http://git-wip-us.apache.org/repos/asf/hbase-site/blob/65565d77/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.FlushedSequenceIdFlusher.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.FlushedSequenceIdFlusher.html b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.FlushedSequenceIdFlusher.html new file mode 100644 index 0000000..ba6e4aa --- /dev/null +++ b/devapidocs/src-html/org/apache/hadoop/hbase/master/ServerManager.FlushedSequenceIdFlusher.html @@ -0,0 +1,1244 @@ +<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd";> +<html lang="en"> +<head> +<title>Source code</title> +<link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style"> +</head> +<body> +<div class="sourceContainer"> +<pre><span class="sourceLineNo">001</span>/*<a name="line.1"></a> +<span class="sourceLineNo">002</span> *<a name="line.2"></a> +<span class="sourceLineNo">003</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.3"></a> +<span class="sourceLineNo">004</span> * or more contributor license agreements. See the NOTICE file<a name="line.4"></a> +<span class="sourceLineNo">005</span> * distributed with this work for additional information<a name="line.5"></a> +<span class="sourceLineNo">006</span> * regarding copyright ownership. The ASF licenses this file<a name="line.6"></a> +<span class="sourceLineNo">007</span> * to you under the Apache License, Version 2.0 (the<a name="line.7"></a> +<span class="sourceLineNo">008</span> * "License"); you may not use this file except in compliance<a name="line.8"></a> +<span class="sourceLineNo">009</span> * with the License. You may obtain a copy of the License at<a name="line.9"></a> +<span class="sourceLineNo">010</span> *<a name="line.10"></a> +<span class="sourceLineNo">011</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.11"></a> +<span class="sourceLineNo">012</span> *<a name="line.12"></a> +<span class="sourceLineNo">013</span> * Unless required by applicable law or agreed to in writing, software<a name="line.13"></a> +<span class="sourceLineNo">014</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.14"></a> +<span class="sourceLineNo">015</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.15"></a> +<span class="sourceLineNo">016</span> * See the License for the specific language governing permissions and<a name="line.16"></a> +<span class="sourceLineNo">017</span> * limitations under the License.<a name="line.17"></a> +<span class="sourceLineNo">018</span> */<a name="line.18"></a> +<span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.master;<a name="line.19"></a> +<span class="sourceLineNo">020</span><a name="line.20"></a> +<span class="sourceLineNo">021</span>import static org.apache.hadoop.hbase.util.CollectionUtils.computeIfAbsent;<a name="line.21"></a> +<span class="sourceLineNo">022</span><a name="line.22"></a> +<span class="sourceLineNo">023</span>import java.io.IOException;<a name="line.23"></a> +<span class="sourceLineNo">024</span>import java.net.InetAddress;<a name="line.24"></a> +<span class="sourceLineNo">025</span>import java.util.ArrayList;<a name="line.25"></a> +<span class="sourceLineNo">026</span>import java.util.Collections;<a name="line.26"></a> +<span class="sourceLineNo">027</span>import java.util.HashMap;<a name="line.27"></a> +<span class="sourceLineNo">028</span>import java.util.Iterator;<a name="line.28"></a> +<span class="sourceLineNo">029</span>import java.util.List;<a name="line.29"></a> +<span class="sourceLineNo">030</span>import java.util.Map;<a name="line.30"></a> +<span class="sourceLineNo">031</span>import java.util.Map.Entry;<a name="line.31"></a> +<span class="sourceLineNo">032</span>import java.util.Set;<a name="line.32"></a> +<span class="sourceLineNo">033</span>import java.util.concurrent.ConcurrentNavigableMap;<a name="line.33"></a> +<span class="sourceLineNo">034</span>import java.util.concurrent.ConcurrentSkipListMap;<a name="line.34"></a> +<span class="sourceLineNo">035</span>import java.util.concurrent.CopyOnWriteArrayList;<a name="line.35"></a> +<span class="sourceLineNo">036</span>import java.util.concurrent.atomic.AtomicBoolean;<a name="line.36"></a> +<span class="sourceLineNo">037</span>import java.util.function.Predicate;<a name="line.37"></a> +<span class="sourceLineNo">038</span>import org.apache.hadoop.conf.Configuration;<a name="line.38"></a> +<span class="sourceLineNo">039</span>import org.apache.hadoop.fs.FSDataInputStream;<a name="line.39"></a> +<span class="sourceLineNo">040</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.40"></a> +<span class="sourceLineNo">041</span>import org.apache.hadoop.fs.FileSystem;<a name="line.41"></a> +<span class="sourceLineNo">042</span>import org.apache.hadoop.fs.Path;<a name="line.42"></a> +<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.ClockOutOfSyncException;<a name="line.43"></a> +<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.HConstants;<a name="line.44"></a> +<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.NotServingRegionException;<a name="line.45"></a> +<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.RegionMetrics;<a name="line.46"></a> +<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.ScheduledChore;<a name="line.47"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.ServerMetrics;<a name="line.48"></a> +<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.ServerMetricsBuilder;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.ServerName;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.YouAreDeadException;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.client.ClusterConnection;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.client.RetriesExhaustedException;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.ipc.HBaseRpcController;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.ipc.RpcControllerFactory;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.master.assignment.RegionStates;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.monitoring.MonitoredTask;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.util.FSUtils;<a name="line.61"></a> +<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.62"></a> +<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.63"></a> +<span class="sourceLineNo">064</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.64"></a> +<span class="sourceLineNo">065</span>import org.apache.zookeeper.KeeperException;<a name="line.65"></a> +<span class="sourceLineNo">066</span>import org.slf4j.Logger;<a name="line.66"></a> +<span class="sourceLineNo">067</span>import org.slf4j.LoggerFactory;<a name="line.67"></a> +<span class="sourceLineNo">068</span><a name="line.68"></a> +<span class="sourceLineNo">069</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.69"></a> +<span class="sourceLineNo">070</span>import org.apache.hbase.thirdparty.com.google.protobuf.ByteString;<a name="line.70"></a> +<span class="sourceLineNo">071</span>import org.apache.hbase.thirdparty.com.google.protobuf.UnsafeByteOperations;<a name="line.71"></a> +<span class="sourceLineNo">072</span><a name="line.72"></a> +<span class="sourceLineNo">073</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.73"></a> +<span class="sourceLineNo">074</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.AdminService;<a name="line.74"></a> +<span class="sourceLineNo">075</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.RegionStoreSequenceIds;<a name="line.75"></a> +<span class="sourceLineNo">076</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.ClusterStatusProtos.StoreSequenceId;<a name="line.76"></a> +<span class="sourceLineNo">077</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.FlushedRegionSequenceId;<a name="line.77"></a> +<span class="sourceLineNo">078</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.FlushedSequenceId;<a name="line.78"></a> +<span class="sourceLineNo">079</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.FlushedStoreSequenceId;<a name="line.79"></a> +<span class="sourceLineNo">080</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionServerStartupRequest;<a name="line.80"></a> +<span class="sourceLineNo">081</span><a name="line.81"></a> +<span class="sourceLineNo">082</span>/**<a name="line.82"></a> +<span class="sourceLineNo">083</span> * The ServerManager class manages info about region servers.<a name="line.83"></a> +<span class="sourceLineNo">084</span> * <p><a name="line.84"></a> +<span class="sourceLineNo">085</span> * Maintains lists of online and dead servers. Processes the startups,<a name="line.85"></a> +<span class="sourceLineNo">086</span> * shutdowns, and deaths of region servers.<a name="line.86"></a> +<span class="sourceLineNo">087</span> * <p><a name="line.87"></a> +<span class="sourceLineNo">088</span> * Servers are distinguished in two different ways. A given server has a<a name="line.88"></a> +<span class="sourceLineNo">089</span> * location, specified by hostname and port, and of which there can only be one<a name="line.89"></a> +<span class="sourceLineNo">090</span> * online at any given time. A server instance is specified by the location<a name="line.90"></a> +<span class="sourceLineNo">091</span> * (hostname and port) as well as the startcode (timestamp from when the server<a name="line.91"></a> +<span class="sourceLineNo">092</span> * was started). This is used to differentiate a restarted instance of a given<a name="line.92"></a> +<span class="sourceLineNo">093</span> * server from the original instance.<a name="line.93"></a> +<span class="sourceLineNo">094</span> * <p><a name="line.94"></a> +<span class="sourceLineNo">095</span> * If a sever is known not to be running any more, it is called dead. The dead<a name="line.95"></a> +<span class="sourceLineNo">096</span> * server needs to be handled by a ServerShutdownHandler. If the handler is not<a name="line.96"></a> +<span class="sourceLineNo">097</span> * enabled yet, the server can't be handled right away so it is queued up.<a name="line.97"></a> +<span class="sourceLineNo">098</span> * After the handler is enabled, the server will be submitted to a handler to handle.<a name="line.98"></a> +<span class="sourceLineNo">099</span> * However, the handler may be just partially enabled. If so,<a name="line.99"></a> +<span class="sourceLineNo">100</span> * the server cannot be fully processed, and be queued up for further processing.<a name="line.100"></a> +<span class="sourceLineNo">101</span> * A server is fully processed only after the handler is fully enabled<a name="line.101"></a> +<span class="sourceLineNo">102</span> * and has completed the handling.<a name="line.102"></a> +<span class="sourceLineNo">103</span> */<a name="line.103"></a> +<span class="sourceLineNo">104</span>@InterfaceAudience.Private<a name="line.104"></a> +<span class="sourceLineNo">105</span>public class ServerManager {<a name="line.105"></a> +<span class="sourceLineNo">106</span> public static final String WAIT_ON_REGIONSERVERS_MAXTOSTART =<a name="line.106"></a> +<span class="sourceLineNo">107</span> "hbase.master.wait.on.regionservers.maxtostart";<a name="line.107"></a> +<span class="sourceLineNo">108</span><a name="line.108"></a> +<span class="sourceLineNo">109</span> public static final String WAIT_ON_REGIONSERVERS_MINTOSTART =<a name="line.109"></a> +<span class="sourceLineNo">110</span> "hbase.master.wait.on.regionservers.mintostart";<a name="line.110"></a> +<span class="sourceLineNo">111</span><a name="line.111"></a> +<span class="sourceLineNo">112</span> public static final String WAIT_ON_REGIONSERVERS_TIMEOUT =<a name="line.112"></a> +<span class="sourceLineNo">113</span> "hbase.master.wait.on.regionservers.timeout";<a name="line.113"></a> +<span class="sourceLineNo">114</span><a name="line.114"></a> +<span class="sourceLineNo">115</span> public static final String WAIT_ON_REGIONSERVERS_INTERVAL =<a name="line.115"></a> +<span class="sourceLineNo">116</span> "hbase.master.wait.on.regionservers.interval";<a name="line.116"></a> +<span class="sourceLineNo">117</span><a name="line.117"></a> +<span class="sourceLineNo">118</span> /**<a name="line.118"></a> +<span class="sourceLineNo">119</span> * see HBASE-20727<a name="line.119"></a> +<span class="sourceLineNo">120</span> * if set to true, flushedSequenceIdByRegion and storeFlushedSequenceIdsByRegion<a name="line.120"></a> +<span class="sourceLineNo">121</span> * will be persisted to HDFS and loaded when master restart to speed up log split<a name="line.121"></a> +<span class="sourceLineNo">122</span> */<a name="line.122"></a> +<span class="sourceLineNo">123</span> public static final String PERSIST_FLUSHEDSEQUENCEID =<a name="line.123"></a> +<span class="sourceLineNo">124</span> "hbase.master.persist.flushedsequenceid.enabled";<a name="line.124"></a> +<span class="sourceLineNo">125</span><a name="line.125"></a> +<span class="sourceLineNo">126</span> public static final boolean PERSIST_FLUSHEDSEQUENCEID_DEFAULT = true;<a name="line.126"></a> +<span class="sourceLineNo">127</span><a name="line.127"></a> +<span class="sourceLineNo">128</span> public static final String FLUSHEDSEQUENCEID_FLUSHER_INTERVAL =<a name="line.128"></a> +<span class="sourceLineNo">129</span> "hbase.master.flushedsequenceid.flusher.interval";<a name="line.129"></a> +<span class="sourceLineNo">130</span><a name="line.130"></a> +<span class="sourceLineNo">131</span> public static final int FLUSHEDSEQUENCEID_FLUSHER_INTERVAL_DEFAULT =<a name="line.131"></a> +<span class="sourceLineNo">132</span> 3 * 60 * 60 * 1000; // 3 hours<a name="line.132"></a> +<span class="sourceLineNo">133</span><a name="line.133"></a> +<span class="sourceLineNo">134</span> private static final Logger LOG = LoggerFactory.getLogger(ServerManager.class);<a name="line.134"></a> +<span class="sourceLineNo">135</span><a name="line.135"></a> +<span class="sourceLineNo">136</span> // Set if we are to shutdown the cluster.<a name="line.136"></a> +<span class="sourceLineNo">137</span> private AtomicBoolean clusterShutdown = new AtomicBoolean(false);<a name="line.137"></a> +<span class="sourceLineNo">138</span><a name="line.138"></a> +<span class="sourceLineNo">139</span> /**<a name="line.139"></a> +<span class="sourceLineNo">140</span> * The last flushed sequence id for a region.<a name="line.140"></a> +<span class="sourceLineNo">141</span> */<a name="line.141"></a> +<span class="sourceLineNo">142</span> private final ConcurrentNavigableMap<byte[], Long> flushedSequenceIdByRegion =<a name="line.142"></a> +<span class="sourceLineNo">143</span> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.143"></a> +<span class="sourceLineNo">144</span><a name="line.144"></a> +<span class="sourceLineNo">145</span> private boolean persistFlushedSequenceId = true;<a name="line.145"></a> +<span class="sourceLineNo">146</span> private volatile boolean isFlushSeqIdPersistInProgress = false;<a name="line.146"></a> +<span class="sourceLineNo">147</span> /** File on hdfs to store last flushed sequence id of regions */<a name="line.147"></a> +<span class="sourceLineNo">148</span> private static final String LAST_FLUSHED_SEQ_ID_FILE = ".lastflushedseqids";<a name="line.148"></a> +<span class="sourceLineNo">149</span> private FlushedSequenceIdFlusher flushedSeqIdFlusher;<a name="line.149"></a> +<span class="sourceLineNo">150</span><a name="line.150"></a> +<span class="sourceLineNo">151</span><a name="line.151"></a> +<span class="sourceLineNo">152</span> /**<a name="line.152"></a> +<span class="sourceLineNo">153</span> * The last flushed sequence id for a store in a region.<a name="line.153"></a> +<span class="sourceLineNo">154</span> */<a name="line.154"></a> +<span class="sourceLineNo">155</span> private final ConcurrentNavigableMap<byte[], ConcurrentNavigableMap<byte[], Long>><a name="line.155"></a> +<span class="sourceLineNo">156</span> storeFlushedSequenceIdsByRegion = new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR);<a name="line.156"></a> +<span class="sourceLineNo">157</span><a name="line.157"></a> +<span class="sourceLineNo">158</span> /** Map of registered servers to their current load */<a name="line.158"></a> +<span class="sourceLineNo">159</span> private final ConcurrentNavigableMap<ServerName, ServerMetrics> onlineServers =<a name="line.159"></a> +<span class="sourceLineNo">160</span> new ConcurrentSkipListMap<>();<a name="line.160"></a> +<span class="sourceLineNo">161</span><a name="line.161"></a> +<span class="sourceLineNo">162</span> /**<a name="line.162"></a> +<span class="sourceLineNo">163</span> * Map of admin interfaces per registered regionserver; these interfaces we use to control<a name="line.163"></a> +<span class="sourceLineNo">164</span> * regionservers out on the cluster<a name="line.164"></a> +<span class="sourceLineNo">165</span> */<a name="line.165"></a> +<span class="sourceLineNo">166</span> private final Map<ServerName, AdminService.BlockingInterface> rsAdmins = new HashMap<>();<a name="line.166"></a> +<span class="sourceLineNo">167</span><a name="line.167"></a> +<span class="sourceLineNo">168</span> /** List of region servers that should not get any more new regions. */<a name="line.168"></a> +<span class="sourceLineNo">169</span> private final ArrayList<ServerName> drainingServers = new ArrayList<>();<a name="line.169"></a> +<span class="sourceLineNo">170</span><a name="line.170"></a> +<span class="sourceLineNo">171</span> private final MasterServices master;<a name="line.171"></a> +<span class="sourceLineNo">172</span> private final ClusterConnection connection;<a name="line.172"></a> +<span class="sourceLineNo">173</span><a name="line.173"></a> +<span class="sourceLineNo">174</span> private final DeadServer deadservers = new DeadServer();<a name="line.174"></a> +<span class="sourceLineNo">175</span><a name="line.175"></a> +<span class="sourceLineNo">176</span> private final long maxSkew;<a name="line.176"></a> +<span class="sourceLineNo">177</span> private final long warningSkew;<a name="line.177"></a> +<span class="sourceLineNo">178</span><a name="line.178"></a> +<span class="sourceLineNo">179</span> private final RpcControllerFactory rpcControllerFactory;<a name="line.179"></a> +<span class="sourceLineNo">180</span><a name="line.180"></a> +<span class="sourceLineNo">181</span> /** Listeners that are called on server events. */<a name="line.181"></a> +<span class="sourceLineNo">182</span> private List<ServerListener> listeners = new CopyOnWriteArrayList<>();<a name="line.182"></a> +<span class="sourceLineNo">183</span><a name="line.183"></a> +<span class="sourceLineNo">184</span> /**<a name="line.184"></a> +<span class="sourceLineNo">185</span> * Constructor.<a name="line.185"></a> +<span class="sourceLineNo">186</span> */<a name="line.186"></a> +<span class="sourceLineNo">187</span> public ServerManager(final MasterServices master) {<a name="line.187"></a> +<span class="sourceLineNo">188</span> this.master = master;<a name="line.188"></a> +<span class="sourceLineNo">189</span> Configuration c = master.getConfiguration();<a name="line.189"></a> +<span class="sourceLineNo">190</span> maxSkew = c.getLong("hbase.master.maxclockskew", 30000);<a name="line.190"></a> +<span class="sourceLineNo">191</span> warningSkew = c.getLong("hbase.master.warningclockskew", 10000);<a name="line.191"></a> +<span class="sourceLineNo">192</span> this.connection = master.getClusterConnection();<a name="line.192"></a> +<span class="sourceLineNo">193</span> this.rpcControllerFactory = this.connection == null? null: connection.getRpcControllerFactory();<a name="line.193"></a> +<span class="sourceLineNo">194</span> persistFlushedSequenceId = c.getBoolean(PERSIST_FLUSHEDSEQUENCEID,<a name="line.194"></a> +<span class="sourceLineNo">195</span> PERSIST_FLUSHEDSEQUENCEID_DEFAULT);<a name="line.195"></a> +<span class="sourceLineNo">196</span> }<a name="line.196"></a> +<span class="sourceLineNo">197</span><a name="line.197"></a> +<span class="sourceLineNo">198</span> /**<a name="line.198"></a> +<span class="sourceLineNo">199</span> * Add the listener to the notification list.<a name="line.199"></a> +<span class="sourceLineNo">200</span> * @param listener The ServerListener to register<a name="line.200"></a> +<span class="sourceLineNo">201</span> */<a name="line.201"></a> +<span class="sourceLineNo">202</span> public void registerListener(final ServerListener listener) {<a name="line.202"></a> +<span class="sourceLineNo">203</span> this.listeners.add(listener);<a name="line.203"></a> +<span class="sourceLineNo">204</span> }<a name="line.204"></a> +<span class="sourceLineNo">205</span><a name="line.205"></a> +<span class="sourceLineNo">206</span> /**<a name="line.206"></a> +<span class="sourceLineNo">207</span> * Remove the listener from the notification list.<a name="line.207"></a> +<span class="sourceLineNo">208</span> * @param listener The ServerListener to unregister<a name="line.208"></a> +<span class="sourceLineNo">209</span> */<a name="line.209"></a> +<span class="sourceLineNo">210</span> public boolean unregisterListener(final ServerListener listener) {<a name="line.210"></a> +<span class="sourceLineNo">211</span> return this.listeners.remove(listener);<a name="line.211"></a> +<span class="sourceLineNo">212</span> }<a name="line.212"></a> +<span class="sourceLineNo">213</span><a name="line.213"></a> +<span class="sourceLineNo">214</span> /**<a name="line.214"></a> +<span class="sourceLineNo">215</span> * Let the server manager know a new regionserver has come online<a name="line.215"></a> +<span class="sourceLineNo">216</span> * @param request the startup request<a name="line.216"></a> +<span class="sourceLineNo">217</span> * @param versionNumber the version of the new regionserver<a name="line.217"></a> +<span class="sourceLineNo">218</span> * @param ia the InetAddress from which request is received<a name="line.218"></a> +<span class="sourceLineNo">219</span> * @return The ServerName we know this server as.<a name="line.219"></a> +<span class="sourceLineNo">220</span> * @throws IOException<a name="line.220"></a> +<span class="sourceLineNo">221</span> */<a name="line.221"></a> +<span class="sourceLineNo">222</span> ServerName regionServerStartup(RegionServerStartupRequest request, int versionNumber,<a name="line.222"></a> +<span class="sourceLineNo">223</span> InetAddress ia) throws IOException {<a name="line.223"></a> +<span class="sourceLineNo">224</span> // Test for case where we get a region startup message from a regionserver<a name="line.224"></a> +<span class="sourceLineNo">225</span> // that has been quickly restarted but whose znode expiration handler has<a name="line.225"></a> +<span class="sourceLineNo">226</span> // not yet run, or from a server whose fail we are currently processing.<a name="line.226"></a> +<span class="sourceLineNo">227</span> // Test its host+port combo is present in serverAddressToServerInfo. If it<a name="line.227"></a> +<span class="sourceLineNo">228</span> // is, reject the server and trigger its expiration. The next time it comes<a name="line.228"></a> +<span class="sourceLineNo">229</span> // in, it should have been removed from serverAddressToServerInfo and queued<a name="line.229"></a> +<span class="sourceLineNo">230</span> // for processing by ProcessServerShutdown.<a name="line.230"></a> +<span class="sourceLineNo">231</span><a name="line.231"></a> +<span class="sourceLineNo">232</span> final String hostname = request.hasUseThisHostnameInstead() ?<a name="line.232"></a> +<span class="sourceLineNo">233</span> request.getUseThisHostnameInstead() :ia.getHostName();<a name="line.233"></a> +<span class="sourceLineNo">234</span> ServerName sn = ServerName.valueOf(hostname, request.getPort(),<a name="line.234"></a> +<span class="sourceLineNo">235</span> request.getServerStartCode());<a name="line.235"></a> +<span class="sourceLineNo">236</span> checkClockSkew(sn, request.getServerCurrentTime());<a name="line.236"></a> +<span class="sourceLineNo">237</span> checkIsDead(sn, "STARTUP");<a name="line.237"></a> +<span class="sourceLineNo">238</span> if (!checkAndRecordNewServer(sn, ServerMetricsBuilder.of(sn, versionNumber))) {<a name="line.238"></a> +<span class="sourceLineNo">239</span> LOG.warn("THIS SHOULD NOT HAPPEN, RegionServerStartup"<a name="line.239"></a> +<span class="sourceLineNo">240</span> + " could not record the server: " + sn);<a name="line.240"></a> +<span class="sourceLineNo">241</span> }<a name="line.241"></a> +<span class="sourceLineNo">242</span> return sn;<a name="line.242"></a> +<span class="sourceLineNo">243</span> }<a name="line.243"></a> +<span class="sourceLineNo">244</span><a name="line.244"></a> +<span class="sourceLineNo">245</span> /**<a name="line.245"></a> +<span class="sourceLineNo">246</span> * Updates last flushed sequence Ids for the regions on server sn<a name="line.246"></a> +<span class="sourceLineNo">247</span> * @param sn<a name="line.247"></a> +<span class="sourceLineNo">248</span> * @param hsl<a name="line.248"></a> +<span class="sourceLineNo">249</span> */<a name="line.249"></a> +<span class="sourceLineNo">250</span> private void updateLastFlushedSequenceIds(ServerName sn, ServerMetrics hsl) {<a name="line.250"></a> +<span class="sourceLineNo">251</span> for (Entry<byte[], RegionMetrics> entry : hsl.getRegionMetrics().entrySet()) {<a name="line.251"></a> +<span class="sourceLineNo">252</span> byte[] encodedRegionName = Bytes.toBytes(RegionInfo.encodeRegionName(entry.getKey()));<a name="line.252"></a> +<span class="sourceLineNo">253</span> Long existingValue = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.253"></a> +<span class="sourceLineNo">254</span> long l = entry.getValue().getCompletedSequenceId();<a name="line.254"></a> +<span class="sourceLineNo">255</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.255"></a> +<span class="sourceLineNo">256</span> if (LOG.isTraceEnabled()) {<a name="line.256"></a> +<span class="sourceLineNo">257</span> LOG.trace(Bytes.toString(encodedRegionName) + ", existingValue=" + existingValue +<a name="line.257"></a> +<span class="sourceLineNo">258</span> ", completeSequenceId=" + l);<a name="line.258"></a> +<span class="sourceLineNo">259</span> }<a name="line.259"></a> +<span class="sourceLineNo">260</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue)) {<a name="line.260"></a> +<span class="sourceLineNo">261</span> flushedSequenceIdByRegion.put(encodedRegionName, l);<a name="line.261"></a> +<span class="sourceLineNo">262</span> } else if (l != HConstants.NO_SEQNUM && l < existingValue) {<a name="line.262"></a> +<span class="sourceLineNo">263</span> LOG.warn("RegionServer " + sn + " indicates a last flushed sequence id ("<a name="line.263"></a> +<span class="sourceLineNo">264</span> + l + ") that is less than the previous last flushed sequence id ("<a name="line.264"></a> +<span class="sourceLineNo">265</span> + existingValue + ") for region " + Bytes.toString(entry.getKey()) + " Ignoring.");<a name="line.265"></a> +<span class="sourceLineNo">266</span> }<a name="line.266"></a> +<span class="sourceLineNo">267</span> ConcurrentNavigableMap<byte[], Long> storeFlushedSequenceId =<a name="line.267"></a> +<span class="sourceLineNo">268</span> computeIfAbsent(storeFlushedSequenceIdsByRegion, encodedRegionName,<a name="line.268"></a> +<span class="sourceLineNo">269</span> () -> new ConcurrentSkipListMap<>(Bytes.BYTES_COMPARATOR));<a name="line.269"></a> +<span class="sourceLineNo">270</span> for (Entry<byte[], Long> storeSeqId : entry.getValue().getStoreSequenceId().entrySet()) {<a name="line.270"></a> +<span class="sourceLineNo">271</span> byte[] family = storeSeqId.getKey();<a name="line.271"></a> +<span class="sourceLineNo">272</span> existingValue = storeFlushedSequenceId.get(family);<a name="line.272"></a> +<span class="sourceLineNo">273</span> l = storeSeqId.getValue();<a name="line.273"></a> +<span class="sourceLineNo">274</span> if (LOG.isTraceEnabled()) {<a name="line.274"></a> +<span class="sourceLineNo">275</span> LOG.trace(Bytes.toString(encodedRegionName) + ", family=" + Bytes.toString(family) +<a name="line.275"></a> +<span class="sourceLineNo">276</span> ", existingValue=" + existingValue + ", completeSequenceId=" + l);<a name="line.276"></a> +<span class="sourceLineNo">277</span> }<a name="line.277"></a> +<span class="sourceLineNo">278</span> // Don't let smaller sequence ids override greater sequence ids.<a name="line.278"></a> +<span class="sourceLineNo">279</span> if (existingValue == null || (l != HConstants.NO_SEQNUM && l > existingValue.longValue())) {<a name="line.279"></a> +<span class="sourceLineNo">280</span> storeFlushedSequenceId.put(family, l);<a name="line.280"></a> +<span class="sourceLineNo">281</span> }<a name="line.281"></a> +<span class="sourceLineNo">282</span> }<a name="line.282"></a> +<span class="sourceLineNo">283</span> }<a name="line.283"></a> +<span class="sourceLineNo">284</span> }<a name="line.284"></a> +<span class="sourceLineNo">285</span><a name="line.285"></a> +<span class="sourceLineNo">286</span> @VisibleForTesting<a name="line.286"></a> +<span class="sourceLineNo">287</span> public void regionServerReport(ServerName sn,<a name="line.287"></a> +<span class="sourceLineNo">288</span> ServerMetrics sl) throws YouAreDeadException {<a name="line.288"></a> +<span class="sourceLineNo">289</span> checkIsDead(sn, "REPORT");<a name="line.289"></a> +<span class="sourceLineNo">290</span> if (null == this.onlineServers.replace(sn, sl)) {<a name="line.290"></a> +<span class="sourceLineNo">291</span> // Already have this host+port combo and its just different start code?<a name="line.291"></a> +<span class="sourceLineNo">292</span> // Just let the server in. Presume master joining a running cluster.<a name="line.292"></a> +<span class="sourceLineNo">293</span> // recordNewServer is what happens at the end of reportServerStartup.<a name="line.293"></a> +<span class="sourceLineNo">294</span> // The only thing we are skipping is passing back to the regionserver<a name="line.294"></a> +<span class="sourceLineNo">295</span> // the ServerName to use. Here we presume a master has already done<a name="line.295"></a> +<span class="sourceLineNo">296</span> // that so we'll press on with whatever it gave us for ServerName.<a name="line.296"></a> +<span class="sourceLineNo">297</span> if (!checkAndRecordNewServer(sn, sl)) {<a name="line.297"></a> +<span class="sourceLineNo">298</span> LOG.info("RegionServerReport ignored, could not record the server: " + sn);<a name="line.298"></a> +<span class="sourceLineNo">299</span> return; // Not recorded, so no need to move on<a name="line.299"></a> +<span class="sourceLineNo">300</span> }<a name="line.300"></a> +<span class="sourceLineNo">301</span> }<a name="line.301"></a> +<span class="sourceLineNo">302</span> updateLastFlushedSequenceIds(sn, sl);<a name="line.302"></a> +<span class="sourceLineNo">303</span> }<a name="line.303"></a> +<span class="sourceLineNo">304</span><a name="line.304"></a> +<span class="sourceLineNo">305</span> /**<a name="line.305"></a> +<span class="sourceLineNo">306</span> * Check is a server of same host and port already exists,<a name="line.306"></a> +<span class="sourceLineNo">307</span> * if not, or the existed one got a smaller start code, record it.<a name="line.307"></a> +<span class="sourceLineNo">308</span> *<a name="line.308"></a> +<span class="sourceLineNo">309</span> * @param serverName the server to check and record<a name="line.309"></a> +<span class="sourceLineNo">310</span> * @param sl the server load on the server<a name="line.310"></a> +<span class="sourceLineNo">311</span> * @return true if the server is recorded, otherwise, false<a name="line.311"></a> +<span class="sourceLineNo">312</span> */<a name="line.312"></a> +<span class="sourceLineNo">313</span> boolean checkAndRecordNewServer(final ServerName serverName, final ServerMetrics sl) {<a name="line.313"></a> +<span class="sourceLineNo">314</span> ServerName existingServer = null;<a name="line.314"></a> +<span class="sourceLineNo">315</span> synchronized (this.onlineServers) {<a name="line.315"></a> +<span class="sourceLineNo">316</span> existingServer = findServerWithSameHostnamePortWithLock(serverName);<a name="line.316"></a> +<span class="sourceLineNo">317</span> if (existingServer != null && (existingServer.getStartcode() > serverName.getStartcode())) {<a name="line.317"></a> +<span class="sourceLineNo">318</span> LOG.info("Server serverName=" + serverName + " rejected; we already have "<a name="line.318"></a> +<span class="sourceLineNo">319</span> + existingServer.toString() + " registered with same hostname and port");<a name="line.319"></a> +<span class="sourceLineNo">320</span> return false;<a name="line.320"></a> +<span class="sourceLineNo">321</span> }<a name="line.321"></a> +<span class="sourceLineNo">322</span> recordNewServerWithLock(serverName, sl);<a name="line.322"></a> +<span class="sourceLineNo">323</span> }<a name="line.323"></a> +<span class="sourceLineNo">324</span><a name="line.324"></a> +<span class="sourceLineNo">325</span> // Tell our listeners that a server was added<a name="line.325"></a> +<span class="sourceLineNo">326</span> if (!this.listeners.isEmpty()) {<a name="line.326"></a> +<span class="sourceLineNo">327</span> for (ServerListener listener : this.listeners) {<a name="line.327"></a> +<span class="sourceLineNo">328</span> listener.serverAdded(serverName);<a name="line.328"></a> +<span class="sourceLineNo">329</span> }<a name="line.329"></a> +<span class="sourceLineNo">330</span> }<a name="line.330"></a> +<span class="sourceLineNo">331</span><a name="line.331"></a> +<span class="sourceLineNo">332</span> // Note that we assume that same ts means same server, and don't expire in that case.<a name="line.332"></a> +<span class="sourceLineNo">333</span> // TODO: ts can theoretically collide due to clock shifts, so this is a bit hacky.<a name="line.333"></a> +<span class="sourceLineNo">334</span> if (existingServer != null &&<a name="line.334"></a> +<span class="sourceLineNo">335</span> (existingServer.getStartcode() < serverName.getStartcode())) {<a name="line.335"></a> +<span class="sourceLineNo">336</span> LOG.info("Triggering server recovery; existingServer " +<a name="line.336"></a> +<span class="sourceLineNo">337</span> existingServer + " looks stale, new server:" + serverName);<a name="line.337"></a> +<span class="sourceLineNo">338</span> expireServer(existingServer);<a name="line.338"></a> +<span class="sourceLineNo">339</span> }<a name="line.339"></a> +<span class="sourceLineNo">340</span> return true;<a name="line.340"></a> +<span class="sourceLineNo">341</span> }<a name="line.341"></a> +<span class="sourceLineNo">342</span><a name="line.342"></a> +<span class="sourceLineNo">343</span> /**<a name="line.343"></a> +<span class="sourceLineNo">344</span> * Find out the region servers crashed between the crash of the previous master instance and the<a name="line.344"></a> +<span class="sourceLineNo">345</span> * current master instance and schedule SCP for them.<a name="line.345"></a> +<span class="sourceLineNo">346</span> * <p/><a name="line.346"></a> +<span class="sourceLineNo">347</span> * Since the {@code RegionServerTracker} has already helped us to construct the online servers set<a name="line.347"></a> +<span class="sourceLineNo">348</span> * by scanning zookeeper, now we can compare the online servers with {@code liveServersFromWALDir}<a name="line.348"></a> +<span class="sourceLineNo">349</span> * to find out whether there are servers which are already dead.<a name="line.349"></a> +<span class="sourceLineNo">350</span> * <p/><a name="line.350"></a> +<span class="sourceLineNo">351</span> * Must be called inside the initialization method of {@code RegionServerTracker} to avoid<a name="line.351"></a> +<span class="sourceLineNo">352</span> * concurrency issue.<a name="line.352"></a> +<span class="sourceLineNo">353</span> * @param deadServersFromPE the region servers which already have SCP associated.<a name="line.353"></a> +<span class="sourceLineNo">354</span> * @param liveServersFromWALDir the live region servers from wal directory.<a name="line.354"></a> +<span class="sourceLineNo">355</span> */<a name="line.355"></a> +<span class="sourceLineNo">356</span> void findOutDeadServersAndProcess(Set<ServerName> deadServersFromPE,<a name="line.356"></a> +<span class="sourceLineNo">357</span> Set<ServerName> liveServersFromWALDir) {<a name="line.357"></a> +<span class="sourceLineNo">358</span> deadServersFromPE.forEach(deadservers::add);<a name="line.358"></a> +<span class="sourceLineNo">359</span> liveServersFromWALDir.stream().filter(sn -> !onlineServers.containsKey(sn))<a name="line.359"></a> +<span class="sourceLineNo">360</span> .forEach(this::expireServer);<a name="line.360"></a> +<span class="sourceLineNo">361</span> }<a name="line.361"></a> +<span class="sourceLineNo">362</span><a name="line.362"></a> +<span class="sourceLineNo">363</span> /**<a name="line.363"></a> +<span class="sourceLineNo">364</span> * Checks if the clock skew between the server and the master. If the clock skew exceeds the<a name="line.364"></a> +<span class="sourceLineNo">365</span> * configured max, it will throw an exception; if it exceeds the configured warning threshold,<a name="line.365"></a> +<span class="sourceLineNo">366</span> * it will log a warning but start normally.<a name="line.366"></a> +<span class="sourceLineNo">367</span> * @param serverName Incoming servers's name<a name="line.367"></a> +<span class="sourceLineNo">368</span> * @param serverCurrentTime<a name="line.368"></a> +<span class="sourceLineNo">369</span> * @throws ClockOutOfSyncException if the skew exceeds the configured max value<a name="line.369"></a> +<span class="sourceLineNo">370</span> */<a name="line.370"></a> +<span class="sourceLineNo">371</span> private void checkClockSkew(final ServerName serverName, final long serverCurrentTime)<a name="line.371"></a> +<span class="sourceLineNo">372</span> throws ClockOutOfSyncException {<a name="line.372"></a> +<span class="sourceLineNo">373</span> long skew = Math.abs(System.currentTimeMillis() - serverCurrentTime);<a name="line.373"></a> +<span class="sourceLineNo">374</span> if (skew > maxSkew) {<a name="line.374"></a> +<span class="sourceLineNo">375</span> String message = "Server " + serverName + " has been " +<a name="line.375"></a> +<span class="sourceLineNo">376</span> "rejected; Reported time is too far out of sync with master. " +<a name="line.376"></a> +<span class="sourceLineNo">377</span> "Time difference of " + skew + "ms > max allowed of " + maxSkew + "ms";<a name="line.377"></a> +<span class="sourceLineNo">378</span> LOG.warn(message);<a name="line.378"></a> +<span class="sourceLineNo">379</span> throw new ClockOutOfSyncException(message);<a name="line.379"></a> +<span class="sourceLineNo">380</span> } else if (skew > warningSkew){<a name="line.380"></a> +<span class="sourceLineNo">381</span> String message = "Reported time for server " + serverName + " is out of sync with master " +<a name="line.381"></a> +<span class="sourceLineNo">382</span> "by " + skew + "ms. (Warning threshold is " + warningSkew + "ms; " +<a name="line.382"></a> +<span class="sourceLineNo">383</span> "error threshold is " + maxSkew + "ms)";<a name="line.383"></a> +<span class="sourceLineNo">384</span> LOG.warn(message);<a name="line.384"></a> +<span class="sourceLineNo">385</span> }<a name="line.385"></a> +<span class="sourceLineNo">386</span> }<a name="line.386"></a> +<span class="sourceLineNo">387</span><a name="line.387"></a> +<span class="sourceLineNo">388</span> /**<a name="line.388"></a> +<span class="sourceLineNo">389</span> * If this server is on the dead list, reject it with a YouAreDeadException.<a name="line.389"></a> +<span class="sourceLineNo">390</span> * If it was dead but came back with a new start code, remove the old entry<a name="line.390"></a> +<span class="sourceLineNo">391</span> * from the dead list.<a name="line.391"></a> +<span class="sourceLineNo">392</span> * @param what START or REPORT<a name="line.392"></a> +<span class="sourceLineNo">393</span> */<a name="line.393"></a> +<span class="sourceLineNo">394</span> private void checkIsDead(final ServerName serverName, final String what)<a name="line.394"></a> +<span class="sourceLineNo">395</span> throws YouAreDeadException {<a name="line.395"></a> +<span class="sourceLineNo">396</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.396"></a> +<span class="sourceLineNo">397</span> // host name, port and start code all match with existing one of the<a name="line.397"></a> +<span class="sourceLineNo">398</span> // dead servers. So, this server must be dead.<a name="line.398"></a> +<span class="sourceLineNo">399</span> String message = "Server " + what + " rejected; currently processing " +<a name="line.399"></a> +<span class="sourceLineNo">400</span> serverName + " as dead server";<a name="line.400"></a> +<span class="sourceLineNo">401</span> LOG.debug(message);<a name="line.401"></a> +<span class="sourceLineNo">402</span> throw new YouAreDeadException(message);<a name="line.402"></a> +<span class="sourceLineNo">403</span> }<a name="line.403"></a> +<span class="sourceLineNo">404</span> // remove dead server with same hostname and port of newly checking in rs after master<a name="line.404"></a> +<span class="sourceLineNo">405</span> // initialization.See HBASE-5916 for more information.<a name="line.405"></a> +<span class="sourceLineNo">406</span> if ((this.master == null || this.master.isInitialized())<a name="line.406"></a> +<span class="sourceLineNo">407</span> && this.deadservers.cleanPreviousInstance(serverName)) {<a name="line.407"></a> +<span class="sourceLineNo">408</span> // This server has now become alive after we marked it as dead.<a name="line.408"></a> +<span class="sourceLineNo">409</span> // We removed it's previous entry from the dead list to reflect it.<a name="line.409"></a> +<span class="sourceLineNo">410</span> LOG.debug(what + ":" + " Server " + serverName + " came back up," +<a name="line.410"></a> +<span class="sourceLineNo">411</span> " removed it from the dead servers list");<a name="line.411"></a> +<span class="sourceLineNo">412</span> }<a name="line.412"></a> +<span class="sourceLineNo">413</span> }<a name="line.413"></a> +<span class="sourceLineNo">414</span><a name="line.414"></a> +<span class="sourceLineNo">415</span> /**<a name="line.415"></a> +<span class="sourceLineNo">416</span> * Assumes onlineServers is locked.<a name="line.416"></a> +<span class="sourceLineNo">417</span> * @return ServerName with matching hostname and port.<a name="line.417"></a> +<span class="sourceLineNo">418</span> */<a name="line.418"></a> +<span class="sourceLineNo">419</span> private ServerName findServerWithSameHostnamePortWithLock(<a name="line.419"></a> +<span class="sourceLineNo">420</span> final ServerName serverName) {<a name="line.420"></a> +<span class="sourceLineNo">421</span> ServerName end = ServerName.valueOf(serverName.getHostname(), serverName.getPort(),<a name="line.421"></a> +<span class="sourceLineNo">422</span> Long.MAX_VALUE);<a name="line.422"></a> +<span class="sourceLineNo">423</span><a name="line.423"></a> +<span class="sourceLineNo">424</span> ServerName r = onlineServers.lowerKey(end);<a name="line.424"></a> +<span class="sourceLineNo">425</span> if (r != null) {<a name="line.425"></a> +<span class="sourceLineNo">426</span> if (ServerName.isSameAddress(r, serverName)) {<a name="line.426"></a> +<span class="sourceLineNo">427</span> return r;<a name="line.427"></a> +<span class="sourceLineNo">428</span> }<a name="line.428"></a> +<span class="sourceLineNo">429</span> }<a name="line.429"></a> +<span class="sourceLineNo">430</span> return null;<a name="line.430"></a> +<span class="sourceLineNo">431</span> }<a name="line.431"></a> +<span class="sourceLineNo">432</span><a name="line.432"></a> +<span class="sourceLineNo">433</span> /**<a name="line.433"></a> +<span class="sourceLineNo">434</span> * Adds the onlineServers list. onlineServers should be locked.<a name="line.434"></a> +<span class="sourceLineNo">435</span> * @param serverName The remote servers name.<a name="line.435"></a> +<span class="sourceLineNo">436</span> */<a name="line.436"></a> +<span class="sourceLineNo">437</span> @VisibleForTesting<a name="line.437"></a> +<span class="sourceLineNo">438</span> void recordNewServerWithLock(final ServerName serverName, final ServerMetrics sl) {<a name="line.438"></a> +<span class="sourceLineNo">439</span> LOG.info("Registering regionserver=" + serverName);<a name="line.439"></a> +<span class="sourceLineNo">440</span> this.onlineServers.put(serverName, sl);<a name="line.440"></a> +<span class="sourceLineNo">441</span> this.rsAdmins.remove(serverName);<a name="line.441"></a> +<span class="sourceLineNo">442</span> }<a name="line.442"></a> +<span class="sourceLineNo">443</span><a name="line.443"></a> +<span class="sourceLineNo">444</span> @VisibleForTesting<a name="line.444"></a> +<span class="sourceLineNo">445</span> public ConcurrentNavigableMap<byte[], Long> getFlushedSequenceIdByRegion() {<a name="line.445"></a> +<span class="sourceLineNo">446</span> return flushedSequenceIdByRegion;<a name="line.446"></a> +<span class="sourceLineNo">447</span> }<a name="line.447"></a> +<span class="sourceLineNo">448</span><a name="line.448"></a> +<span class="sourceLineNo">449</span> public RegionStoreSequenceIds getLastFlushedSequenceId(byte[] encodedRegionName) {<a name="line.449"></a> +<span class="sourceLineNo">450</span> RegionStoreSequenceIds.Builder builder = RegionStoreSequenceIds.newBuilder();<a name="line.450"></a> +<span class="sourceLineNo">451</span> Long seqId = flushedSequenceIdByRegion.get(encodedRegionName);<a name="line.451"></a> +<span class="sourceLineNo">452</span> builder.setLastFlushedSequenceId(seqId != null ? seqId.longValue() : HConstants.NO_SEQNUM);<a name="line.452"></a> +<span class="sourceLineNo">453</span> Map<byte[], Long> storeFlushedSequenceId =<a name="line.453"></a> +<span class="sourceLineNo">454</span> storeFlushedSequenceIdsByRegion.get(encodedRegionName);<a name="line.454"></a> +<span class="sourceLineNo">455</span> if (storeFlushedSequenceId != null) {<a name="line.455"></a> +<span class="sourceLineNo">456</span> for (Map.Entry<byte[], Long> entry : storeFlushedSequenceId.entrySet()) {<a name="line.456"></a> +<span class="sourceLineNo">457</span> builder.addStoreSequenceId(StoreSequenceId.newBuilder()<a name="line.457"></a> +<span class="sourceLineNo">458</span> .setFamilyName(UnsafeByteOperations.unsafeWrap(entry.getKey()))<a name="line.458"></a> +<span class="sourceLineNo">459</span> .setSequenceId(entry.getValue().longValue()).build());<a name="line.459"></a> +<span class="sourceLineNo">460</span> }<a name="line.460"></a> +<span class="sourceLineNo">461</span> }<a name="line.461"></a> +<span class="sourceLineNo">462</span> return builder.build();<a name="line.462"></a> +<span class="sourceLineNo">463</span> }<a name="line.463"></a> +<span class="sourceLineNo">464</span><a name="line.464"></a> +<span class="sourceLineNo">465</span> /**<a name="line.465"></a> +<span class="sourceLineNo">466</span> * @param serverName<a name="line.466"></a> +<span class="sourceLineNo">467</span> * @return ServerMetrics if serverName is known else null<a name="line.467"></a> +<span class="sourceLineNo">468</span> */<a name="line.468"></a> +<span class="sourceLineNo">469</span> public ServerMetrics getLoad(final ServerName serverName) {<a name="line.469"></a> +<span class="sourceLineNo">470</span> return this.onlineServers.get(serverName);<a name="line.470"></a> +<span class="sourceLineNo">471</span> }<a name="line.471"></a> +<span class="sourceLineNo">472</span><a name="line.472"></a> +<span class="sourceLineNo">473</span> /**<a name="line.473"></a> +<span class="sourceLineNo">474</span> * Compute the average load across all region servers.<a name="line.474"></a> +<span class="sourceLineNo">475</span> * Currently, this uses a very naive computation - just uses the number of<a name="line.475"></a> +<span class="sourceLineNo">476</span> * regions being served, ignoring stats about number of requests.<a name="line.476"></a> +<span class="sourceLineNo">477</span> * @return the average load<a name="line.477"></a> +<span class="sourceLineNo">478</span> */<a name="line.478"></a> +<span class="sourceLineNo">479</span> public double getAverageLoad() {<a name="line.479"></a> +<span class="sourceLineNo">480</span> int totalLoad = 0;<a name="line.480"></a> +<span class="sourceLineNo">481</span> int numServers = 0;<a name="line.481"></a> +<span class="sourceLineNo">482</span> for (ServerMetrics sl : this.onlineServers.values()) {<a name="line.482"></a> +<span class="sourceLineNo">483</span> numServers++;<a name="line.483"></a> +<span class="sourceLineNo">484</span> totalLoad += sl.getRegionMetrics().size();<a name="line.484"></a> +<span class="sourceLineNo">485</span> }<a name="line.485"></a> +<span class="sourceLineNo">486</span> return numServers == 0 ? 0 :<a name="line.486"></a> +<span class="sourceLineNo">487</span> (double)totalLoad / (double)numServers;<a name="line.487"></a> +<span class="sourceLineNo">488</span> }<a name="line.488"></a> +<span class="sourceLineNo">489</span><a name="line.489"></a> +<span class="sourceLineNo">490</span> /** @return the count of active regionservers */<a name="line.490"></a> +<span class="sourceLineNo">491</span> public int countOfRegionServers() {<a name="line.491"></a> +<span class="sourceLineNo">492</span> // Presumes onlineServers is a concurrent map<a name="line.492"></a> +<span class="sourceLineNo">493</span> return this.onlineServers.size();<a name="line.493"></a> +<span class="sourceLineNo">494</span> }<a name="line.494"></a> +<span class="sourceLineNo">495</span><a name="line.495"></a> +<span class="sourceLineNo">496</span> /**<a name="line.496"></a> +<span class="sourceLineNo">497</span> * @return Read-only map of servers to serverinfo<a name="line.497"></a> +<span class="sourceLineNo">498</span> */<a name="line.498"></a> +<span class="sourceLineNo">499</span> public Map<ServerName, ServerMetrics> getOnlineServers() {<a name="line.499"></a> +<span class="sourceLineNo">500</span> // Presumption is that iterating the returned Map is OK.<a name="line.500"></a> +<span class="sourceLineNo">501</span> synchronized (this.onlineServers) {<a name="line.501"></a> +<span class="sourceLineNo">502</span> return Collections.unmodifiableMap(this.onlineServers);<a name="line.502"></a> +<span class="sourceLineNo">503</span> }<a name="line.503"></a> +<span class="sourceLineNo">504</span> }<a name="line.504"></a> +<span class="sourceLineNo">505</span><a name="line.505"></a> +<span class="sourceLineNo">506</span> public DeadServer getDeadServers() {<a name="line.506"></a> +<span class="sourceLineNo">507</span> return this.deadservers;<a name="line.507"></a> +<span class="sourceLineNo">508</span> }<a name="line.508"></a> +<span class="sourceLineNo">509</span><a name="line.509"></a> +<span class="sourceLineNo">510</span> /**<a name="line.510"></a> +<span class="sourceLineNo">511</span> * Checks if any dead servers are currently in progress.<a name="line.511"></a> +<span class="sourceLineNo">512</span> * @return true if any RS are being processed as dead, false if not<a name="line.512"></a> +<span class="sourceLineNo">513</span> */<a name="line.513"></a> +<span class="sourceLineNo">514</span> public boolean areDeadServersInProgress() {<a name="line.514"></a> +<span class="sourceLineNo">515</span> return this.deadservers.areDeadServersInProgress();<a name="line.515"></a> +<span class="sourceLineNo">516</span> }<a name="line.516"></a> +<span class="sourceLineNo">517</span><a name="line.517"></a> +<span class="sourceLineNo">518</span> void letRegionServersShutdown() {<a name="line.518"></a> +<span class="sourceLineNo">519</span> long previousLogTime = 0;<a name="line.519"></a> +<span class="sourceLineNo">520</span> ServerName sn = master.getServerName();<a name="line.520"></a> +<span class="sourceLineNo">521</span> ZKWatcher zkw = master.getZooKeeper();<a name="line.521"></a> +<span class="sourceLineNo">522</span> int onlineServersCt;<a name="line.522"></a> +<span class="sourceLineNo">523</span> while ((onlineServersCt = onlineServers.size()) > 0){<a name="line.523"></a> +<span class="sourceLineNo">524</span><a name="line.524"></a> +<span class="sourceLineNo">525</span> if (System.currentTimeMillis() > (previousLogTime + 1000)) {<a name="line.525"></a> +<span class="sourceLineNo">526</span> Set<ServerName> remainingServers = onlineServers.keySet();<a name="line.526"></a> +<span class="sourceLineNo">527</span> synchronized (onlineServers) {<a name="line.527"></a> +<span class="sourceLineNo">528</span> if (remainingServers.size() == 1 && remainingServers.contains(sn)) {<a name="line.528"></a> +<span class="sourceLineNo">529</span> // Master will delete itself later.<a name="line.529"></a> +<span class="sourceLineNo">530</span> return;<a name="line.530"></a> +<span class="sourceLineNo">531</span> }<a name="line.531"></a> +<span class="sourceLineNo">532</span> }<a name="line.532"></a> +<span class="sourceLineNo">533</span> StringBuilder sb = new StringBuilder();<a name="line.533"></a> +<span class="sourceLineNo">534</span> // It's ok here to not sync on onlineServers - merely logging<a name="line.534"></a> +<span class="sourceLineNo">535</span> for (ServerName key : remainingServers) {<a name="line.535"></a> +<span class="sourceLineNo">536</span> if (sb.length() > 0) {<a name="line.536"></a> +<span class="sourceLineNo">537</span> sb.append(", ");<a name="line.537"></a> +<span class="sourceLineNo">538</span> }<a name="line.538"></a> +<span class="sourceLineNo">539</span> sb.append(key);<a name="line.539"></a> +<span class="sourceLineNo">540</span> }<a name="line.540"></a> +<span class="sourceLineNo">541</span> LOG.info("Waiting on regionserver(s) " + sb.toString());<a name="line.541"></a> +<span class="sourceLineNo">542</span> previousLogTime = System.currentTimeMillis();<a name="line.542"></a> +<span class="sourceLineNo">543</span> }<a name="line.543"></a> +<span class="sourceLineNo">544</span><a name="line.544"></a> +<span class="sourceLineNo">545</span> try {<a name="line.545"></a> +<span class="sourceLineNo">546</span> List<String> servers = getRegionServersInZK(zkw);<a name="line.546"></a> +<span class="sourceLineNo">547</span> if (servers == null || servers.isEmpty() || (servers.size() == 1<a name="line.547"></a> +<span class="sourceLineNo">548</span> && servers.contains(sn.toString()))) {<a name="line.548"></a> +<span class="sourceLineNo">549</span> LOG.info("ZK shows there is only the master self online, exiting now");<a name="line.549"></a> +<span class="sourceLineNo">550</span> // Master could have lost some ZK events, no need to wait more.<a name="line.550"></a> +<span class="sourceLineNo">551</span> break;<a name="line.551"></a> +<span class="sourceLineNo">552</span> }<a name="line.552"></a> +<span class="sourceLineNo">553</span> } catch (KeeperException ke) {<a name="line.553"></a> +<span class="sourceLineNo">554</span> LOG.warn("Failed to list regionservers", ke);<a name="line.554"></a> +<span class="sourceLineNo">555</span> // ZK is malfunctioning, don't hang here<a name="line.555"></a> +<span class="sourceLineNo">556</span> break;<a name="line.556"></a> +<span class="sourceLineNo">557</span> }<a name="line.557"></a> +<span class="sourceLineNo">558</span> synchronized (onlineServers) {<a name="line.558"></a> +<span class="sourceLineNo">559</span> try {<a name="line.559"></a> +<span class="sourceLineNo">560</span> if (onlineServersCt == onlineServers.size()) onlineServers.wait(100);<a name="line.560"></a> +<span class="sourceLineNo">561</span> } catch (InterruptedException ignored) {<a name="line.561"></a> +<span class="sourceLineNo">562</span> // continue<a name="line.562"></a> +<span class="sourceLineNo">563</span> }<a name="line.563"></a> +<span class="sourceLineNo">564</span> }<a name="line.564"></a> +<span class="sourceLineNo">565</span> }<a name="line.565"></a> +<span class="sourceLineNo">566</span> }<a name="line.566"></a> +<span class="sourceLineNo">567</span><a name="line.567"></a> +<span class="sourceLineNo">568</span> private List<String> getRegionServersInZK(final ZKWatcher zkw)<a name="line.568"></a> +<span class="sourceLineNo">569</span> throws KeeperException {<a name="line.569"></a> +<span class="sourceLineNo">570</span> return ZKUtil.listChildrenNoWatch(zkw, zkw.getZNodePaths().rsZNode);<a name="line.570"></a> +<span class="sourceLineNo">571</span> }<a name="line.571"></a> +<span class="sourceLineNo">572</span><a name="line.572"></a> +<span class="sourceLineNo">573</span> /**<a name="line.573"></a> +<span class="sourceLineNo">574</span> * Expire the passed server. Add it to list of dead servers and queue a shutdown processing.<a name="line.574"></a> +<span class="sourceLineNo">575</span> * @return True if we queued a ServerCrashProcedure else false if we did not (could happen for<a name="line.575"></a> +<span class="sourceLineNo">576</span> * many reasons including the fact that its this server that is going down or we already<a name="line.576"></a> +<span class="sourceLineNo">577</span> * have queued an SCP for this server or SCP processing is currently disabled because we<a name="line.577"></a> +<span class="sourceLineNo">578</span> * are in startup phase).<a name="line.578"></a> +<span class="sourceLineNo">579</span> */<a name="line.579"></a> +<span class="sourceLineNo">580</span> public synchronized boolean expireServer(final ServerName serverName) {<a name="line.580"></a> +<span class="sourceLineNo">581</span> // THIS server is going down... can't handle our own expiration.<a name="line.581"></a> +<span class="sourceLineNo">582</span> if (serverName.equals(master.getServerName())) {<a name="line.582"></a> +<span class="sourceLineNo">583</span> if (!(master.isAborted() || master.isStopped())) {<a name="line.583"></a> +<span class="sourceLineNo">584</span> master.stop("We lost our znode?");<a name="line.584"></a> +<span class="sourceLineNo">585</span> }<a name="line.585"></a> +<span class="sourceLineNo">586</span> return false;<a name="line.586"></a> +<span class="sourceLineNo">587</span> }<a name="line.587"></a> +<span class="sourceLineNo">588</span> if (this.deadservers.isDeadServer(serverName)) {<a name="line.588"></a> +<span class="sourceLineNo">589</span> LOG.warn("Expiration called on {} but crash processing already in progress", serverName);<a name="line.589"></a> +<span class="sourceLineNo">590</span> return false;<a name="line.590"></a> +<span class="sourceLineNo">591</span> }<a name="line.591"></a> +<span class="sourceLineNo">592</span> moveFromOnlineToDeadServers(serverName);<a name="line.592"></a> +<span class="sourceLineNo">593</span><a name="line.593"></a> +<span class="sourceLineNo">594</span> // If cluster is going down, yes, servers are going to be expiring; don't<a name="line.594"></a> +<span class="sourceLineNo">595</span> // process as a dead server<a name="line.595"></a> +<span class="sourceLineNo">596</span> if (isClusterShutdown()) {<a name="line.596"></a> +<span class="sourceLineNo">597</span> LOG.info("Cluster shutdown set; " + serverName +<a name="line.597"></a> +<span class="sourceLineNo">598</span> " expired; onlineServers=" + this.onlineServers.size());<a name="line.598"></a> +<span class="sourceLineNo">599</span> if (this.onlineServers.isEmpty()) {<a name="line.599"></a> +<span class="sourceLineNo">600</span> master.stop("Cluster shutdown set; onlineServer=0");<a name="line.600"></a> +<span class="sourceLineNo">601</span> }<a name="line.601"></a> +<span class="sourceLineNo">602</span> return false;<a name="line.602"></a> +<span class="sourceLineNo">603</span> }<a name="line.603"></a> +<span class="sourceLineNo">604</span> LOG.info("Processing expiration of " + serverName + " on " + this.master.getServerName());<a name="line.604"></a> +<span class="sourceLineNo">605</span> master.getAssignmentManager().submitServerCrash(serverName, true);<a name="line.605"></a> +<span class="sourceLineNo">606</span><a name="line.606"></a> +<span class="sourceLineNo">607</span> // Tell our listeners that a server was removed<a name="line.607"></a> +<span class="sourceLineNo">608</span> if (!this.listeners.isEmpty()) {<a name="line.608"></a> +<span class="sourceLineNo">609</span> for (ServerListener listener : this.listeners) {<a name="line.609"></a> +<span class="sourceLineNo">610</span> listener.serverRemoved(serverName);<a name="line.610"></a> +<span class="sourceLineNo">611</span> }<a name="line.611"></a> +<span class="sourceLineNo">612</span> }<a name="line.612"></a> +<span class="sourceLineNo">613</span> // trigger a persist of flushedSeqId<a name="line.613"></a> +<span class="sourceLineNo">614</span> if (flushedSeqIdFlusher != null) {<a name="line.614"></a> +<span class="sourceLineNo">615</span> flushedSeqIdFlusher.triggerNow();<a name="line.615"></a> +<span class="sourceLineNo">616</span> }<a name="line.616"></a> +<span class="sourceLineNo">617</span> return true;<a name="line.617"></a> +<span class="sourceLineNo">618</span> }<a name="line.618"></a> +<span class="sourceLineNo">619</span><a name="line.619"></a> +<span class="sourceLineNo">620</span> @VisibleForTesting<a name="line.620"></a> +<span class="sourceLineNo">621</span> public void moveFromOnlineToDeadServers(final ServerName sn) {<a name="line.621"></a> +<span class="sourceLineNo">622</span> synchronized (onlineServers) {<a name="line.622"></a> +<span class="sourceLineNo">623</span> if (!this.onlineServers.containsKey(sn)) {<a name="line.623"></a> +<span class="sourceLineNo">624</span> LOG.warn("Expiration of " + sn + " but server not online");<a name="line.624"></a> +<span class="sourceLineNo">625</span> }<a name="line.625"></a> +<span class="sourceLineNo">626</span> // Remove the server from the known servers lists and update load info BUT<a name="line.626"></a> +<span class="sourceLineNo">627</span> // add to deadservers first; do this so it'll show in dead servers list if<a name="line.627"></a> +<span class="sourceLineNo">628</span> // not in online servers list.<a name="line.628"></a> +<span class="sourceLineNo">629</span> this.deadservers.add(sn);<a name="line.629"></a> +<span class="sourceLineNo">630</span> this.onlineServers.remove(sn);<a name="line.630"></a> +<span class="sourceLineNo">631</span> onlineServers.notifyAll();<a name="line.631"></a> +<span class="sourceLineNo">632</span> }<a name="line.632"></a> +<span class="sourceLineNo">633</span> this.rsAdmins.remove(sn);<a name="line.633"></a> +<span class="sourceLineNo">634</span> }<a name="line.634"></a> +<span class="sourceLineNo">635</span><a name="line.635"></a> +<span class="sourceLineNo">636</span> /*<a name="line.636"></a> +<span class="sourceLineNo">637</span> * Remove the server from the drain list.<a name="line.637"></a> +<span class="sourceLineNo">638</span> */<a name="line.638"></a> +<span class="sourceLineNo">639</span> public synchronized boolean removeServerFromDrainList(final ServerName sn) {<a name="line.639"></a> +<span class="sourceLineNo">640</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.640"></a> +<span class="sourceLineNo">641</span> // <hostname> , <port> , <startcode><a name="line.641"></a> +<span class="sourceLineNo">642</span><a name="line.642"></a> +<span class="sourceLineNo">643</span> if (!this.isServerOnline(sn)) {<a name="line.643"></a> +<span class="sourceLineNo">644</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.644"></a> +<span class="sourceLineNo">645</span> "Removing from draining list anyway, as requested.");<a name="line.645"></a> +<span class="sourceLineNo">646</span> }<a name="line.646"></a> +<span class="sourceLineNo">647</span> // Remove the server from the draining servers lists.<a name="line.647"></a> +<span class="sourceLineNo">648</span> return this.drainingServers.remove(sn);<a name="line.648"></a> +<span class="sourceLineNo">649</span> }<a name="line.649"></a> +<span class="sourceLineNo">650</span><a name="line.650"></a> +<span class="sourceLineNo">651</span> /**<a name="line.651"></a> +<span class="sourceLineNo">652</span> * Add the server to the drain list.<a name="line.652"></a> +<span class="sourceLineNo">653</span> * @param sn<a name="line.653"></a> +<span class="sourceLineNo">654</span> * @return True if the server is added or the server is already on the drain list.<a name="line.654"></a> +<span class="sourceLineNo">655</span> */<a name="line.655"></a> +<span class="sourceLineNo">656</span> public synchronized boolean addServerToDrainList(final ServerName sn) {<a name="line.656"></a> +<span class="sourceLineNo">657</span> // Warn if the server (sn) is not online. ServerName is of the form:<a name="line.657"></a> +<span class="sourceLineNo">658</span> // <hostname> , <port> , <startcode><a name="line.658"></a> +<span class="sourceLineNo">659</span><a name="line.659"></a> +<span class="sourceLineNo">660</span> if (!this.isServerOnline(sn)) {<a name="line.660"></a> +<span class="sourceLineNo">661</span> LOG.warn("Server " + sn + " is not currently online. " +<a name="line.661"></a> +<span class="sourceLineNo">662</span> "Ignoring request to add it to draining list.");<a name="line.662"></a> +<span class="sourceLineNo">663</span> return false;<a name="line.663"></a> +<span class="sourceLineNo">664</span> }<a name="line.664"></a> +<span class="sourceLineNo">665</span> // Add the server to the draining servers lists, if it's not already in<a name="line.665"></a> +<span class="sourceLineNo">666</span> // it.<a name="line.666"></a> +<span class="sourceLineNo">667</span> if (this.drainingServers.contains(sn)) {<a name="line.667"></a> +<span class="sourceLineNo">668</span> LOG.warn("Server " + sn + " is already in the draining server list." +<a name="line.668"></a> +<span class="sourceLineNo">669</span> "Ignoring request to add it again.");<a name="line.669"></a> +<span class="sourceLineNo">670</span> return true;<a name="line.670"></a> +<span class="sourceLineNo">671</span> }<a name="line.671"></a> +<span class="sourceLineNo">672</span> LOG.info("Server " + sn + " added to draining server list.");<a name="line.672"></a> +<span class="sourceLineNo">673</span> return this.drainingServers.add(sn);<a name="line.673"></a> +<span class="sourceLineNo">674</span> }<a name="line.674"></a> +<span class="sourceLineNo">675</span><a name="line.675"></a> +<span class="sourceLineNo">676</span> // RPC methods to region servers<a name="line.676"></a> +<span class="sourceLineNo">677</span><a name="line.677"></a> +<span class="sourceLineNo">678</span> private HBaseRpcController newRpcController() {<a name="line.678"></a> +<span class="sourceLineNo">679</span> return rpcControllerFactory == null ? null : rpcControllerFactory.newController();<a name="line.679"></a> +<span class="sourceLineNo">680</span> }<a name="line.680"></a> +<span class="sourceLineNo">681</span><a name="line.681"></a> +<span class="sourceLineNo">682</span> /**<a name="line.682"></a> +<span class="sourceLineNo">683</span> * Sends a WARMUP RPC to the specified server to warmup the specified region.<a name="line.683"></a> +<span class="sourceLineNo">684</span> * <p><a name="line.684"></a> +<span class="sourceLineNo">685</span> * A region server could reject the close request because it either does not<a name="line.685"></a> +<span class="sourceLineNo">686</span> * have the specified region or the region is being split.<a name="line.686"></a> +<span class="sourceLineNo">687</span> * @param server server to warmup a region<a name="line.687"></a> +<span class="sourceLineNo">688</span> * @param region region to warmup<a name="line.688"></a> +<span class="sourceLineNo">689</span> */<a name="line.689"></a> +<span class="sourceLineNo">690</span> public void sendRegionWarmup(ServerName server,<a name="line.690"></a> +<span class="sourceLineNo">691</span> RegionInfo region) {<a name="line.691"></a> +<span class="sourceLineNo">692</span> if (server == null) return;<a name="line.692"></a> +<span class="sourceLineNo">693</span> try {<a name="line.693"></a> +<span class="sourceLineNo">694</span> AdminService.BlockingInterface admin = getRsAdmin(server);<a name="line.694"></a> +<span class="sourceLineNo">695</span> HBaseRpcController controller = newRpcController();<a name="line.695"></a> +<span class="sourceLineNo">696</span> ProtobufUtil.warmupRegion(controller, admin, region);<a name="line.696"></a> +<span class="sourceLineNo">697</span> } catch (IOException e) {<a name="line.697"></a> +<span class="sourceLineNo">698</span> LOG.error("Received exception in RPC for warmup server:" +<a name="line.698"></a> +<span class="sourceLineNo">699</span> server + "region: " + region +<a name="line.699"></a> +<span class="sourceLineNo">700</span> "exception: " + e);<a name="line.700"></a> +<span class="sourceLineNo">701</span> }<a name="line.701"></a> +<span class="sourceLineNo">702</span> }<a name="line.702"></a> +<span class="sourceLineNo">703</span><a name="line.703"></a> +<span class="sourceLineNo">704</span> /**<a name="line.704"></a> +<span class="sourceLineNo">705</span> * Contacts a region server and waits up to timeout ms<a name="line.705"></a> +<span class="sourceLineNo">706</span> * to close the region. This bypasses the active hmaster.<a name="line.706"></a> +<span class="sourceLineNo">707</span> */<a name="line.707"></a> +<span class="sourceLineNo">708</span> public static void closeRegionSilentlyAndWait(ClusterConnection connection,<a name="line.708"></a> +<span class="sourceLineNo">709</span> ServerName server, RegionInfo region, long timeout) throws IOException, InterruptedException {<a name="line.709"></a> +<span class="sourceLineNo">710</span> AdminService.BlockingInterface rs = connection.getAdmin(server);<a name="line.710"></a> +<span class="sourceLineNo">711</span> HBaseRpcController controller = connection.getRpcControllerFactory().newController();<a name="line.711"></a> +<span class="sourceLineNo">712</span> try {<a name="line.712"></a> +<span class="sourceLineNo">713</span> ProtobufUtil.closeRegion(controller, rs, server, region.getRegionName());<a name="line.713"></a> +<span class="sourceLineNo">714</span> } catch (IOException e) {<a name="line.714"></a> +<span class="sourceLineNo">715</span> LOG.warn("Exception when closing region: " + region.getRegionNameAsString(), e);<a name="line.715"></a> +<span class="sourceLineNo">716</span> }<a name="line.716"></a> +<span class="sourceLineNo">717</span> long expiration = timeout + System.currentTimeMillis();<a name="line.717"></a> +<span class="sourceLineNo">718</span> while (System.currentTimeMillis() < expiration) {<a name="line.718"></a> +<span class="sourceLineNo">719</span> controller.reset();<a name="line.719"></a> +<span class="sourceLineNo">720</span> try {<a name="line.720"></a> +<span class="sourceLineNo">721</span> RegionInfo rsRegion =<a name="line.721"></a> +<span class="sourceLineNo">722</span> ProtobufUtil.getRegionInfo(controller, rs, region.getRegionName());<a name="line.722"></a> +<span class="sourceLineNo">723</span> if (rsRegion == null) return;<a name="line.723"></a> +<span class="sourceLineNo">724</span> } catch (IOException ioe) {<a name="line.724"></a> +<span class="sourceLineNo">725</span> if (ioe instanceof NotServingRegionException) // no need to retry again<a name="line.725"></a> +<span class="sourceLineNo">726</span> return;<a name="line.726"></a> +<span class="sourceLineNo">727</span> LOG.warn("Exception when retrieving regioninfo from: "<a name="line.727"></a> +<span class="sourceLineNo">728</span> + region.getRegionNameAsString(), ioe);<a name="line.728"></a> +<span class="sourceLineNo">729</span> }<a name="line.729"></a> +<span class="sourceLineNo">730</span> Thread.sleep(1000);<a name="line.730"></a> +<span class="sourceLineNo">731</span> }<a name="line.731"></a> +<span class="sourceLineNo">732</span> throw new IOException("Region " + region + " failed to close within"<a name="line.732"></a> +<span class="sourceLineNo">733</span> + " timeout " + timeout);<a name="line.733"></a> +<span class="sourceLineNo">734</span> }<a name="line.734"></a> +<span class="sourceLineNo">735</span><a name="line.735"></a> +<span class="sourceLineNo">736</span> /**<a name="line.736"></a> +<span class="sourceLineNo">737</span> * @param sn<a name="line.737"></a> +<span class="sourceLineNo">738</span> * @return Admin interface for the remote regionserver named <code>sn</code><a name="line.738"></a> +<span class="sourceLineNo">739</span> * @throws IOException<a name="line.739"></a> +<span class="sourceLineNo">740</span> * @throws RetriesExhaustedException wrapping a ConnectException if failed<a name="line.740"></a> +<span class="sourceLineNo">741</span> */<a name="line.741"></a> +<span class="sourceLineNo">742</span> public AdminService.BlockingInterface getRsAdmin(final ServerName sn)<a name="line.742"></a> +<span class="sourceLineNo">743</span> throws IOException {<a name="line.743"></a> +<span class="sourceLineNo">744</span> AdminService.BlockingInterface admin = this.rsAdmins.get(sn);<a name="line.744"></a> +<span class="sourceLineNo">745</span> if (admin == null) {<a name="line.745"></a> +<span class="sourceLineNo">746</span> LOG.debug("New admin connection to " + sn.toString());<a name="line.746"></a> +<span class="sourceLineNo">747</span> if (sn.equals(master.getServerName()) && master instanceof HRegionServer) {<a name="line.747"></a> +<span class="sourceLineNo">748</span> // A master is also a region server now, see HBASE-10569 for details<a name="line.748"></a> +<span class="sourceLineNo">749</span> admin = ((HRegionServer)master).getRSRpcServices();<a name="line.749"></a> +<span class="sourceLineNo">750</span> } else {<a name="line.750"></a> +<span class="sourceLineNo">751</span> admin = this.connection.getAdmin(sn);<a name="line.751"></a> +<span class="sourceLineNo">752</span> }<a name="line.752"></a> +<span class="sourceLineNo">753</span> this.rsAdmins.put(sn, admin);<a name="line.753"></a> +<span class="sourceLineNo">754</span> }<a name="line.754"></a> +<span class="sourceLineNo">755</span> return admin;<a name="line.755"></a> +<span class="sourceLineNo">756</span> }<a name="line.756"></a> +<span class="sourceLineNo">757</span><a name="line.757"></a> +<span class="sourceLineNo">758</span> /**<a name="line.758"></a> +<span class="sourceLineNo">759</span> * Calculate min necessary to start. This is not an absolute. It is just<a name="line.759"></a> +<span class="sourceLineNo">760</span> * a friction that will cause us hang around a bit longer waiting on<a name="line.760"></a> +<span class="sourceLineNo">761</span> * RegionServers to check-in.<a name="line.761"></a> +<span class="sourceLineNo">762</span> */<a name="line.762"></a> +<span class="sourceLineNo">763</span> private int getMinToStart() {<a name="line.763"></a> +<span class="sourceLineNo">764</span> // One server should be enough to get us off the ground.<a name="line.764"></a> +<span class="sourceLineNo">765</span> int requiredMinToStart = 1;<a name="line.765"></a> +<span class="sourceLineNo">766</span> if (LoadBalancer.isTablesOnMaster(master.getConfiguration())) {<a name="line.766"></a> +<span class="sourceLineNo">767</span> if (LoadBalancer.isSystemTablesOnlyOnMaster(master.getConfiguration())) {<a name="line.767"></a> +<span class="sourceLineNo">768</span> // If Master is carrying regions but NOT user-space regions, it<a name="line.768"></a> +<span class="sourceLineNo">769</span> // still shows as a 'server'. We need at least one more server to check<a name="line.769"></a> +<span class="sourceLineNo">770</span> // in before we can start up so set defaultMinToStart to 2.<a name="line.770"></a> +<span class="sourceLineNo">771</span> requiredMinToStart = requiredMinToStart + 1;<a name="line.771"></a> +<span class="sourceLineNo">772</span> }<a name="line.772"></a> +<span class="sourceLineNo">773</span> }<a name="line.773"></a> +<span class="sourceLineNo">774</span> int minToStart = this.master.getConfiguration().getInt(WAIT_ON_REGIONSERVERS_MINTOSTART, -1);<a name="line.774"></a> +<span class="sourceLineNo">775</span> // Ensure we are never less than requiredMinToStart else stuff won't work.<a name="line.775"></a> +<span class="sourceLineNo">776</span> return minToStart == -1 || minToStart < requiredMinToStart? requiredMinToStart: minToStart;<a name="line.776"></a> +<span class="sourceLineNo">777</span> }<a name="line.777"></a> +<span class="sourceLineNo">778</span><a name="line.778"></a> +<span class="sourceLineNo">779</span> /**<a name="line.779"></a> +<span class="sourceLineNo">780</span> * Wait for the region servers to report in.<a name="line.780"></a> +<span class="sourceLineNo">781</span> * We will wait until one of this condition is met:<a name="line.781"></a> +<span class="sourceLineNo">782</span> * - the master is stopped<a name="line.782"></a> +<span class="sourceLineNo">783</span> * - the 'hbase.master.wait.on.regionservers.maxtostart' number of<a name="line.783"></a> +<span class="sourceLineNo">784</span> * region servers is reached<a name="line.784"></a> +<span class="sourceLineNo">785</span> * - the 'hbase.master.wait.on.regionservers.mintostart' is reached AND<a name="line.785"></a> +<span class="sourceLineNo">786</span> * there have been no new region server in for<a name="line.786"></a> +<span class="sourceLineNo">787</span> * 'hbase.master.wait.on.regionservers.interval' time AND<a name="line.787"></a> +<span class="sourceLineNo">788</span> * the 'hbase.master.wait.on.regionservers.timeout' is reached<a name="line.788"></a> +<span class="sourceLineNo">789</span> *<a name="line.789"></a> +<span class="sourceLineNo">790</span> * @throws InterruptedException<a name="line.790"></a> +<span class="sourceLineNo">791</span> */<a name="line.791"></a> +<span class="sourceLineNo">792</span> public void waitForRegionServers(MonitoredTask status) throws InterruptedException {<a name="line.792"></a> +<span class="sourceLineNo">793</span> final long interval = this.master.getConfiguration().<a name="line.793"></a> +<span class="sourceLineNo">794</span> getLong(WAIT_ON_REGIONSERVERS_INTERVAL, 1500);<a name="line.794"></a> +<span class="sourceLineNo">795</span> final long timeout = this.master.getConfiguration().<a name="line.795"></a> +<span class="sourceLineNo">796</span> getLong(WAIT_ON_REGIONSERVERS_TIMEOUT, 4500);<a name="line.796"></a> +<span class="sourceLineNo">797</span> // Min is not an absolute; just a friction making us wait longer on server checkin.<a name="line.797"></a> +<span class="sourceLineNo">798</span> int minToStart = getMinToStart();<a name="line.798"></a> +<span class="sourceLineNo">799</span> int maxToStart = this.master.getConfiguration().<a name="line.799"></a> +<span class="sourceLineNo">800</span> getInt(WAIT_ON_REGIONSERVERS_MAXTOSTART, Integer.MAX_VALUE);<a name="line.800"></a> +<span class="sourceLineNo">801</span> if (maxToStart < minToStart) {<a name="line.801"></a> +<span class="sourceLineNo">802</span> LOG.warn(String.format("The value of '%s' (%d) is set less than '%s' (%d), ignoring.",<a name="line.802"></a> +<span class="sourceLineNo">803</span> WAIT_ON_REGIONSERVERS_MAXTOSTART, maxToStart,<a name="line.803"></a> +<span class="sourceLineNo">804</span> WAIT_ON_REGIONSERVERS_MINTOSTART, minToStart));<a name="line.804"></a> +<span class="sourceLineNo">805</span> maxToStart = Integer.MAX_VALUE;<a name="line.805"></a> +<span class="sourceLineNo">806</span> }<a name="line.806"></a> +<span class="sourceLineNo">807</span><a name="line.807"></a> +<span class="sourceLineNo">808</span> long now = System.currentTimeMillis();<a name="line.808"></a> +<span class="sourceLineNo">809</span> final long startTime = now;<a name="line.809"></a> +<span class="sourceLineNo">810</span> long slept = 0;<a name="line.810"></a> +<span class="sourceLineNo">811</span> long lastLogTime = 0;<a name="line.811"></a> +<span class="sourceLineNo">812</span> long lastCountChange = startTime;<a name="line.812"></a> +<span class="sourceLineNo">813</span> int count = countOfRegionServers();<a name="line.813"></a> +<span class="sourceLineNo">814</span> int oldCount = 0;<a name="line.814"></a> +<span class="sourceLineNo">815</span> // This while test is a little hard to read. We try to comment it in below but in essence:<a name="line.815"></a> +<span class="sourceLineNo">816</span> // Wait if Master is not stopped and the number of regionservers that have checked-in is<a name="line.816"></a> +<span class="sourceLineNo">817</span> // less than the maxToStart. Both of these conditions will be true near universally.<a name="line.817"></a> +<span class="sourceLineNo">818</span> // Next, we will keep cycling if ANY of the following three conditions are true:<a name="line.818"></a> +<span class="sourceLineNo">819</span> // 1. The time since a regionserver registered is < interval (means servers are actively checking in).<a name="line.819"></a> +<span class="sourceLineNo">820</span> // 2. We are under the total timeout.<a name="line.820"></a> +<span class="sourceLineNo">821</span> // 3. The count of servers is < minimum.<a name="line.821"></a> +<span class="sourceLineNo">822</span> for (ServerListener listener: this.listeners) {<a name="line.822"></a> +<span class="sourceLineNo">823</span> listener.waiting();<a name="line.823"></a> +<span class="sourceLineNo">824</span> }<a name="line.824"></a> +<span class="sourceLineNo">825</span> while (!this.master.isStopped() && !isClusterShutdown() && count < maxToStart &&<a name="line.825"></a> +<span class="sourceLineNo">826</span> ((lastCountChange + interval) > now || timeout > slept || count < minToStart)) {<a name="line.826"></a> +<span class="sourceLineNo">827</span> // Log some info at every interval time or if there is a change<a name="line.827"></a> +<span class="sourceLineNo">828</span> if (oldCount != count || lastLogTime + interval < now) {<a name="line.828"></a> +<span class="sourceLineNo">829</span> lastLogTime = now;<a name="line.829"></a> +<span class="sourceLineNo">830</span> String msg =<a name="line.830"></a> +<span class="sourceLineNo">831</span> "Waiting on regionserver count=" + count + "; waited="+<a name="line.831"></a> +<span class="sourceLineNo">832</span> slept + "ms, expecting min=" + minToStart + " server(s), max="+ getStrForMax(maxToStart) +<a name="line.832"></a> +<span class="sourceLineNo">833</span> " server(s), " + "timeout=" + timeout + "ms, lastChange=" + (lastCountChange - now) + "ms";<a name="line.833"></a> +<span class="sourceLineNo">834</span> LOG.info(msg);<a name="line.834"></a> +<span class="sourceLineNo">835</span> status.setStatus(msg);<a name="line.835"></a> +<span class="sourceLineNo">836</span> }<a name="line.836"></a> +<span class="sourceLineNo">837</span><a name="line.837"></a> +<span class="sourceLineNo">838</span> // We sleep for some time<a name="line.838"></a> +<span class="sourceLineNo">839</span> final long sleepTime = 50;<a name="line.839"></a> +<span class="sourceLineNo">840</span> Thread.sleep(sleepTime);<a name="line.840"></a> +<span class="sourceLineNo">841</span> now = System.currentTimeMillis();<a name="line.841"></a> +<span class="sourceLineNo">842</span> slept = now - startTime;<a name="line.842"></a> +<span class="sourceLineNo">843</span><a name="line.843"></a> +<span class="sourceLineNo">844</span> oldCount = count;<a name="line.844"></a> +<span class="sourceLineNo">845</span> count = countOfRegionServers();<a name="line.845"></a> +<span class="sourceLineNo">846</span> if (count != oldCount) {<a name="line.846"></a> +<span class="sourceLineNo">847</span> lastCountChange = now;<a name="line.847"></a> +<span class="sourceLineNo">848</span> }<a name="line.848"></a> +<span class="sourceLineNo">849</span> }<a name="line.849"></a> +<span class="sourceLineNo">850</span> // Did we exit the loop because cluster is going down?<a name="line.850"></a> +<span class="sourceLineNo">851</span> if (isClusterShutdown()) {<a name="line.851"></a> +<span class="sourceLineNo">852</span> this.master.stop("Cluster shutdown");<a name="line.852"></a> +<span class="sourceLineNo">853</span> }<a name="line.853"></a> +<span class="sourceLineNo">854</span> LOG.info("Finished waiting on RegionServer count=" + count + "; waited=" + slept + "ms," +<a name="line.854"></a> +<span class="sourceLineNo">855</span> " expected min=" + minToStart + " server(s), max=" + getStrForMax(maxToStart) + " server(s),"+<a name="line.855"></a> +<span class="sourceLineNo">856</span> " master is "+ (this.master.isStopped() ? "stopped.": "running"));<a name="line.856"></a> +<span class="sourceLineNo">857</span> }<a name="line.857"></a> +<span class="sourceLineNo">858</span><a nam
<TRUNCATED>
