blob: a061e313b0684aa4653ec38bb1f7afe61c2a65f4 [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><a name="line.18"></a>
<span class="sourceLineNo">019</span>package org.apache.hadoop.hbase.regionserver.handler;<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.io.InterruptedIOException;<a name="line.22"></a>
<span class="sourceLineNo">023</span>import org.apache.hadoop.conf.Configuration;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import org.apache.hadoop.hbase.HConstants;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import org.apache.hadoop.hbase.Server;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import org.apache.hadoop.hbase.client.AsyncClusterConnection;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import org.apache.hadoop.hbase.executor.EventHandler;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import org.apache.hadoop.hbase.executor.EventType;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import org.apache.hadoop.hbase.regionserver.HRegion;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import org.apache.hadoop.hbase.util.FutureUtils;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import org.apache.hadoop.hbase.util.RetryCounter;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.util.RetryCounterFactory;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.util.ServerRegionReplicaUtil;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.slf4j.Logger;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.slf4j.LoggerFactory;<a name="line.37"></a>
<span class="sourceLineNo">038</span><a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.FlushRegionResponse;<a name="line.39"></a>
<span class="sourceLineNo">040</span><a name="line.40"></a>
<span class="sourceLineNo">041</span>/**<a name="line.41"></a>
<span class="sourceLineNo">042</span> * HBASE-11580: With the async wal approach (HBASE-11568), the edits are not persisted to WAL in<a name="line.42"></a>
<span class="sourceLineNo">043</span> * secondary region replicas. This means that a secondary region replica can serve some edits from<a name="line.43"></a>
<span class="sourceLineNo">044</span> * it's memstore that are still not flushed from primary. We do not want to allow secondary<a name="line.44"></a>
<span class="sourceLineNo">045</span> * region's seqId to go back in time, when this secondary region is opened elsewhere after a<a name="line.45"></a>
<span class="sourceLineNo">046</span> * crash or region move. We will trigger a flush cache in the primary region replica and wait<a name="line.46"></a>
<span class="sourceLineNo">047</span> * for observing a complete flush cycle before marking the region readsEnabled. This handler does<a name="line.47"></a>
<span class="sourceLineNo">048</span> * the flushing of the primary region replica and ensures that regular region opening is not<a name="line.48"></a>
<span class="sourceLineNo">049</span> * blocked while the secondary replica is blocked on flush.<a name="line.49"></a>
<span class="sourceLineNo">050</span> */<a name="line.50"></a>
<span class="sourceLineNo">051</span>@InterfaceAudience.Private<a name="line.51"></a>
<span class="sourceLineNo">052</span>public class RegionReplicaFlushHandler extends EventHandler {<a name="line.52"></a>
<span class="sourceLineNo">053</span> private static final Logger LOG = LoggerFactory.getLogger(RegionReplicaFlushHandler.class);<a name="line.53"></a>
<span class="sourceLineNo">054</span><a name="line.54"></a>
<span class="sourceLineNo">055</span> private final AsyncClusterConnection connection;<a name="line.55"></a>
<span class="sourceLineNo">056</span><a name="line.56"></a>
<span class="sourceLineNo">057</span> private final HRegion region;<a name="line.57"></a>
<span class="sourceLineNo">058</span><a name="line.58"></a>
<span class="sourceLineNo">059</span> public RegionReplicaFlushHandler(Server server, HRegion region) {<a name="line.59"></a>
<span class="sourceLineNo">060</span> super(server, EventType.RS_REGION_REPLICA_FLUSH);<a name="line.60"></a>
<span class="sourceLineNo">061</span> this.connection = server.getAsyncClusterConnection();<a name="line.61"></a>
<span class="sourceLineNo">062</span> this.region = region;<a name="line.62"></a>
<span class="sourceLineNo">063</span> }<a name="line.63"></a>
<span class="sourceLineNo">064</span><a name="line.64"></a>
<span class="sourceLineNo">065</span> @Override<a name="line.65"></a>
<span class="sourceLineNo">066</span> public void process() throws IOException {<a name="line.66"></a>
<span class="sourceLineNo">067</span> triggerFlushInPrimaryRegion(region);<a name="line.67"></a>
<span class="sourceLineNo">068</span> }<a name="line.68"></a>
<span class="sourceLineNo">069</span><a name="line.69"></a>
<span class="sourceLineNo">070</span> @Override<a name="line.70"></a>
<span class="sourceLineNo">071</span> protected void handleException(Throwable t) {<a name="line.71"></a>
<span class="sourceLineNo">072</span> if (t instanceof InterruptedIOException || t instanceof InterruptedException) {<a name="line.72"></a>
<span class="sourceLineNo">073</span> LOG.error("Caught throwable while processing event " + eventType, t);<a name="line.73"></a>
<span class="sourceLineNo">074</span> } else if (t instanceof RuntimeException) {<a name="line.74"></a>
<span class="sourceLineNo">075</span> server.abort("Server aborting", t);<a name="line.75"></a>
<span class="sourceLineNo">076</span> } else {<a name="line.76"></a>
<span class="sourceLineNo">077</span> // something fishy since we cannot flush the primary region until all retries (retries from<a name="line.77"></a>
<span class="sourceLineNo">078</span> // rpc times 35 trigger). We cannot close the region since there is no such mechanism to<a name="line.78"></a>
<span class="sourceLineNo">079</span> // close a region without master triggering it. We just abort the server for now.<a name="line.79"></a>
<span class="sourceLineNo">080</span> server.abort("ServerAborting because an exception was thrown", t);<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><a name="line.83"></a>
<span class="sourceLineNo">084</span> private int getRetriesCount(Configuration conf) {<a name="line.84"></a>
<span class="sourceLineNo">085</span> int numRetries = conf.getInt(HConstants.HBASE_CLIENT_RETRIES_NUMBER,<a name="line.85"></a>
<span class="sourceLineNo">086</span> HConstants.DEFAULT_HBASE_CLIENT_RETRIES_NUMBER);<a name="line.86"></a>
<span class="sourceLineNo">087</span> if (numRetries &gt; 10) {<a name="line.87"></a>
<span class="sourceLineNo">088</span> int mult = conf.getInt(HConstants.HBASE_CLIENT_SERVERSIDE_RETRIES_MULTIPLIER,<a name="line.88"></a>
<span class="sourceLineNo">089</span> HConstants.DEFAULT_HBASE_CLIENT_SERVERSIDE_RETRIES_MULTIPLIER);<a name="line.89"></a>
<span class="sourceLineNo">090</span> numRetries = numRetries / mult; // reset if HRS has multiplied this already<a name="line.90"></a>
<span class="sourceLineNo">091</span> }<a name="line.91"></a>
<span class="sourceLineNo">092</span> return numRetries;<a name="line.92"></a>
<span class="sourceLineNo">093</span> }<a name="line.93"></a>
<span class="sourceLineNo">094</span><a name="line.94"></a>
<span class="sourceLineNo">095</span> void triggerFlushInPrimaryRegion(final HRegion region) throws IOException {<a name="line.95"></a>
<span class="sourceLineNo">096</span> long pause = connection.getConfiguration().getLong(HConstants.HBASE_CLIENT_PAUSE,<a name="line.96"></a>
<span class="sourceLineNo">097</span> HConstants.DEFAULT_HBASE_CLIENT_PAUSE);<a name="line.97"></a>
<span class="sourceLineNo">098</span><a name="line.98"></a>
<span class="sourceLineNo">099</span> int maxAttempts = getRetriesCount(connection.getConfiguration());<a name="line.99"></a>
<span class="sourceLineNo">100</span> RetryCounter counter = new RetryCounterFactory(maxAttempts, (int)pause).create();<a name="line.100"></a>
<span class="sourceLineNo">101</span><a name="line.101"></a>
<span class="sourceLineNo">102</span> if (LOG.isDebugEnabled()) {<a name="line.102"></a>
<span class="sourceLineNo">103</span> LOG.debug("RPC'ing to primary region replica " +<a name="line.103"></a>
<span class="sourceLineNo">104</span> ServerRegionReplicaUtil.getRegionInfoForDefaultReplica(region.getRegionInfo()) + " from " +<a name="line.104"></a>
<span class="sourceLineNo">105</span> region.getRegionInfo() + " to trigger FLUSH");<a name="line.105"></a>
<span class="sourceLineNo">106</span> }<a name="line.106"></a>
<span class="sourceLineNo">107</span> while (!region.isClosing() &amp;&amp; !region.isClosed()<a name="line.107"></a>
<span class="sourceLineNo">108</span> &amp;&amp; !server.isAborted() &amp;&amp; !server.isStopped()) {<a name="line.108"></a>
<span class="sourceLineNo">109</span> // TODO: flushRegion() is a blocking call waiting for the flush to complete. Ideally we<a name="line.109"></a>
<span class="sourceLineNo">110</span> // do not have to wait for the whole flush here, just initiate it.<a name="line.110"></a>
<span class="sourceLineNo">111</span> FlushRegionResponse response;<a name="line.111"></a>
<span class="sourceLineNo">112</span> try {<a name="line.112"></a>
<span class="sourceLineNo">113</span> response = FutureUtils.get(connection.flush(ServerRegionReplicaUtil<a name="line.113"></a>
<span class="sourceLineNo">114</span> .getRegionInfoForDefaultReplica(region.getRegionInfo()).getRegionName(), true));<a name="line.114"></a>
<span class="sourceLineNo">115</span> } catch (IOException e) {<a name="line.115"></a>
<span class="sourceLineNo">116</span> if (e instanceof TableNotFoundException || FutureUtils<a name="line.116"></a>
<span class="sourceLineNo">117</span> .get(connection.getAdmin().isTableDisabled(region.getRegionInfo().getTable()))) {<a name="line.117"></a>
<span class="sourceLineNo">118</span> return;<a name="line.118"></a>
<span class="sourceLineNo">119</span> }<a name="line.119"></a>
<span class="sourceLineNo">120</span> if (!counter.shouldRetry()) {<a name="line.120"></a>
<span class="sourceLineNo">121</span> throw e;<a name="line.121"></a>
<span class="sourceLineNo">122</span> }<a name="line.122"></a>
<span class="sourceLineNo">123</span> // The reason that why we need to retry here is that, the retry for asynchronous admin<a name="line.123"></a>
<span class="sourceLineNo">124</span> // request is much simpler than the normal operation, if we failed to locate the region once<a name="line.124"></a>
<span class="sourceLineNo">125</span> // then we will throw the exception out and will not try to relocate again. So here we need<a name="line.125"></a>
<span class="sourceLineNo">126</span> // to add some retries by ourselves to prevent shutting down the region server too<a name="line.126"></a>
<span class="sourceLineNo">127</span> // frequent...<a name="line.127"></a>
<span class="sourceLineNo">128</span> LOG.debug("Failed to trigger a flush of primary region replica {} of region {}, retry={}",<a name="line.128"></a>
<span class="sourceLineNo">129</span> ServerRegionReplicaUtil.getRegionInfoForDefaultReplica(region.getRegionInfo())<a name="line.129"></a>
<span class="sourceLineNo">130</span> .getRegionNameAsString(),<a name="line.130"></a>
<span class="sourceLineNo">131</span> region.getRegionInfo().getRegionNameAsString(), counter.getAttemptTimes(), e);<a name="line.131"></a>
<span class="sourceLineNo">132</span> try {<a name="line.132"></a>
<span class="sourceLineNo">133</span> counter.sleepUntilNextRetry();<a name="line.133"></a>
<span class="sourceLineNo">134</span> } catch (InterruptedException e1) {<a name="line.134"></a>
<span class="sourceLineNo">135</span> throw new InterruptedIOException(e1.getMessage());<a name="line.135"></a>
<span class="sourceLineNo">136</span> }<a name="line.136"></a>
<span class="sourceLineNo">137</span> continue;<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> if (response.getFlushed()) {<a name="line.140"></a>
<span class="sourceLineNo">141</span> // then we have to wait for seeing the flush entry. All reads will be rejected until we see<a name="line.141"></a>
<span class="sourceLineNo">142</span> // a complete flush cycle or replay a region open event<a name="line.142"></a>
<span class="sourceLineNo">143</span> if (LOG.isDebugEnabled()) {<a name="line.143"></a>
<span class="sourceLineNo">144</span> LOG.debug("Triggered flush of primary region replica " +<a name="line.144"></a>
<span class="sourceLineNo">145</span> ServerRegionReplicaUtil.getRegionInfoForDefaultReplica(region.getRegionInfo())<a name="line.145"></a>
<span class="sourceLineNo">146</span> .getRegionNameAsString() +<a name="line.146"></a>
<span class="sourceLineNo">147</span> " for " + region.getRegionInfo().getEncodedName() +<a name="line.147"></a>
<span class="sourceLineNo">148</span> "; now waiting and blocking reads until completes a full flush cycle");<a name="line.148"></a>
<span class="sourceLineNo">149</span> }<a name="line.149"></a>
<span class="sourceLineNo">150</span> region.setReadsEnabled(true);<a name="line.150"></a>
<span class="sourceLineNo">151</span> break;<a name="line.151"></a>
<span class="sourceLineNo">152</span> } else {<a name="line.152"></a>
<span class="sourceLineNo">153</span> if (response.hasWroteFlushWalMarker()) {<a name="line.153"></a>
<span class="sourceLineNo">154</span> if (response.getWroteFlushWalMarker()) {<a name="line.154"></a>
<span class="sourceLineNo">155</span> if (LOG.isDebugEnabled()) {<a name="line.155"></a>
<span class="sourceLineNo">156</span> LOG.debug("Triggered empty flush marker (memstore empty) on primary region replica " +<a name="line.156"></a>
<span class="sourceLineNo">157</span> ServerRegionReplicaUtil.getRegionInfoForDefaultReplica(region.getRegionInfo()).<a name="line.157"></a>
<span class="sourceLineNo">158</span> getRegionNameAsString() + " for " + region.getRegionInfo().getEncodedName() +<a name="line.158"></a>
<span class="sourceLineNo">159</span> "; now waiting and blocking reads until observing a flush marker");<a name="line.159"></a>
<span class="sourceLineNo">160</span> }<a name="line.160"></a>
<span class="sourceLineNo">161</span> region.setReadsEnabled(true);<a name="line.161"></a>
<span class="sourceLineNo">162</span> break;<a name="line.162"></a>
<span class="sourceLineNo">163</span> } else {<a name="line.163"></a>
<span class="sourceLineNo">164</span> // somehow we were not able to get the primary to write the flush request. It may be<a name="line.164"></a>
<span class="sourceLineNo">165</span> // closing or already flushing. Retry flush again after some sleep.<a name="line.165"></a>
<span class="sourceLineNo">166</span> if (!counter.shouldRetry()) {<a name="line.166"></a>
<span class="sourceLineNo">167</span> throw new IOException("Cannot cause primary to flush or drop a wal marker after " +<a name="line.167"></a>
<span class="sourceLineNo">168</span> counter.getAttemptTimes() + " retries. Failing opening of this region replica " +<a name="line.168"></a>
<span class="sourceLineNo">169</span> region.getRegionInfo().getRegionNameAsString());<a name="line.169"></a>
<span class="sourceLineNo">170</span> } else {<a name="line.170"></a>
<span class="sourceLineNo">171</span> LOG.warn(<a name="line.171"></a>
<span class="sourceLineNo">172</span> "Cannot cause primary replica {} to flush or drop a wal marker " +<a name="line.172"></a>
<span class="sourceLineNo">173</span> "for region replica {}, retry={}",<a name="line.173"></a>
<span class="sourceLineNo">174</span> ServerRegionReplicaUtil.getRegionInfoForDefaultReplica(region.getRegionInfo())<a name="line.174"></a>
<span class="sourceLineNo">175</span> .getRegionNameAsString(),<a name="line.175"></a>
<span class="sourceLineNo">176</span> region.getRegionInfo().getRegionNameAsString(), counter.getAttemptTimes());<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> } else {<a name="line.179"></a>
<span class="sourceLineNo">180</span> // nothing to do. Are we dealing with an old server?<a name="line.180"></a>
<span class="sourceLineNo">181</span> LOG.warn(<a name="line.181"></a>
<span class="sourceLineNo">182</span> "Was not able to trigger a flush from primary region due to old server version? " +<a name="line.182"></a>
<span class="sourceLineNo">183</span> "Continuing to open the secondary region replica: " +<a name="line.183"></a>
<span class="sourceLineNo">184</span> region.getRegionInfo().getRegionNameAsString());<a name="line.184"></a>
<span class="sourceLineNo">185</span> break;<a name="line.185"></a>
<span class="sourceLineNo">186</span> }<a name="line.186"></a>
<span class="sourceLineNo">187</span> }<a name="line.187"></a>
<span class="sourceLineNo">188</span> try {<a name="line.188"></a>
<span class="sourceLineNo">189</span> counter.sleepUntilNextRetry();<a name="line.189"></a>
<span class="sourceLineNo">190</span> } catch (InterruptedException e) {<a name="line.190"></a>
<span class="sourceLineNo">191</span> throw new InterruptedIOException(e.getMessage());<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> region.setReadsEnabled(true);<a name="line.194"></a>
<span class="sourceLineNo">195</span> }<a name="line.195"></a>
<span class="sourceLineNo">196</span>}<a name="line.196"></a>
</pre>
</div>
</body>
</html>