blob: e017f46de3c3b0ac2c72e3a64b749a1c45d840f6 [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.replication.regionserver;<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.List;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.Map;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.Optional;<a name="line.25"></a>
<span class="sourceLineNo">026</span>import java.util.TreeMap;<a name="line.26"></a>
<span class="sourceLineNo">027</span>import java.util.concurrent.CompletableFuture;<a name="line.27"></a>
<span class="sourceLineNo">028</span>import java.util.concurrent.ExecutionException;<a name="line.28"></a>
<span class="sourceLineNo">029</span>import java.util.concurrent.TimeUnit;<a name="line.29"></a>
<span class="sourceLineNo">030</span>import java.util.concurrent.atomic.AtomicInteger;<a name="line.30"></a>
<span class="sourceLineNo">031</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.31"></a>
<span class="sourceLineNo">032</span>import java.util.concurrent.atomic.AtomicReference;<a name="line.32"></a>
<span class="sourceLineNo">033</span>import org.apache.hadoop.conf.Configuration;<a name="line.33"></a>
<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.CellUtil;<a name="line.34"></a>
<span class="sourceLineNo">035</span>import org.apache.hadoop.hbase.HBaseIOException;<a name="line.35"></a>
<span class="sourceLineNo">036</span>import org.apache.hadoop.hbase.HConstants;<a name="line.36"></a>
<span class="sourceLineNo">037</span>import org.apache.hadoop.hbase.RegionLocations;<a name="line.37"></a>
<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.TableDescriptors;<a name="line.38"></a>
<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.TableName;<a name="line.39"></a>
<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.TableNotFoundException;<a name="line.40"></a>
<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.client.AsyncClusterConnection;<a name="line.41"></a>
<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.client.RegionInfo;<a name="line.42"></a>
<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.client.RegionReplicaUtil;<a name="line.43"></a>
<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.client.TableDescriptor;<a name="line.44"></a>
<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.replication.HBaseReplicationEndpoint;<a name="line.45"></a>
<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.replication.WALEntryFilter;<a name="line.46"></a>
<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.util.AtomicUtils;<a name="line.47"></a>
<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.48"></a>
<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.util.FutureUtils;<a name="line.49"></a>
<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.50"></a>
<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.util.RetryCounter;<a name="line.51"></a>
<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.util.RetryCounterFactory;<a name="line.52"></a>
<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.wal.WAL.Entry;<a name="line.53"></a>
<span class="sourceLineNo">054</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.54"></a>
<span class="sourceLineNo">055</span>import org.slf4j.Logger;<a name="line.55"></a>
<span class="sourceLineNo">056</span>import org.slf4j.LoggerFactory;<a name="line.56"></a>
<span class="sourceLineNo">057</span><a name="line.57"></a>
<span class="sourceLineNo">058</span>import org.apache.hbase.thirdparty.com.google.common.cache.Cache;<a name="line.58"></a>
<span class="sourceLineNo">059</span>import org.apache.hbase.thirdparty.com.google.common.cache.CacheBuilder;<a name="line.59"></a>
<span class="sourceLineNo">060</span>import org.apache.hbase.thirdparty.com.google.common.cache.CacheLoader;<a name="line.60"></a>
<span class="sourceLineNo">061</span>import org.apache.hbase.thirdparty.com.google.common.cache.LoadingCache;<a name="line.61"></a>
<span class="sourceLineNo">062</span><a name="line.62"></a>
<span class="sourceLineNo">063</span>/**<a name="line.63"></a>
<span class="sourceLineNo">064</span> * A {@link org.apache.hadoop.hbase.replication.ReplicationEndpoint} endpoint which receives the WAL<a name="line.64"></a>
<span class="sourceLineNo">065</span> * edits from the WAL, and sends the edits to replicas of regions.<a name="line.65"></a>
<span class="sourceLineNo">066</span> */<a name="line.66"></a>
<span class="sourceLineNo">067</span>@InterfaceAudience.Private<a name="line.67"></a>
<span class="sourceLineNo">068</span>public class RegionReplicaReplicationEndpoint extends HBaseReplicationEndpoint {<a name="line.68"></a>
<span class="sourceLineNo">069</span><a name="line.69"></a>
<span class="sourceLineNo">070</span> private static final Logger LOG = LoggerFactory.getLogger(RegionReplicaReplicationEndpoint.class);<a name="line.70"></a>
<span class="sourceLineNo">071</span><a name="line.71"></a>
<span class="sourceLineNo">072</span> // Can be configured differently than hbase.client.retries.number<a name="line.72"></a>
<span class="sourceLineNo">073</span> private static String CLIENT_RETRIES_NUMBER =<a name="line.73"></a>
<span class="sourceLineNo">074</span> "hbase.region.replica.replication.client.retries.number";<a name="line.74"></a>
<span class="sourceLineNo">075</span><a name="line.75"></a>
<span class="sourceLineNo">076</span> private Configuration conf;<a name="line.76"></a>
<span class="sourceLineNo">077</span> private AsyncClusterConnection connection;<a name="line.77"></a>
<span class="sourceLineNo">078</span> private TableDescriptors tableDescriptors;<a name="line.78"></a>
<span class="sourceLineNo">079</span><a name="line.79"></a>
<span class="sourceLineNo">080</span> private int numRetries;<a name="line.80"></a>
<span class="sourceLineNo">081</span><a name="line.81"></a>
<span class="sourceLineNo">082</span> private long operationTimeoutNs;<a name="line.82"></a>
<span class="sourceLineNo">083</span><a name="line.83"></a>
<span class="sourceLineNo">084</span> private LoadingCache&lt;TableName, Optional&lt;TableDescriptor&gt;&gt; tableDescriptorCache;<a name="line.84"></a>
<span class="sourceLineNo">085</span><a name="line.85"></a>
<span class="sourceLineNo">086</span> private Cache&lt;TableName, TableName&gt; disabledTableCache;<a name="line.86"></a>
<span class="sourceLineNo">087</span><a name="line.87"></a>
<span class="sourceLineNo">088</span> private final RetryCounterFactory retryCounterFactory =<a name="line.88"></a>
<span class="sourceLineNo">089</span> new RetryCounterFactory(Integer.MAX_VALUE, 1000, 60000);<a name="line.89"></a>
<span class="sourceLineNo">090</span><a name="line.90"></a>
<span class="sourceLineNo">091</span> @Override<a name="line.91"></a>
<span class="sourceLineNo">092</span> public void init(Context context) throws IOException {<a name="line.92"></a>
<span class="sourceLineNo">093</span> super.init(context);<a name="line.93"></a>
<span class="sourceLineNo">094</span> this.conf = context.getConfiguration();<a name="line.94"></a>
<span class="sourceLineNo">095</span> this.tableDescriptors = context.getTableDescriptors();<a name="line.95"></a>
<span class="sourceLineNo">096</span> int memstoreReplicationEnabledCacheExpiryMs = conf<a name="line.96"></a>
<span class="sourceLineNo">097</span> .getInt("hbase.region.replica.replication.cache.memstoreReplicationEnabled.expiryMs", 5000);<a name="line.97"></a>
<span class="sourceLineNo">098</span> // A cache for the table "memstore replication enabled" flag.<a name="line.98"></a>
<span class="sourceLineNo">099</span> // It has a default expiry of 5 sec. This means that if the table is altered<a name="line.99"></a>
<span class="sourceLineNo">100</span> // with a different flag value, we might miss to replicate for that amount of<a name="line.100"></a>
<span class="sourceLineNo">101</span> // time. But this cache avoid the slow lookup and parsing of the TableDescriptor.<a name="line.101"></a>
<span class="sourceLineNo">102</span> tableDescriptorCache = CacheBuilder.newBuilder()<a name="line.102"></a>
<span class="sourceLineNo">103</span> .expireAfterWrite(memstoreReplicationEnabledCacheExpiryMs, TimeUnit.MILLISECONDS)<a name="line.103"></a>
<span class="sourceLineNo">104</span> .initialCapacity(10).maximumSize(1000)<a name="line.104"></a>
<span class="sourceLineNo">105</span> .build(new CacheLoader&lt;TableName, Optional&lt;TableDescriptor&gt;&gt;() {<a name="line.105"></a>
<span class="sourceLineNo">106</span><a name="line.106"></a>
<span class="sourceLineNo">107</span> @Override<a name="line.107"></a>
<span class="sourceLineNo">108</span> public Optional&lt;TableDescriptor&gt; load(TableName tableName) throws Exception {<a name="line.108"></a>
<span class="sourceLineNo">109</span> // check if the table requires memstore replication<a name="line.109"></a>
<span class="sourceLineNo">110</span> // some unit-test drop the table, so we should do a bypass check and always replicate.<a name="line.110"></a>
<span class="sourceLineNo">111</span> return Optional.ofNullable(tableDescriptors.get(tableName));<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> int nonExistentTableCacheExpiryMs =<a name="line.114"></a>
<span class="sourceLineNo">115</span> conf.getInt("hbase.region.replica.replication.cache.disabledAndDroppedTables.expiryMs", 5000);<a name="line.115"></a>
<span class="sourceLineNo">116</span> // A cache for non existing tables that have a default expiry of 5 sec. This means that if the<a name="line.116"></a>
<span class="sourceLineNo">117</span> // table is created again with the same name, we might miss to replicate for that amount of<a name="line.117"></a>
<span class="sourceLineNo">118</span> // time. But this cache prevents overloading meta requests for every edit from a deleted file.<a name="line.118"></a>
<span class="sourceLineNo">119</span> disabledTableCache = CacheBuilder.newBuilder()<a name="line.119"></a>
<span class="sourceLineNo">120</span> .expireAfterWrite(nonExistentTableCacheExpiryMs, TimeUnit.MILLISECONDS).initialCapacity(10)<a name="line.120"></a>
<span class="sourceLineNo">121</span> .maximumSize(1000).build();<a name="line.121"></a>
<span class="sourceLineNo">122</span> // HRS multiplies client retries by 10 globally for meta operations, but we do not want this.<a name="line.122"></a>
<span class="sourceLineNo">123</span> // We are resetting it here because we want default number of retries (35) rather than 10 times<a name="line.123"></a>
<span class="sourceLineNo">124</span> // that which makes very long retries for disabled tables etc.<a name="line.124"></a>
<span class="sourceLineNo">125</span> int defaultNumRetries = conf.getInt(HConstants.HBASE_CLIENT_RETRIES_NUMBER,<a name="line.125"></a>
<span class="sourceLineNo">126</span> HConstants.DEFAULT_HBASE_CLIENT_RETRIES_NUMBER);<a name="line.126"></a>
<span class="sourceLineNo">127</span> if (defaultNumRetries &gt; 10) {<a name="line.127"></a>
<span class="sourceLineNo">128</span> int mult = conf.getInt(HConstants.HBASE_CLIENT_SERVERSIDE_RETRIES_MULTIPLIER,<a name="line.128"></a>
<span class="sourceLineNo">129</span> HConstants.DEFAULT_HBASE_CLIENT_SERVERSIDE_RETRIES_MULTIPLIER);<a name="line.129"></a>
<span class="sourceLineNo">130</span> defaultNumRetries = defaultNumRetries / mult; // reset if HRS has multiplied this already<a name="line.130"></a>
<span class="sourceLineNo">131</span> }<a name="line.131"></a>
<span class="sourceLineNo">132</span> this.numRetries = conf.getInt(CLIENT_RETRIES_NUMBER, defaultNumRetries);<a name="line.132"></a>
<span class="sourceLineNo">133</span> // use the regular RPC timeout for replica replication RPC's<a name="line.133"></a>
<span class="sourceLineNo">134</span> this.operationTimeoutNs =<a name="line.134"></a>
<span class="sourceLineNo">135</span> TimeUnit.MILLISECONDS.toNanos(conf.getInt(HConstants.HBASE_CLIENT_OPERATION_TIMEOUT,<a name="line.135"></a>
<span class="sourceLineNo">136</span> HConstants.DEFAULT_HBASE_CLIENT_OPERATION_TIMEOUT));<a name="line.136"></a>
<span class="sourceLineNo">137</span> this.connection = context.getServer().getAsyncClusterConnection();<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> /**<a name="line.140"></a>
<span class="sourceLineNo">141</span> * returns true if the specified entry must be replicated. We should always replicate meta<a name="line.141"></a>
<span class="sourceLineNo">142</span> * operations (e.g. flush) and use the user HTD flag to decide whether or not replicate the<a name="line.142"></a>
<span class="sourceLineNo">143</span> * memstore.<a name="line.143"></a>
<span class="sourceLineNo">144</span> */<a name="line.144"></a>
<span class="sourceLineNo">145</span> private boolean requiresReplication(Optional&lt;TableDescriptor&gt; tableDesc,<a name="line.145"></a>
<span class="sourceLineNo">146</span> Entry entry) {<a name="line.146"></a>
<span class="sourceLineNo">147</span> // empty edit does not need to be replicated<a name="line.147"></a>
<span class="sourceLineNo">148</span> if (entry.getEdit().isEmpty() || !tableDesc.isPresent()) {<a name="line.148"></a>
<span class="sourceLineNo">149</span> return false;<a name="line.149"></a>
<span class="sourceLineNo">150</span> }<a name="line.150"></a>
<span class="sourceLineNo">151</span> // meta edits (e.g. flush) must be always replicated<a name="line.151"></a>
<span class="sourceLineNo">152</span> return entry.getEdit().isMetaEdit() || tableDesc.get().hasRegionMemStoreReplication();<a name="line.152"></a>
<span class="sourceLineNo">153</span> }<a name="line.153"></a>
<span class="sourceLineNo">154</span><a name="line.154"></a>
<span class="sourceLineNo">155</span> private void getRegionLocations(CompletableFuture&lt;RegionLocations&gt; future,<a name="line.155"></a>
<span class="sourceLineNo">156</span> TableDescriptor tableDesc, byte[] encodedRegionName, byte[] row, boolean reload) {<a name="line.156"></a>
<span class="sourceLineNo">157</span> FutureUtils.addListener(connection.getRegionLocations(tableDesc.getTableName(), row, reload),<a name="line.157"></a>
<span class="sourceLineNo">158</span> (locs, e) -&gt; {<a name="line.158"></a>
<span class="sourceLineNo">159</span> if (e != null) {<a name="line.159"></a>
<span class="sourceLineNo">160</span> future.completeExceptionally(e);<a name="line.160"></a>
<span class="sourceLineNo">161</span> return;<a name="line.161"></a>
<span class="sourceLineNo">162</span> }<a name="line.162"></a>
<span class="sourceLineNo">163</span> // if we are not loading from cache, just return<a name="line.163"></a>
<span class="sourceLineNo">164</span> if (reload) {<a name="line.164"></a>
<span class="sourceLineNo">165</span> future.complete(locs);<a name="line.165"></a>
<span class="sourceLineNo">166</span> return;<a name="line.166"></a>
<span class="sourceLineNo">167</span> }<a name="line.167"></a>
<span class="sourceLineNo">168</span> // check if the number of region replicas is correct, and also the primary region name<a name="line.168"></a>
<span class="sourceLineNo">169</span> // matches.<a name="line.169"></a>
<span class="sourceLineNo">170</span> if (locs.size() == tableDesc.getRegionReplication() &amp;&amp;<a name="line.170"></a>
<span class="sourceLineNo">171</span> locs.getDefaultRegionLocation() != null &amp;&amp;<a name="line.171"></a>
<span class="sourceLineNo">172</span> Bytes.equals(locs.getDefaultRegionLocation().getRegion().getEncodedNameAsBytes(),<a name="line.172"></a>
<span class="sourceLineNo">173</span> encodedRegionName)) {<a name="line.173"></a>
<span class="sourceLineNo">174</span> future.complete(locs);<a name="line.174"></a>
<span class="sourceLineNo">175</span> } else {<a name="line.175"></a>
<span class="sourceLineNo">176</span> // reload again as the information in cache maybe stale<a name="line.176"></a>
<span class="sourceLineNo">177</span> getRegionLocations(future, tableDesc, encodedRegionName, row, true);<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> }<a name="line.180"></a>
<span class="sourceLineNo">181</span><a name="line.181"></a>
<span class="sourceLineNo">182</span> private void replicate(CompletableFuture&lt;Long&gt; future, RegionLocations locs,<a name="line.182"></a>
<span class="sourceLineNo">183</span> TableDescriptor tableDesc, byte[] encodedRegionName, byte[] row, List&lt;Entry&gt; entries) {<a name="line.183"></a>
<span class="sourceLineNo">184</span> if (locs.size() == 1) {<a name="line.184"></a>
<span class="sourceLineNo">185</span> // Could this happen?<a name="line.185"></a>
<span class="sourceLineNo">186</span> future.complete(Long.valueOf(entries.size()));<a name="line.186"></a>
<span class="sourceLineNo">187</span> return;<a name="line.187"></a>
<span class="sourceLineNo">188</span> }<a name="line.188"></a>
<span class="sourceLineNo">189</span> RegionInfo defaultReplica = locs.getDefaultRegionLocation().getRegion();<a name="line.189"></a>
<span class="sourceLineNo">190</span> if (!Bytes.equals(defaultReplica.getEncodedNameAsBytes(), encodedRegionName)) {<a name="line.190"></a>
<span class="sourceLineNo">191</span> // the region name is not equal, this usually means the region has been split or merged, so<a name="line.191"></a>
<span class="sourceLineNo">192</span> // give up replicating as the new region(s) should already have all the data of the parent<a name="line.192"></a>
<span class="sourceLineNo">193</span> // region(s).<a name="line.193"></a>
<span class="sourceLineNo">194</span> if (LOG.isTraceEnabled()) {<a name="line.194"></a>
<span class="sourceLineNo">195</span> LOG.trace(<a name="line.195"></a>
<span class="sourceLineNo">196</span> "Skipping {} entries in table {} because located region {} is different than" +<a name="line.196"></a>
<span class="sourceLineNo">197</span> " the original region {} from WALEdit",<a name="line.197"></a>
<span class="sourceLineNo">198</span> tableDesc.getTableName(), defaultReplica.getEncodedName(),<a name="line.198"></a>
<span class="sourceLineNo">199</span> Bytes.toStringBinary(encodedRegionName));<a name="line.199"></a>
<span class="sourceLineNo">200</span> }<a name="line.200"></a>
<span class="sourceLineNo">201</span> future.complete(Long.valueOf(entries.size()));<a name="line.201"></a>
<span class="sourceLineNo">202</span> return;<a name="line.202"></a>
<span class="sourceLineNo">203</span> }<a name="line.203"></a>
<span class="sourceLineNo">204</span> AtomicReference&lt;Throwable&gt; error = new AtomicReference&lt;&gt;();<a name="line.204"></a>
<span class="sourceLineNo">205</span> AtomicInteger remainingTasks = new AtomicInteger(locs.size() - 1);<a name="line.205"></a>
<span class="sourceLineNo">206</span> AtomicLong skippedEdits = new AtomicLong(0);<a name="line.206"></a>
<span class="sourceLineNo">207</span><a name="line.207"></a>
<span class="sourceLineNo">208</span> for (int i = 1, n = locs.size(); i &lt; n; i++) {<a name="line.208"></a>
<span class="sourceLineNo">209</span> // Do not use the elements other than the default replica as they may be null. We will fail<a name="line.209"></a>
<span class="sourceLineNo">210</span> // earlier if the location for default replica is null.<a name="line.210"></a>
<span class="sourceLineNo">211</span> final RegionInfo replica = RegionReplicaUtil.getRegionInfoForReplica(defaultReplica, i);<a name="line.211"></a>
<span class="sourceLineNo">212</span> FutureUtils<a name="line.212"></a>
<span class="sourceLineNo">213</span> .addListener(connection.replay(tableDesc.getTableName(), replica.getEncodedNameAsBytes(),<a name="line.213"></a>
<span class="sourceLineNo">214</span> row, entries, replica.getReplicaId(), numRetries, operationTimeoutNs), (r, e) -&gt; {<a name="line.214"></a>
<span class="sourceLineNo">215</span> if (e != null) {<a name="line.215"></a>
<span class="sourceLineNo">216</span> LOG.warn("Failed to replicate to {}", replica, e);<a name="line.216"></a>
<span class="sourceLineNo">217</span> error.compareAndSet(null, e);<a name="line.217"></a>
<span class="sourceLineNo">218</span> } else {<a name="line.218"></a>
<span class="sourceLineNo">219</span> AtomicUtils.updateMax(skippedEdits, r.longValue());<a name="line.219"></a>
<span class="sourceLineNo">220</span> }<a name="line.220"></a>
<span class="sourceLineNo">221</span> if (remainingTasks.decrementAndGet() == 0) {<a name="line.221"></a>
<span class="sourceLineNo">222</span> if (error.get() != null) {<a name="line.222"></a>
<span class="sourceLineNo">223</span> future.completeExceptionally(error.get());<a name="line.223"></a>
<span class="sourceLineNo">224</span> } else {<a name="line.224"></a>
<span class="sourceLineNo">225</span> future.complete(skippedEdits.get());<a name="line.225"></a>
<span class="sourceLineNo">226</span> }<a name="line.226"></a>
<span class="sourceLineNo">227</span> }<a name="line.227"></a>
<span class="sourceLineNo">228</span> });<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><a name="line.231"></a>
<span class="sourceLineNo">232</span> private void logSkipped(TableName tableName, List&lt;Entry&gt; entries, String reason) {<a name="line.232"></a>
<span class="sourceLineNo">233</span> if (LOG.isTraceEnabled()) {<a name="line.233"></a>
<span class="sourceLineNo">234</span> LOG.trace("Skipping {} entries because table {} is {}", entries.size(), tableName, reason);<a name="line.234"></a>
<span class="sourceLineNo">235</span> for (Entry entry : entries) {<a name="line.235"></a>
<span class="sourceLineNo">236</span> LOG.trace("Skipping : {}", entry);<a name="line.236"></a>
<span class="sourceLineNo">237</span> }<a name="line.237"></a>
<span class="sourceLineNo">238</span> }<a name="line.238"></a>
<span class="sourceLineNo">239</span> }<a name="line.239"></a>
<span class="sourceLineNo">240</span><a name="line.240"></a>
<span class="sourceLineNo">241</span> private CompletableFuture&lt;Long&gt; replicate(TableDescriptor tableDesc, byte[] encodedRegionName,<a name="line.241"></a>
<span class="sourceLineNo">242</span> List&lt;Entry&gt; entries) {<a name="line.242"></a>
<span class="sourceLineNo">243</span> if (disabledTableCache.getIfPresent(tableDesc.getTableName()) != null) {<a name="line.243"></a>
<span class="sourceLineNo">244</span> logSkipped(tableDesc.getTableName(), entries, "cached as a disabled table");<a name="line.244"></a>
<span class="sourceLineNo">245</span> return CompletableFuture.completedFuture(Long.valueOf(entries.size()));<a name="line.245"></a>
<span class="sourceLineNo">246</span> }<a name="line.246"></a>
<span class="sourceLineNo">247</span> byte[] row = CellUtil.cloneRow(entries.get(0).getEdit().getCells().get(0));<a name="line.247"></a>
<span class="sourceLineNo">248</span> CompletableFuture&lt;RegionLocations&gt; locateFuture = new CompletableFuture&lt;&gt;();<a name="line.248"></a>
<span class="sourceLineNo">249</span> getRegionLocations(locateFuture, tableDesc, encodedRegionName, row, false);<a name="line.249"></a>
<span class="sourceLineNo">250</span> CompletableFuture&lt;Long&gt; future = new CompletableFuture&lt;&gt;();<a name="line.250"></a>
<span class="sourceLineNo">251</span> FutureUtils.addListener(locateFuture, (locs, error) -&gt; {<a name="line.251"></a>
<span class="sourceLineNo">252</span> if (error != null) {<a name="line.252"></a>
<span class="sourceLineNo">253</span> future.completeExceptionally(error);<a name="line.253"></a>
<span class="sourceLineNo">254</span> } else if (locs.getDefaultRegionLocation() == null) {<a name="line.254"></a>
<span class="sourceLineNo">255</span> future.completeExceptionally(<a name="line.255"></a>
<span class="sourceLineNo">256</span> new HBaseIOException("No location found for default replica of table=" +<a name="line.256"></a>
<span class="sourceLineNo">257</span> tableDesc.getTableName() + " row='" + Bytes.toStringBinary(row) + "'"));<a name="line.257"></a>
<span class="sourceLineNo">258</span> } else {<a name="line.258"></a>
<span class="sourceLineNo">259</span> replicate(future, locs, tableDesc, encodedRegionName, row, entries);<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> return future;<a name="line.262"></a>
<span class="sourceLineNo">263</span> }<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 boolean replicate(ReplicateContext replicateContext) {<a name="line.266"></a>
<span class="sourceLineNo">267</span> Map&lt;byte[], Pair&lt;TableDescriptor, List&lt;Entry&gt;&gt;&gt; encodedRegionName2Entries =<a name="line.267"></a>
<span class="sourceLineNo">268</span> new TreeMap&lt;&gt;(Bytes.BYTES_COMPARATOR);<a name="line.268"></a>
<span class="sourceLineNo">269</span> long skippedEdits = 0;<a name="line.269"></a>
<span class="sourceLineNo">270</span> RetryCounter retryCounter = retryCounterFactory.create();<a name="line.270"></a>
<span class="sourceLineNo">271</span> outer: while (isRunning()) {<a name="line.271"></a>
<span class="sourceLineNo">272</span> encodedRegionName2Entries.clear();<a name="line.272"></a>
<span class="sourceLineNo">273</span> skippedEdits = 0;<a name="line.273"></a>
<span class="sourceLineNo">274</span> for (Entry entry : replicateContext.getEntries()) {<a name="line.274"></a>
<span class="sourceLineNo">275</span> Optional&lt;TableDescriptor&gt; tableDesc;<a name="line.275"></a>
<span class="sourceLineNo">276</span> try {<a name="line.276"></a>
<span class="sourceLineNo">277</span> tableDesc = tableDescriptorCache.get(entry.getKey().getTableName());<a name="line.277"></a>
<span class="sourceLineNo">278</span> } catch (ExecutionException e) {<a name="line.278"></a>
<span class="sourceLineNo">279</span> LOG.warn("Failed to load table descriptor for {}, attempts={}",<a name="line.279"></a>
<span class="sourceLineNo">280</span> entry.getKey().getTableName(), retryCounter.getAttemptTimes(), e.getCause());<a name="line.280"></a>
<span class="sourceLineNo">281</span> if (!retryCounter.shouldRetry()) {<a name="line.281"></a>
<span class="sourceLineNo">282</span> return false;<a name="line.282"></a>
<span class="sourceLineNo">283</span> }<a name="line.283"></a>
<span class="sourceLineNo">284</span> try {<a name="line.284"></a>
<span class="sourceLineNo">285</span> retryCounter.sleepUntilNextRetry();<a name="line.285"></a>
<span class="sourceLineNo">286</span> } catch (InterruptedException e1) {<a name="line.286"></a>
<span class="sourceLineNo">287</span> // restore the interrupted state<a name="line.287"></a>
<span class="sourceLineNo">288</span> Thread.currentThread().interrupt();<a name="line.288"></a>
<span class="sourceLineNo">289</span> return false;<a name="line.289"></a>
<span class="sourceLineNo">290</span> }<a name="line.290"></a>
<span class="sourceLineNo">291</span> continue outer;<a name="line.291"></a>
<span class="sourceLineNo">292</span> }<a name="line.292"></a>
<span class="sourceLineNo">293</span> if (!requiresReplication(tableDesc, entry)) {<a name="line.293"></a>
<span class="sourceLineNo">294</span> skippedEdits++;<a name="line.294"></a>
<span class="sourceLineNo">295</span> continue;<a name="line.295"></a>
<span class="sourceLineNo">296</span> }<a name="line.296"></a>
<span class="sourceLineNo">297</span> byte[] encodedRegionName = entry.getKey().getEncodedRegionName();<a name="line.297"></a>
<span class="sourceLineNo">298</span> encodedRegionName2Entries<a name="line.298"></a>
<span class="sourceLineNo">299</span> .computeIfAbsent(encodedRegionName, k -&gt; Pair.newPair(tableDesc.get(), new ArrayList&lt;&gt;()))<a name="line.299"></a>
<span class="sourceLineNo">300</span> .getSecond().add(entry);<a name="line.300"></a>
<span class="sourceLineNo">301</span> }<a name="line.301"></a>
<span class="sourceLineNo">302</span> break;<a name="line.302"></a>
<span class="sourceLineNo">303</span> }<a name="line.303"></a>
<span class="sourceLineNo">304</span> // send the request to regions<a name="line.304"></a>
<span class="sourceLineNo">305</span> retryCounter = retryCounterFactory.create();<a name="line.305"></a>
<span class="sourceLineNo">306</span> while (isRunning()) {<a name="line.306"></a>
<span class="sourceLineNo">307</span> List&lt;Pair&lt;CompletableFuture&lt;Long&gt;, byte[]&gt;&gt; futureAndEncodedRegionNameList =<a name="line.307"></a>
<span class="sourceLineNo">308</span> new ArrayList&lt;Pair&lt;CompletableFuture&lt;Long&gt;, byte[]&gt;&gt;();<a name="line.308"></a>
<span class="sourceLineNo">309</span> for (Map.Entry&lt;byte[], Pair&lt;TableDescriptor, List&lt;Entry&gt;&gt;&gt; entry : encodedRegionName2Entries<a name="line.309"></a>
<span class="sourceLineNo">310</span> .entrySet()) {<a name="line.310"></a>
<span class="sourceLineNo">311</span> CompletableFuture&lt;Long&gt; future =<a name="line.311"></a>
<span class="sourceLineNo">312</span> replicate(entry.getValue().getFirst(), entry.getKey(), entry.getValue().getSecond());<a name="line.312"></a>
<span class="sourceLineNo">313</span> futureAndEncodedRegionNameList.add(Pair.newPair(future, entry.getKey()));<a name="line.313"></a>
<span class="sourceLineNo">314</span> }<a name="line.314"></a>
<span class="sourceLineNo">315</span> for (Pair&lt;CompletableFuture&lt;Long&gt;, byte[]&gt; pair : futureAndEncodedRegionNameList) {<a name="line.315"></a>
<span class="sourceLineNo">316</span> byte[] encodedRegionName = pair.getSecond();<a name="line.316"></a>
<span class="sourceLineNo">317</span> try {<a name="line.317"></a>
<span class="sourceLineNo">318</span> skippedEdits += pair.getFirst().get().longValue();<a name="line.318"></a>
<span class="sourceLineNo">319</span> encodedRegionName2Entries.remove(encodedRegionName);<a name="line.319"></a>
<span class="sourceLineNo">320</span> } catch (InterruptedException e) {<a name="line.320"></a>
<span class="sourceLineNo">321</span> // restore the interrupted state<a name="line.321"></a>
<span class="sourceLineNo">322</span> Thread.currentThread().interrupt();<a name="line.322"></a>
<span class="sourceLineNo">323</span> return false;<a name="line.323"></a>
<span class="sourceLineNo">324</span> } catch (ExecutionException e) {<a name="line.324"></a>
<span class="sourceLineNo">325</span> Pair&lt;TableDescriptor, List&lt;Entry&gt;&gt; tableAndEntries =<a name="line.325"></a>
<span class="sourceLineNo">326</span> encodedRegionName2Entries.get(encodedRegionName);<a name="line.326"></a>
<span class="sourceLineNo">327</span> TableName tableName = tableAndEntries.getFirst().getTableName();<a name="line.327"></a>
<span class="sourceLineNo">328</span> List&lt;Entry&gt; entries = tableAndEntries.getSecond();<a name="line.328"></a>
<span class="sourceLineNo">329</span> Throwable cause = e.getCause();<a name="line.329"></a>
<span class="sourceLineNo">330</span> // The table can be disabled or dropped at this time. For disabled tables, we have no<a name="line.330"></a>
<span class="sourceLineNo">331</span> // cheap mechanism to detect this case because meta does not contain this information.<a name="line.331"></a>
<span class="sourceLineNo">332</span> // ClusterConnection.isTableDisabled() is a zk call which we cannot do for every replay<a name="line.332"></a>
<span class="sourceLineNo">333</span> // RPC. So instead we start the replay RPC with retries and check whether the table is<a name="line.333"></a>
<span class="sourceLineNo">334</span> // dropped or disabled which might cause SocketTimeoutException, or<a name="line.334"></a>
<span class="sourceLineNo">335</span> // RetriesExhaustedException or similar if we get IOE.<a name="line.335"></a>
<span class="sourceLineNo">336</span> if (cause instanceof TableNotFoundException) {<a name="line.336"></a>
<span class="sourceLineNo">337</span> // add to cache that the table does not exist<a name="line.337"></a>
<span class="sourceLineNo">338</span> tableDescriptorCache.put(tableName, Optional.empty());<a name="line.338"></a>
<span class="sourceLineNo">339</span> logSkipped(tableName, entries, "dropped");<a name="line.339"></a>
<span class="sourceLineNo">340</span> skippedEdits += entries.size();<a name="line.340"></a>
<span class="sourceLineNo">341</span> encodedRegionName2Entries.remove(encodedRegionName);<a name="line.341"></a>
<span class="sourceLineNo">342</span> continue;<a name="line.342"></a>
<span class="sourceLineNo">343</span> }<a name="line.343"></a>
<span class="sourceLineNo">344</span> boolean disabled = false;<a name="line.344"></a>
<span class="sourceLineNo">345</span> try {<a name="line.345"></a>
<span class="sourceLineNo">346</span> disabled = connection.getAdmin().isTableDisabled(tableName).get();<a name="line.346"></a>
<span class="sourceLineNo">347</span> } catch (InterruptedException e1) {<a name="line.347"></a>
<span class="sourceLineNo">348</span> // restore the interrupted state<a name="line.348"></a>
<span class="sourceLineNo">349</span> Thread.currentThread().interrupt();<a name="line.349"></a>
<span class="sourceLineNo">350</span> return false;<a name="line.350"></a>
<span class="sourceLineNo">351</span> } catch (ExecutionException e1) {<a name="line.351"></a>
<span class="sourceLineNo">352</span> LOG.warn("Failed to test whether {} is disabled, assume it is not disabled", tableName,<a name="line.352"></a>
<span class="sourceLineNo">353</span> e1.getCause());<a name="line.353"></a>
<span class="sourceLineNo">354</span> }<a name="line.354"></a>
<span class="sourceLineNo">355</span> if (disabled) {<a name="line.355"></a>
<span class="sourceLineNo">356</span> disabledTableCache.put(tableName, tableName);<a name="line.356"></a>
<span class="sourceLineNo">357</span> logSkipped(tableName, entries, "disabled");<a name="line.357"></a>
<span class="sourceLineNo">358</span> skippedEdits += entries.size();<a name="line.358"></a>
<span class="sourceLineNo">359</span> encodedRegionName2Entries.remove(encodedRegionName);<a name="line.359"></a>
<span class="sourceLineNo">360</span> continue;<a name="line.360"></a>
<span class="sourceLineNo">361</span> }<a name="line.361"></a>
<span class="sourceLineNo">362</span> LOG.warn("Failed to replicate {} entries for region {} of table {}", entries.size(),<a name="line.362"></a>
<span class="sourceLineNo">363</span> Bytes.toStringBinary(encodedRegionName), tableName);<a name="line.363"></a>
<span class="sourceLineNo">364</span> }<a name="line.364"></a>
<span class="sourceLineNo">365</span> }<a name="line.365"></a>
<span class="sourceLineNo">366</span> // we have done<a name="line.366"></a>
<span class="sourceLineNo">367</span> if (encodedRegionName2Entries.isEmpty()) {<a name="line.367"></a>
<span class="sourceLineNo">368</span> ctx.getMetrics().incrLogEditsFiltered(skippedEdits);<a name="line.368"></a>
<span class="sourceLineNo">369</span> return true;<a name="line.369"></a>
<span class="sourceLineNo">370</span> } else {<a name="line.370"></a>
<span class="sourceLineNo">371</span> LOG.warn("Failed to replicate all entris, retry={}", retryCounter.getAttemptTimes());<a name="line.371"></a>
<span class="sourceLineNo">372</span> if (!retryCounter.shouldRetry()) {<a name="line.372"></a>
<span class="sourceLineNo">373</span> return false;<a name="line.373"></a>
<span class="sourceLineNo">374</span> }<a name="line.374"></a>
<span class="sourceLineNo">375</span> try {<a name="line.375"></a>
<span class="sourceLineNo">376</span> retryCounter.sleepUntilNextRetry();<a name="line.376"></a>
<span class="sourceLineNo">377</span> } catch (InterruptedException e) {<a name="line.377"></a>
<span class="sourceLineNo">378</span> // restore the interrupted state<a name="line.378"></a>
<span class="sourceLineNo">379</span> Thread.currentThread().interrupt();<a name="line.379"></a>
<span class="sourceLineNo">380</span> return false;<a name="line.380"></a>
<span class="sourceLineNo">381</span> }<a name="line.381"></a>
<span class="sourceLineNo">382</span> }<a name="line.382"></a>
<span class="sourceLineNo">383</span> }<a name="line.383"></a>
<span class="sourceLineNo">384</span><a name="line.384"></a>
<span class="sourceLineNo">385</span> return false;<a name="line.385"></a>
<span class="sourceLineNo">386</span> }<a name="line.386"></a>
<span class="sourceLineNo">387</span><a name="line.387"></a>
<span class="sourceLineNo">388</span> @Override<a name="line.388"></a>
<span class="sourceLineNo">389</span> public boolean canReplicateToSameCluster() {<a name="line.389"></a>
<span class="sourceLineNo">390</span> return true;<a name="line.390"></a>
<span class="sourceLineNo">391</span> }<a name="line.391"></a>
<span class="sourceLineNo">392</span><a name="line.392"></a>
<span class="sourceLineNo">393</span> @Override<a name="line.393"></a>
<span class="sourceLineNo">394</span> protected WALEntryFilter getScopeWALEntryFilter() {<a name="line.394"></a>
<span class="sourceLineNo">395</span> // we do not care about scope. We replicate everything.<a name="line.395"></a>
<span class="sourceLineNo">396</span> return null;<a name="line.396"></a>
<span class="sourceLineNo">397</span> }<a name="line.397"></a>
<span class="sourceLineNo">398</span>}<a name="line.398"></a>
</pre>
</div>
</body>
</html>