line-no">1042</span><span id="line-1042"> region.getShortNameToLog(), retry, ioe);</span> |
| <span class="source-line-no">1043</span><span id="line-1043"> toThrow = ioe;</span> |
| <span class="source-line-no">1044</span><span id="line-1044"> }</span> |
| <span class="source-line-no">1045</span><span id="line-1045"> }</span> |
| <span class="source-line-no">1046</span><span id="line-1046"> }</span> |
| <span class="source-line-no">1047</span><span id="line-1047"> }</span> |
| <span class="source-line-no">1048</span><span id="line-1048"> waitForRegionMovement(assignmentFutures, failedRegions, sourceGroupName, retry);</span> |
| <span class="source-line-no">1049</span><span id="line-1049"> if (failedRegions.isEmpty()) {</span> |
| <span class="source-line-no">1050</span><span id="line-1050"> LOG.info("All regions from {} are moved back to {}", movedServerNames, sourceGroupName);</span> |
| <span class="source-line-no">1051</span><span id="line-1051"> return;</span> |
| <span class="source-line-no">1052</span><span id="line-1052"> } else {</span> |
| <span class="source-line-no">1053</span><span id="line-1053"> try {</span> |
| <span class="source-line-no">1054</span><span id="line-1054"> wait(1000);</span> |
| <span class="source-line-no">1055</span><span id="line-1055"> } catch (InterruptedException e) {</span> |
| <span class="source-line-no">1056</span><span id="line-1056"> LOG.warn("Sleep interrupted", e);</span> |
| <span class="source-line-no">1057</span><span id="line-1057"> Thread.currentThread().interrupt();</span> |
| <span class="source-line-no">1058</span><span id="line-1058"> }</span> |
| <span class="source-line-no">1059</span><span id="line-1059"> retry++;</span> |
| <span class="source-line-no">1060</span><span id="line-1060"> }</span> |
| <span class="source-line-no">1061</span><span id="line-1061"> } while (</span> |
| <span class="source-line-no">1062</span><span id="line-1062"> !failedRegions.isEmpty() && retry <= masterServices.getConfiguration()</span> |
| <span class="source-line-no">1063</span><span id="line-1063"> .getInt(FAILED_MOVE_MAX_RETRY, DEFAULT_MAX_RETRY_VALUE)</span> |
| <span class="source-line-no">1064</span><span id="line-1064"> );</span> |
| <span class="source-line-no">1065</span><span id="line-1065"></span> |
| <span class="source-line-no">1066</span><span id="line-1066"> // has up to max retry time or there are no more regions to move</span> |
| <span class="source-line-no">1067</span><span id="line-1067"> if (!failedRegions.isEmpty()) {</span> |
| <span class="source-line-no">1068</span><span id="line-1068"> // print failed moved regions, for later process conveniently</span> |
| <span class="source-line-no">1069</span><span id="line-1069"> String msg = String.format("move regions for group %s failed, failed regions: %s",</span> |
| <span class="source-line-no">1070</span><span id="line-1070"> sourceGroupName, failedRegions);</span> |
| <span class="source-line-no">1071</span><span id="line-1071"> LOG.error(msg);</span> |
| <span class="source-line-no">1072</span><span id="line-1072"> throw new DoNotRetryIOException(</span> |
| <span class="source-line-no">1073</span><span id="line-1073"> msg + ", just record the last failed region's cause, more details in server log", toThrow);</span> |
| <span class="source-line-no">1074</span><span id="line-1074"> }</span> |
| <span class="source-line-no">1075</span><span id="line-1075"> }</span> |
| <span class="source-line-no">1076</span><span id="line-1076"></span> |
| <span class="source-line-no">1077</span><span id="line-1077"> /**</span> |
| <span class="source-line-no">1078</span><span id="line-1078"> * Wait for all the region move to complete. Keep waiting for other region movement completion</span> |
| <span class="source-line-no">1079</span><span id="line-1079"> * even if some region movement fails.</span> |
| <span class="source-line-no">1080</span><span id="line-1080"> */</span> |
| <span class="source-line-no">1081</span><span id="line-1081"> private void waitForRegionMovement(List<Pair<RegionInfo, Future<byte[]>>> regionMoveFutures,</span> |
| <span class="source-line-no">1082</span><span id="line-1082"> Set<String> failedRegions, String sourceGroupName, int retryCount) {</span> |
| <span class="source-line-no">1083</span><span id="line-1083"> LOG.info("Moving {} region(s) to group {}, current retry={}", regionMoveFutures.size(),</span> |
| <span class="source-line-no">1084</span><span id="line-1084"> sourceGroupName, retryCount);</span> |
| <span class="source-line-no">1085</span><span id="line-1085"> for (Pair<RegionInfo, Future<byte[]>> pair : regionMoveFutures) {</span> |
| <span class="source-line-no">1086</span><span id="line-1086"> try {</span> |
| <span class="source-line-no">1087</span><span id="line-1087"> pair.getSecond().get();</span> |
| <span class="source-line-no">1088</span><span id="line-1088"> if (</span> |
| <span class="source-line-no">1089</span><span id="line-1089"> masterServices.getAssignmentManager().getRegionStates().getRegionState(pair.getFirst())</span> |
| <span class="source-line-no">1090</span><span id="line-1090"> .isFailedOpen()</span> |
| <span class="source-line-no">1091</span><span id="line-1091"> ) {</span> |
| <span class="source-line-no">1092</span><span id="line-1092"> failedRegions.add(pair.getFirst().getRegionNameAsString());</span> |
| <span class="source-line-no">1093</span><span id="line-1093"> }</span> |
| <span class="source-line-no">1094</span><span id="line-1094"> } catch (InterruptedException e) {</span> |
| <span class="source-line-no">1095</span><span id="line-1095"> // Dont return form there lets wait for other regions to complete movement.</span> |
| <span class="source-line-no">1096</span><span id="line-1096"> failedRegions.add(pair.getFirst().getRegionNameAsString());</span> |
| <span class="source-line-no">1097</span><span id="line-1097"> LOG.warn("Sleep interrupted", e);</span> |
| <span class="source-line-no">1098</span><span id="line-1098"> } catch (Exception e) {</span> |
| <span class="source-line-no">1099</span><span id="line-1099"> failedRegions.add(pair.getFirst().getRegionNameAsString());</span> |
| <span class="source-line-no">1100</span><span id="line-1100"> LOG.error("Move region {} to group {} failed, will retry on next attempt",</span> |
| <span class="source-line-no">1101</span><span id="line-1101"> pair.getFirst().getShortNameToLog(), sourceGroupName, e);</span> |
| <span class="source-line-no">1102</span><span id="line-1102"> }</span> |
| <span class="source-line-no">1103</span><span id="line-1103"> }</span> |
| <span class="source-line-no">1104</span><span id="line-1104"> }</span> |
| <span class="source-line-no">1105</span><span id="line-1105"></span> |
| <span class="source-line-no">1106</span><span id="line-1106"> private boolean isTableInGroup(TableName tableName, String groupName,</span> |
| <span class="source-line-no">1107</span><span id="line-1107"> Set<TableName> tablesInGroupCache) throws IOException {</span> |
| <span class="source-line-no">1108</span><span id="line-1108"> if (tablesInGroupCache.contains(tableName)) {</span> |
| <span class="source-line-no">1109</span><span id="line-1109"> return true;</span> |
| <span class="source-line-no">1110</span><span id="line-1110"> }</span> |
| <span class="source-line-no">1111</span><span id="line-1111"> if (</span> |
| <span class="source-line-no">1112</span><span id="line-1112"> RSGroupUtil.getRSGroupInfo(masterServices, this, tableName).map(RSGroupInfo::getName)</span> |
| <span class="source-line-no">1113</span><span id="line-1113"> .orElse(RSGroupInfo.DEFAULT_GROUP).equals(groupName)</span> |
| <span class="source-line-no">1114</span><span id="line-1114"> ) {</span> |
| <span class="source-line-no">1115</span><span id="line-1115"> tablesInGroupCache.add(tableName);</span> |
| <span class="source-line-no">1116</span><span id="line-1116"> return true;</span> |
| <span class="source-line-no">1117</span><span id="line-1117"> }</span> |
| <span class="source-line-no">1118</span><span id="line-1118"> return false;</span> |
| <span class="source-line-no">1119</span><span id="line-1119"> }</span> |
| <span class="source-line-no">1120</span><span id="line-1120"></span> |
| <span class="source-line-no">1121</span><span id="line-1121"> private Map<String, RegionState> rsGroupGetRegionsInTransition(String groupName)</span> |
| <span class="source-line-no">1122</span><span id="line-1122"> throws IOException {</span> |
| <span class="source-line-no">1123</span><span id="line-1123"> Map<String, RegionState> rit = Maps.newTreeMap();</span> |
| <span class="source-line-no">1124</span><span id="line-1124"> Set<TableName> tablesInGroupCache = new HashSet<>();</span> |
| <span class="source-line-no">1125</span><span id="line-1125"> for (RegionStateNode regionNode : masterServices.getAssignmentManager()</span> |
| <span class="source-line-no">1126</span><span id="line-1126"> .getRegionsInTransition()) {</span> |
| <span class="source-line-no">1127</span><span id="line-1127"> TableName tn = regionNode.getTable();</span> |
| <span class="source-line-no">1128</span><span id="line-1128"> if (isTableInGroup(tn, groupName, tablesInGroupCache)) {</span> |
| <span class="source-line-no">1129</span><span id="line-1129"> rit.put(regionNode.getRegionInfo().getEncodedName(), regionNode.toRegionState());</span> |
| <span class="source-line-no">1130</span><span id="line-1130"> }</span> |
| <span class="source-line-no">1131</span><span id="line-1131"> }</span> |
| <span class="source-line-no">1132</span><span id="line-1132"> return rit;</span> |
| <span class="source-line-no">1133</span><span id="line-1133"> }</span> |
| <span class="source-line-no">1134</span><span id="line-1134"></span> |
| <span class="source-line-no">1135</span><span id="line-1135"> /**</span> |
| <span class="source-line-no">1136</span><span id="line-1136"> * This is an EXPENSIVE clone. Cloning though is the safest thing to do. Can't let out original</span> |
| <span class="source-line-no">1137</span><span id="line-1137"> * since it can change and at least the load balancer wants to iterate this exported list. Load</span> |
| <span class="source-line-no">1138</span><span id="line-1138"> * balancer should iterate over this list because cloned list will ignore disabled table and split</span> |
| <span class="source-line-no">1139</span><span id="line-1139"> * parent region cases. This method is invoked by {@link #balanceRSGroup}</span> |
| <span class="source-line-no">1140</span><span id="line-1140"> * @return A clone of current assignments for this group.</span> |
| <span class="source-line-no">1141</span><span id="line-1141"> */</span> |
| <span class="source-line-no">1142</span><span id="line-1142"> Map<TableName, Map<ServerName, List<RegionInfo>>> getRSGroupAssignmentsByTable(</span> |
| <span class="source-line-no">1143</span><span id="line-1143"> TableStateManager tableStateManager, String groupName) throws IOException {</span> |
| <span class="source-line-no">1144</span><span id="line-1144"> Map<TableName, Map<ServerName, List<RegionInfo>>> result = Maps.newHashMap();</span> |
| <span class="source-line-no">1145</span><span id="line-1145"> Set<TableName> tablesInGroupCache = new HashSet<>();</span> |
| <span class="source-line-no">1146</span><span id="line-1146"> for (Map.Entry<RegionInfo, ServerName> entry : masterServices.getAssignmentManager()</span> |
| <span class="source-line-no">1147</span><span id="line-1147"> .getRegionStates().getRegionAssignments().entrySet()) {</span> |
| <span class="source-line-no">1148</span><span id="line-1148"> RegionInfo region = entry.getKey();</span> |
| <span class="source-line-no">1149</span><span id="line-1149"> TableName tn = region.getTable();</span> |
| <span class="source-line-no">1150</span><span id="line-1150"> ServerName server = entry.getValue();</span> |
| <span class="source-line-no">1151</span><span id="line-1151"> if (isTableInGroup(tn, groupName, tablesInGroupCache)) {</span> |
| <span class="source-line-no">1152</span><span id="line-1152"> if (</span> |
| <span class="source-line-no">1153</span><span id="line-1153"> tableStateManager.isTableState(tn, TableState.State.DISABLED, TableState.State.DISABLING)</span> |
| <span class="source-line-no">1154</span><span id="line-1154"> ) {</span> |
| <span class="source-line-no">1155</span><span id="line-1155"> continue;</span> |
| <span class="source-line-no">1156</span><span id="line-1156"> }</span> |
| <span class="source-line-no">1157</span><span id="line-1157"> if (region.isSplitParent()) {</span> |
| <span class="source-line-no">1158</span><span id="line-1158"> continue;</span> |
| <span class="source-line-no">1159</span><span id="line-1159"> }</span> |
| <span class="source-line-no">1160</span><span id="line-1160"> result.computeIfAbsent(tn, k -> new HashMap<>())</span> |
| <span class="source-line-no">1161</span><span id="line-1161"> .computeIfAbsent(server, k -> new ArrayList<>()).add(region);</span> |
| <span class="source-line-no">1162</span><span id="line-1162"> }</span> |
| <span class="source-line-no">1163</span><span id="line-1163"> }</span> |
| <span class="source-line-no">1164</span><span id="line-1164"> RSGroupInfo rsGroupInfo = getRSGroupInfo(groupName);</span> |
| <span class="source-line-no">1165</span><span id="line-1165"> for (ServerName serverName : masterServices.getServerManager().getOnlineServers().keySet()) {</span> |
| <span class="source-line-no">1166</span><span id="line-1166"> if (rsGroupInfo.containsServer(serverName.getAddress())) {</span> |
| <span class="source-line-no">1167</span><span id="line-1167"> for (Map<ServerName, List<RegionInfo>> map : result.values()) {</span> |
| <span class="source-line-no">1168</span><span id="line-1168"> map.computeIfAbsent(serverName, k -> Collections.emptyList());</span> |
| <span class="source-line-no">1169</span><span id="line-1169"> }</span> |
| <span class="source-line-no">1170</span><span id="line-1170"> }</span> |
| <span class="source-line-no">1171</span><span id="line-1171"> }</span> |
| <span class="source-line-no">1172</span><span id="line-1172"> return result;</span> |
| <span class="source-line-no">1173</span><span id="line-1173"> }</span> |
| <span class="source-line-no">1174</span><span id="line-1174"></span> |
| <span class="source-line-no">1175</span><span id="line-1175"> @Override</span> |
| <span class="source-line-no">1176</span><span id="line-1176"> public BalanceResponse balanceRSGroup(String groupName, BalanceRequest request)</span> |
| <span class="source-line-no">1177</span><span id="line-1177"> throws IOException {</span> |
| <span class="source-line-no">1178</span><span id="line-1178"> ServerManager serverManager = masterServices.getServerManager();</span> |
| <span class="source-line-no">1179</span><span id="line-1179"> LoadBalancer balancer = masterServices.getLoadBalancer();</span> |
| <span class="source-line-no">1180</span><span id="line-1180"> getRSGroupInfo(groupName);</span> |
| <span class="source-line-no">1181</span><span id="line-1181"></span> |
| <span class="source-line-no">1182</span><span id="line-1182"> BalanceResponse.Builder responseBuilder = BalanceResponse.newBuilder();</span> |
| <span class="source-line-no">1183</span><span id="line-1183"></span> |
| <span class="source-line-no">1184</span><span id="line-1184"> synchronized (balancer) {</span> |
| <span class="source-line-no">1185</span><span id="line-1185"> // If balance not true, don't run balancer.</span> |
| <span class="source-line-no">1186</span><span id="line-1186"> if (!masterServices.isBalancerOn() && !request.isDryRun()) {</span> |
| <span class="source-line-no">1187</span><span id="line-1187"> return responseBuilder.build();</span> |
| <span class="source-line-no">1188</span><span id="line-1188"> }</span> |
| <span class="source-line-no">1189</span><span id="line-1189"></span> |
| <span class="source-line-no">1190</span><span id="line-1190"> // Only allow one balance run at at time.</span> |
| <span class="source-line-no">1191</span><span id="line-1191"> Map<String, RegionState> groupRIT = rsGroupGetRegionsInTransition(groupName);</span> |
| <span class="source-line-no">1192</span><span id="line-1192"> if (groupRIT.size() > 0 && !request.isIgnoreRegionsInTransition()) {</span> |
| <span class="source-line-no">1193</span><span id="line-1193"> LOG.debug("Not running balancer because {} region(s) in transition: {}", groupRIT.size(),</span> |
| <span class="source-line-no">1194</span><span id="line-1194"> StringUtils.abbreviate(masterServices.getAssignmentManager().getRegionStates()</span> |
| <span class="source-line-no">1195</span><span id="line-1195"> .getRegionsInTransition().toString(), 256));</span> |
| <span class="source-line-no">1196</span><span id="line-1196"> return responseBuilder.build();</span> |
| <span class="source-line-no">1197</span><span id="line-1197"> }</span> |
| <span class="source-line-no">1198</span><span id="line-1198"></span> |
| <span class="source-line-no">1199</span><span id="line-1199"> if (serverManager.areDeadServersInProgress()) {</span> |
| <span class="source-line-no">1200</span><span id="line-1200"> LOG.debug("Not running balancer because processing dead regionserver(s): {}",</span> |
| <span class="source-line-no">1201</span><span id="line-1201"> serverManager.getDeadServers());</span> |
| <span class="source-line-no">1202</span><span id="line-1202"> return responseBuilder.build();</span> |
| <span class="source-line-no">1203</span><span id="line-1203"> }</span> |
| <span class="source-line-no">1204</span><span id="line-1204"></span> |
| <span class="source-line-no">1205</span><span id="line-1205"> // We balance per group instead of per table</span> |
| <span class="source-line-no">1206</span><span id="line-1206"> Map<TableName, Map<ServerName, List<RegionInfo>>> assignmentsByTable =</span> |
| <span class="source-line-no">1207</span><span id="line-1207"> getRSGroupAssignmentsByTable(masterServices.getTableStateManager(), groupName);</span> |
| <span class="source-line-no">1208</span><span id="line-1208"> List<RegionPlan> plans = balancer.balanceCluster(assignmentsByTable);</span> |
| <span class="source-line-no">1209</span><span id="line-1209"> boolean balancerRan = !plans.isEmpty();</span> |
| <span class="source-line-no">1210</span><span id="line-1210"></span> |
| <span class="source-line-no">1211</span><span id="line-1211"> responseBuilder.setBalancerRan(balancerRan).setMovesCalculated(plans.size());</span> |
| <span class="source-line-no">1212</span><span id="line-1212"></span> |
| <span class="source-line-no">1213</span><span id="line-1213"> if (balancerRan && !request.isDryRun()) {</span> |
| <span class="source-line-no">1214</span><span id="line-1214"> LOG.info("RSGroup balance {} starting with plan count: {}", groupName, plans.size());</span> |
| <span class="source-line-no">1215</span><span id="line-1215"> List<RegionPlan> executed = masterServices.executeRegionPlansWithThrottling(plans);</span> |
| <span class="source-line-no">1216</span><span id="line-1216"> responseBuilder.setMovesExecuted(executed.size());</span> |
| <span class="source-line-no">1217</span><span id="line-1217"> LOG.info("RSGroup balance " + groupName + " completed");</span> |
| <span class="source-line-no">1218</span><span id="line-1218"> }</span> |
| <span class="source-line-no">1219</span><span id="line-1219"></span> |
| <span class="source-line-no">1220</span><span id="line-1220"> return responseBuilder.build();</span> |
| <span class="source-line-no">1221</span><span id="line-1221"> }</span> |
| <span class="source-line-no">1222</span><span id="line-1222"> }</span> |
| <span class="source-line-no">1223</span><span id="line-1223"></span> |
| <span class="source-line-no">1224</span><span id="line-1224"> private void moveTablesAndWait(Set<TableName> tables, String targetGroup) throws IOException {</span> |
| <span class="source-line-no">1225</span><span id="line-1225"> LOG.debug("Moving {} tables to target group {}", tables.size(), targetGroup);</span> |
| <span class="source-line-no">1226</span><span id="line-1226"> List<Long> procIds = new ArrayList<Long>();</span> |
| <span class="source-line-no">1227</span><span id="line-1227"> for (TableName tableName : tables) {</span> |
| <span class="source-line-no">1228</span><span id="line-1228"> TableDescriptor oldTd = masterServices.getTableDescriptors().get(tableName);</span> |
| <span class="source-line-no">1229</span><span id="line-1229"> if (oldTd == null) {</span> |
| <span class="source-line-no">1230</span><span id="line-1230"> continue;</span> |
| <span class="source-line-no">1231</span><span id="line-1231"> }</span> |
| <span class="source-line-no">1232</span><span id="line-1232"> TableDescriptor newTd =</span> |
| <span class="source-line-no">1233</span><span id="line-1233"> TableDescriptorBuilder.newBuilder(oldTd).setRegionServerGroup(targetGroup).build();</span> |
| <span class="source-line-no">1234</span><span id="line-1234"> procIds.add(</span> |
| <span class="source-line-no">1235</span><span id="line-1235"> masterServices.modifyTable(tableName, newTd, HConstants.NO_NONCE, HConstants.NO_NONCE));</span> |
| <span class="source-line-no">1236</span><span id="line-1236"> }</span> |
| <span class="source-line-no">1237</span><span id="line-1237"> for (long procId : procIds) {</span> |
| <span class="source-line-no">1238</span><span id="line-1238"> Procedure<?> proc = masterServices.getMasterProcedureExecutor().getProcedure(procId);</span> |
| <span class="source-line-no">1239</span><span id="line-1239"> if (proc == null) {</span> |
| <span class="source-line-no">1240</span><span id="line-1240"> continue;</span> |
| <span class="source-line-no">1241</span><span id="line-1241"> }</span> |
| <span class="source-line-no">1242</span><span id="line-1242"> ProcedureSyncWait.waitForProcedureToCompleteIOE(masterServices.getMasterProcedureExecutor(),</span> |
| <span class="source-line-no">1243</span><span id="line-1243"> proc, Long.MAX_VALUE);</span> |
| <span class="source-line-no">1244</span><span id="line-1244"> }</span> |
| <span class="source-line-no">1245</span><span id="line-1245"> LOG.info("Move tables done: moved {} tables to {}", tables.size(), targetGroup);</span> |
| <span class="source-line-no">1246</span><span id="line-1246"> if (LOG.isDebugEnabled()) {</span> |
| <span class="source-line-no">1247</span><span id="line-1247"> LOG.debug("Tables moved to {}: {}", targetGroup, tables);</span> |
| <span class="source-line-no">1248</span><span id="line-1248"> }</span> |
| <span class="source-line-no">1249</span><span id="line-1249"> }</span> |
| <span class="source-line-no">1250</span><span id="line-1250"></span> |
| <span class="source-line-no">1251</span><span id="line-1251"> @Override</span> |
| <span class="source-line-no">1252</span><span id="line-1252"> public void setRSGroup(Set<TableName> tables, String groupName) throws IOException {</span> |
| <span class="source-line-no">1253</span><span id="line-1253"> getRSGroupInfo(groupName);</span> |
| <span class="source-line-no">1254</span><span id="line-1254"> moveTablesAndWait(tables, groupName);</span> |
| <span class="source-line-no">1255</span><span id="line-1255"> }</span> |
| <span class="source-line-no">1256</span><span id="line-1256"></span> |
| <span class="source-line-no">1257</span><span id="line-1257"> public void moveServers(Set<Address> servers, String targetGroupName) throws IOException {</span> |
| <span class="source-line-no">1258</span><span id="line-1258"> if (servers == null) {</span> |
| <span class="source-line-no">1259</span><span id="line-1259"> throw new ConstraintException("The list of servers to move cannot be null.");</span> |
| <span class="source-line-no">1260</span><span id="line-1260"> }</span> |
| <span class="source-line-no">1261</span><span id="line-1261"> if (servers.isEmpty()) {</span> |
| <span class="source-line-no">1262</span><span id="line-1262"> // For some reason this difference between null servers and isEmpty is important distinction.</span> |
| <span class="source-line-no">1263</span><span id="line-1263"> // TODO. Why? Stuff breaks if I equate them.</span> |
| <span class="source-line-no">1264</span><span id="line-1264"> return;</span> |
| <span class="source-line-no">1265</span><span id="line-1265"> }</span> |
| <span class="source-line-no">1266</span><span id="line-1266"> if (StringUtils.isEmpty(targetGroupName)) {</span> |
| <span class="source-line-no">1267</span><span id="line-1267"> throw new ConstraintException("RSGroup cannot be null.");</span> |
| <span class="source-line-no">1268</span><span id="line-1268"> }</span> |
| <span class="source-line-no">1269</span><span id="line-1269"></span> |
| <span class="source-line-no">1270</span><span id="line-1270"> // Hold a lock on the manager instance while moving servers to prevent</span> |
| <span class="source-line-no">1271</span><span id="line-1271"> // another writer changing our state while we are working.</span> |
| <span class="source-line-no">1272</span><span id="line-1272"> synchronized (this) {</span> |
| <span class="source-line-no">1273</span><span id="line-1273"> // Presume first server's source group. Later ensure all servers are from this group.</span> |
| <span class="source-line-no">1274</span><span id="line-1274"> Address firstServer = servers.iterator().next();</span> |
| <span class="source-line-no">1275</span><span id="line-1275"> RSGroupInfo srcGrp = getRSGroupOfServer(firstServer);</span> |
| <span class="source-line-no">1276</span><span id="line-1276"> if (srcGrp == null) {</span> |
| <span class="source-line-no">1277</span><span id="line-1277"> // Be careful. This exception message is tested for in TestRSGroupAdmin2...</span> |
| <span class="source-line-no">1278</span><span id="line-1278"> throw new ConstraintException(</span> |
| <span class="source-line-no">1279</span><span id="line-1279"> "Server " + firstServer + " is either offline or it does not exist.");</span> |
| <span class="source-line-no">1280</span><span id="line-1280"> }</span> |
| <span class="source-line-no">1281</span><span id="line-1281"></span> |
| <span class="source-line-no">1282</span><span id="line-1282"> // Only move online servers (when moving from 'default') or servers from other</span> |
| <span class="source-line-no">1283</span><span id="line-1283"> // groups. This prevents bogus servers from entering groups</span> |
| <span class="source-line-no">1284</span><span id="line-1284"> if (RSGroupInfo.DEFAULT_GROUP.equals(srcGrp.getName())) {</span> |
| <span class="source-line-no">1285</span><span id="line-1285"> if (srcGrp.getServers().size() <= servers.size()) {</span> |
| <span class="source-line-no">1286</span><span id="line-1286"> throw new ConstraintException(KEEP_ONE_SERVER_IN_DEFAULT_ERROR_MESSAGE);</span> |
| <span class="source-line-no">1287</span><span id="line-1287"> }</span> |
| <span class="source-line-no">1288</span><span id="line-1288"> checkOnlineServersOnly(servers);</span> |
| <span class="source-line-no">1289</span><span id="line-1289"> }</span> |
| <span class="source-line-no">1290</span><span id="line-1290"> // Ensure all servers are of same rsgroup.</span> |
| <span class="source-line-no">1291</span><span id="line-1291"> for (Address server : servers) {</span> |
| <span class="source-line-no">1292</span><span id="line-1292"> String tmpGroup = getRSGroupOfServer(server).getName();</span> |
| <span class="source-line-no">1293</span><span id="line-1293"> if (!tmpGroup.equals(srcGrp.getName())) {</span> |
| <span class="source-line-no">1294</span><span id="line-1294"> throw new ConstraintException("Move server request should only come from one source "</span> |
| <span class="source-line-no">1295</span><span id="line-1295"> + "RSGroup. Expecting only " + srcGrp.getName() + " but contains " + tmpGroup);</span> |
| <span class="source-line-no">1296</span><span id="line-1296"> }</span> |
| <span class="source-line-no">1297</span><span id="line-1297"> }</span> |
| <span class="source-line-no">1298</span><span id="line-1298"> if (srcGrp.getServers().size() <= servers.size()) {</span> |
| <span class="source-line-no">1299</span><span id="line-1299"> // check if there are still tables reference this group</span> |
| <span class="source-line-no">1300</span><span id="line-1300"> for (TableDescriptor td : masterServices.getTableDescriptors().getAll().values()) {</span> |
| <span class="source-line-no">1301</span><span id="line-1301"> Optional<String> optGroupName = td.getRegionServerGroup();</span> |
| <span class="source-line-no">1302</span><span id="line-1302"> if (optGroupName.isPresent() && optGroupName.get().equals(srcGrp.getName())) {</span> |
| <span class="source-line-no">1303</span><span id="line-1303"> throw new ConstraintException(</span> |
| <span class="source-line-no">1304</span><span id="line-1304"> "Cannot leave a RSGroup " + srcGrp.getName() + " that contains tables('"</span> |
| <span class="source-line-no">1305</span><span id="line-1305"> + td.getTableName() + "' at least) without servers to host them.");</span> |
| <span class="source-line-no">1306</span><span id="line-1306"> }</span> |
| <span class="source-line-no">1307</span><span id="line-1307"> }</span> |
| <span class="source-line-no">1308</span><span id="line-1308"> }</span> |
| <span class="source-line-no">1309</span><span id="line-1309"></span> |
| <span class="source-line-no">1310</span><span id="line-1310"> // MovedServers may be < passed in 'servers'.</span> |
| <span class="source-line-no">1311</span><span id="line-1311"> Set<Address> movedServers = moveServers(servers, srcGrp.getName(), targetGroupName);</span> |
| <span class="source-line-no">1312</span><span id="line-1312"> moveServerRegionsFromGroup(movedServers, srcGrp.getServers(), targetGroupName,</span> |
| <span class="source-line-no">1313</span><span id="line-1313"> srcGrp.getName());</span> |
| <span class="source-line-no">1314</span><span id="line-1314"> LOG.info("Move servers done: moved {} servers from {} to {}", movedServers.size(),</span> |
| <span class="source-line-no">1315</span><span id="line-1315"> srcGrp.getName(), targetGroupName);</span> |
| <span class="source-line-no">1316</span><span id="line-1316"> if (LOG.isDebugEnabled()) {</span> |
| <span class="source-line-no">1317</span><span id="line-1317"> LOG.debug("Servers moved from {} to {}: {}", srcGrp.getName(), targetGroupName,</span> |
| <span class="source-line-no">1318</span><span id="line-1318"> movedServers);</span> |
| <span class="source-line-no">1319</span><span id="line-1319"> }</span> |
| <span class="source-line-no">1320</span><span id="line-1320"> }</span> |
| <span class="source-line-no">1321</span><span id="line-1321"> }</span> |
| <span class="source-line-no">1322</span><span id="line-1322"></span> |
| <span class="source-line-no">1323</span><span id="line-1323"> @Override</span> |
| <span class="source-line-no">1324</span><span id="line-1324"> public String determineRSGroupInfoForTable(TableName tableName) {</span> |
| <span class="source-line-no">1325</span><span id="line-1325"> return script.getRSGroup(tableName.getNamespaceAsString(), tableName.getQualifierAsString());</span> |
| <span class="source-line-no">1326</span><span id="line-1326"> }</span> |
| <span class="source-line-no">1327</span><span id="line-1327"></span> |
| <span class="source-line-no">1328</span><span id="line-1328"> @Override</span> |
| <span class="source-line-no">1329</span><span id="line-1329"> public synchronized void renameRSGroup(String oldName, String newName) throws IOException {</span> |
| <span class="source-line-no">1330</span><span id="line-1330"> if (oldName.equals(RSGroupInfo.DEFAULT_GROUP)) {</span> |
| <span class="source-line-no">1331</span><span id="line-1331"> throw new ConstraintException(RSGroupInfo.DEFAULT_GROUP + " can't be rename");</span> |
| <span class="source-line-no">1332</span><span id="line-1332"> }</span> |
| <span class="source-line-no">1333</span><span id="line-1333"> checkGroupName(newName);</span> |
| <span class="source-line-no">1334</span><span id="line-1334"> // getRSGroupInfo validates old RSGroup existence.</span> |
| <span class="source-line-no">1335</span><span id="line-1335"> RSGroupInfo oldRSG = getRSGroupInfo(oldName);</span> |
| <span class="source-line-no">1336</span><span id="line-1336"> Map<String, RSGroupInfo> rsGroupMap = holder.groupName2Group;</span> |
| <span class="source-line-no">1337</span><span id="line-1337"> if (rsGroupMap.containsKey(newName)) {</span> |
| <span class="source-line-no">1338</span><span id="line-1338"> throw new ConstraintException("Group already exists: " + newName);</span> |
| <span class="source-line-no">1339</span><span id="line-1339"> }</span> |
| <span class="source-line-no">1340</span><span id="line-1340"></span> |
| <span class="source-line-no">1341</span><span id="line-1341"> Map<String, RSGroupInfo> newGroupMap = Maps.newHashMap(rsGroupMap);</span> |
| <span class="source-line-no">1342</span><span id="line-1342"> newGroupMap.remove(oldRSG.getName());</span> |
| <span class="source-line-no">1343</span><span id="line-1343"> RSGroupInfo newRSG = new RSGroupInfo(newName, oldRSG.getServers());</span> |
| <span class="source-line-no">1344</span><span id="line-1344"> newGroupMap.put(newName, newRSG);</span> |
| <span class="source-line-no">1345</span><span id="line-1345"> flushConfig(newGroupMap);</span> |
| <span class="source-line-no">1346</span><span id="line-1346"> Set<TableName> updateTables = masterServices.getTableDescriptors().getAll().values().stream()</span> |
| <span class="source-line-no">1347</span><span id="line-1347"> .filter(t -> oldName.equals(t.getRegionServerGroup().orElse(null)))</span> |
| <span class="source-line-no">1348</span><span id="line-1348"> .map(TableDescriptor::getTableName).collect(Collectors.toSet());</span> |
| <span class="source-line-no">1349</span><span id="line-1349"> setRSGroup(updateTables, newName);</span> |
| <span class="source-line-no">1350</span><span id="line-1350"> LOG.info("Rename RSGroup done: {} => {}", oldName, newName);</span> |
| <span class="source-line-no">1351</span><span id="line-1351"> }</span> |
| <span class="source-line-no">1352</span><span id="line-1352"></span> |
| <span class="source-line-no">1353</span><span id="line-1353"> @Override</span> |
| <span class="source-line-no">1354</span><span id="line-1354"> public synchronized void updateRSGroupConfig(String groupName, Map<String, String> configuration)</span> |
| <span class="source-line-no">1355</span><span id="line-1355"> throws IOException {</span> |
| <span class="source-line-no">1356</span><span id="line-1356"> if (RSGroupInfo.DEFAULT_GROUP.equals(groupName)) {</span> |
| <span class="source-line-no">1357</span><span id="line-1357"> // We do not persist anything of default group, therefore, it is not supported to update</span> |
| <span class="source-line-no">1358</span><span id="line-1358"> // default group's configuration which lost once master down.</span> |
| <span class="source-line-no">1359</span><span id="line-1359"> throw new ConstraintException(</span> |
| <span class="source-line-no">1360</span><span id="line-1360"> "configuration of " + RSGroupInfo.DEFAULT_GROUP + " can't be stored persistently");</span> |
| <span class="source-line-no">1361</span><span id="line-1361"> }</span> |
| <span class="source-line-no">1362</span><span id="line-1362"> RSGroupInfo rsGroupInfo = getRSGroupInfo(groupName);</span> |
| <span class="source-line-no">1363</span><span id="line-1363"> rsGroupInfo.getConfiguration().forEach((k, v) -> rsGroupInfo.removeConfiguration(k));</span> |
| <span class="source-line-no">1364</span><span id="line-1364"> configuration.forEach((k, v) -> rsGroupInfo.setConfiguration(k, v));</span> |
| <span class="source-line-no">1365</span><span id="line-1365"> flushConfig();</span> |
| <span class="source-line-no">1366</span><span id="line-1366"> }</span> |
| <span class="source-line-no">1367</span><span id="line-1367">}</span> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </main> |
| </body> |
| </html> |