| <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"> |
| <html lang="en"> |
| <head> |
| <title>Source code</title> |
| <link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style"> |
| </head> |
| <body> |
| <div class="sourceContainer"> |
| <pre><span class="sourceLineNo">001</span>/**<a name="line.1"></a> |
| <span class="sourceLineNo">002</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a> |
| <span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a> |
| <span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a> |
| <span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a> |
| <span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a> |
| <span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a> |
| <span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a> |
| <span class="sourceLineNo">009</span> *<a name="line.9"></a> |
| <span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a> |
| <span class="sourceLineNo">011</span> *<a name="line.11"></a> |
| <span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a> |
| <span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a> |
| <span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a> |
| <span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a> |
| <span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a> |
| <span class="sourceLineNo">017</span> */<a name="line.17"></a> |
| <span class="sourceLineNo">018</span>package org.apache.hadoop.hbase.master;<a name="line.18"></a> |
| <span class="sourceLineNo">019</span><a name="line.19"></a> |
| <span class="sourceLineNo">020</span>import java.io.IOException;<a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import java.io.InterruptedIOException;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span>import java.util.HashSet;<a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.util.Iterator;<a name="line.23"></a> |
| <span class="sourceLineNo">024</span>import java.util.List;<a name="line.24"></a> |
| <span class="sourceLineNo">025</span>import java.util.Set;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import java.util.concurrent.ExecutorService;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import java.util.concurrent.Executors;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span>import java.util.stream.Collectors;<a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.ServerMetrics;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.ServerMetricsBuilder;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.ServerName;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.client.VersionInfoUtil;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.zookeeper.ZKListener;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.zookeeper.ZNodePaths;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import org.apache.zookeeper.KeeperException;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import org.slf4j.Logger;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span>import org.slf4j.LoggerFactory;<a name="line.41"></a> |
| <span class="sourceLineNo">042</span><a name="line.42"></a> |
| <span class="sourceLineNo">043</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.43"></a> |
| <span class="sourceLineNo">044</span><a name="line.44"></a> |
| <span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.45"></a> |
| <span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionServerInfo;<a name="line.46"></a> |
| <span class="sourceLineNo">047</span><a name="line.47"></a> |
| <span class="sourceLineNo">048</span>/**<a name="line.48"></a> |
| <span class="sourceLineNo">049</span> * Tracks the online region servers via ZK.<a name="line.49"></a> |
| <span class="sourceLineNo">050</span> * <p/><a name="line.50"></a> |
| <span class="sourceLineNo">051</span> * Handling of new RSs checking in is done via RPC. This class is only responsible for watching for<a name="line.51"></a> |
| <span class="sourceLineNo">052</span> * expired nodes. It handles listening for changes in the RS node list. The only exception is when<a name="line.52"></a> |
| <span class="sourceLineNo">053</span> * master restart, we will use the list fetched from zk to construct the initial set of live region<a name="line.53"></a> |
| <span class="sourceLineNo">054</span> * servers.<a name="line.54"></a> |
| <span class="sourceLineNo">055</span> * <p/><a name="line.55"></a> |
| <span class="sourceLineNo">056</span> * If an RS node gets deleted, this automatically handles calling of<a name="line.56"></a> |
| <span class="sourceLineNo">057</span> * {@link ServerManager#expireServer(ServerName)}<a name="line.57"></a> |
| <span class="sourceLineNo">058</span> */<a name="line.58"></a> |
| <span class="sourceLineNo">059</span>@InterfaceAudience.Private<a name="line.59"></a> |
| <span class="sourceLineNo">060</span>public class RegionServerTracker extends ZKListener {<a name="line.60"></a> |
| <span class="sourceLineNo">061</span> private static final Logger LOG = LoggerFactory.getLogger(RegionServerTracker.class);<a name="line.61"></a> |
| <span class="sourceLineNo">062</span> private final Set<ServerName> regionServers = new HashSet<>();<a name="line.62"></a> |
| <span class="sourceLineNo">063</span> private final ServerManager serverManager;<a name="line.63"></a> |
| <span class="sourceLineNo">064</span> private final MasterServices server;<a name="line.64"></a> |
| <span class="sourceLineNo">065</span> // As we need to send request to zk when processing the nodeChildrenChanged event, we'd better<a name="line.65"></a> |
| <span class="sourceLineNo">066</span> // move the operation to a single threaded thread pool in order to not block the zk event<a name="line.66"></a> |
| <span class="sourceLineNo">067</span> // processing since all the zk listener across HMaster will be called in one thread sequentially.<a name="line.67"></a> |
| <span class="sourceLineNo">068</span> private final ExecutorService executor;<a name="line.68"></a> |
| <span class="sourceLineNo">069</span><a name="line.69"></a> |
| <span class="sourceLineNo">070</span> public RegionServerTracker(ZKWatcher watcher, MasterServices server,<a name="line.70"></a> |
| <span class="sourceLineNo">071</span> ServerManager serverManager) {<a name="line.71"></a> |
| <span class="sourceLineNo">072</span> super(watcher);<a name="line.72"></a> |
| <span class="sourceLineNo">073</span> this.server = server;<a name="line.73"></a> |
| <span class="sourceLineNo">074</span> this.serverManager = serverManager;<a name="line.74"></a> |
| <span class="sourceLineNo">075</span> this.executor = Executors.newSingleThreadExecutor(<a name="line.75"></a> |
| <span class="sourceLineNo">076</span> new ThreadFactoryBuilder().setDaemon(true).setNameFormat("RegionServerTracker-%d").build());<a name="line.76"></a> |
| <span class="sourceLineNo">077</span> }<a name="line.77"></a> |
| <span class="sourceLineNo">078</span><a name="line.78"></a> |
| <span class="sourceLineNo">079</span> private Pair<ServerName, RegionServerInfo> getServerInfo(String name)<a name="line.79"></a> |
| <span class="sourceLineNo">080</span> throws KeeperException, IOException {<a name="line.80"></a> |
| <span class="sourceLineNo">081</span> ServerName serverName = ServerName.parseServerName(name);<a name="line.81"></a> |
| <span class="sourceLineNo">082</span> String nodePath = ZNodePaths.joinZNode(watcher.getZNodePaths().rsZNode, name);<a name="line.82"></a> |
| <span class="sourceLineNo">083</span> byte[] data;<a name="line.83"></a> |
| <span class="sourceLineNo">084</span> try {<a name="line.84"></a> |
| <span class="sourceLineNo">085</span> data = ZKUtil.getData(watcher, nodePath);<a name="line.85"></a> |
| <span class="sourceLineNo">086</span> } catch (InterruptedException e) {<a name="line.86"></a> |
| <span class="sourceLineNo">087</span> throw (InterruptedIOException) new InterruptedIOException().initCause(e);<a name="line.87"></a> |
| <span class="sourceLineNo">088</span> }<a name="line.88"></a> |
| <span class="sourceLineNo">089</span> if (data == null) {<a name="line.89"></a> |
| <span class="sourceLineNo">090</span> // we should receive a children changed event later and then we will expire it, so we still<a name="line.90"></a> |
| <span class="sourceLineNo">091</span> // need to add it to the region server set.<a name="line.91"></a> |
| <span class="sourceLineNo">092</span> LOG.warn("Server node {} does not exist, already dead?", name);<a name="line.92"></a> |
| <span class="sourceLineNo">093</span> return Pair.newPair(serverName, null);<a name="line.93"></a> |
| <span class="sourceLineNo">094</span> }<a name="line.94"></a> |
| <span class="sourceLineNo">095</span> if (data.length == 0 || !ProtobufUtil.isPBMagicPrefix(data)) {<a name="line.95"></a> |
| <span class="sourceLineNo">096</span> // this should not happen actually, unless we have bugs or someone has messed zk up.<a name="line.96"></a> |
| <span class="sourceLineNo">097</span> LOG.warn("Invalid data for region server node {} on zookeeper, data length = {}", name,<a name="line.97"></a> |
| <span class="sourceLineNo">098</span> data.length);<a name="line.98"></a> |
| <span class="sourceLineNo">099</span> return Pair.newPair(serverName, null);<a name="line.99"></a> |
| <span class="sourceLineNo">100</span> }<a name="line.100"></a> |
| <span class="sourceLineNo">101</span> RegionServerInfo.Builder builder = RegionServerInfo.newBuilder();<a name="line.101"></a> |
| <span class="sourceLineNo">102</span> int magicLen = ProtobufUtil.lengthOfPBMagic();<a name="line.102"></a> |
| <span class="sourceLineNo">103</span> ProtobufUtil.mergeFrom(builder, data, magicLen, data.length - magicLen);<a name="line.103"></a> |
| <span class="sourceLineNo">104</span> return Pair.newPair(serverName, builder.build());<a name="line.104"></a> |
| <span class="sourceLineNo">105</span> }<a name="line.105"></a> |
| <span class="sourceLineNo">106</span><a name="line.106"></a> |
| <span class="sourceLineNo">107</span> /**<a name="line.107"></a> |
| <span class="sourceLineNo">108</span> * Starts the tracking of online RegionServers. All RSes will be tracked after this method is<a name="line.108"></a> |
| <span class="sourceLineNo">109</span> * called.<a name="line.109"></a> |
| <span class="sourceLineNo">110</span> * <p/><a name="line.110"></a> |
| <span class="sourceLineNo">111</span> * In this method, we will also construct the region server sets in {@link ServerManager}. If a<a name="line.111"></a> |
| <span class="sourceLineNo">112</span> * region server is dead between the crash of the previous master instance and the start of the<a name="line.112"></a> |
| <span class="sourceLineNo">113</span> * current master instance, we will schedule a SCP for it. This is done in<a name="line.113"></a> |
| <span class="sourceLineNo">114</span> * {@link ServerManager#findDeadServersAndProcess(Set, Set)}, we call it here under the lock<a name="line.114"></a> |
| <span class="sourceLineNo">115</span> * protection to prevent concurrency issues with server expiration operation.<a name="line.115"></a> |
| <span class="sourceLineNo">116</span> * @param deadServersFromPE the region servers which already have SCP associated.<a name="line.116"></a> |
| <span class="sourceLineNo">117</span> * @param liveServersFromWALDir the live region servers from wal directory.<a name="line.117"></a> |
| <span class="sourceLineNo">118</span> * @param splittingServersFromWALDir Servers whose WALs are being actively 'split'.<a name="line.118"></a> |
| <span class="sourceLineNo">119</span> */<a name="line.119"></a> |
| <span class="sourceLineNo">120</span> public void start(Set<ServerName> deadServersFromPE, Set<ServerName> liveServersFromWALDir,<a name="line.120"></a> |
| <span class="sourceLineNo">121</span> Set<ServerName> splittingServersFromWALDir)<a name="line.121"></a> |
| <span class="sourceLineNo">122</span> throws KeeperException, IOException {<a name="line.122"></a> |
| <span class="sourceLineNo">123</span> LOG.info("Starting RegionServerTracker; {} have existing ServerCrashProcedures, {} " +<a name="line.123"></a> |
| <span class="sourceLineNo">124</span> "possibly 'live' servers, and {} 'splitting'.", deadServersFromPE.size(),<a name="line.124"></a> |
| <span class="sourceLineNo">125</span> liveServersFromWALDir.size(), splittingServersFromWALDir.size());<a name="line.125"></a> |
| <span class="sourceLineNo">126</span> // deadServersFromPE is made from a list of outstanding ServerCrashProcedures.<a name="line.126"></a> |
| <span class="sourceLineNo">127</span> // splittingServersFromWALDir are being actively split -- the directory in the FS ends in<a name="line.127"></a> |
| <span class="sourceLineNo">128</span> // '-SPLITTING'. Each splitting server should have a corresponding SCP. Log if not.<a name="line.128"></a> |
| <span class="sourceLineNo">129</span> splittingServersFromWALDir.stream().filter(s -> !deadServersFromPE.contains(s)).<a name="line.129"></a> |
| <span class="sourceLineNo">130</span> forEach(s -> LOG.error("{} has no matching ServerCrashProcedure", s));<a name="line.130"></a> |
| <span class="sourceLineNo">131</span> //create ServerNode for all possible live servers from wal directory<a name="line.131"></a> |
| <span class="sourceLineNo">132</span> liveServersFromWALDir.stream()<a name="line.132"></a> |
| <span class="sourceLineNo">133</span> .forEach(sn -> server.getAssignmentManager().getRegionStates().getOrCreateServer(sn));<a name="line.133"></a> |
| <span class="sourceLineNo">134</span> watcher.registerListener(this);<a name="line.134"></a> |
| <span class="sourceLineNo">135</span> synchronized (this) {<a name="line.135"></a> |
| <span class="sourceLineNo">136</span> List<String> servers =<a name="line.136"></a> |
| <span class="sourceLineNo">137</span> ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.137"></a> |
| <span class="sourceLineNo">138</span> for (String n : servers) {<a name="line.138"></a> |
| <span class="sourceLineNo">139</span> Pair<ServerName, RegionServerInfo> pair = getServerInfo(n);<a name="line.139"></a> |
| <span class="sourceLineNo">140</span> ServerName serverName = pair.getFirst();<a name="line.140"></a> |
| <span class="sourceLineNo">141</span> RegionServerInfo info = pair.getSecond();<a name="line.141"></a> |
| <span class="sourceLineNo">142</span> regionServers.add(serverName);<a name="line.142"></a> |
| <span class="sourceLineNo">143</span> ServerMetrics serverMetrics = info != null ? ServerMetricsBuilder.of(serverName,<a name="line.143"></a> |
| <span class="sourceLineNo">144</span> VersionInfoUtil.getVersionNumber(info.getVersionInfo()),<a name="line.144"></a> |
| <span class="sourceLineNo">145</span> info.getVersionInfo().getVersion()) : ServerMetricsBuilder.of(serverName);<a name="line.145"></a> |
| <span class="sourceLineNo">146</span> serverManager.checkAndRecordNewServer(serverName, serverMetrics);<a name="line.146"></a> |
| <span class="sourceLineNo">147</span> }<a name="line.147"></a> |
| <span class="sourceLineNo">148</span> serverManager.findDeadServersAndProcess(deadServersFromPE, liveServersFromWALDir);<a name="line.148"></a> |
| <span class="sourceLineNo">149</span> }<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> public void stop() {<a name="line.152"></a> |
| <span class="sourceLineNo">153</span> executor.shutdownNow();<a name="line.153"></a> |
| <span class="sourceLineNo">154</span> }<a name="line.154"></a> |
| <span class="sourceLineNo">155</span><a name="line.155"></a> |
| <span class="sourceLineNo">156</span> private synchronized void refresh() {<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> List<String> names;<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> try {<a name="line.158"></a> |
| <span class="sourceLineNo">159</span> names = ZKUtil.listChildrenAndWatchForNewChildren(watcher, watcher.getZNodePaths().rsZNode);<a name="line.159"></a> |
| <span class="sourceLineNo">160</span> } catch (KeeperException e) {<a name="line.160"></a> |
| <span class="sourceLineNo">161</span> // here we need to abort as we failed to set watcher on the rs node which means that we can<a name="line.161"></a> |
| <span class="sourceLineNo">162</span> // not track the node deleted evetnt any more.<a name="line.162"></a> |
| <span class="sourceLineNo">163</span> server.abort("Unexpected zk exception getting RS nodes", e);<a name="line.163"></a> |
| <span class="sourceLineNo">164</span> return;<a name="line.164"></a> |
| <span class="sourceLineNo">165</span> }<a name="line.165"></a> |
| <span class="sourceLineNo">166</span> Set<ServerName> servers =<a name="line.166"></a> |
| <span class="sourceLineNo">167</span> names.stream().map(ServerName::parseServerName).collect(Collectors.toSet());<a name="line.167"></a> |
| <span class="sourceLineNo">168</span> for (Iterator<ServerName> iter = regionServers.iterator(); iter.hasNext();) {<a name="line.168"></a> |
| <span class="sourceLineNo">169</span> ServerName sn = iter.next();<a name="line.169"></a> |
| <span class="sourceLineNo">170</span> if (!servers.contains(sn)) {<a name="line.170"></a> |
| <span class="sourceLineNo">171</span> LOG.info("RegionServer ephemeral node deleted, processing expiration [{}]", sn);<a name="line.171"></a> |
| <span class="sourceLineNo">172</span> serverManager.expireServer(sn);<a name="line.172"></a> |
| <span class="sourceLineNo">173</span> iter.remove();<a name="line.173"></a> |
| <span class="sourceLineNo">174</span> }<a name="line.174"></a> |
| <span class="sourceLineNo">175</span> }<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> // here we do not need to parse the region server info as it is useless now, we only need the<a name="line.176"></a> |
| <span class="sourceLineNo">177</span> // server name.<a name="line.177"></a> |
| <span class="sourceLineNo">178</span> boolean newServerAdded = false;<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> for (ServerName sn : servers) {<a name="line.179"></a> |
| <span class="sourceLineNo">180</span> if (regionServers.add(sn)) {<a name="line.180"></a> |
| <span class="sourceLineNo">181</span> newServerAdded = true;<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> LOG.info("RegionServer ephemeral node created, adding [" + sn + "]");<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> if (newServerAdded && server.isInitialized()) {<a name="line.185"></a> |
| <span class="sourceLineNo">186</span> // Only call the check to move servers if a RegionServer was added to the cluster; in this<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> // case it could be a server with a new version so it makes sense to run the check.<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> server.checkIfShouldMoveSystemRegionAsync();<a name="line.188"></a> |
| <span class="sourceLineNo">189</span> }<a name="line.189"></a> |
| <span class="sourceLineNo">190</span> }<a name="line.190"></a> |
| <span class="sourceLineNo">191</span><a name="line.191"></a> |
| <span class="sourceLineNo">192</span> @Override<a name="line.192"></a> |
| <span class="sourceLineNo">193</span> public void nodeChildrenChanged(String path) {<a name="line.193"></a> |
| <span class="sourceLineNo">194</span> if (path.equals(watcher.getZNodePaths().rsZNode) && !server.isAborted() &&<a name="line.194"></a> |
| <span class="sourceLineNo">195</span> !server.isStopped()) {<a name="line.195"></a> |
| <span class="sourceLineNo">196</span> executor.execute(this::refresh);<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>}<a name="line.199"></a> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |