Repository: hbase-site Updated Branches: refs/heads/asf-site 9f95f7932 -> b9722a17b
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/b9722a17/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html index c552d8a..a792ab2 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.NodeFailoverWorker.html @@ -48,922 +48,928 @@ <span class="sourceLineNo">040</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.40"></a> <span class="sourceLineNo">041</span>import java.util.concurrent.TimeUnit;<a name="line.41"></a> <span class="sourceLineNo">042</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.42"></a> -<span class="sourceLineNo">043</span><a name="line.43"></a> -<span class="sourceLineNo">044</span>import org.apache.commons.logging.Log;<a name="line.44"></a> -<span class="sourceLineNo">045</span>import org.apache.commons.logging.LogFactory;<a name="line.45"></a> -<span class="sourceLineNo">046</span>import org.apache.hadoop.conf.Configuration;<a name="line.46"></a> -<span class="sourceLineNo">047</span>import org.apache.hadoop.fs.FileSystem;<a name="line.47"></a> -<span class="sourceLineNo">048</span>import org.apache.hadoop.fs.Path;<a name="line.48"></a> -<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.HConstants;<a name="line.49"></a> -<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.50"></a> -<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.Server;<a name="line.51"></a> -<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.TableDescriptors;<a name="line.52"></a> -<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.TableName;<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.apache.hadoop.hbase.client.Connection;<a name="line.55"></a> -<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.56"></a> -<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.57"></a> -<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.regionserver.RegionServerCoprocessorHost;<a name="line.58"></a> -<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.replication.ReplicationEndpoint;<a name="line.59"></a> -<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.60"></a> -<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.61"></a> -<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.62"></a> -<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.replication.ReplicationPeerConfig;<a name="line.63"></a> -<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.64"></a> -<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.65"></a> -<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.replication.ReplicationQueues;<a name="line.66"></a> -<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.67"></a> -<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.68"></a> -<span class="sourceLineNo">069</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.69"></a> -<span class="sourceLineNo">070</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.70"></a> -<span class="sourceLineNo">071</span><a name="line.71"></a> -<span class="sourceLineNo">072</span>import org.apache.hadoop.hbase.shaded.com.google.common.annotations.VisibleForTesting;<a name="line.72"></a> -<span class="sourceLineNo">073</span>import org.apache.hadoop.hbase.shaded.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.73"></a> -<span class="sourceLineNo">074</span><a name="line.74"></a> -<span class="sourceLineNo">075</span>/**<a name="line.75"></a> -<span class="sourceLineNo">076</span> * This class is responsible to manage all the replication<a name="line.76"></a> -<span class="sourceLineNo">077</span> * sources. There are two classes of sources:<a name="line.77"></a> -<span class="sourceLineNo">078</span> * <ul><a name="line.78"></a> -<span class="sourceLineNo">079</span> * <li> Normal sources are persistent and one per peer cluster</li><a name="line.79"></a> -<span class="sourceLineNo">080</span> * <li> Old sources are recovered from a failed region server and our<a name="line.80"></a> -<span class="sourceLineNo">081</span> * only goal is to finish replicating the WAL queue it had up in ZK</li><a name="line.81"></a> -<span class="sourceLineNo">082</span> * </ul><a name="line.82"></a> -<span class="sourceLineNo">083</span> *<a name="line.83"></a> -<span class="sourceLineNo">084</span> * When a region server dies, this class uses a watcher to get notified and it<a name="line.84"></a> -<span class="sourceLineNo">085</span> * tries to grab a lock in order to transfer all the queues in a local<a name="line.85"></a> -<span class="sourceLineNo">086</span> * old source.<a name="line.86"></a> -<span class="sourceLineNo">087</span> *<a name="line.87"></a> -<span class="sourceLineNo">088</span> * This class implements the ReplicationListener interface so that it can track changes in<a name="line.88"></a> -<span class="sourceLineNo">089</span> * replication state.<a name="line.89"></a> -<span class="sourceLineNo">090</span> */<a name="line.90"></a> -<span class="sourceLineNo">091</span>@InterfaceAudience.Private<a name="line.91"></a> -<span class="sourceLineNo">092</span>public class ReplicationSourceManager implements ReplicationListener {<a name="line.92"></a> -<span class="sourceLineNo">093</span> private static final Log LOG =<a name="line.93"></a> -<span class="sourceLineNo">094</span> LogFactory.getLog(ReplicationSourceManager.class);<a name="line.94"></a> -<span class="sourceLineNo">095</span> // List of all the sources that read this RS's logs<a name="line.95"></a> -<span class="sourceLineNo">096</span> private final List<ReplicationSourceInterface> sources;<a name="line.96"></a> -<span class="sourceLineNo">097</span> // List of all the sources we got from died RSs<a name="line.97"></a> -<span class="sourceLineNo">098</span> private final List<ReplicationSourceInterface> oldsources;<a name="line.98"></a> -<span class="sourceLineNo">099</span> private final ReplicationQueues replicationQueues;<a name="line.99"></a> -<span class="sourceLineNo">100</span> private final ReplicationTracker replicationTracker;<a name="line.100"></a> -<span class="sourceLineNo">101</span> private final ReplicationPeers replicationPeers;<a name="line.101"></a> -<span class="sourceLineNo">102</span> // UUID for this cluster<a name="line.102"></a> -<span class="sourceLineNo">103</span> private final UUID clusterId;<a name="line.103"></a> -<span class="sourceLineNo">104</span> // All about stopping<a name="line.104"></a> -<span class="sourceLineNo">105</span> private final Server server;<a name="line.105"></a> -<span class="sourceLineNo">106</span> // All logs we are currently tracking<a name="line.106"></a> -<span class="sourceLineNo">107</span> // Index structure of the map is: peer_id->logPrefix/logGroup->logs<a name="line.107"></a> -<span class="sourceLineNo">108</span> private final Map<String, Map<String, SortedSet<String>>> walsById;<a name="line.108"></a> -<span class="sourceLineNo">109</span> // Logs for recovered sources we are currently tracking<a name="line.109"></a> -<span class="sourceLineNo">110</span> private final Map<String, Map<String, SortedSet<String>>> walsByIdRecoveredQueues;<a name="line.110"></a> -<span class="sourceLineNo">111</span> private final Configuration conf;<a name="line.111"></a> -<span class="sourceLineNo">112</span> private final FileSystem fs;<a name="line.112"></a> -<span class="sourceLineNo">113</span> // The paths to the latest log of each wal group, for new coming peers<a name="line.113"></a> -<span class="sourceLineNo">114</span> private Set<Path> latestPaths;<a name="line.114"></a> -<span class="sourceLineNo">115</span> // Path to the wals directories<a name="line.115"></a> -<span class="sourceLineNo">116</span> private final Path logDir;<a name="line.116"></a> -<span class="sourceLineNo">117</span> // Path to the wal archive<a name="line.117"></a> -<span class="sourceLineNo">118</span> private final Path oldLogDir;<a name="line.118"></a> -<span class="sourceLineNo">119</span> private final WALFileLengthProvider walFileLengthProvider;<a name="line.119"></a> -<span class="sourceLineNo">120</span> // The number of ms that we wait before moving znodes, HBASE-3596<a name="line.120"></a> -<span class="sourceLineNo">121</span> private final long sleepBeforeFailover;<a name="line.121"></a> -<span class="sourceLineNo">122</span> // Homemade executer service for replication<a name="line.122"></a> -<span class="sourceLineNo">123</span> private final ThreadPoolExecutor executor;<a name="line.123"></a> -<span class="sourceLineNo">124</span><a name="line.124"></a> -<span class="sourceLineNo">125</span> private final boolean replicationForBulkLoadDataEnabled;<a name="line.125"></a> -<span class="sourceLineNo">126</span><a name="line.126"></a> -<span class="sourceLineNo">127</span> private Connection connection;<a name="line.127"></a> -<span class="sourceLineNo">128</span> private long replicationWaitTime;<a name="line.128"></a> -<span class="sourceLineNo">129</span><a name="line.129"></a> -<span class="sourceLineNo">130</span> private AtomicLong totalBufferUsed = new AtomicLong();<a name="line.130"></a> -<span class="sourceLineNo">131</span><a name="line.131"></a> -<span class="sourceLineNo">132</span> /**<a name="line.132"></a> -<span class="sourceLineNo">133</span> * Creates a replication manager and sets the watch on all the other registered region servers<a name="line.133"></a> -<span class="sourceLineNo">134</span> * @param replicationQueues the interface for manipulating replication queues<a name="line.134"></a> -<span class="sourceLineNo">135</span> * @param replicationPeers<a name="line.135"></a> -<span class="sourceLineNo">136</span> * @param replicationTracker<a name="line.136"></a> -<span class="sourceLineNo">137</span> * @param conf the configuration to use<a name="line.137"></a> -<span class="sourceLineNo">138</span> * @param server the server for this region server<a name="line.138"></a> -<span class="sourceLineNo">139</span> * @param fs the file system to use<a name="line.139"></a> -<span class="sourceLineNo">140</span> * @param logDir the directory that contains all wal directories of live RSs<a name="line.140"></a> -<span class="sourceLineNo">141</span> * @param oldLogDir the directory where old logs are archived<a name="line.141"></a> -<span class="sourceLineNo">142</span> * @param clusterId<a name="line.142"></a> -<span class="sourceLineNo">143</span> */<a name="line.143"></a> -<span class="sourceLineNo">144</span> public ReplicationSourceManager(ReplicationQueues replicationQueues,<a name="line.144"></a> -<span class="sourceLineNo">145</span> ReplicationPeers replicationPeers, ReplicationTracker replicationTracker, Configuration conf,<a name="line.145"></a> -<span class="sourceLineNo">146</span> Server server, FileSystem fs, Path logDir, Path oldLogDir, UUID clusterId,<a name="line.146"></a> -<span class="sourceLineNo">147</span> WALFileLengthProvider walFileLengthProvider) throws IOException {<a name="line.147"></a> -<span class="sourceLineNo">148</span> //CopyOnWriteArrayList is thread-safe.<a name="line.148"></a> -<span class="sourceLineNo">149</span> //Generally, reading is more than modifying.<a name="line.149"></a> -<span class="sourceLineNo">150</span> this.sources = new CopyOnWriteArrayList<>();<a name="line.150"></a> -<span class="sourceLineNo">151</span> this.replicationQueues = replicationQueues;<a name="line.151"></a> -<span class="sourceLineNo">152</span> this.replicationPeers = replicationPeers;<a name="line.152"></a> -<span class="sourceLineNo">153</span> this.replicationTracker = replicationTracker;<a name="line.153"></a> -<span class="sourceLineNo">154</span> this.server = server;<a name="line.154"></a> -<span class="sourceLineNo">155</span> this.walsById = new HashMap<>();<a name="line.155"></a> -<span class="sourceLineNo">156</span> this.walsByIdRecoveredQueues = new ConcurrentHashMap<>();<a name="line.156"></a> -<span class="sourceLineNo">157</span> this.oldsources = new CopyOnWriteArrayList<>();<a name="line.157"></a> -<span class="sourceLineNo">158</span> this.conf = conf;<a name="line.158"></a> -<span class="sourceLineNo">159</span> this.fs = fs;<a name="line.159"></a> -<span class="sourceLineNo">160</span> this.logDir = logDir;<a name="line.160"></a> -<span class="sourceLineNo">161</span> this.oldLogDir = oldLogDir;<a name="line.161"></a> -<span class="sourceLineNo">162</span> this.sleepBeforeFailover =<a name="line.162"></a> -<span class="sourceLineNo">163</span> conf.getLong("replication.sleep.before.failover", 30000); // 30 seconds<a name="line.163"></a> -<span class="sourceLineNo">164</span> this.clusterId = clusterId;<a name="line.164"></a> -<span class="sourceLineNo">165</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.165"></a> -<span class="sourceLineNo">166</span> this.replicationTracker.registerListener(this);<a name="line.166"></a> -<span class="sourceLineNo">167</span> this.replicationPeers.getAllPeerIds();<a name="line.167"></a> -<span class="sourceLineNo">168</span> // It's preferable to failover 1 RS at a time, but with good zk servers<a name="line.168"></a> -<span class="sourceLineNo">169</span> // more could be processed at the same time.<a name="line.169"></a> -<span class="sourceLineNo">170</span> int nbWorkers = conf.getInt("replication.executor.workers", 1);<a name="line.170"></a> -<span class="sourceLineNo">171</span> // use a short 100ms sleep since this could be done inline with a RS startup<a name="line.171"></a> -<span class="sourceLineNo">172</span> // even if we fail, other region servers can take care of it<a name="line.172"></a> -<span class="sourceLineNo">173</span> this.executor = new ThreadPoolExecutor(nbWorkers, nbWorkers,<a name="line.173"></a> -<span class="sourceLineNo">174</span> 100, TimeUnit.MILLISECONDS, new LinkedBlockingQueue<>());<a name="line.174"></a> -<span class="sourceLineNo">175</span> ThreadFactoryBuilder tfb = new ThreadFactoryBuilder();<a name="line.175"></a> -<span class="sourceLineNo">176</span> tfb.setNameFormat("ReplicationExecutor-%d");<a name="line.176"></a> -<span class="sourceLineNo">177</span> tfb.setDaemon(true);<a name="line.177"></a> -<span class="sourceLineNo">178</span> this.executor.setThreadFactory(tfb.build());<a name="line.178"></a> -<span class="sourceLineNo">179</span> this.latestPaths = new HashSet<Path>();<a name="line.179"></a> -<span class="sourceLineNo">180</span> replicationForBulkLoadDataEnabled =<a name="line.180"></a> -<span class="sourceLineNo">181</span> conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.181"></a> -<span class="sourceLineNo">182</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.182"></a> -<span class="sourceLineNo">183</span> this.replicationWaitTime = conf.getLong(HConstants.REPLICATION_SERIALLY_WAITING_KEY,<a name="line.183"></a> -<span class="sourceLineNo">184</span> HConstants.REPLICATION_SERIALLY_WAITING_DEFAULT);<a name="line.184"></a> -<span class="sourceLineNo">185</span> connection = ConnectionFactory.createConnection(conf);<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> /**<a name="line.188"></a> -<span class="sourceLineNo">189</span> * Provide the id of the peer and a log key and this method will figure which<a name="line.189"></a> -<span class="sourceLineNo">190</span> * wal it belongs to and will log, for this region server, the current<a name="line.190"></a> -<span class="sourceLineNo">191</span> * position. It will also clean old logs from the queue.<a name="line.191"></a> -<span class="sourceLineNo">192</span> * @param log Path to the log currently being replicated from<a name="line.192"></a> -<span class="sourceLineNo">193</span> * replication status in zookeeper. It will also delete older entries.<a name="line.193"></a> -<span class="sourceLineNo">194</span> * @param id id of the peer cluster<a name="line.194"></a> -<span class="sourceLineNo">195</span> * @param position current location in the log<a name="line.195"></a> -<span class="sourceLineNo">196</span> * @param queueRecovered indicates if this queue comes from another region server<a name="line.196"></a> -<span class="sourceLineNo">197</span> * @param holdLogInZK if true then the log is retained in ZK<a name="line.197"></a> -<span class="sourceLineNo">198</span> */<a name="line.198"></a> -<span class="sourceLineNo">199</span> public void logPositionAndCleanOldLogs(Path log, String id, long position,<a name="line.199"></a> -<span class="sourceLineNo">200</span> boolean queueRecovered, boolean holdLogInZK) {<a name="line.200"></a> -<span class="sourceLineNo">201</span> String fileName = log.getName();<a name="line.201"></a> -<span class="sourceLineNo">202</span> this.replicationQueues.setLogPosition(id, fileName, position);<a name="line.202"></a> -<span class="sourceLineNo">203</span> if (holdLogInZK) {<a name="line.203"></a> -<span class="sourceLineNo">204</span> return;<a name="line.204"></a> -<span class="sourceLineNo">205</span> }<a name="line.205"></a> -<span class="sourceLineNo">206</span> cleanOldLogs(fileName, id, queueRecovered);<a name="line.206"></a> -<span class="sourceLineNo">207</span> }<a name="line.207"></a> -<span class="sourceLineNo">208</span><a name="line.208"></a> -<span class="sourceLineNo">209</span> /**<a name="line.209"></a> -<span class="sourceLineNo">210</span> * Cleans a log file and all older files from ZK. Called when we are sure that a<a name="line.210"></a> -<span class="sourceLineNo">211</span> * log file is closed and has no more entries.<a name="line.211"></a> -<span class="sourceLineNo">212</span> * @param key Path to the log<a name="line.212"></a> -<span class="sourceLineNo">213</span> * @param id id of the peer cluster<a name="line.213"></a> -<span class="sourceLineNo">214</span> * @param queueRecovered Whether this is a recovered queue<a name="line.214"></a> -<span class="sourceLineNo">215</span> */<a name="line.215"></a> -<span class="sourceLineNo">216</span> public void cleanOldLogs(String key, String id, boolean queueRecovered) {<a name="line.216"></a> -<span class="sourceLineNo">217</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(key);<a name="line.217"></a> -<span class="sourceLineNo">218</span> if (queueRecovered) {<a name="line.218"></a> -<span class="sourceLineNo">219</span> SortedSet<String> wals = walsByIdRecoveredQueues.get(id).get(logPrefix);<a name="line.219"></a> -<span class="sourceLineNo">220</span> if (wals != null && !wals.first().equals(key)) {<a name="line.220"></a> -<span class="sourceLineNo">221</span> cleanOldLogs(wals, key, id);<a name="line.221"></a> -<span class="sourceLineNo">222</span> }<a name="line.222"></a> -<span class="sourceLineNo">223</span> } else {<a name="line.223"></a> -<span class="sourceLineNo">224</span> synchronized (this.walsById) {<a name="line.224"></a> -<span class="sourceLineNo">225</span> SortedSet<String> wals = walsById.get(id).get(logPrefix);<a name="line.225"></a> -<span class="sourceLineNo">226</span> if (wals != null && !wals.first().equals(key)) {<a name="line.226"></a> -<span class="sourceLineNo">227</span> cleanOldLogs(wals, key, id);<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><a name="line.232"></a> -<span class="sourceLineNo">233</span> private void cleanOldLogs(SortedSet<String> wals, String key, String id) {<a name="line.233"></a> -<span class="sourceLineNo">234</span> SortedSet<String> walSet = wals.headSet(key);<a name="line.234"></a> -<span class="sourceLineNo">235</span> LOG.debug("Removing " + walSet.size() + " logs in the list: " + walSet);<a name="line.235"></a> -<span class="sourceLineNo">236</span> for (String wal : walSet) {<a name="line.236"></a> -<span class="sourceLineNo">237</span> this.replicationQueues.removeLog(id, wal);<a name="line.237"></a> -<span class="sourceLineNo">238</span> }<a name="line.238"></a> -<span class="sourceLineNo">239</span> walSet.clear();<a name="line.239"></a> -<span class="sourceLineNo">240</span> }<a name="line.240"></a> -<span class="sourceLineNo">241</span><a name="line.241"></a> -<span class="sourceLineNo">242</span> /**<a name="line.242"></a> -<span class="sourceLineNo">243</span> * Adds a normal source per registered peer cluster and tries to process all<a name="line.243"></a> -<span class="sourceLineNo">244</span> * old region server wal queues<a name="line.244"></a> -<span class="sourceLineNo">245</span> */<a name="line.245"></a> -<span class="sourceLineNo">246</span> void init() throws IOException, ReplicationException {<a name="line.246"></a> -<span class="sourceLineNo">247</span> for (String id : this.replicationPeers.getConnectedPeerIds()) {<a name="line.247"></a> -<span class="sourceLineNo">248</span> addSource(id);<a name="line.248"></a> -<span class="sourceLineNo">249</span> if (replicationForBulkLoadDataEnabled) {<a name="line.249"></a> -<span class="sourceLineNo">250</span> // Check if peer exists in hfile-refs queue, if not add it. This can happen in the case<a name="line.250"></a> -<span class="sourceLineNo">251</span> // when a peer was added before replication for bulk loaded data was enabled.<a name="line.251"></a> -<span class="sourceLineNo">252</span> this.replicationQueues.addPeerToHFileRefs(id);<a name="line.252"></a> -<span class="sourceLineNo">253</span> }<a name="line.253"></a> -<span class="sourceLineNo">254</span> }<a name="line.254"></a> -<span class="sourceLineNo">255</span> AdoptAbandonedQueuesWorker adoptionWorker = new AdoptAbandonedQueuesWorker();<a name="line.255"></a> -<span class="sourceLineNo">256</span> try {<a name="line.256"></a> -<span class="sourceLineNo">257</span> this.executor.execute(adoptionWorker);<a name="line.257"></a> -<span class="sourceLineNo">258</span> } catch (RejectedExecutionException ex) {<a name="line.258"></a> -<span class="sourceLineNo">259</span> LOG.info("Cancelling the adoption of abandoned queues because of " + ex.getMessage());<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> /**<a name="line.263"></a> -<span class="sourceLineNo">264</span> * Add sources for the given peer cluster on this region server. For the newly added peer, we only<a name="line.264"></a> -<span class="sourceLineNo">265</span> * need to enqueue the latest log of each wal group and do replication<a name="line.265"></a> -<span class="sourceLineNo">266</span> * @param id the id of the peer cluster<a name="line.266"></a> -<span class="sourceLineNo">267</span> * @return the source that was created<a name="line.267"></a> -<span class="sourceLineNo">268</span> * @throws IOException<a name="line.268"></a> -<span class="sourceLineNo">269</span> */<a name="line.269"></a> -<span class="sourceLineNo">270</span> @VisibleForTesting<a name="line.270"></a> -<span class="sourceLineNo">271</span> ReplicationSourceInterface addSource(String id) throws IOException, ReplicationException {<a name="line.271"></a> -<span class="sourceLineNo">272</span> ReplicationPeerConfig peerConfig = replicationPeers.getReplicationPeerConfig(id);<a name="line.272"></a> -<span class="sourceLineNo">273</span> ReplicationPeer peer = replicationPeers.getConnectedPeer(id);<a name="line.273"></a> -<span class="sourceLineNo">274</span> ReplicationSourceInterface src = getReplicationSource(this.conf, this.fs, this,<a name="line.274"></a> -<span class="sourceLineNo">275</span> this.replicationQueues, this.replicationPeers, server, id, this.clusterId, peerConfig, peer,<a name="line.275"></a> -<span class="sourceLineNo">276</span> walFileLengthProvider);<a name="line.276"></a> -<span class="sourceLineNo">277</span> synchronized (this.walsById) {<a name="line.277"></a> -<span class="sourceLineNo">278</span> this.sources.add(src);<a name="line.278"></a> -<span class="sourceLineNo">279</span> Map<String, SortedSet<String>> walsByGroup = new HashMap<>();<a name="line.279"></a> -<span class="sourceLineNo">280</span> this.walsById.put(id, walsByGroup);<a name="line.280"></a> -<span class="sourceLineNo">281</span> // Add the latest wal to that source's queue<a name="line.281"></a> -<span class="sourceLineNo">282</span> synchronized (latestPaths) {<a name="line.282"></a> -<span class="sourceLineNo">283</span> if (this.latestPaths.size() > 0) {<a name="line.283"></a> -<span class="sourceLineNo">284</span> for (Path logPath : latestPaths) {<a name="line.284"></a> -<span class="sourceLineNo">285</span> String name = logPath.getName();<a name="line.285"></a> -<span class="sourceLineNo">286</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(name);<a name="line.286"></a> -<span class="sourceLineNo">287</span> SortedSet<String> logs = new TreeSet<>();<a name="line.287"></a> -<span class="sourceLineNo">288</span> logs.add(name);<a name="line.288"></a> -<span class="sourceLineNo">289</span> walsByGroup.put(walPrefix, logs);<a name="line.289"></a> -<span class="sourceLineNo">290</span> try {<a name="line.290"></a> -<span class="sourceLineNo">291</span> this.replicationQueues.addLog(id, name);<a name="line.291"></a> -<span class="sourceLineNo">292</span> } catch (ReplicationException e) {<a name="line.292"></a> -<span class="sourceLineNo">293</span> String message =<a name="line.293"></a> -<span class="sourceLineNo">294</span> "Cannot add log to queue when creating a new source, queueId=" + id<a name="line.294"></a> -<span class="sourceLineNo">295</span> + ", filename=" + name;<a name="line.295"></a> -<span class="sourceLineNo">296</span> server.stop(message);<a name="line.296"></a> -<span class="sourceLineNo">297</span> throw e;<a name="line.297"></a> -<span class="sourceLineNo">298</span> }<a name="line.298"></a> -<span class="sourceLineNo">299</span> src.enqueueLog(logPath);<a name="line.299"></a> -<span class="sourceLineNo">300</span> }<a name="line.300"></a> -<span class="sourceLineNo">301</span> }<a name="line.301"></a> -<span class="sourceLineNo">302</span> }<a name="line.302"></a> -<span class="sourceLineNo">303</span> }<a name="line.303"></a> -<span class="sourceLineNo">304</span> src.startup();<a name="line.304"></a> -<span class="sourceLineNo">305</span> return src;<a name="line.305"></a> -<span class="sourceLineNo">306</span> }<a name="line.306"></a> -<span class="sourceLineNo">307</span><a name="line.307"></a> -<span class="sourceLineNo">308</span> /**<a name="line.308"></a> -<span class="sourceLineNo">309</span> * Delete a complete queue of wals associated with a peer cluster<a name="line.309"></a> -<span class="sourceLineNo">310</span> * @param peerId Id of the peer cluster queue of wals to delete<a name="line.310"></a> -<span class="sourceLineNo">311</span> */<a name="line.311"></a> -<span class="sourceLineNo">312</span> public void deleteSource(String peerId, boolean closeConnection) {<a name="line.312"></a> -<span class="sourceLineNo">313</span> this.replicationQueues.removeQueue(peerId);<a name="line.313"></a> -<span class="sourceLineNo">314</span> if (closeConnection) {<a name="line.314"></a> -<span class="sourceLineNo">315</span> this.replicationPeers.peerDisconnected(peerId);<a name="line.315"></a> -<span class="sourceLineNo">316</span> }<a name="line.316"></a> -<span class="sourceLineNo">317</span> }<a name="line.317"></a> -<span class="sourceLineNo">318</span><a name="line.318"></a> -<span class="sourceLineNo">319</span> /**<a name="line.319"></a> -<span class="sourceLineNo">320</span> * Terminate the replication on this region server<a name="line.320"></a> -<span class="sourceLineNo">321</span> */<a name="line.321"></a> -<span class="sourceLineNo">322</span> public void join() {<a name="line.322"></a> -<span class="sourceLineNo">323</span> this.executor.shutdown();<a name="line.323"></a> -<span class="sourceLineNo">324</span> for (ReplicationSourceInterface source : this.sources) {<a name="line.324"></a> -<span class="sourceLineNo">325</span> source.terminate("Region server is closing");<a name="line.325"></a> -<span class="sourceLineNo">326</span> }<a name="line.326"></a> -<span class="sourceLineNo">327</span> }<a name="line.327"></a> -<span class="sourceLineNo">328</span><a name="line.328"></a> -<span class="sourceLineNo">329</span> /**<a name="line.329"></a> -<span class="sourceLineNo">330</span> * Get a copy of the wals of the first source on this rs<a name="line.330"></a> -<span class="sourceLineNo">331</span> * @return a sorted set of wal names<a name="line.331"></a> -<span class="sourceLineNo">332</span> */<a name="line.332"></a> -<span class="sourceLineNo">333</span> @VisibleForTesting<a name="line.333"></a> -<span class="sourceLineNo">334</span> Map<String, Map<String, SortedSet<String>>> getWALs() {<a name="line.334"></a> -<span class="sourceLineNo">335</span> return Collections.unmodifiableMap(walsById);<a name="line.335"></a> -<span class="sourceLineNo">336</span> }<a name="line.336"></a> -<span class="sourceLineNo">337</span><a name="line.337"></a> -<span class="sourceLineNo">338</span> /**<a name="line.338"></a> -<span class="sourceLineNo">339</span> * Get a copy of the wals of the recovered sources on this rs<a name="line.339"></a> -<span class="sourceLineNo">340</span> * @return a sorted set of wal names<a name="line.340"></a> -<span class="sourceLineNo">341</span> */<a name="line.341"></a> -<span class="sourceLineNo">342</span> @VisibleForTesting<a name="line.342"></a> -<span class="sourceLineNo">343</span> Map<String, Map<String, SortedSet<String>>> getWalsByIdRecoveredQueues() {<a name="line.343"></a> -<span class="sourceLineNo">344</span> return Collections.unmodifiableMap(walsByIdRecoveredQueues);<a name="line.344"></a> -<span class="sourceLineNo">345</span> }<a name="line.345"></a> -<span class="sourceLineNo">346</span><a name="line.346"></a> -<span class="sourceLineNo">347</span> /**<a name="line.347"></a> -<span class="sourceLineNo">348</span> * Get a list of all the normal sources of this rs<a name="line.348"></a> -<span class="sourceLineNo">349</span> * @return lis of all sources<a name="line.349"></a> -<span class="sourceLineNo">350</span> */<a name="line.350"></a> -<span class="sourceLineNo">351</span> public List<ReplicationSourceInterface> getSources() {<a name="line.351"></a> -<span class="sourceLineNo">352</span> return this.sources;<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> /**<a name="line.355"></a> -<span class="sourceLineNo">356</span> * Get a list of all the old sources of this rs<a name="line.356"></a> -<span class="sourceLineNo">357</span> * @return list of all old sources<a name="line.357"></a> -<span class="sourceLineNo">358</span> */<a name="line.358"></a> -<span class="sourceLineNo">359</span> public List<ReplicationSourceInterface> getOldSources() {<a name="line.359"></a> -<span class="sourceLineNo">360</span> return this.oldsources;<a name="line.360"></a> -<span class="sourceLineNo">361</span> }<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> * Get the normal source for a given peer<a name="line.364"></a> -<span class="sourceLineNo">365</span> * @param peerId<a name="line.365"></a> -<span class="sourceLineNo">366</span> * @return the normal source for the give peer if it exists, otherwise null.<a name="line.366"></a> -<span class="sourceLineNo">367</span> */<a name="line.367"></a> -<span class="sourceLineNo">368</span> public ReplicationSourceInterface getSource(String peerId) {<a name="line.368"></a> -<span class="sourceLineNo">369</span> return getSources().stream().filter(s -> s.getPeerId().equals(peerId)).findFirst().orElse(null);<a name="line.369"></a> -<span class="sourceLineNo">370</span> }<a name="line.370"></a> -<span class="sourceLineNo">371</span><a name="line.371"></a> -<span class="sourceLineNo">372</span> @VisibleForTesting<a name="line.372"></a> -<span class="sourceLineNo">373</span> List<String> getAllQueues() {<a name="line.373"></a> -<span class="sourceLineNo">374</span> return replicationQueues.getAllQueues();<a name="line.374"></a> -<span class="sourceLineNo">375</span> }<a name="line.375"></a> -<span class="sourceLineNo">376</span><a name="line.376"></a> -<span class="sourceLineNo">377</span> void preLogRoll(Path newLog) throws IOException {<a name="line.377"></a> -<span class="sourceLineNo">378</span> recordLog(newLog);<a name="line.378"></a> -<span class="sourceLineNo">379</span> String logName = newLog.getName();<a name="line.379"></a> -<span class="sourceLineNo">380</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.380"></a> -<span class="sourceLineNo">381</span> synchronized (latestPaths) {<a name="line.381"></a> -<span class="sourceLineNo">382</span> Iterator<Path> iterator = latestPaths.iterator();<a name="line.382"></a> -<span class="sourceLineNo">383</span> while (iterator.hasNext()) {<a name="line.383"></a> -<span class="sourceLineNo">384</span> Path path = iterator.next();<a name="line.384"></a> -<span class="sourceLineNo">385</span> if (path.getName().contains(logPrefix)) {<a name="line.385"></a> -<span class="sourceLineNo">386</span> iterator.remove();<a name="line.386"></a> -<span class="sourceLineNo">387</span> break;<a name="line.387"></a> -<span class="sourceLineNo">388</span> }<a name="line.388"></a> -<span class="sourceLineNo">389</span> }<a name="line.389"></a> -<span class="sourceLineNo">390</span> this.latestPaths.add(newLog);<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><a name="line.393"></a> -<span class="sourceLineNo">394</span> /**<a name="line.394"></a> -<span class="sourceLineNo">395</span> * Check and enqueue the given log to the correct source. If there's still no source for the<a name="line.395"></a> -<span class="sourceLineNo">396</span> * group to which the given log belongs, create one<a name="line.396"></a> -<span class="sourceLineNo">397</span> * @param logPath the log path to check and enqueue<a name="line.397"></a> -<span class="sourceLineNo">398</span> * @throws IOException<a name="line.398"></a> -<span class="sourceLineNo">399</span> */<a name="line.399"></a> -<span class="sourceLineNo">400</span> private void recordLog(Path logPath) throws IOException {<a name="line.400"></a> -<span class="sourceLineNo">401</span> String logName = logPath.getName();<a name="line.401"></a> -<span class="sourceLineNo">402</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.402"></a> -<span class="sourceLineNo">403</span> // update replication queues on ZK<a name="line.403"></a> -<span class="sourceLineNo">404</span> // synchronize on replicationPeers to avoid adding source for the to-be-removed peer<a name="line.404"></a> -<span class="sourceLineNo">405</span> synchronized (replicationPeers) {<a name="line.405"></a> -<span class="sourceLineNo">406</span> for (String id : replicationPeers.getConnectedPeerIds()) {<a name="line.406"></a> -<span class="sourceLineNo">407</span> try {<a name="line.407"></a> -<span class="sourceLineNo">408</span> this.replicationQueues.addLog(id, logName);<a name="line.408"></a> -<span class="sourceLineNo">409</span> } catch (ReplicationException e) {<a name="line.409"></a> -<span class="sourceLineNo">410</span> throw new IOException("Cannot add log to replication queue"<a name="line.410"></a> -<span class="sourceLineNo">411</span> + " when creating a new source, queueId=" + id + ", filename=" + logName, e);<a name="line.411"></a> -<span class="sourceLineNo">412</span> }<a name="line.412"></a> -<span class="sourceLineNo">413</span> }<a name="line.413"></a> -<span class="sourceLineNo">414</span> }<a name="line.414"></a> -<span class="sourceLineNo">415</span> // update walsById map<a name="line.415"></a> -<span class="sourceLineNo">416</span> synchronized (walsById) {<a name="line.416"></a> -<span class="sourceLineNo">417</span> for (Map.Entry<String, Map<String, SortedSet<String>>> entry : this.walsById.entrySet()) {<a name="line.417"></a> -<span class="sourceLineNo">418</span> String peerId = entry.getKey();<a name="line.418"></a> -<span class="sourceLineNo">419</span> Map<String, SortedSet<String>> walsByPrefix = entry.getValue();<a name="line.419"></a> -<span class="sourceLineNo">420</span> boolean existingPrefix = false;<a name="line.420"></a> -<span class="sourceLineNo">421</span> for (Map.Entry<String, SortedSet<String>> walsEntry : walsByPrefix.entrySet()) {<a name="line.421"></a> -<span class="sourceLineNo">422</span> SortedSet<String> wals = walsEntry.getValue();<a name="line.422"></a> -<span class="sourceLineNo">423</span> if (this.sources.isEmpty()) {<a name="line.423"></a> -<span class="sourceLineNo">424</span> // If there's no slaves, don't need to keep the old wals since<a name="line.424"></a> -<span class="sourceLineNo">425</span> // we only consider the last one when a new slave comes in<a name="line.425"></a> -<span class="sourceLineNo">426</span> wals.clear();<a name="line.426"></a> -<span class="sourceLineNo">427</span> }<a name="line.427"></a> -<span class="sourceLineNo">428</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.428"></a> -<span class="sourceLineNo">429</span> wals.add(logName);<a name="line.429"></a> -<span class="sourceLineNo">430</span> existingPrefix = true;<a name="line.430"></a> -<span class="sourceLineNo">431</span> }<a name="line.431"></a> -<span class="sourceLineNo">432</span> }<a name="line.432"></a> -<span class="sourceLineNo">433</span> if (!existingPrefix) {<a name="line.433"></a> -<span class="sourceLineNo">434</span> // The new log belongs to a new group, add it into this peer<a name="line.434"></a> -<span class="sourceLineNo">435</span> LOG.debug("Start tracking logs for wal group " + logPrefix + " for peer " + peerId);<a name="line.435"></a> -<span class="sourceLineNo">436</span> SortedSet<String> wals = new TreeSet<>();<a name="line.436"></a> -<span class="sourceLineNo">437</span> wals.add(logName);<a name="line.437"></a> -<span class="sourceLineNo">438</span> walsByPrefix.put(logPrefix, wals);<a name="line.438"></a> -<span class="sourceLineNo">439</span> }<a name="line.439"></a> -<span class="sourceLineNo">440</span> }<a name="line.440"></a> -<span class="sourceLineNo">441</span> }<a name="line.441"></a> -<span class="sourceLineNo">442</span> }<a name="line.442"></a> -<span class="sourceLineNo">443</span><a name="line.443"></a> -<span class="sourceLineNo">444</span> void postLogRoll(Path newLog) throws IOException {<a name="line.444"></a> -<span class="sourceLineNo">445</span> // This only updates the sources we own, not the recovered ones<a name="line.445"></a> -<span class="sourceLineNo">446</span> for (ReplicationSourceInterface source : this.sources) {<a name="line.446"></a> -<span class="sourceLineNo">447</span> source.enqueueLog(newLog);<a name="line.447"></a> -<span class="sourceLineNo">448</span> }<a name="line.448"></a> -<span class="sourceLineNo">449</span> }<a name="line.449"></a> -<span class="sourceLineNo">450</span><a name="line.450"></a> -<span class="sourceLineNo">451</span> @VisibleForTesting<a name="line.451"></a> -<span class="sourceLineNo">452</span> public AtomicLong getTotalBufferUsed() {<a name="line.452"></a> -<span class="sourceLineNo">453</span> return totalBufferUsed;<a name="line.453"></a> -<span class="sourceLineNo">454</span> }<a name="line.454"></a> -<span class="sourceLineNo">455</span><a name="line.455"></a> -<span class="sourceLineNo">456</span> /**<a name="line.456"></a> -<span class="sourceLineNo">457</span> * Factory method to create a replication source<a name="line.457"></a> -<span class="sourceLineNo">458</span> * @param conf the configuration to use<a name="line.458"></a> -<span class="sourceLineNo">459</span> * @param fs the file system to use<a name="line.459"></a> -<span class="sourceLineNo">460</span> * @param manager the manager to use<a name="line.460"></a> -<span class="sourceLineNo">461</span> * @param server the server object for this region server<a name="line.461"></a> -<span class="sourceLineNo">462</span> * @param peerId the id of the peer cluster<a name="line.462"></a> -<span class="sourceLineNo">463</span> * @return the created source<a name="line.463"></a> -<span class="sourceLineNo">464</span> * @throws IOException<a name="line.464"></a> -<span class="sourceLineNo">465</span> */<a name="line.465"></a> -<span class="sourceLineNo">466</span> private ReplicationSourceInterface getReplicationSource(Configuration conf, FileSystem fs,<a name="line.466"></a> -<span class="sourceLineNo">467</span> ReplicationSourceManager manager, ReplicationQueues replicationQueues,<a name="line.467"></a> -<span class="sourceLineNo">468</span> ReplicationPeers replicationPeers, Server server, String peerId, UUID clusterId,<a name="line.468"></a> -<span class="sourceLineNo">469</span> ReplicationPeerConfig peerConfig, ReplicationPeer replicationPeer,<a name="line.469"></a> -<span class="sourceLineNo">470</span> WALFileLengthProvider walFileLengthProvider) throws IOException {<a name="line.470"></a> -<span class="sourceLineNo">471</span> RegionServerCoprocessorHost rsServerHost = null;<a name="line.471"></a> -<span class="sourceLineNo">472</span> TableDescriptors tableDescriptors = null;<a name="line.472"></a> -<span class="sourceLineNo">473</span> if (server instanceof HRegionServer) {<a name="line.473"></a> -<span class="sourceLineNo">474</span> rsServerHost = ((HRegionServer) server).getRegionServerCoprocessorHost();<a name="line.474"></a> -<span class="sourceLineNo">475</span> tableDescriptors = ((HRegionServer) server).getTableDescriptors();<a name="line.475"></a> -<span class="sourceLineNo">476</span> }<a name="line.476"></a> -<span class="sourceLineNo">477</span><a name="line.477"></a> -<span class="sourceLineNo">478</span> ReplicationSourceInterface src = ReplicationSourceFactory.create(conf, peerId);<a name="line.478"></a> -<span class="sourceLineNo">479</span><a name="line.479"></a> -<span class="sourceLineNo">480</span> ReplicationEndpoint replicationEndpoint = null;<a name="line.480"></a> -<span class="sourceLineNo">481</span> try {<a name="line.481"></a> -<span class="sourceLineNo">482</span> String replicationEndpointImpl = peerConfig.getReplicationEndpointImpl();<a name="line.482"></a> -<span class="sourceLineNo">483</span> if (replicationEndpointImpl == null) {<a name="line.483"></a> -<span class="sourceLineNo">484</span> // Default to HBase inter-cluster replication endpoint<a name="line.484"></a> -<span class="sourceLineNo">485</span> replicationEndpointImpl = HBaseInterClusterReplicationEndpoint.class.getName();<a name="line.485"></a> -<span class="sourceLineNo">486</span> }<a name="line.486"></a> -<span class="sourceLineNo">487</span> @SuppressWarnings("rawtypes")<a name="line.487"></a> -<span class="sourceLineNo">488</span> Class c = Class.forName(replicationEndpointImpl);<a name="line.488"></a> -<span class="sourceLineNo">489</span> replicationEndpoint = (ReplicationEndpoint) c.newInstance();<a name="line.489"></a> -<span class="sourceLineNo">490</span> if(rsServerHost != null) {<a name="line.490"></a> -<span class="sourceLineNo">491</span> ReplicationEndpoint newReplicationEndPoint = rsServerHost<a name="line.491"></a> -<span class="sourceLineNo">492</span> .postCreateReplicationEndPoint(replicationEndpoint);<a name="line.492"></a> -<span class="sourceLineNo">493</span> if(newReplicationEndPoint != null) {<a name="line.493"></a> -<span class="sourceLineNo">494</span> // Override the newly created endpoint from the hook with configured end point<a name="line.494"></a> -<span class="sourceLineNo">495</span> replicationEndpoint = newReplicationEndPoint;<a name="line.495"></a> -<span class="sourceLineNo">496</span> }<a name="line.496"></a> -<span class="sourceLineNo">497</span> }<a name="line.497"></a> -<span class="sourceLineNo">498</span> } catch (Exception e) {<a name="line.498"></a> -<span class="sourceLineNo">499</span> LOG.warn("Passed replication endpoint implementation throws errors"<a name="line.499"></a> -<span class="sourceLineNo">500</span> + " while initializing ReplicationSource for peer: " + peerId, e);<a name="line.500"></a> -<span class="sourceLineNo">501</span> throw new IOException(e);<a name="line.501"></a> -<span class="sourceLineNo">502</span> }<a name="line.502"></a> -<span class="sourceLineNo">503</span><a name="line.503"></a> -<span class="sourceLineNo">504</span> MetricsSource metrics = new MetricsSource(peerId);<a name="line.504"></a> -<span class="sourceLineNo">505</span> // init replication source<a name="line.505"></a> -<span class="sourceLineNo">506</span> src.init(conf, fs, manager, replicationQueues, replicationPeers, server, peerId, clusterId,<a name="line.506"></a> -<span class="sourceLineNo">507</span> replicationEndpoint, walFileLengthProvider, metrics);<a name="line.507"></a> -<span class="sourceLineNo">508</span><a name="line.508"></a> -<span class="sourceLineNo">509</span> // init replication endpoint<a name="line.509"></a> -<span class="sourceLineNo">510</span> replicationEndpoint.init(new ReplicationEndpoint.Context(conf, replicationPeer.getConfiguration(),<a name="line.510"></a> -<span class="sourceLineNo">511</span> fs, peerId, clusterId, replicationPeer, metrics, tableDescriptors, server));<a name="line.511"></a> -<span class="sourceLineNo">512</span><a name="line.512"></a> -<span class="sourceLineNo">513</span> return src;<a name="line.513"></a> -<span class="sourceLineNo">514</span> }<a name="line.514"></a> -<span class="sourceLineNo">515</span><a name="line.515"></a> -<span class="sourceLineNo">516</span> /**<a name="line.516"></a> -<span class="sourceLineNo">517</span> * Transfer all the queues of the specified to this region server.<a name="line.517"></a> -<span class="sourceLineNo">518</span> * First it tries to grab a lock and if it works it will move the<a name="line.518"></a> -<span class="sourceLineNo">519</span> * znodes and finally will delete the old znodes.<a name="line.519"></a> -<span class="sourceLineNo">520</span> *<a name="line.520"></a> -<span class="sourceLineNo">521</span> * It creates one old source for any type of source of the old rs.<a name="line.521"></a> -<span class="sourceLineNo">522</span> * @param rsZnode<a name="line.522"></a> -<span class="sourceLineNo">523</span> */<a name="line.523"></a> -<span class="sourceLineNo">524</span> private void transferQueues(String rsZnode) {<a name="line.524"></a> -<span class="sourceLineNo">525</span> NodeFailoverWorker transfer =<a name="line.525"></a> -<span class="sourceLineNo">526</span> new NodeFailoverWorker(rsZnode, this.replicationQueues, this.replicationPeers,<a name="line.526"></a> -<span class="sourceLineNo">527</span> this.clusterId);<a name="line.527"></a> -<span class="sourceLineNo">528</span> try {<a name="line.528"></a> -<span class="sourceLineNo">529</span> this.executor.execute(transfer);<a name="line.529"></a> -<span class="sourceLineNo">530</span> } catch (RejectedExecutionException ex) {<a name="line.530"></a> -<span class="sourceLineNo">531</span> LOG.info("Cancelling the transfer of " + rsZnode + " because of " + ex.getMessage());<a name="line.531"></a> -<span class="sourceLineNo">532</span> }<a name="line.532"></a> -<span class="sourceLineNo">533</span> }<a name="line.533"></a> -<span class="sourceLineNo">534</span><a name="line.534"></a> -<span class="sourceLineNo">535</span> /**<a name="line.535"></a> -<span class="sourceLineNo">536</span> * Clear the references to the specified old source<a name="line.536"></a> -<span class="sourceLineNo">537</span> * @param src source to clear<a name="line.537"></a> -<span class="sourceLineNo">538</span> */<a name="line.538"></a> -<span class="sourceLineNo">539</span> public void closeRecoveredQueue(ReplicationSourceInterface src) {<a name="line.539"></a> -<span class="sourceLineNo">540</span> LOG.info("Done with the recovered queue " + src.getPeerClusterZnode());<a name="line.540"></a> -<span class="sourceLineNo">541</span> if (src instanceof ReplicationSource) {<a name="line.541"></a> -<span class="sourceLineNo">542</span> ((ReplicationSource) src).getSourceMetrics().clear();<a name="line.542"></a> -<span class="sourceLineNo">543</span> }<a name="line.543"></a> -<span class="sourceLineNo">544</span> this.oldsources.remove(src);<a name="line.544"></a> -<span class="sourceLineNo">545</span> deleteSource(src.getPeerClusterZnode(), false);<a name="line.545"></a> -<span class="sourceLineNo">546</span> this.walsByIdRecoveredQueues.remove(src.getPeerClusterZnode());<a name="line.546"></a> -<span class="sourceLineNo">547</span> }<a name="line.547"></a> -<span class="sourceLineNo">548</span><a name="line.548"></a> -<span class="sourceLineNo">549</span> /**<a name="line.549"></a> -<span class="sourceLineNo">550</span> * Clear the references to the specified old source<a name="line.550"></a> -<span class="sourceLineNo">551</span> * @param src source to clear<a name="line.551"></a> -<span class="sourceLineNo">552</span> */<a name="line.552"></a> -<span class="sourceLineNo">553</span> public void closeQueue(ReplicationSourceInterface src) {<a name="line.553"></a> -<span class="sourceLineNo">554</span> LOG.info("Done with the queue " + src.getPeerClusterZnode());<a name="line.554"></a> -<span class="sourceLineNo">555</span> src.getSourceMetrics().clear();<a name="line.555"></a> -<span class="sourceLineNo">556</span> this.sources.remove(src);<a name="line.556"></a> -<span class="sourceLineNo">557</span> deleteSource(src.getPeerClusterZnode(), true);<a name="line.557"></a> -<span class="sourceLineNo">558</span> this.walsById.remove(src.getPeerClusterZnode());<a name="line.558"></a> -<span class="sourceLineNo">559</span> }<a name="line.559"></a> -<span class="sourceLineNo">560</span><a name="line.560"></a> -<span class="sourceLineNo">561</span> /**<a name="line.561"></a> -<span class="sourceLineNo">562</span> * Thie method first deletes all the recovered sources for the specified<a name="line.562"></a> -<span class="sourceLineNo">563</span> * id, then deletes the normal source (deleting all related data in ZK).<a name="line.563"></a> -<span class="sourceLineNo">564</span> * @param id The id of the peer cluster<a name="line.564"></a> -<span class="sourceLineNo">565</span> */<a name="line.565"></a> -<span class="sourceLineNo">566</span> public void removePeer(String id) {<a name="line.566"></a> -<span class="sourceLineNo">567</span> LOG.info("Closing the following queue " + id + ", currently have "<a name="line.567"></a> -<span class="sourceLineNo">568</span> + sources.size() + " and another "<a name="line.568"></a> -<span class="sourceLineNo">569</span> + oldsources.size() + " that were recovered");<a name="line.569"></a> -<span class="sourceLineNo">570</span> String terminateMessage = "Replication stream was removed by a user";<a name="line.570"></a> -<span class="sourceLineNo">571</span> List<ReplicationSourceInterface> oldSourcesToDelete = new ArrayList<>();<a name="line.571"></a> -<span class="sourceLineNo">572</span> // synchronized on oldsources to avoid adding recovered source for the to-be-removed peer<a name="line.572"></a> -<span class="sourceLineNo">573</span> // see NodeFailoverWorker.run<a name="line.573"></a> -<span class="sourceLineNo">574</span> synchronized (oldsources) {<a name="line.574"></a> -<span class="sourceLineNo">575</span> // First close all the recovered sources for this peer<a name="line.575"></a> -<span class="sourceLineNo">576</span> for (ReplicationSourceInterface src : oldsources) {<a name="line.576"></a> -<span class="sourceLineNo">577</span> if (id.equals(src.getPeerId())) {<a name="line.577"></a> -<span class="sourceLineNo">578</span> oldSourcesToDelete.add(src);<a name="line.578"></a> -<span class="sourceLineNo">579</span> }<a name="line.579"></a> -<span class="sourceLineNo">580</span> }<a name="line.580"></a> -<span class="sourceLineNo">581</span> for (ReplicationSourceInterface src : oldSourcesToDelete) {<a name="line.581"></a> -<span class="sourceLineNo">582</span> src.terminate(terminateMessage);<a name="line.582"></a> -<span class="sourceLineNo">583</span> closeRecoveredQueue(src);<a name="line.583"></a> -<span class="sourceLineNo">584</span> }<a name="line.584"></a> -<span class="sourceLineNo">585</span> }<a name="line.585"></a> -<span class="sourceLineNo">586</span> LOG.info("Number of deleted recovered sources for " + id + ": "<a name="line.586"></a> -<span class="sourceLineNo">587</span> + oldSourcesToDelete.size());<a name="line.587"></a> -<span class="sourceLineNo">588</span> // Now look for the one on this cluster<a name="line.588"></a> -<span class="sourceLineNo">589</span> List<ReplicationSourceInterface> srcToRemove = new ArrayList<>();<a name="line.589"></a> -<span class="sourceLineNo">590</span> // synchronize on replicationPeers to avoid adding source for the to-be-removed peer<a name="line.590"></a> -<span class="sourceLineNo">591</span> synchronized (this.replicationPeers) {<a name="line.591"></a> -<span class="sourceLineNo">592</span> for (ReplicationSourceInterface src : this.sources) {<a name="line.592"></a> -<span class="sourceLineNo">593</span> if (id.equals(src.getPeerId())) {<a name="line.593"></a> -<span class="sourceLineNo">594</span> srcToRemove.add(src);<a name="line.594"></a> -<span class="sourceLineNo">595</span> }<a name="line.595"></a> -<span class="sourceLineNo">596</span> }<a name="line.596"></a> -<span class="sourceLineNo">597</span> if (srcToRemove.isEmpty()) {<a name="line.597"></a> -<span class="sourceLineNo">598</span> LOG.error("The peer we wanted to remove is missing a ReplicationSourceInterface. " +<a name="line.598"></a> -<span class="sourceLineNo">599</span> "This could mean that ReplicationSourceInterface initialization failed for this peer " +<a name="line.599"></a> -<span class="sourceLineNo">600</span> "and that replication on this peer may not be caught up. peerId=" + id);<a name="line.600"></a> -<span class="sourceLineNo">601</span> }<a name="line.601"></a> -<span class="sourceLineNo">602</span> for (ReplicationSourceInterface toRemove : srcToRemove) {<a name="line.602"></a> -<span class="sourceLineNo">603</span> toRemove.terminate(terminateMessage);<a name="line.603"></a> -<span class="sourceLineNo">604</span> closeQueue(toRemove);<a name="line.604"></a> -<span class="sourceLineNo">605</span> }<a name="line.605"></a> -<span class="sourceLineNo">606</span> deleteSource(id, true);<a name="line.606"></a> -<span class="sourceLineNo">607</span> }<a name="line.607"></a> -<span class="sourceLineNo">608</span> }<a name="line.608"></a> -<span class="sourceLineNo">609</span><a name="line.609"></a> -<span class="sourceLineNo">610</span> @Override<a name="line.610"></a> -<span class="sourceLineNo">611</span> public void regionServerRemoved(String regionserver) {<a name="line.611"></a> -<span class="sourceLineNo">612</span> transferQueues(regionserver);<a name="line.612"></a> -<span class="sourceLineNo">613</span> }<a name="line.613"></a> -<span class="sourceLineNo">614</span><a name="line.614"></a> -<span class="sourceLineNo">615</span> @Override<a name="line.615"></a> -<span class="sourceLineNo">616</span> public void peerRemoved(String peerId) {<a name="line.616"></a> -<span class="sourceLineNo">617</span> removePeer(peerId);<a name="line.617"></a> -<span class="sourceLineNo">618</span> this.replicationQueues.removePeerFromHFileRefs(peerId);<a name="line.618"></a> +<span class="sourceLineNo">043</span>import org.apache.commons.logging.Log;<a name="line.43"></a> +<span class="sourceLineNo">044</span>import org.apache.commons.logging.LogFactory;<a name="line.44"></a> +<span class="sourceLineNo">045</span>import org.apache.hadoop.conf.Configuration;<a name="line.45"></a> +<span class="sourceLineNo">046</span>import org.apache.hadoop.fs.FileSystem;<a name="line.46"></a> +<span class="sourceLineNo">047</span>import org.apache.hadoop.fs.Path;<a name="line.47"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.HConstants;<a name="line.48"></a> +<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.MetaTableAccessor;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.Server;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.TableDescriptors;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.TableName;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.client.Connection;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.client.ConnectionFactory;<a name="line.54"></a> +<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.regionserver.HRegionServer;<a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.regionserver.RegionServerCoprocessorHost;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.ReplicationEndpoint;<a name="line.57"></a> +<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.58"></a> +<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.59"></a> +<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.60"></a> +<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.ReplicationPeerConfig;<a name="line.61"></a> +<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.62"></a> +<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.63"></a> +<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.replication.ReplicationQueues;<a name="line.64"></a> +<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.65"></a> +<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.66"></a> +<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.67"></a> +<span class="sourceLineNo">068</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.68"></a> +<span class="sourceLineNo">069</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.69"></a> +<span class="sourceLineNo">070</span><a name="line.70"></a> +<span class="sourceLineNo">071</span>import org.apache.hadoop.hbase.shaded.com.google.common.annotations.VisibleForTesting;<a name="line.71"></a> +<span class="sourceLineNo">072</span>import org.apache.hadoop.hbase.shaded.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.72"></a> +<span class="sourceLineNo">073</span><a name="line.73"></a> +<span class="sourceLineNo">074</span>/**<a name="line.74"></a> +<span class="sourceLineNo">075</span> * This class is responsible to manage all the replication<a name="line.75"></a> +<span class="sourceLineNo">076</span> * sources. There are two classes of sources:<a name="line.76"></a> +<span class="sourceLineNo">077</span> * <ul><a name="line.77"></a> +<span class="sourceLineNo">078</span> * <li> Normal sources are persistent and one per peer cluster</li><a name="line.78"></a> +<span class="sourceLineNo">079</span> * <li> Old sources are recovered from a failed region server and our<a name="line.79"></a> +<span class="sourceLineNo">080</span> * only goal is to finish replicating the WAL queue it had up in ZK</li><a name="line.80"></a> +<span class="sourceLineNo">081</span> * </ul><a name="line.81"></a> +<span class="sourceLineNo">082</span> *<a name="line.82"></a> +<span class="sourceLineNo">083</span> * When a region server dies, this class uses a watcher to get notified and it<a name="line.83"></a> +<span class="sourceLineNo">084</span> * tries to grab a lock in order to transfer all the queues in a local<a name="line.84"></a> +<span class="sourceLineNo">085</span> * old source.<a name="line.85"></a> +<span class="sourceLineNo">086</span> *<a name="line.86"></a> +<span class="sourceLineNo">087</span> * This class implements the ReplicationListener interface so that it can track changes in<a name="line.87"></a> +<span class="sourceLineNo">088</span> * replication state.<a name="line.88"></a> +<span class="sourceLineNo">089</span> */<a name="line.89"></a> +<span class="sourceLineNo">090</span>@InterfaceAudience.Private<a name="line.90"></a> +<span class="sourceLineNo">091</span>public class ReplicationSourceManager implements ReplicationListener {<a name="line.91"></a> +<span class="sourceLineNo">092</span> private static final Log LOG =<a name="line.92"></a> +<span class="sourceLineNo">093</span> LogFactory.getLog(ReplicationSourceManager.class);<a name="line.93"></a> +<span class="sourceLineNo">094</span> // List of all the sources that read this RS's logs<a name="line.94"></a> +<span class="sourceLineNo">095</span> private final List<ReplicationSourceInterface> sources;<a name="line.95"></a> +<span class="sourceLineNo">096</span> // List of all the sources we got from died RSs<a name="line.96"></a> +<span class="sourceLineNo">097</span> private final List<ReplicationSourceInterface> oldsources;<a name="line.97"></a> +<span class="sourceLineNo">098</span> private final ReplicationQueues replicationQueues;<a name="line.98"></a> +<span class="sourceLineNo">099</span> private final ReplicationTracker replicationTracker;<a name="line.99"></a> +<span class="sourceLineNo">100</span> private final ReplicationPeers replicationPeers;<a name="line.100"></a> +<span class="sourceLineNo">101</span> // UUID for this cluster<a name="line.101"></a> +<span class="sourceLineNo">102</span> private final UUID clusterId;<a name="line.102"></a> +<span class="sourceLineNo">103</span> // All about stopping<a name="line.103"></a> +<span class="sourceLineNo">104</span> private final Server server;<a name="line.104"></a> +<span class="sourceLineNo">105</span> // All logs we are currently tracking<a name="line.105"></a> +<span class="sourceLineNo">106</span> // Index structure of the map is: peer_id->logPrefix/logGroup->logs<a name="line.106"></a> +<span class="sourceLineNo">107</span> private final Map<String, Map<String, SortedSet<String>>> walsById;<a name="line.107"></a> +<span class="sourceLineNo">108</span> // Logs for recovered sources we are currently tracking<a name="line.108"></a> +<span class="sourceLineNo">109</span> private final Map<String, Map<String, SortedSet<String>>> walsByIdRecoveredQueues;<a name="line.109"></a> +<span class="sourceLineNo">110</span> private final Configuration conf;<a name="line.110"></a> +<span class="sourceLineNo">111</span> private final FileSystem fs;<a name="line.111"></a> +<span class="sourceLineNo">112</span> // The paths to the latest log of each wal group, for new coming peers<a name="line.112"></a> +<span class="sourceLineNo">113</span> private Set<Path> latestPaths;<a name="line.113"></a> +<span class="sourceLineNo">114</span> // Path to the wals directories<a name="line.114"></a> +<span class="sourceLineNo">115</span> private final Path logDir;<a name="line.115"></a> +<span class="sourceLineNo">116</span> // Path to the wal archive<a name="line.116"></a> +<span class="sourceLineNo">117</span> private final Path oldLogDir;<a name="line.117"></a> +<span class="sourceLineNo">118</span> private final WALFileLengthProvider walFileLengthProvider;<a name="line.118"></a> +<span class="sourceLineNo">119</span> // The number of ms that we wait before moving znodes, HBASE-3596<a name="line.119"></a> +<span class="sourceLineNo">120</span> private final long sleepBeforeFailover;<a name="line.120"></a> +<span class="sourceLineNo">121</span> // Homemade executer service for replication<a name="line.121"></a> +<span class="sourceLineNo">122</span> private final ThreadPoolExecutor executor;<a name="line.122"></a> +<span class="sourceLineNo">123</span><a name="line.123"></a> +<span class="sourceLineNo">124</span> private final boolean replicationForBulkLoadDataEnabled;<a name="line.124"></a> +<span class="sourceLineNo">125</span><a name="line.125"></a> +<span class="sourceLineNo">126</span> private Connection connection;<a name="line.126"></a> +<span class="sourceLineNo">127</span> private long replicationWaitTime;<a name="line.127"></a> +<span class="sourceLineNo">128</span><a name="line.128"></a> +<span class="sourceLineNo">129</span> private AtomicLong totalBufferUsed = new AtomicLong();<a name="line.129"></a> +<span class="sourceLineNo">130</span><a name="line.130"></a> +<span class="sourceLineNo">131</span> /**<a name="line.131"></a> +<span class="sourceLineNo">132</span> * Creates a replication manager and sets the watch on all the other registered region servers<a name="line.132"></a> +<span class="sourceLineNo">133</span> * @param replicationQueues the interface for manipulating replication queues<a name="line.133"></a> +<span class="sourceLineNo">134</span> * @param replicationPeers<a name="line.134"></a> +<span class="sourceLineNo">135</span> * @param replicationTracker<a name="line.135"></a> +<span class="sourceLineNo">136</span> * @param conf the configuration to use<a name="line.136"></a> +<span class="sourceLineNo">137</span> * @param server the server for this region server<a name="line.137"></a> +<span class="sourceLineNo">138</span> * @param fs the file system to use<a name="line.138"></a> +<span class="sourceLineNo">139</span> * @param logDir the directory that contains all wal directories of live RSs<a name="line.139"></a> +<span class="sourceLineNo">140</span> * @param oldLogDir the directory where old logs are archived<a name="line.140"></a> +<span class="sourceLineNo">141</span> * @param clusterId<a name="line.141"></a> +<span class="sourceLineNo">142</span> */<a name="line.142"></a> +<span class="sourceLineNo">143</span> public ReplicationSourceManager(ReplicationQueues replicationQueues,<a name="line.143"></a> +<span class="sourceLineNo">144</span> ReplicationPeers replicationPeers, ReplicationTracker replicationTracker, Configuration conf,<a name="line.144"></a> +<span class="sourceLineNo">145</span> Server server, FileSystem fs, Path logDir, Path oldLogDir, UUID clusterId,<a name="line.145"></a> +<span class="sourceLineNo">146</span> WALFileLengthProvider walFileLengthProvider) throws IOException {<a name="line.146"></a> +<span class="sourceLineNo">147</span> //CopyOnWriteArrayList is thread-safe.<a name="line.147"></a> +<span class="sourceLineNo">148</span> //Generally, reading is more than modifying.<a name="line.148"></a> +<span class="sourceLineNo">149</span> this.sources = new CopyOnWriteArrayList<>();<a name="line.149"></a> +<span class="sourceLineNo">150</span> this.replicationQueues = replicationQueues;<a name="line.150"></a> +<span class="sourceLineNo">151</span> this.replicationPeers = replicationPeers;<a name="line.151"></a> +<span class="sourceLineNo">152</span> this.replicationTracker = replicationTracker;<a name="line.152"></a> +<span class="sourceLineNo">153</span> this.server = server;<a name="line.153"></a> +<span class="sourceLineNo">154</span> this.walsById = new HashMap<>();<a name="line.154"></a> +<span class="sourceLineNo">155</span> this.walsByIdRecoveredQueues = new ConcurrentHashMap<>();<a name="line.155"></a> +<span class="sourceLineNo">156</span> this.oldsources = new CopyOnWriteArrayList<>();<a name="line.156"></a> +<span class="sourceLineNo">157</span> this.conf = conf;<a name="line.157"></a> +<span class="sourceLineNo">158</span> this.fs = fs;<a name="line.158"></a> +<span class="sourceLineNo">159</span> this.logDir = logDir;<a name="line.159"></a> +<span class="sourceLineNo">160</span> this.oldLogDir = oldLogDir;<a name="line.160"></a> +<span class="sourceLineNo">161</span> this.sleepBeforeFailover =<a name="line.161"></a> +<span class="sourceLineNo">162</span> conf.getLong("replication.sleep.before.failover", 30000); // 30 seconds<a name="line.162"></a> +<span class="sourceLineNo">163</span> this.clusterId = clusterId;<a name="line.163"></a> +<span class="sourceLineNo">164</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.164"></a> +<span class="sourceLineNo">165</span> this.replicationTracker.registerListener(this);<a name="line.165"></a> +<span class="sourceLineNo">166</span> this.replicationPeers.getAllPeerIds();<a name="line.166"></a> +<span class="sourceLineNo">167</span> // It's preferable to failover 1 RS at a time, but with good zk servers<a name="line.167"></a> +<span class="sourceLineNo">168</span> // more could be processed at the same time.<a name="line.168"></a> +<span class="sourceLineNo">169</span> int nbWorkers = conf.getInt("replication.executor.workers", 1);<a name="line.169"></a> +<span class="sourceLineNo">170</span> // use a short 100ms sleep since this could be done inline with a RS startup<a name="line.170"></a> +<span class="sourceLineNo">171</span> // even if we fail, other region servers can take care of it<a name="line.171"></a> +<span class="sourceLineNo">172</span> this.executor = new ThreadPoolExecutor(nbWorkers, nbWorkers,<a name="line.172"></a> +<span class="sourceLineNo">173</span> 100, TimeUnit.MILLISECONDS, new LinkedBlockingQueue<>());<a name="line.173"></a> +<span class="sourceLineNo">174</span> ThreadFactoryBuilder tfb = new ThreadFactoryBuilder();<a name="line.174"></a> +<span class="sourceLineNo">175</span> tfb.setNameFormat("ReplicationExecutor-%d");<a name="line.175"></a> +<span class="sourceLineNo">176</span> tfb.setDaemon(true);<a name="line.176"></a> +<span class="sourceLineNo">177</span> this.executor.setThreadFactory(tfb.build());<a name="line.177"></a> +<span class="sourceLineNo">178</span> this.latestPaths = new HashSet<Path>();<a name="line.178"></a> +<span class="sourceLineNo">179</span> replicationForBulkLoadDataEnabled =<a name="line.179"></a> +<span class="sourceLineNo">180</span> conf.getBoolean(HConstants.REPLICATION_BULKLOAD_ENABLE_KEY,<a name="line.180"></a> +<span class="sourceLineNo">181</span> HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.181"></a> +<span class="sourceLineNo">182</span> this.replicationWaitTime = conf.getLong(HConstants.REPLICATION_SERIALLY_WAITING_KEY,<a name="line.182"></a> +<span class="sourceLineNo">183</span> HConstants.REPLICATION_SERIALLY_WAITING_DEFAULT);<a name="line.183"></a> +<span class="sourceLineNo">184</span> connection = ConnectionFactory.createConnection(conf);<a name="line.184"></a> +<span class="sourceLineNo">185</span> }<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> * Provide the id of the peer and a log key and this method will figure which<a name="line.188"></a> +<span class="sourceLineNo">189</span> * wal it belongs to and will log, for this region server, the current<a name="line.189"></a> +<span class="sourceLineNo">190</span> * position. It will also clean old logs from the queue.<a name="line.190"></a> +<span class="sourceLineNo">191</span> * @param log Path to the log currently being replicated from<a name="line.191"></a> +<span class="sourceLineNo">192</span> * replication status in zookeeper. It will also delete older entries.<a name="line.192"></a> +<span class="sourceLineNo">193</span> * @param id id of the peer cluster<a name="line.193"></a> +<span class="sourceLineNo">194</span> * @param position current location in the log<a name="line.194"></a> +<span class="sourceLineNo">195</span> * @param queueRecovered indicates if this queue comes from another region server<a name="line.195"></a> +<span class="sourceLineNo">196</span> * @param holdLogInZK if true then the log is retained in ZK<a name="line.196"></a> +<span class="sourceLineNo">197</span> */<a name="line.197"></a> +<span class="sourceLineNo">198</span> public void logPositionAndCleanOldLogs(Path log, String id, long position,<a name="line.198"></a> +<span class="sourceLineNo">199</span> boolean queueRecovered, boolean holdLogInZK) {<a name="line.199"></a> +<span class="sourceLineNo">200</span> String fileName = log.getName();<a name="line.200"></a> +<span class="sourceLineNo">201</span> this.replicationQueues.setLogPosition(id, fileName, position);<a name="line.201"></a> +<span class="sourceLineNo">202</span> if (holdLogInZK) {<a name="line.202"></a> +<span class="sourceLineNo">203</span> return;<a name="line.203"></a> +<span class="sourceLineNo">204</span> }<a name="line.204"></a> +<span class="sourceLineNo">205</span> cleanOldLogs(fileName, id, queueRecovered);<a name="line.205"></a> +<span class="sourceLineNo">206</span> }<a name="line.206"></a> +<span class="sourceLineNo">207</span><a name="line.207"></a> +<span class="sourceLineNo">208</span> /**<a name="line.208"></a> +<span class="sourceLineNo">209</span> * Cleans a log file and all older files from ZK. Called when we are sure that a<a name="line.209"></a> +<span class="sourceLineNo">210</span> * log file is closed and has no more entries.<a name="line.210"></a> +<span class="sourceLineNo">211</span> * @param key Path to the log<a name="line.211"></a> +<span class="sourceLineNo">212</span> * @param id id of the peer cluster<a name="line.212"></a> +<span class="sourceLineNo">213</span> * @param queueRecovered Whether this is a recovered queue<a name="line.213"></a> +<span class="sourceLineNo">214</span> */<a name="line.214"></a> +<span class="sourceLineNo">215</span> public void cleanOldLogs(String key, String id, boolean queueRecovered) {<a name="line.215"></a> +<span class="sourceLineNo">216</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(key);<a name="line.216"></a> +<span class="sourceLineNo">217</span> if (queueRecovered) {<a name="line.217"></a> +<span class="sourceLineNo">218</span> SortedSet<String> wals = walsByIdRecoveredQueues.get(id).get(logPrefix);<a name="line.218"></a> +<span class="sourceLineNo">219</span> if (wals != null && !wals.first().equals(key)) {<a name="line.219"></a> +<span class="sourceLineNo">220</span> cleanOldLogs(wals, key, id);<a name="line.220"></a> +<span class="sourceLineNo">221</span> }<a name="line.221"></a> +<span class="sourceLineNo">222</span> } else {<a name="line.222"></a> +<span class="sourceLineNo">223</span> synchronized (this.walsById) {<a name="line.223"></a> +<span class="sourceLineNo">224</span> SortedSet<String> wals = walsById.get(id).get(logPrefix);<a name="line.224"></a> +<span class="sourceLineNo">225</span> if (wals != null && !wals.first().equals(key)) {<a name="line.225"></a> +<span class="sourceLineNo">226</span> cleanOldLogs(wals, key, id);<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 cleanOldLogs(SortedSet<String> wals, String key, String id) {<a name="line.232"></a> +<span class="sourceLineNo">233</span> SortedSet<String> walSet = wals.headSet(key);<a name="line.233"></a> +<span class="sourceLineNo">234</span> LOG.debug("Removing " + walSet.size() + " logs in the list: " + walSet);<a name="line.234"></a> +<span class="sourceLineNo">235</span> for (String wal : walSet) {<a name="line.235"></a> +<span class="sourceLineNo">236</span> this.replicationQueues.removeLog(id, wal);<a name="line.236"></a> +<span class="sourceLineNo">237</span> }<a name="line.237"></a> +<span class="sourceLineNo">238</span> walSet.clear();<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> /**<a name="line.241"></a> +<span class="sourceLineNo">242</span> * Adds a normal source per registered peer cluster and tries to process all<a name="line.242"></a> +<span class="sourceLineNo">243</span> * old region server wal queues<a name="line.243"></a> +<span class="sourceLineNo">244</span> */<a name="line.244"></a> +<span class="sourceLineNo">245</span> void init() throws IOException, ReplicationException {<a name="line.245"></a> +<span class="sourceLineNo">246</span> for (String id : this.replicationPeers.getConnectedPeerIds()) {<a name="line.246"></a> +<span class="sourceLineNo">247</span> addSource(id);<a name="line.247"></a> +<span class="sourceLineNo">248</span> if (replicationForBulkLoadDataEnabled) {<a name="line.248"></a> +<span class="sourceLineNo">249</span> // Check if peer exists in hfile-refs queue, if not add it. This can happen in the case<a name="line.249"></a> +<span class="sourceLineNo">250</span> // when a peer was added before replication for bulk loaded data was enabled.<a name="line.250"></a> +<span class="sourceLineNo">251</span> this.replicationQueues.addPeerToHFileRefs(id);<a name="line.251"></a> +<span class="sourceLineNo">252</span> }<a name="line.252"></a> +<span class="sourceLineNo">253</span> }<a name="line.253"></a> +<span class="sourceLineNo">254</span> AdoptAbandonedQueuesWorker adoptionWorker = new AdoptAbandonedQueuesWorker();<a name="line.254"></a> +<span class="sourceLineNo">255</span> try {<a name="line.255"></a> +<span class="sourceLineNo">256</span> this.executor.execute(adoptionWorker);<a name="line.256"></a> +<span class="sourceLineNo">257</span> } catch (RejectedExecutionException ex) {<a name="line.257"></a> +<span class="sourceLineNo">258</span> LOG.info("Cancelling the adoption of abandoned queues because of " + ex.getMessage());<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> * Add sources for the given peer cluster on this region server. For the newly added peer, we only<a name="line.263"></a> +<span class="sourceLineNo">264</span> * need to enqueue the latest log of each wal group and do replication<a name="line.264"></a> +<span class="sourceLineNo">265</span> * @param id the id of the peer cluster<a name="line.265"></a> +<span class="sourceLineNo">266</span> * @return the source that was created<a name="line.266"></a> +<span class="sourceLineNo">267</span> * @throws IOException<a name="line.267"></a> +<span class="sourceLineNo">268</span> */<a name="line.268"></a> +<span class="sourceLineNo">269</span> @VisibleForTesting<a name="line.269"></a> +<span class="sourceLineNo">270</span> ReplicationSourceInterface addSource(String id) throws IOException, ReplicationException {<a name="line.270"></a> +<span class="sourceLineNo">271</span> ReplicationPeerConfig peerConfig = replicationPeers.getReplicationPeerConfig(id);<a name="line.271"></a> +<span class="sourceLineNo">272</span> ReplicationPeer peer = replicationPeers.getConnectedPeer(id);<a name="line.272"></a> +<span class="sourceLineNo">273</span> ReplicationSourceInterface src = getReplicationSource(this.conf, this.fs, this,<a name="line.273"></a> +<span class="sourceLineNo">274</span> this.replicationQueues, this.replicationPeers, server, id, this.clusterId, peerConfig, peer,<a name="line.274"></a> +<span class="sourceLineNo">275</span> walFileLengthProvider);<a name="line.275"></a> +<span class="sourceLineNo">276</span> synchronized (this.walsById) {<a name="line.276"></a> +<span class="sourceLineNo">277</span> this.sources.add(src);<a name="line.277"></a> +<span class="sourceLineNo">278</span> Map<String, SortedSet<String>> walsByGroup = new HashMap<>();<a name="line.278"></a> +<span class="sourceLineNo">279</span> this.walsById.put(id, walsByGroup);<a name="line.279"></a> +<span class="sourceLineNo">280</span> // Add the latest wal to that source's queue<a name="line.280"></a> +<span class="sourceLineNo">281</span> synchronized (latestPaths) {<a name="line.281"></a> +<span class="sourceLineNo">282</span> if (this.latestPaths.size() > 0) {<a name="line.282"></a> +<span class="sourceLineNo">283</span> for (Path logPath : latestPaths) {<a name="line.283"></a> +<span class="sourceLineNo">284</span> String name = logPath.getName();<a name="line.284"></a> +<span class="sourceLineNo">285</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(name);<a name="line.285"></a> +<span class="sourceLineNo">286</span> SortedSet<String> logs = new TreeSet<>();<a name="line.286"></a> +<span class="sourceLineNo">287</span> logs.add(name);<a name="line.287"></a> +<span class="sourceLineNo">288</span> walsByGroup.put(walPrefix, logs);<a name="line.288"></a> +<span class="sourceLineNo">289</span> try {<a name="line.289"></a> +<span class="sourceLineNo">290</span> this.replicationQueues.addLog(id, name);<a name="line.290"></a> +<span class="sourceLineNo">291</span> } catch (ReplicationException e) {<a name="line.291"></a> +<span class="sourceLineNo">292</span> String message =<a name="line.292"></a> +<span class="sourceLineNo">293</span> "Cannot add log to queue when creating a new source, queueId=" + id<a name="line.293"></a> +<span class="sourceLineNo">294</span> + ", filename=" + name;<a name="line.294"></a> +<span class="sourceLineNo">295</span> server.stop(message);<a name="line.295"></a> +<span class="sourceLineNo">296</span> throw e;<a name="line.296"></a> +<span class="sourceLineNo">297</span> }<a name="line.297"></a> +<span class="sourceLineNo">298</span> src.enqueueLog(logPath);<a name="line.298"></a> +<span class="sourceLineNo">299</span> }<a name="line.299"></a> +<span class="sourceLineNo">300</span> }<a name="line.300"></a> +<span class="sourceLineNo">301</span> }<a name="line.301"></a> +<span class="sourceLineNo">302</span> }<a name="line.302"></a> +<span class="sourceLineNo">303</span> src.startup();<a name="line.303"></a> +<span class="sourceLineNo">304</span> return src;<a name="line.304"></a> +<span class="sourceLineNo">305</span> }<a name="line.305"></a> +<span class="sourceLineNo">306</span><a name="line.306"></a> +<span class="sourceLineNo">307</span> @VisibleForTesting<a name="line.307"></a> +<span class="sourceLineNo">308</span> int getSizeOfLatestPath() {<a name="line.308"></a> +<span class="sourceLineNo">309</span> synchronized (latestPaths) {<a name="line.309"></a> +<span class="sourceLineNo">310</span> return latestPaths.size();<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><a name="line.313"></a> +<span class="sourceLineNo">314</span> /**<a name="line.314"></a> +<span class="sourceLineNo">315</span> * Delete a complete queue of wals associated with a peer cluster<a name="line.315"></a> +<span class="sourceLineNo">316</span> * @param peerId Id of the peer cluster queue of wals to delete<a name="line.316"></a> +<span class="sourceLineNo">317</span> */<a name="line.317"></a> +<span class="sourceLineNo">318</span> public void deleteSource(String peerId, boolean closeConnection) {<a name="line.318"></a> +<span class="sourceLineNo">319</span> this.replicationQueues.removeQueue(peerId);<a name="line.319"></a> +<span class="sourceLineNo">320</span> if (closeConnection) {<a name="line.320"></a> +<span class="sourceLineNo">321</span> this.replicationPeers.peerDisconnected(peerId);<a name="line.321"></a> +<span class="sourceLineNo">322</span> }<a name="line.322"></a> +<span class="sourceLineNo">323</span> }<a name="line.323"></a> +<span class="sourceLineNo">324</span><a name="line.324"></a> +<span class="sourceLineNo">325</span> /**<a name="line.325"></a> +<span class="sourceLineNo">326</span> * Terminate the replication on this region server<a name="line.326"></a> +<span class="sourceLineNo">327</span> */<a name="line.327"></a> +<span class="sourceLineNo">328</span> public void join() {<a name="line.328"></a> +<sp <TRUNCATED>