blob: 0e38344f2c4e000f43db6135c597a3faee769cd2 [file] [log] [blame]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html lang="en">
<head>
<title>Source code</title>
<link rel="stylesheet" type="text/css" href="../../../../../../stylesheet.css" title="Style">
</head>
<body>
<div class="sourceContainer">
<pre><span class="sourceLineNo">001</span>/**<a name="line.1"></a>
<span class="sourceLineNo">002</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a>
<span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a>
<span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a>
<span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a>
<span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a>
<span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a>
<span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a>
<span class="sourceLineNo">009</span> *<a name="line.9"></a>
<span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a>
<span class="sourceLineNo">011</span> *<a name="line.11"></a>
<span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a>
<span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a>
<span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a>
<span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a>
<span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a>
<span class="sourceLineNo">017</span> */<a name="line.17"></a>
<span class="sourceLineNo">018</span>package org.apache.hadoop.hbase.procedure;<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.util.Arrays;<a name="line.21"></a>
<span class="sourceLineNo">022</span>import java.util.List;<a name="line.22"></a>
<span class="sourceLineNo">023</span><a name="line.23"></a>
<span class="sourceLineNo">024</span>import org.apache.hadoop.hbase.zookeeper.ZKWatcher;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.errorhandling.ForeignException;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.zookeeper.ZKUtil;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.zookeeper.ZNodePaths;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.zookeeper.KeeperException;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.slf4j.Logger;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.slf4j.LoggerFactory;<a name="line.33"></a>
<span class="sourceLineNo">034</span><a name="line.34"></a>
<span class="sourceLineNo">035</span>/**<a name="line.35"></a>
<span class="sourceLineNo">036</span> * ZooKeeper based controller for a procedure member.<a name="line.36"></a>
<span class="sourceLineNo">037</span> * &lt;p&gt;<a name="line.37"></a>
<span class="sourceLineNo">038</span> * There can only be one {@link ZKProcedureMemberRpcs} per procedure type per member,<a name="line.38"></a>
<span class="sourceLineNo">039</span> * since each procedure type is bound to a single set of znodes. You can have multiple<a name="line.39"></a>
<span class="sourceLineNo">040</span> * {@link ZKProcedureMemberRpcs} on the same server, each serving a different member<a name="line.40"></a>
<span class="sourceLineNo">041</span> * name, but each individual rpcs is still bound to a single member name (and since they are<a name="line.41"></a>
<span class="sourceLineNo">042</span> * used to determine global progress, its important to not get this wrong).<a name="line.42"></a>
<span class="sourceLineNo">043</span> * &lt;p&gt;<a name="line.43"></a>
<span class="sourceLineNo">044</span> * To make this slightly more confusing, you can run multiple, concurrent procedures at the same<a name="line.44"></a>
<span class="sourceLineNo">045</span> * time (as long as they have different types), from the same controller, but the same node name<a name="line.45"></a>
<span class="sourceLineNo">046</span> * must be used for each procedure (though there is no conflict between the two procedure as long<a name="line.46"></a>
<span class="sourceLineNo">047</span> * as they have distinct names).<a name="line.47"></a>
<span class="sourceLineNo">048</span> * &lt;p&gt;<a name="line.48"></a>
<span class="sourceLineNo">049</span> * There is no real error recovery with this mechanism currently -- if any the coordinator fails,<a name="line.49"></a>
<span class="sourceLineNo">050</span> * its re-initialization will delete the znodes and require all in progress subprocedures to start<a name="line.50"></a>
<span class="sourceLineNo">051</span> * anew.<a name="line.51"></a>
<span class="sourceLineNo">052</span> */<a name="line.52"></a>
<span class="sourceLineNo">053</span>@InterfaceAudience.Private<a name="line.53"></a>
<span class="sourceLineNo">054</span>public class ZKProcedureMemberRpcs implements ProcedureMemberRpcs {<a name="line.54"></a>
<span class="sourceLineNo">055</span> private static final Logger LOG = LoggerFactory.getLogger(ZKProcedureMemberRpcs.class);<a name="line.55"></a>
<span class="sourceLineNo">056</span><a name="line.56"></a>
<span class="sourceLineNo">057</span> private final ZKProcedureUtil zkController;<a name="line.57"></a>
<span class="sourceLineNo">058</span><a name="line.58"></a>
<span class="sourceLineNo">059</span> protected ProcedureMember member;<a name="line.59"></a>
<span class="sourceLineNo">060</span> private String memberName;<a name="line.60"></a>
<span class="sourceLineNo">061</span><a name="line.61"></a>
<span class="sourceLineNo">062</span> /**<a name="line.62"></a>
<span class="sourceLineNo">063</span> * Must call {@link #start(String, ProcedureMember)} before this can be used.<a name="line.63"></a>
<span class="sourceLineNo">064</span> * @param watcher {@link ZKWatcher} to be owned by &lt;tt&gt;this&lt;/tt&gt;. Closed via<a name="line.64"></a>
<span class="sourceLineNo">065</span> * {@link #close()}.<a name="line.65"></a>
<span class="sourceLineNo">066</span> * @param procType name of the znode describing the procedure type<a name="line.66"></a>
<span class="sourceLineNo">067</span> * @throws KeeperException if we can't reach zookeeper<a name="line.67"></a>
<span class="sourceLineNo">068</span> */<a name="line.68"></a>
<span class="sourceLineNo">069</span> public ZKProcedureMemberRpcs(final ZKWatcher watcher, final String procType)<a name="line.69"></a>
<span class="sourceLineNo">070</span> throws KeeperException {<a name="line.70"></a>
<span class="sourceLineNo">071</span> this.zkController = new ZKProcedureUtil(watcher, procType) {<a name="line.71"></a>
<span class="sourceLineNo">072</span> @Override<a name="line.72"></a>
<span class="sourceLineNo">073</span> public void nodeCreated(String path) {<a name="line.73"></a>
<span class="sourceLineNo">074</span> if (!isInProcedurePath(path)) {<a name="line.74"></a>
<span class="sourceLineNo">075</span> return;<a name="line.75"></a>
<span class="sourceLineNo">076</span> }<a name="line.76"></a>
<span class="sourceLineNo">077</span><a name="line.77"></a>
<span class="sourceLineNo">078</span> LOG.info("Received created event:" + path);<a name="line.78"></a>
<span class="sourceLineNo">079</span> // if it is a simple start/end/abort then we just rewatch the node<a name="line.79"></a>
<span class="sourceLineNo">080</span> if (isAcquiredNode(path)) {<a name="line.80"></a>
<span class="sourceLineNo">081</span> waitForNewProcedures();<a name="line.81"></a>
<span class="sourceLineNo">082</span> return;<a name="line.82"></a>
<span class="sourceLineNo">083</span> } else if (isAbortNode(path)) {<a name="line.83"></a>
<span class="sourceLineNo">084</span> watchForAbortedProcedures();<a name="line.84"></a>
<span class="sourceLineNo">085</span> return;<a name="line.85"></a>
<span class="sourceLineNo">086</span> }<a name="line.86"></a>
<span class="sourceLineNo">087</span> String parent = ZKUtil.getParent(path);<a name="line.87"></a>
<span class="sourceLineNo">088</span> // if its the end barrier, the procedure can be completed<a name="line.88"></a>
<span class="sourceLineNo">089</span> if (isReachedNode(parent)) {<a name="line.89"></a>
<span class="sourceLineNo">090</span> receivedReachedGlobalBarrier(path);<a name="line.90"></a>
<span class="sourceLineNo">091</span> return;<a name="line.91"></a>
<span class="sourceLineNo">092</span> } else if (isAbortNode(parent)) {<a name="line.92"></a>
<span class="sourceLineNo">093</span> abort(path);<a name="line.93"></a>
<span class="sourceLineNo">094</span> return;<a name="line.94"></a>
<span class="sourceLineNo">095</span> } else if (isAcquiredNode(parent)) {<a name="line.95"></a>
<span class="sourceLineNo">096</span> startNewSubprocedure(path);<a name="line.96"></a>
<span class="sourceLineNo">097</span> } else {<a name="line.97"></a>
<span class="sourceLineNo">098</span> LOG.debug("Ignoring created notification for node:" + path);<a name="line.98"></a>
<span class="sourceLineNo">099</span> }<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 nodeChildrenChanged(String path) {<a name="line.103"></a>
<span class="sourceLineNo">104</span> if (path.equals(this.acquiredZnode)) {<a name="line.104"></a>
<span class="sourceLineNo">105</span> LOG.info("Received procedure start children changed event: " + path);<a name="line.105"></a>
<span class="sourceLineNo">106</span> waitForNewProcedures();<a name="line.106"></a>
<span class="sourceLineNo">107</span> } else if (path.equals(this.abortZnode)) {<a name="line.107"></a>
<span class="sourceLineNo">108</span> LOG.info("Received procedure abort children changed event: " + path);<a name="line.108"></a>
<span class="sourceLineNo">109</span> watchForAbortedProcedures();<a name="line.109"></a>
<span class="sourceLineNo">110</span> }<a name="line.110"></a>
<span class="sourceLineNo">111</span> }<a name="line.111"></a>
<span class="sourceLineNo">112</span> };<a name="line.112"></a>
<span class="sourceLineNo">113</span> }<a name="line.113"></a>
<span class="sourceLineNo">114</span><a name="line.114"></a>
<span class="sourceLineNo">115</span> public ZKProcedureUtil getZkController() {<a name="line.115"></a>
<span class="sourceLineNo">116</span> return zkController;<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> @Override<a name="line.119"></a>
<span class="sourceLineNo">120</span> public String getMemberName() {<a name="line.120"></a>
<span class="sourceLineNo">121</span> return memberName;<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 name="line.124"></a>
<span class="sourceLineNo">125</span> * Pass along the procedure global barrier notification to any listeners<a name="line.125"></a>
<span class="sourceLineNo">126</span> * @param path full znode path that cause the notification<a name="line.126"></a>
<span class="sourceLineNo">127</span> */<a name="line.127"></a>
<span class="sourceLineNo">128</span> private void receivedReachedGlobalBarrier(String path) {<a name="line.128"></a>
<span class="sourceLineNo">129</span> LOG.debug("Received reached global barrier:" + path);<a name="line.129"></a>
<span class="sourceLineNo">130</span> String procName = ZKUtil.getNodeName(path);<a name="line.130"></a>
<span class="sourceLineNo">131</span> this.member.receivedReachedGlobalBarrier(procName);<a name="line.131"></a>
<span class="sourceLineNo">132</span> }<a name="line.132"></a>
<span class="sourceLineNo">133</span><a name="line.133"></a>
<span class="sourceLineNo">134</span> private void watchForAbortedProcedures() {<a name="line.134"></a>
<span class="sourceLineNo">135</span> LOG.debug("Checking for aborted procedures on node: '" + zkController.getAbortZnode() + "'");<a name="line.135"></a>
<span class="sourceLineNo">136</span> try {<a name="line.136"></a>
<span class="sourceLineNo">137</span> // this is the list of the currently aborted procedues<a name="line.137"></a>
<span class="sourceLineNo">138</span> List&lt;String&gt; children = ZKUtil.listChildrenAndWatchForNewChildren(zkController.getWatcher(),<a name="line.138"></a>
<span class="sourceLineNo">139</span> zkController.getAbortZnode());<a name="line.139"></a>
<span class="sourceLineNo">140</span> if (children == null || children.isEmpty()) {<a name="line.140"></a>
<span class="sourceLineNo">141</span> return;<a name="line.141"></a>
<span class="sourceLineNo">142</span> }<a name="line.142"></a>
<span class="sourceLineNo">143</span> for (String node : children) {<a name="line.143"></a>
<span class="sourceLineNo">144</span> String abortNode = ZNodePaths.joinZNode(zkController.getAbortZnode(), node);<a name="line.144"></a>
<span class="sourceLineNo">145</span> abort(abortNode);<a name="line.145"></a>
<span class="sourceLineNo">146</span> }<a name="line.146"></a>
<span class="sourceLineNo">147</span> } catch (KeeperException e) {<a name="line.147"></a>
<span class="sourceLineNo">148</span> member.controllerConnectionFailure("Failed to list children for abort node:"<a name="line.148"></a>
<span class="sourceLineNo">149</span> + zkController.getAbortZnode(), e, null);<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> private void waitForNewProcedures() {<a name="line.153"></a>
<span class="sourceLineNo">154</span> // watch for new procedues that we need to start subprocedures for<a name="line.154"></a>
<span class="sourceLineNo">155</span> LOG.debug("Looking for new procedures under znode:'" + zkController.getAcquiredBarrier() + "'");<a name="line.155"></a>
<span class="sourceLineNo">156</span> List&lt;String&gt; runningProcedures = null;<a name="line.156"></a>
<span class="sourceLineNo">157</span> try {<a name="line.157"></a>
<span class="sourceLineNo">158</span> runningProcedures = ZKUtil.listChildrenAndWatchForNewChildren(zkController.getWatcher(),<a name="line.158"></a>
<span class="sourceLineNo">159</span> zkController.getAcquiredBarrier());<a name="line.159"></a>
<span class="sourceLineNo">160</span> if (runningProcedures == null) {<a name="line.160"></a>
<span class="sourceLineNo">161</span> LOG.debug("No running procedures.");<a name="line.161"></a>
<span class="sourceLineNo">162</span> return;<a name="line.162"></a>
<span class="sourceLineNo">163</span> }<a name="line.163"></a>
<span class="sourceLineNo">164</span> } catch (KeeperException e) {<a name="line.164"></a>
<span class="sourceLineNo">165</span> member.controllerConnectionFailure("General failure when watching for new procedures",<a name="line.165"></a>
<span class="sourceLineNo">166</span> e, null);<a name="line.166"></a>
<span class="sourceLineNo">167</span> }<a name="line.167"></a>
<span class="sourceLineNo">168</span> if (runningProcedures == null) {<a name="line.168"></a>
<span class="sourceLineNo">169</span> LOG.debug("No running procedures.");<a name="line.169"></a>
<span class="sourceLineNo">170</span> return;<a name="line.170"></a>
<span class="sourceLineNo">171</span> }<a name="line.171"></a>
<span class="sourceLineNo">172</span> for (String procName : runningProcedures) {<a name="line.172"></a>
<span class="sourceLineNo">173</span> // then read in the procedure information<a name="line.173"></a>
<span class="sourceLineNo">174</span> String path = ZNodePaths.joinZNode(zkController.getAcquiredBarrier(), procName);<a name="line.174"></a>
<span class="sourceLineNo">175</span> startNewSubprocedure(path);<a name="line.175"></a>
<span class="sourceLineNo">176</span> }<a name="line.176"></a>
<span class="sourceLineNo">177</span> }<a name="line.177"></a>
<span class="sourceLineNo">178</span><a name="line.178"></a>
<span class="sourceLineNo">179</span> /**<a name="line.179"></a>
<span class="sourceLineNo">180</span> * Kick off a new sub-procedure on the listener with the data stored in the passed znode.<a name="line.180"></a>
<span class="sourceLineNo">181</span> * &lt;p&gt;<a name="line.181"></a>
<span class="sourceLineNo">182</span> * Will attempt to create the same procedure multiple times if an procedure znode with the same<a name="line.182"></a>
<span class="sourceLineNo">183</span> * name is created. It is left up the coordinator to ensure this doesn't occur.<a name="line.183"></a>
<span class="sourceLineNo">184</span> * @param path full path to the znode for the procedure to start<a name="line.184"></a>
<span class="sourceLineNo">185</span> */<a name="line.185"></a>
<span class="sourceLineNo">186</span> private synchronized void startNewSubprocedure(String path) {<a name="line.186"></a>
<span class="sourceLineNo">187</span> LOG.debug("Found procedure znode: " + path);<a name="line.187"></a>
<span class="sourceLineNo">188</span> String opName = ZKUtil.getNodeName(path);<a name="line.188"></a>
<span class="sourceLineNo">189</span> // start watching for an abort notification for the procedure<a name="line.189"></a>
<span class="sourceLineNo">190</span> String abortZNode = zkController.getAbortZNode(opName);<a name="line.190"></a>
<span class="sourceLineNo">191</span> try {<a name="line.191"></a>
<span class="sourceLineNo">192</span> if (ZKUtil.watchAndCheckExists(zkController.getWatcher(), abortZNode)) {<a name="line.192"></a>
<span class="sourceLineNo">193</span> LOG.debug("Not starting:" + opName + " because we already have an abort notification.");<a name="line.193"></a>
<span class="sourceLineNo">194</span> return;<a name="line.194"></a>
<span class="sourceLineNo">195</span> }<a name="line.195"></a>
<span class="sourceLineNo">196</span> } catch (KeeperException e) {<a name="line.196"></a>
<span class="sourceLineNo">197</span> member.controllerConnectionFailure("Failed to get the abort znode (" + abortZNode<a name="line.197"></a>
<span class="sourceLineNo">198</span> + ") for procedure :" + opName, e, opName);<a name="line.198"></a>
<span class="sourceLineNo">199</span> return;<a name="line.199"></a>
<span class="sourceLineNo">200</span> }<a name="line.200"></a>
<span class="sourceLineNo">201</span><a name="line.201"></a>
<span class="sourceLineNo">202</span> // get the data for the procedure<a name="line.202"></a>
<span class="sourceLineNo">203</span> Subprocedure subproc = null;<a name="line.203"></a>
<span class="sourceLineNo">204</span> try {<a name="line.204"></a>
<span class="sourceLineNo">205</span> byte[] data = ZKUtil.getData(zkController.getWatcher(), path);<a name="line.205"></a>
<span class="sourceLineNo">206</span> if (!ProtobufUtil.isPBMagicPrefix(data)) {<a name="line.206"></a>
<span class="sourceLineNo">207</span> String msg = "Data in for starting procedure " + opName +<a name="line.207"></a>
<span class="sourceLineNo">208</span> " is illegally formatted (no pb magic). " +<a name="line.208"></a>
<span class="sourceLineNo">209</span> "Killing the procedure: " + Bytes.toString(data);<a name="line.209"></a>
<span class="sourceLineNo">210</span> LOG.error(msg);<a name="line.210"></a>
<span class="sourceLineNo">211</span> throw new IllegalArgumentException(msg);<a name="line.211"></a>
<span class="sourceLineNo">212</span> }<a name="line.212"></a>
<span class="sourceLineNo">213</span> LOG.debug("start proc data length is " + data.length);<a name="line.213"></a>
<span class="sourceLineNo">214</span> data = Arrays.copyOfRange(data, ProtobufUtil.lengthOfPBMagic(), data.length);<a name="line.214"></a>
<span class="sourceLineNo">215</span> LOG.debug("Found data for znode:" + path);<a name="line.215"></a>
<span class="sourceLineNo">216</span> subproc = member.createSubprocedure(opName, data);<a name="line.216"></a>
<span class="sourceLineNo">217</span> member.submitSubprocedure(subproc);<a name="line.217"></a>
<span class="sourceLineNo">218</span> } catch (IllegalArgumentException iae ) {<a name="line.218"></a>
<span class="sourceLineNo">219</span> LOG.error("Illegal argument exception", iae);<a name="line.219"></a>
<span class="sourceLineNo">220</span> sendMemberAborted(subproc, new ForeignException(getMemberName(), iae));<a name="line.220"></a>
<span class="sourceLineNo">221</span> } catch (IllegalStateException ise) {<a name="line.221"></a>
<span class="sourceLineNo">222</span> LOG.error("Illegal state exception ", ise);<a name="line.222"></a>
<span class="sourceLineNo">223</span> sendMemberAborted(subproc, new ForeignException(getMemberName(), ise));<a name="line.223"></a>
<span class="sourceLineNo">224</span> } catch (KeeperException e) {<a name="line.224"></a>
<span class="sourceLineNo">225</span> member.controllerConnectionFailure("Failed to get data for new procedure:" + opName,<a name="line.225"></a>
<span class="sourceLineNo">226</span> e, opName);<a name="line.226"></a>
<span class="sourceLineNo">227</span> } catch (InterruptedException e) {<a name="line.227"></a>
<span class="sourceLineNo">228</span> member.controllerConnectionFailure("Failed to get data for new procedure:" + opName,<a name="line.228"></a>
<span class="sourceLineNo">229</span> e, opName);<a name="line.229"></a>
<span class="sourceLineNo">230</span> Thread.currentThread().interrupt();<a name="line.230"></a>
<span class="sourceLineNo">231</span> }<a name="line.231"></a>
<span class="sourceLineNo">232</span> }<a name="line.232"></a>
<span class="sourceLineNo">233</span><a name="line.233"></a>
<span class="sourceLineNo">234</span> /**<a name="line.234"></a>
<span class="sourceLineNo">235</span> * This attempts to create an acquired state znode for the procedure (snapshot name).<a name="line.235"></a>
<span class="sourceLineNo">236</span> *<a name="line.236"></a>
<span class="sourceLineNo">237</span> * It then looks for the reached znode to trigger in-barrier execution. If not present we<a name="line.237"></a>
<span class="sourceLineNo">238</span> * have a watcher, if present then trigger the in-barrier action.<a name="line.238"></a>
<span class="sourceLineNo">239</span> */<a name="line.239"></a>
<span class="sourceLineNo">240</span> @Override<a name="line.240"></a>
<span class="sourceLineNo">241</span> public void sendMemberAcquired(Subprocedure sub) throws IOException {<a name="line.241"></a>
<span class="sourceLineNo">242</span> String procName = sub.getName();<a name="line.242"></a>
<span class="sourceLineNo">243</span> try {<a name="line.243"></a>
<span class="sourceLineNo">244</span> LOG.debug("Member: '" + memberName + "' joining acquired barrier for procedure (" + procName<a name="line.244"></a>
<span class="sourceLineNo">245</span> + ") in zk");<a name="line.245"></a>
<span class="sourceLineNo">246</span> String acquiredZNode = ZNodePaths.joinZNode(ZKProcedureUtil.getAcquireBarrierNode(<a name="line.246"></a>
<span class="sourceLineNo">247</span> zkController, procName), memberName);<a name="line.247"></a>
<span class="sourceLineNo">248</span> ZKUtil.createAndFailSilent(zkController.getWatcher(), acquiredZNode);<a name="line.248"></a>
<span class="sourceLineNo">249</span><a name="line.249"></a>
<span class="sourceLineNo">250</span> // watch for the complete node for this snapshot<a name="line.250"></a>
<span class="sourceLineNo">251</span> String reachedBarrier = zkController.getReachedBarrierNode(procName);<a name="line.251"></a>
<span class="sourceLineNo">252</span> LOG.debug("Watch for global barrier reached:" + reachedBarrier);<a name="line.252"></a>
<span class="sourceLineNo">253</span> if (ZKUtil.watchAndCheckExists(zkController.getWatcher(), reachedBarrier)) {<a name="line.253"></a>
<span class="sourceLineNo">254</span> receivedReachedGlobalBarrier(reachedBarrier);<a name="line.254"></a>
<span class="sourceLineNo">255</span> }<a name="line.255"></a>
<span class="sourceLineNo">256</span> } catch (KeeperException e) {<a name="line.256"></a>
<span class="sourceLineNo">257</span> member.controllerConnectionFailure("Failed to acquire barrier for procedure: "<a name="line.257"></a>
<span class="sourceLineNo">258</span> + procName + " and member: " + memberName, e, procName);<a name="line.258"></a>
<span class="sourceLineNo">259</span> }<a name="line.259"></a>
<span class="sourceLineNo">260</span> }<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> * This acts as the ack for a completed procedure<a name="line.263"></a>
<span class="sourceLineNo">264</span> */<a name="line.264"></a>
<span class="sourceLineNo">265</span> @Override<a name="line.265"></a>
<span class="sourceLineNo">266</span> public void sendMemberCompleted(Subprocedure sub, byte[] data) throws IOException {<a name="line.266"></a>
<span class="sourceLineNo">267</span> String procName = sub.getName();<a name="line.267"></a>
<span class="sourceLineNo">268</span> LOG.debug("Marking procedure '" + procName + "' completed for member '" + memberName<a name="line.268"></a>
<span class="sourceLineNo">269</span> + "' in zk");<a name="line.269"></a>
<span class="sourceLineNo">270</span> String joinPath =<a name="line.270"></a>
<span class="sourceLineNo">271</span> ZNodePaths.joinZNode(zkController.getReachedBarrierNode(procName), memberName);<a name="line.271"></a>
<span class="sourceLineNo">272</span> // ProtobufUtil.prependPBMagic does not take care of null<a name="line.272"></a>
<span class="sourceLineNo">273</span> if (data == null) {<a name="line.273"></a>
<span class="sourceLineNo">274</span> data = new byte[0];<a name="line.274"></a>
<span class="sourceLineNo">275</span> }<a name="line.275"></a>
<span class="sourceLineNo">276</span> try {<a name="line.276"></a>
<span class="sourceLineNo">277</span> ZKUtil.createAndFailSilent(zkController.getWatcher(), joinPath,<a name="line.277"></a>
<span class="sourceLineNo">278</span> ProtobufUtil.prependPBMagic(data));<a name="line.278"></a>
<span class="sourceLineNo">279</span> } catch (KeeperException e) {<a name="line.279"></a>
<span class="sourceLineNo">280</span> member.controllerConnectionFailure("Failed to post zk node:" + joinPath<a name="line.280"></a>
<span class="sourceLineNo">281</span> + " to join procedure barrier.", e, procName);<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> * This should be called by the member and should write a serialized root cause exception as<a name="line.286"></a>
<span class="sourceLineNo">287</span> * to the abort znode.<a name="line.287"></a>
<span class="sourceLineNo">288</span> */<a name="line.288"></a>
<span class="sourceLineNo">289</span> @Override<a name="line.289"></a>
<span class="sourceLineNo">290</span> public void sendMemberAborted(Subprocedure sub, ForeignException ee) {<a name="line.290"></a>
<span class="sourceLineNo">291</span> if (sub == null) {<a name="line.291"></a>
<span class="sourceLineNo">292</span> LOG.error("Failed due to null subprocedure", ee);<a name="line.292"></a>
<span class="sourceLineNo">293</span> return;<a name="line.293"></a>
<span class="sourceLineNo">294</span> }<a name="line.294"></a>
<span class="sourceLineNo">295</span> String procName = sub.getName();<a name="line.295"></a>
<span class="sourceLineNo">296</span> LOG.debug("Aborting procedure (" + procName + ") in zk");<a name="line.296"></a>
<span class="sourceLineNo">297</span> String procAbortZNode = zkController.getAbortZNode(procName);<a name="line.297"></a>
<span class="sourceLineNo">298</span> try {<a name="line.298"></a>
<span class="sourceLineNo">299</span> String source = (ee.getSource() == null) ? memberName: ee.getSource();<a name="line.299"></a>
<span class="sourceLineNo">300</span> byte[] errorInfo = ProtobufUtil.prependPBMagic(ForeignException.serialize(source, ee));<a name="line.300"></a>
<span class="sourceLineNo">301</span> ZKUtil.createAndFailSilent(zkController.getWatcher(), procAbortZNode, errorInfo);<a name="line.301"></a>
<span class="sourceLineNo">302</span> LOG.debug("Finished creating abort znode:" + procAbortZNode);<a name="line.302"></a>
<span class="sourceLineNo">303</span> } catch (KeeperException e) {<a name="line.303"></a>
<span class="sourceLineNo">304</span> // possible that we get this error for the procedure if we already reset the zk state, but in<a name="line.304"></a>
<span class="sourceLineNo">305</span> // that case we should still get an error for that procedure anyways<a name="line.305"></a>
<span class="sourceLineNo">306</span> zkController.logZKTree(zkController.getBaseZnode());<a name="line.306"></a>
<span class="sourceLineNo">307</span> member.controllerConnectionFailure("Failed to post zk node:" + procAbortZNode<a name="line.307"></a>
<span class="sourceLineNo">308</span> + " to abort procedure", e, procName);<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><a name="line.311"></a>
<span class="sourceLineNo">312</span> /**<a name="line.312"></a>
<span class="sourceLineNo">313</span> * Pass along the found abort notification to the listener<a name="line.313"></a>
<span class="sourceLineNo">314</span> * @param abortZNode full znode path to the failed procedure information<a name="line.314"></a>
<span class="sourceLineNo">315</span> */<a name="line.315"></a>
<span class="sourceLineNo">316</span> protected void abort(String abortZNode) {<a name="line.316"></a>
<span class="sourceLineNo">317</span> LOG.debug("Aborting procedure member for znode " + abortZNode);<a name="line.317"></a>
<span class="sourceLineNo">318</span> String opName = ZKUtil.getNodeName(abortZNode);<a name="line.318"></a>
<span class="sourceLineNo">319</span> try {<a name="line.319"></a>
<span class="sourceLineNo">320</span> byte[] data = ZKUtil.getData(zkController.getWatcher(), abortZNode);<a name="line.320"></a>
<span class="sourceLineNo">321</span><a name="line.321"></a>
<span class="sourceLineNo">322</span> // figure out the data we need to pass<a name="line.322"></a>
<span class="sourceLineNo">323</span> ForeignException ee;<a name="line.323"></a>
<span class="sourceLineNo">324</span> try {<a name="line.324"></a>
<span class="sourceLineNo">325</span> if (data == null || data.length == 0) {<a name="line.325"></a>
<span class="sourceLineNo">326</span> // ignore<a name="line.326"></a>
<span class="sourceLineNo">327</span> return;<a name="line.327"></a>
<span class="sourceLineNo">328</span> } else if (!ProtobufUtil.isPBMagicPrefix(data)) {<a name="line.328"></a>
<span class="sourceLineNo">329</span> String msg = "Illegally formatted data in abort node for proc " + opName<a name="line.329"></a>
<span class="sourceLineNo">330</span> + ". Killing the procedure.";<a name="line.330"></a>
<span class="sourceLineNo">331</span> LOG.error(msg);<a name="line.331"></a>
<span class="sourceLineNo">332</span> // we got a remote exception, but we can't describe it so just return exn from here<a name="line.332"></a>
<span class="sourceLineNo">333</span> ee = new ForeignException(getMemberName(), new IllegalArgumentException(msg));<a name="line.333"></a>
<span class="sourceLineNo">334</span> } else {<a name="line.334"></a>
<span class="sourceLineNo">335</span> data = Arrays.copyOfRange(data, ProtobufUtil.lengthOfPBMagic(), data.length);<a name="line.335"></a>
<span class="sourceLineNo">336</span> ee = ForeignException.deserialize(data);<a name="line.336"></a>
<span class="sourceLineNo">337</span> }<a name="line.337"></a>
<span class="sourceLineNo">338</span> } catch (IOException e) {<a name="line.338"></a>
<span class="sourceLineNo">339</span> LOG.warn("Got an error notification for op:" + opName<a name="line.339"></a>
<span class="sourceLineNo">340</span> + " but we can't read the information. Killing the procedure.");<a name="line.340"></a>
<span class="sourceLineNo">341</span> // we got a remote exception, but we can't describe it so just return exn from here<a name="line.341"></a>
<span class="sourceLineNo">342</span> ee = new ForeignException(getMemberName(), e);<a name="line.342"></a>
<span class="sourceLineNo">343</span> }<a name="line.343"></a>
<span class="sourceLineNo">344</span><a name="line.344"></a>
<span class="sourceLineNo">345</span> this.member.receiveAbortProcedure(opName, ee);<a name="line.345"></a>
<span class="sourceLineNo">346</span> } catch (KeeperException e) {<a name="line.346"></a>
<span class="sourceLineNo">347</span> member.controllerConnectionFailure("Failed to get data for abort znode:" + abortZNode<a name="line.347"></a>
<span class="sourceLineNo">348</span> + zkController.getAbortZnode(), e, opName);<a name="line.348"></a>
<span class="sourceLineNo">349</span> } catch (InterruptedException e) {<a name="line.349"></a>
<span class="sourceLineNo">350</span> LOG.warn("abort already in progress", e);<a name="line.350"></a>
<span class="sourceLineNo">351</span> Thread.currentThread().interrupt();<a name="line.351"></a>
<span class="sourceLineNo">352</span> }<a name="line.352"></a>
<span class="sourceLineNo">353</span> }<a name="line.353"></a>
<span class="sourceLineNo">354</span><a name="line.354"></a>
<span class="sourceLineNo">355</span> @Override<a name="line.355"></a>
<span class="sourceLineNo">356</span> public void start(final String memberName, final ProcedureMember listener) {<a name="line.356"></a>
<span class="sourceLineNo">357</span> LOG.debug("Starting procedure member '" + memberName + "'");<a name="line.357"></a>
<span class="sourceLineNo">358</span> this.member = listener;<a name="line.358"></a>
<span class="sourceLineNo">359</span> this.memberName = memberName;<a name="line.359"></a>
<span class="sourceLineNo">360</span> watchForAbortedProcedures();<a name="line.360"></a>
<span class="sourceLineNo">361</span> waitForNewProcedures();<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> @Override<a name="line.364"></a>
<span class="sourceLineNo">365</span> public void close() throws IOException {<a name="line.365"></a>
<span class="sourceLineNo">366</span> zkController.close();<a name="line.366"></a>
<span class="sourceLineNo">367</span> }<a name="line.367"></a>
<span class="sourceLineNo">368</span><a name="line.368"></a>
<span class="sourceLineNo">369</span>}<a name="line.369"></a>
</pre>
</div>
</body>
</html>