| <!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><a name="line.18"></a> |
| <span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.replication;<a name="line.19"></a> |
| <span class="sourceLineNo">020</span><a name="line.20"></a> |
| <span class="sourceLineNo">021</span>import java.io.IOException;<a name="line.21"></a> |
| <span class="sourceLineNo">022</span>import java.util.ArrayList;<a name="line.22"></a> |
| <span class="sourceLineNo">023</span>import java.util.Collections;<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.Map;<a name="line.25"></a> |
| <span class="sourceLineNo">026</span>import java.util.UUID;<a name="line.26"></a> |
| <span class="sourceLineNo">027</span>import java.util.concurrent.ThreadLocalRandom;<a name="line.27"></a> |
| <span class="sourceLineNo">028</span><a name="line.28"></a> |
| <span class="sourceLineNo">029</span>import org.apache.hadoop.conf.Configuration;<a name="line.29"></a> |
| <span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.HBaseConfiguration;<a name="line.30"></a> |
| <span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.client.AsyncClusterConnection;<a name="line.31"></a> |
| <span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.client.AsyncRegionServerAdmin;<a name="line.32"></a> |
| <span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.client.ClusterConnectionFactory;<a name="line.33"></a> |
| <span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.security.User;<a name="line.34"></a> |
| <span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.zookeeper.ZKListener;<a name="line.35"></a> |
| <span class="sourceLineNo">036</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.36"></a> |
| <span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.Abortable;<a name="line.37"></a> |
| <span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.ServerName;<a name="line.38"></a> |
| <span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.zookeeper.ZKClusterId;<a name="line.39"></a> |
| <span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.40"></a> |
| <span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.41"></a> |
| <span class="sourceLineNo">042</span>import org.apache.zookeeper.KeeperException;<a name="line.42"></a> |
| <span class="sourceLineNo">043</span>import org.apache.zookeeper.KeeperException.AuthFailedException;<a name="line.43"></a> |
| <span class="sourceLineNo">044</span>import org.apache.zookeeper.KeeperException.ConnectionLossException;<a name="line.44"></a> |
| <span class="sourceLineNo">045</span>import org.apache.zookeeper.KeeperException.SessionExpiredException;<a name="line.45"></a> |
| <span class="sourceLineNo">046</span>import org.slf4j.Logger;<a name="line.46"></a> |
| <span class="sourceLineNo">047</span>import org.slf4j.LoggerFactory;<a name="line.47"></a> |
| <span class="sourceLineNo">048</span><a name="line.48"></a> |
| <span class="sourceLineNo">049</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.49"></a> |
| <span class="sourceLineNo">050</span>import org.apache.hbase.thirdparty.com.google.common.collect.Maps;<a name="line.50"></a> |
| <span class="sourceLineNo">051</span><a name="line.51"></a> |
| <span class="sourceLineNo">052</span>/**<a name="line.52"></a> |
| <span class="sourceLineNo">053</span> * A {@link BaseReplicationEndpoint} for replication endpoints whose<a name="line.53"></a> |
| <span class="sourceLineNo">054</span> * target cluster is an HBase cluster.<a name="line.54"></a> |
| <span class="sourceLineNo">055</span> */<a name="line.55"></a> |
| <span class="sourceLineNo">056</span>@InterfaceAudience.Private<a name="line.56"></a> |
| <span class="sourceLineNo">057</span>public abstract class HBaseReplicationEndpoint extends BaseReplicationEndpoint<a name="line.57"></a> |
| <span class="sourceLineNo">058</span> implements Abortable {<a name="line.58"></a> |
| <span class="sourceLineNo">059</span><a name="line.59"></a> |
| <span class="sourceLineNo">060</span> private static final Logger LOG = LoggerFactory.getLogger(HBaseReplicationEndpoint.class);<a name="line.60"></a> |
| <span class="sourceLineNo">061</span><a name="line.61"></a> |
| <span class="sourceLineNo">062</span> private ZKWatcher zkw = null;<a name="line.62"></a> |
| <span class="sourceLineNo">063</span><a name="line.63"></a> |
| <span class="sourceLineNo">064</span> protected Configuration conf;<a name="line.64"></a> |
| <span class="sourceLineNo">065</span><a name="line.65"></a> |
| <span class="sourceLineNo">066</span> protected AsyncClusterConnection conn;<a name="line.66"></a> |
| <span class="sourceLineNo">067</span><a name="line.67"></a> |
| <span class="sourceLineNo">068</span> /**<a name="line.68"></a> |
| <span class="sourceLineNo">069</span> * Default maximum number of times a replication sink can be reported as bad before<a name="line.69"></a> |
| <span class="sourceLineNo">070</span> * it will no longer be provided as a sink for replication without the pool of<a name="line.70"></a> |
| <span class="sourceLineNo">071</span> * replication sinks being refreshed.<a name="line.71"></a> |
| <span class="sourceLineNo">072</span> */<a name="line.72"></a> |
| <span class="sourceLineNo">073</span> public static final int DEFAULT_BAD_SINK_THRESHOLD = 3;<a name="line.73"></a> |
| <span class="sourceLineNo">074</span><a name="line.74"></a> |
| <span class="sourceLineNo">075</span> /**<a name="line.75"></a> |
| <span class="sourceLineNo">076</span> * Default ratio of the total number of peer cluster region servers to consider<a name="line.76"></a> |
| <span class="sourceLineNo">077</span> * replicating to.<a name="line.77"></a> |
| <span class="sourceLineNo">078</span> */<a name="line.78"></a> |
| <span class="sourceLineNo">079</span> public static final float DEFAULT_REPLICATION_SOURCE_RATIO = 0.5f;<a name="line.79"></a> |
| <span class="sourceLineNo">080</span><a name="line.80"></a> |
| <span class="sourceLineNo">081</span> // Ratio of total number of potential peer region servers to be used<a name="line.81"></a> |
| <span class="sourceLineNo">082</span> private float ratio;<a name="line.82"></a> |
| <span class="sourceLineNo">083</span><a name="line.83"></a> |
| <span class="sourceLineNo">084</span> // Maximum number of times a sink can be reported as bad before the pool of<a name="line.84"></a> |
| <span class="sourceLineNo">085</span> // replication sinks is refreshed<a name="line.85"></a> |
| <span class="sourceLineNo">086</span> private int badSinkThreshold;<a name="line.86"></a> |
| <span class="sourceLineNo">087</span> // Count of "bad replication sink" reports per peer sink<a name="line.87"></a> |
| <span class="sourceLineNo">088</span> private Map<ServerName, Integer> badReportCounts;<a name="line.88"></a> |
| <span class="sourceLineNo">089</span><a name="line.89"></a> |
| <span class="sourceLineNo">090</span> private List<ServerName> sinkServers = new ArrayList<>(0);<a name="line.90"></a> |
| <span class="sourceLineNo">091</span><a name="line.91"></a> |
| <span class="sourceLineNo">092</span> /*<a name="line.92"></a> |
| <span class="sourceLineNo">093</span> * Some implementations of HBaseInterClusterReplicationEndpoint may require instantiate different<a name="line.93"></a> |
| <span class="sourceLineNo">094</span> * Connection implementations, or initialize it in a different way, so defining createConnection<a name="line.94"></a> |
| <span class="sourceLineNo">095</span> * as protected for possible overridings.<a name="line.95"></a> |
| <span class="sourceLineNo">096</span> */<a name="line.96"></a> |
| <span class="sourceLineNo">097</span> protected AsyncClusterConnection createConnection(Configuration conf) throws IOException {<a name="line.97"></a> |
| <span class="sourceLineNo">098</span> return ClusterConnectionFactory.createAsyncClusterConnection(conf,<a name="line.98"></a> |
| <span class="sourceLineNo">099</span> null, User.getCurrent());<a name="line.99"></a> |
| <span class="sourceLineNo">100</span> }<a name="line.100"></a> |
| <span class="sourceLineNo">101</span><a name="line.101"></a> |
| <span class="sourceLineNo">102</span> @Override<a name="line.102"></a> |
| <span class="sourceLineNo">103</span> public void init(Context context) throws IOException {<a name="line.103"></a> |
| <span class="sourceLineNo">104</span> super.init(context);<a name="line.104"></a> |
| <span class="sourceLineNo">105</span> this.conf = HBaseConfiguration.create(ctx.getConfiguration());<a name="line.105"></a> |
| <span class="sourceLineNo">106</span> // TODO: This connection is replication specific or we should make it particular to<a name="line.106"></a> |
| <span class="sourceLineNo">107</span> // replication and make replication specific settings such as compression or codec to use<a name="line.107"></a> |
| <span class="sourceLineNo">108</span> // passing Cells.<a name="line.108"></a> |
| <span class="sourceLineNo">109</span> this.conn = createConnection(this.conf);<a name="line.109"></a> |
| <span class="sourceLineNo">110</span> this.ratio =<a name="line.110"></a> |
| <span class="sourceLineNo">111</span> ctx.getConfiguration().getFloat("replication.source.ratio", DEFAULT_REPLICATION_SOURCE_RATIO);<a name="line.111"></a> |
| <span class="sourceLineNo">112</span> this.badSinkThreshold =<a name="line.112"></a> |
| <span class="sourceLineNo">113</span> ctx.getConfiguration().getInt("replication.bad.sink.threshold", DEFAULT_BAD_SINK_THRESHOLD);<a name="line.113"></a> |
| <span class="sourceLineNo">114</span> this.badReportCounts = Maps.newHashMap();<a name="line.114"></a> |
| <span class="sourceLineNo">115</span> }<a name="line.115"></a> |
| <span class="sourceLineNo">116</span><a name="line.116"></a> |
| <span class="sourceLineNo">117</span> protected synchronized void disconnect() {<a name="line.117"></a> |
| <span class="sourceLineNo">118</span> if (zkw != null) {<a name="line.118"></a> |
| <span class="sourceLineNo">119</span> zkw.close();<a name="line.119"></a> |
| <span class="sourceLineNo">120</span> }<a name="line.120"></a> |
| <span class="sourceLineNo">121</span> }<a name="line.121"></a> |
| <span class="sourceLineNo">122</span><a name="line.122"></a> |
| <span class="sourceLineNo">123</span> /**<a name="line.123"></a> |
| <span class="sourceLineNo">124</span> * A private method used to re-establish a zookeeper session with a peer cluster.<a name="line.124"></a> |
| <span class="sourceLineNo">125</span> * @param ke<a name="line.125"></a> |
| <span class="sourceLineNo">126</span> */<a name="line.126"></a> |
| <span class="sourceLineNo">127</span> private void reconnect(KeeperException ke) {<a name="line.127"></a> |
| <span class="sourceLineNo">128</span> if (ke instanceof ConnectionLossException || ke instanceof SessionExpiredException<a name="line.128"></a> |
| <span class="sourceLineNo">129</span> || ke instanceof AuthFailedException) {<a name="line.129"></a> |
| <span class="sourceLineNo">130</span> String clusterKey = ctx.getPeerConfig().getClusterKey();<a name="line.130"></a> |
| <span class="sourceLineNo">131</span> LOG.warn("Lost the ZooKeeper connection for peer " + clusterKey, ke);<a name="line.131"></a> |
| <span class="sourceLineNo">132</span> try {<a name="line.132"></a> |
| <span class="sourceLineNo">133</span> reloadZkWatcher();<a name="line.133"></a> |
| <span class="sourceLineNo">134</span> } catch (IOException io) {<a name="line.134"></a> |
| <span class="sourceLineNo">135</span> LOG.warn("Creation of ZookeeperWatcher failed for peer " + clusterKey, io);<a name="line.135"></a> |
| <span class="sourceLineNo">136</span> }<a name="line.136"></a> |
| <span class="sourceLineNo">137</span> }<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> @Override<a name="line.140"></a> |
| <span class="sourceLineNo">141</span> public void start() {<a name="line.141"></a> |
| <span class="sourceLineNo">142</span> startAsync();<a name="line.142"></a> |
| <span class="sourceLineNo">143</span> }<a name="line.143"></a> |
| <span class="sourceLineNo">144</span><a name="line.144"></a> |
| <span class="sourceLineNo">145</span> @Override<a name="line.145"></a> |
| <span class="sourceLineNo">146</span> public void stop() {<a name="line.146"></a> |
| <span class="sourceLineNo">147</span> stopAsync();<a name="line.147"></a> |
| <span class="sourceLineNo">148</span> }<a name="line.148"></a> |
| <span class="sourceLineNo">149</span><a name="line.149"></a> |
| <span class="sourceLineNo">150</span> @Override<a name="line.150"></a> |
| <span class="sourceLineNo">151</span> protected void doStart() {<a name="line.151"></a> |
| <span class="sourceLineNo">152</span> try {<a name="line.152"></a> |
| <span class="sourceLineNo">153</span> reloadZkWatcher();<a name="line.153"></a> |
| <span class="sourceLineNo">154</span> notifyStarted();<a name="line.154"></a> |
| <span class="sourceLineNo">155</span> } catch (IOException e) {<a name="line.155"></a> |
| <span class="sourceLineNo">156</span> notifyFailed(e);<a name="line.156"></a> |
| <span class="sourceLineNo">157</span> }<a name="line.157"></a> |
| <span class="sourceLineNo">158</span> }<a name="line.158"></a> |
| <span class="sourceLineNo">159</span><a name="line.159"></a> |
| <span class="sourceLineNo">160</span> @Override<a name="line.160"></a> |
| <span class="sourceLineNo">161</span> protected void doStop() {<a name="line.161"></a> |
| <span class="sourceLineNo">162</span> disconnect();<a name="line.162"></a> |
| <span class="sourceLineNo">163</span> notifyStopped();<a name="line.163"></a> |
| <span class="sourceLineNo">164</span> }<a name="line.164"></a> |
| <span class="sourceLineNo">165</span><a name="line.165"></a> |
| <span class="sourceLineNo">166</span> @Override<a name="line.166"></a> |
| <span class="sourceLineNo">167</span> // Synchronize peer cluster connection attempts to avoid races and rate<a name="line.167"></a> |
| <span class="sourceLineNo">168</span> // limit connections when multiple replication sources try to connect to<a name="line.168"></a> |
| <span class="sourceLineNo">169</span> // the peer cluster. If the peer cluster is down we can get out of control<a name="line.169"></a> |
| <span class="sourceLineNo">170</span> // over time.<a name="line.170"></a> |
| <span class="sourceLineNo">171</span> public synchronized UUID getPeerUUID() {<a name="line.171"></a> |
| <span class="sourceLineNo">172</span> UUID peerUUID = null;<a name="line.172"></a> |
| <span class="sourceLineNo">173</span> try {<a name="line.173"></a> |
| <span class="sourceLineNo">174</span> peerUUID = ZKClusterId.getUUIDForCluster(zkw);<a name="line.174"></a> |
| <span class="sourceLineNo">175</span> } catch (KeeperException ke) {<a name="line.175"></a> |
| <span class="sourceLineNo">176</span> reconnect(ke);<a name="line.176"></a> |
| <span class="sourceLineNo">177</span> }<a name="line.177"></a> |
| <span class="sourceLineNo">178</span> return peerUUID;<a name="line.178"></a> |
| <span class="sourceLineNo">179</span> }<a name="line.179"></a> |
| <span class="sourceLineNo">180</span><a name="line.180"></a> |
| <span class="sourceLineNo">181</span> /**<a name="line.181"></a> |
| <span class="sourceLineNo">182</span> * Closes the current ZKW (if not null) and creates a new one<a name="line.182"></a> |
| <span class="sourceLineNo">183</span> * @throws IOException If anything goes wrong connecting<a name="line.183"></a> |
| <span class="sourceLineNo">184</span> */<a name="line.184"></a> |
| <span class="sourceLineNo">185</span> private synchronized void reloadZkWatcher() throws IOException {<a name="line.185"></a> |
| <span class="sourceLineNo">186</span> if (zkw != null) {<a name="line.186"></a> |
| <span class="sourceLineNo">187</span> zkw.close();<a name="line.187"></a> |
| <span class="sourceLineNo">188</span> }<a name="line.188"></a> |
| <span class="sourceLineNo">189</span> zkw = new ZKWatcher(ctx.getConfiguration(),<a name="line.189"></a> |
| <span class="sourceLineNo">190</span> "connection to cluster: " + ctx.getPeerId(), this);<a name="line.190"></a> |
| <span class="sourceLineNo">191</span> zkw.registerListener(new PeerRegionServerListener(this));<a name="line.191"></a> |
| <span class="sourceLineNo">192</span> }<a name="line.192"></a> |
| <span class="sourceLineNo">193</span><a name="line.193"></a> |
| <span class="sourceLineNo">194</span> @Override<a name="line.194"></a> |
| <span class="sourceLineNo">195</span> public void abort(String why, Throwable e) {<a name="line.195"></a> |
| <span class="sourceLineNo">196</span> LOG.error("The HBaseReplicationEndpoint corresponding to peer " + ctx.getPeerId()<a name="line.196"></a> |
| <span class="sourceLineNo">197</span> + " was aborted for the following reason(s):" + why, e);<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> |
| <span class="sourceLineNo">200</span> @Override<a name="line.200"></a> |
| <span class="sourceLineNo">201</span> public boolean isAborted() {<a name="line.201"></a> |
| <span class="sourceLineNo">202</span> // Currently this is never "Aborted", we just log when the abort method is called.<a name="line.202"></a> |
| <span class="sourceLineNo">203</span> return false;<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> * Get the list of all the region servers from the specified peer<a name="line.207"></a> |
| <span class="sourceLineNo">208</span> *<a name="line.208"></a> |
| <span class="sourceLineNo">209</span> * @return list of region server addresses or an empty list if the slave is unavailable<a name="line.209"></a> |
| <span class="sourceLineNo">210</span> */<a name="line.210"></a> |
| <span class="sourceLineNo">211</span> protected List<ServerName> fetchSlavesAddresses() {<a name="line.211"></a> |
| <span class="sourceLineNo">212</span> List<String> children = null;<a name="line.212"></a> |
| <span class="sourceLineNo">213</span> try {<a name="line.213"></a> |
| <span class="sourceLineNo">214</span> children = ZKUtil.listChildrenAndWatchForNewChildren(zkw, zkw.getZNodePaths().rsZNode);<a name="line.214"></a> |
| <span class="sourceLineNo">215</span> } catch (KeeperException ke) {<a name="line.215"></a> |
| <span class="sourceLineNo">216</span> if (LOG.isDebugEnabled()) {<a name="line.216"></a> |
| <span class="sourceLineNo">217</span> LOG.debug("Fetch slaves addresses failed", ke);<a name="line.217"></a> |
| <span class="sourceLineNo">218</span> }<a name="line.218"></a> |
| <span class="sourceLineNo">219</span> reconnect(ke);<a name="line.219"></a> |
| <span class="sourceLineNo">220</span> }<a name="line.220"></a> |
| <span class="sourceLineNo">221</span> if (children == null) {<a name="line.221"></a> |
| <span class="sourceLineNo">222</span> return Collections.emptyList();<a name="line.222"></a> |
| <span class="sourceLineNo">223</span> }<a name="line.223"></a> |
| <span class="sourceLineNo">224</span> List<ServerName> addresses = new ArrayList<>(children.size());<a name="line.224"></a> |
| <span class="sourceLineNo">225</span> for (String child : children) {<a name="line.225"></a> |
| <span class="sourceLineNo">226</span> addresses.add(ServerName.parseServerName(child));<a name="line.226"></a> |
| <span class="sourceLineNo">227</span> }<a name="line.227"></a> |
| <span class="sourceLineNo">228</span> return addresses;<a name="line.228"></a> |
| <span class="sourceLineNo">229</span> }<a name="line.229"></a> |
| <span class="sourceLineNo">230</span><a name="line.230"></a> |
| <span class="sourceLineNo">231</span> protected synchronized void chooseSinks() {<a name="line.231"></a> |
| <span class="sourceLineNo">232</span> List<ServerName> slaveAddresses = fetchSlavesAddresses();<a name="line.232"></a> |
| <span class="sourceLineNo">233</span> if (slaveAddresses.isEmpty()) {<a name="line.233"></a> |
| <span class="sourceLineNo">234</span> LOG.warn("No sinks available at peer. Will not be able to replicate");<a name="line.234"></a> |
| <span class="sourceLineNo">235</span> }<a name="line.235"></a> |
| <span class="sourceLineNo">236</span> Collections.shuffle(slaveAddresses, ThreadLocalRandom.current());<a name="line.236"></a> |
| <span class="sourceLineNo">237</span> int numSinks = (int) Math.ceil(slaveAddresses.size() * ratio);<a name="line.237"></a> |
| <span class="sourceLineNo">238</span> this.sinkServers = slaveAddresses.subList(0, numSinks);<a name="line.238"></a> |
| <span class="sourceLineNo">239</span> badReportCounts.clear();<a name="line.239"></a> |
| <span class="sourceLineNo">240</span> }<a name="line.240"></a> |
| <span class="sourceLineNo">241</span><a name="line.241"></a> |
| <span class="sourceLineNo">242</span> protected synchronized int getNumSinks() {<a name="line.242"></a> |
| <span class="sourceLineNo">243</span> return sinkServers.size();<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> /**<a name="line.246"></a> |
| <span class="sourceLineNo">247</span> * Get a randomly-chosen replication sink to replicate to.<a name="line.247"></a> |
| <span class="sourceLineNo">248</span> * @return a replication sink to replicate to<a name="line.248"></a> |
| <span class="sourceLineNo">249</span> */<a name="line.249"></a> |
| <span class="sourceLineNo">250</span> protected synchronized SinkPeer getReplicationSink() throws IOException {<a name="line.250"></a> |
| <span class="sourceLineNo">251</span> if (sinkServers.isEmpty()) {<a name="line.251"></a> |
| <span class="sourceLineNo">252</span> LOG.info("Current list of sinks is out of date or empty, updating");<a name="line.252"></a> |
| <span class="sourceLineNo">253</span> chooseSinks();<a name="line.253"></a> |
| <span class="sourceLineNo">254</span> }<a name="line.254"></a> |
| <span class="sourceLineNo">255</span> if (sinkServers.isEmpty()) {<a name="line.255"></a> |
| <span class="sourceLineNo">256</span> throw new IOException("No replication sinks are available");<a name="line.256"></a> |
| <span class="sourceLineNo">257</span> }<a name="line.257"></a> |
| <span class="sourceLineNo">258</span> ServerName serverName =<a name="line.258"></a> |
| <span class="sourceLineNo">259</span> sinkServers.get(ThreadLocalRandom.current().nextInt(sinkServers.size()));<a name="line.259"></a> |
| <span class="sourceLineNo">260</span> return new SinkPeer(serverName, conn.getRegionServerAdmin(serverName));<a name="line.260"></a> |
| <span class="sourceLineNo">261</span> }<a name="line.261"></a> |
| <span class="sourceLineNo">262</span><a name="line.262"></a> |
| <span class="sourceLineNo">263</span> /**<a name="line.263"></a> |
| <span class="sourceLineNo">264</span> * Report a {@code SinkPeer} as being bad (i.e. an attempt to replicate to it<a name="line.264"></a> |
| <span class="sourceLineNo">265</span> * failed). If a single SinkPeer is reported as bad more than<a name="line.265"></a> |
| <span class="sourceLineNo">266</span> * replication.bad.sink.threshold times, it will be removed<a name="line.266"></a> |
| <span class="sourceLineNo">267</span> * from the pool of potential replication targets.<a name="line.267"></a> |
| <span class="sourceLineNo">268</span> *<a name="line.268"></a> |
| <span class="sourceLineNo">269</span> * @param sinkPeer The SinkPeer that had a failed replication attempt on it<a name="line.269"></a> |
| <span class="sourceLineNo">270</span> */<a name="line.270"></a> |
| <span class="sourceLineNo">271</span> protected synchronized void reportBadSink(SinkPeer sinkPeer) {<a name="line.271"></a> |
| <span class="sourceLineNo">272</span> ServerName serverName = sinkPeer.getServerName();<a name="line.272"></a> |
| <span class="sourceLineNo">273</span> int badReportCount = badReportCounts.compute(serverName, (k, v) -> v == null ? 1 : v + 1);<a name="line.273"></a> |
| <span class="sourceLineNo">274</span> if (badReportCount > badSinkThreshold) {<a name="line.274"></a> |
| <span class="sourceLineNo">275</span> this.sinkServers.remove(serverName);<a name="line.275"></a> |
| <span class="sourceLineNo">276</span> if (sinkServers.isEmpty()) {<a name="line.276"></a> |
| <span class="sourceLineNo">277</span> chooseSinks();<a name="line.277"></a> |
| <span class="sourceLineNo">278</span> }<a name="line.278"></a> |
| <span class="sourceLineNo">279</span> }<a name="line.279"></a> |
| <span class="sourceLineNo">280</span> }<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> * Report that a {@code SinkPeer} successfully replicated a chunk of data.<a name="line.283"></a> |
| <span class="sourceLineNo">284</span> *<a name="line.284"></a> |
| <span class="sourceLineNo">285</span> * @param sinkPeer<a name="line.285"></a> |
| <span class="sourceLineNo">286</span> * The SinkPeer that had a failed replication attempt on it<a name="line.286"></a> |
| <span class="sourceLineNo">287</span> */<a name="line.287"></a> |
| <span class="sourceLineNo">288</span> protected synchronized void reportSinkSuccess(SinkPeer sinkPeer) {<a name="line.288"></a> |
| <span class="sourceLineNo">289</span> badReportCounts.remove(sinkPeer.getServerName());<a name="line.289"></a> |
| <span class="sourceLineNo">290</span> }<a name="line.290"></a> |
| <span class="sourceLineNo">291</span><a name="line.291"></a> |
| <span class="sourceLineNo">292</span> @VisibleForTesting<a name="line.292"></a> |
| <span class="sourceLineNo">293</span> List<ServerName> getSinkServers() {<a name="line.293"></a> |
| <span class="sourceLineNo">294</span> return sinkServers;<a name="line.294"></a> |
| <span class="sourceLineNo">295</span> }<a name="line.295"></a> |
| <span class="sourceLineNo">296</span><a name="line.296"></a> |
| <span class="sourceLineNo">297</span> /**<a name="line.297"></a> |
| <span class="sourceLineNo">298</span> * Tracks changes to the list of region servers in a peer's cluster.<a name="line.298"></a> |
| <span class="sourceLineNo">299</span> */<a name="line.299"></a> |
| <span class="sourceLineNo">300</span> public static class PeerRegionServerListener extends ZKListener {<a name="line.300"></a> |
| <span class="sourceLineNo">301</span><a name="line.301"></a> |
| <span class="sourceLineNo">302</span> private final HBaseReplicationEndpoint replicationEndpoint;<a name="line.302"></a> |
| <span class="sourceLineNo">303</span> private final String regionServerListNode;<a name="line.303"></a> |
| <span class="sourceLineNo">304</span><a name="line.304"></a> |
| <span class="sourceLineNo">305</span> public PeerRegionServerListener(HBaseReplicationEndpoint endpoint) {<a name="line.305"></a> |
| <span class="sourceLineNo">306</span> super(endpoint.zkw);<a name="line.306"></a> |
| <span class="sourceLineNo">307</span> this.replicationEndpoint = endpoint;<a name="line.307"></a> |
| <span class="sourceLineNo">308</span> this.regionServerListNode = endpoint.zkw.getZNodePaths().rsZNode;<a name="line.308"></a> |
| <span class="sourceLineNo">309</span> }<a name="line.309"></a> |
| <span class="sourceLineNo">310</span><a name="line.310"></a> |
| <span class="sourceLineNo">311</span> @Override<a name="line.311"></a> |
| <span class="sourceLineNo">312</span> public synchronized void nodeChildrenChanged(String path) {<a name="line.312"></a> |
| <span class="sourceLineNo">313</span> if (path.equals(regionServerListNode)) {<a name="line.313"></a> |
| <span class="sourceLineNo">314</span> LOG.info("Detected change to peer region servers, fetching updated list");<a name="line.314"></a> |
| <span class="sourceLineNo">315</span> replicationEndpoint.chooseSinks();<a name="line.315"></a> |
| <span class="sourceLineNo">316</span> }<a name="line.316"></a> |
| <span class="sourceLineNo">317</span> }<a name="line.317"></a> |
| <span class="sourceLineNo">318</span> }<a name="line.318"></a> |
| <span class="sourceLineNo">319</span><a name="line.319"></a> |
| <span class="sourceLineNo">320</span> /**<a name="line.320"></a> |
| <span class="sourceLineNo">321</span> * Wraps a replication region server sink to provide the ability to identify it.<a name="line.321"></a> |
| <span class="sourceLineNo">322</span> */<a name="line.322"></a> |
| <span class="sourceLineNo">323</span> public static class SinkPeer {<a name="line.323"></a> |
| <span class="sourceLineNo">324</span> private ServerName serverName;<a name="line.324"></a> |
| <span class="sourceLineNo">325</span> private AsyncRegionServerAdmin regionServer;<a name="line.325"></a> |
| <span class="sourceLineNo">326</span><a name="line.326"></a> |
| <span class="sourceLineNo">327</span> public SinkPeer(ServerName serverName, AsyncRegionServerAdmin regionServer) {<a name="line.327"></a> |
| <span class="sourceLineNo">328</span> this.serverName = serverName;<a name="line.328"></a> |
| <span class="sourceLineNo">329</span> this.regionServer = regionServer;<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> ServerName getServerName() {<a name="line.332"></a> |
| <span class="sourceLineNo">333</span> return serverName;<a name="line.333"></a> |
| <span class="sourceLineNo">334</span> }<a name="line.334"></a> |
| <span class="sourceLineNo">335</span><a name="line.335"></a> |
| <span class="sourceLineNo">336</span> public AsyncRegionServerAdmin getRegionServer() {<a name="line.336"></a> |
| <span class="sourceLineNo">337</span> return regionServer;<a name="line.337"></a> |
| <span class="sourceLineNo">338</span> }<a name="line.338"></a> |
| <span class="sourceLineNo">339</span> }<a name="line.339"></a> |
| <span class="sourceLineNo">340</span>}<a name="line.340"></a> |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| </pre> |
| </div> |
| </body> |
| </html> |