http://git-wip-us.apache.org/repos/asf/hbase-site/blob/883dde2f/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSource.LogsComparator.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSource.LogsComparator.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSource.LogsComparator.html index a99b4a7..119472c 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSource.LogsComparator.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSource.LogsComparator.html @@ -109,528 +109,529 @@ <span class="sourceLineNo">101</span> protected FileSystem fs;<a name="line.101"></a> <span class="sourceLineNo">102</span> // id of this cluster<a name="line.102"></a> <span class="sourceLineNo">103</span> private UUID clusterId;<a name="line.103"></a> -<span class="sourceLineNo">104</span> // id of the other cluster<a name="line.104"></a> -<span class="sourceLineNo">105</span> private UUID peerClusterId;<a name="line.105"></a> -<span class="sourceLineNo">106</span> // total number of edits we replicated<a name="line.106"></a> -<span class="sourceLineNo">107</span> private AtomicLong totalReplicatedEdits = new AtomicLong(0);<a name="line.107"></a> -<span class="sourceLineNo">108</span> // The znode we currently play with<a name="line.108"></a> -<span class="sourceLineNo">109</span> protected String queueId;<a name="line.109"></a> -<span class="sourceLineNo">110</span> // Maximum number of retries before taking bold actions<a name="line.110"></a> -<span class="sourceLineNo">111</span> private int maxRetriesMultiplier;<a name="line.111"></a> -<span class="sourceLineNo">112</span> // Indicates if this particular source is running<a name="line.112"></a> -<span class="sourceLineNo">113</span> private volatile boolean sourceRunning = false;<a name="line.113"></a> -<span class="sourceLineNo">114</span> // Metrics for this source<a name="line.114"></a> -<span class="sourceLineNo">115</span> private MetricsSource metrics;<a name="line.115"></a> -<span class="sourceLineNo">116</span> // WARN threshold for the number of queued logs, defaults to 2<a name="line.116"></a> -<span class="sourceLineNo">117</span> private int logQueueWarnThreshold;<a name="line.117"></a> -<span class="sourceLineNo">118</span> // ReplicationEndpoint which will handle the actual replication<a name="line.118"></a> -<span class="sourceLineNo">119</span> private volatile ReplicationEndpoint replicationEndpoint;<a name="line.119"></a> -<span class="sourceLineNo">120</span> // A filter (or a chain of filters) for the WAL entries.<a name="line.120"></a> -<span class="sourceLineNo">121</span> protected WALEntryFilter walEntryFilter;<a name="line.121"></a> -<span class="sourceLineNo">122</span> // throttler<a name="line.122"></a> -<span class="sourceLineNo">123</span> private ReplicationThrottler throttler;<a name="line.123"></a> -<span class="sourceLineNo">124</span> private long defaultBandwidth;<a name="line.124"></a> -<span class="sourceLineNo">125</span> private long currentBandwidth;<a name="line.125"></a> -<span class="sourceLineNo">126</span> private WALFileLengthProvider walFileLengthProvider;<a name="line.126"></a> -<span class="sourceLineNo">127</span> protected final ConcurrentHashMap<String, ReplicationSourceShipper> workerThreads =<a name="line.127"></a> -<span class="sourceLineNo">128</span> new ConcurrentHashMap<>();<a name="line.128"></a> +<span class="sourceLineNo">104</span> // total number of edits we replicated<a name="line.104"></a> +<span class="sourceLineNo">105</span> private AtomicLong totalReplicatedEdits = new AtomicLong(0);<a name="line.105"></a> +<span class="sourceLineNo">106</span> // The znode we currently play with<a name="line.106"></a> +<span class="sourceLineNo">107</span> protected String queueId;<a name="line.107"></a> +<span class="sourceLineNo">108</span> // Maximum number of retries before taking bold actions<a name="line.108"></a> +<span class="sourceLineNo">109</span> private int maxRetriesMultiplier;<a name="line.109"></a> +<span class="sourceLineNo">110</span> // Indicates if this particular source is running<a name="line.110"></a> +<span class="sourceLineNo">111</span> private volatile boolean sourceRunning = false;<a name="line.111"></a> +<span class="sourceLineNo">112</span> // Metrics for this source<a name="line.112"></a> +<span class="sourceLineNo">113</span> private MetricsSource metrics;<a name="line.113"></a> +<span class="sourceLineNo">114</span> // WARN threshold for the number of queued logs, defaults to 2<a name="line.114"></a> +<span class="sourceLineNo">115</span> private int logQueueWarnThreshold;<a name="line.115"></a> +<span class="sourceLineNo">116</span> // ReplicationEndpoint which will handle the actual replication<a name="line.116"></a> +<span class="sourceLineNo">117</span> private volatile ReplicationEndpoint replicationEndpoint;<a name="line.117"></a> +<span class="sourceLineNo">118</span> // A filter (or a chain of filters) for the WAL entries.<a name="line.118"></a> +<span class="sourceLineNo">119</span> protected volatile WALEntryFilter walEntryFilter;<a name="line.119"></a> +<span class="sourceLineNo">120</span> // throttler<a name="line.120"></a> +<span class="sourceLineNo">121</span> private ReplicationThrottler throttler;<a name="line.121"></a> +<span class="sourceLineNo">122</span> private long defaultBandwidth;<a name="line.122"></a> +<span class="sourceLineNo">123</span> private long currentBandwidth;<a name="line.123"></a> +<span class="sourceLineNo">124</span> private WALFileLengthProvider walFileLengthProvider;<a name="line.124"></a> +<span class="sourceLineNo">125</span> protected final ConcurrentHashMap<String, ReplicationSourceShipper> workerThreads =<a name="line.125"></a> +<span class="sourceLineNo">126</span> new ConcurrentHashMap<>();<a name="line.126"></a> +<span class="sourceLineNo">127</span><a name="line.127"></a> +<span class="sourceLineNo">128</span> private AtomicLong totalBufferUsed;<a name="line.128"></a> <span class="sourceLineNo">129</span><a name="line.129"></a> -<span class="sourceLineNo">130</span> private AtomicLong totalBufferUsed;<a name="line.130"></a> -<span class="sourceLineNo">131</span><a name="line.131"></a> -<span class="sourceLineNo">132</span> public static final String WAIT_ON_ENDPOINT_SECONDS =<a name="line.132"></a> -<span class="sourceLineNo">133</span> "hbase.replication.wait.on.endpoint.seconds";<a name="line.133"></a> -<span class="sourceLineNo">134</span> public static final int DEFAULT_WAIT_ON_ENDPOINT_SECONDS = 30;<a name="line.134"></a> -<span class="sourceLineNo">135</span> private int waitOnEndpointSeconds = -1;<a name="line.135"></a> +<span class="sourceLineNo">130</span> public static final String WAIT_ON_ENDPOINT_SECONDS =<a name="line.130"></a> +<span class="sourceLineNo">131</span> "hbase.replication.wait.on.endpoint.seconds";<a name="line.131"></a> +<span class="sourceLineNo">132</span> public static final int DEFAULT_WAIT_ON_ENDPOINT_SECONDS = 30;<a name="line.132"></a> +<span class="sourceLineNo">133</span> private int waitOnEndpointSeconds = -1;<a name="line.133"></a> +<span class="sourceLineNo">134</span><a name="line.134"></a> +<span class="sourceLineNo">135</span> private Thread initThread;<a name="line.135"></a> <span class="sourceLineNo">136</span><a name="line.136"></a> -<span class="sourceLineNo">137</span> private Thread initThread;<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> * Instantiation method used by region servers<a name="line.140"></a> -<span class="sourceLineNo">141</span> * @param conf configuration to use<a name="line.141"></a> -<span class="sourceLineNo">142</span> * @param fs file system to use<a name="line.142"></a> -<span class="sourceLineNo">143</span> * @param manager replication manager to ping to<a name="line.143"></a> -<span class="sourceLineNo">144</span> * @param server the server for this region server<a name="line.144"></a> -<span class="sourceLineNo">145</span> * @param queueId the id of our replication queue<a name="line.145"></a> -<span class="sourceLineNo">146</span> * @param clusterId unique UUID for the cluster<a name="line.146"></a> -<span class="sourceLineNo">147</span> * @param metrics metrics for replication source<a name="line.147"></a> -<span class="sourceLineNo">148</span> */<a name="line.148"></a> -<span class="sourceLineNo">149</span> @Override<a name="line.149"></a> -<span class="sourceLineNo">150</span> public void init(Configuration conf, FileSystem fs, ReplicationSourceManager manager,<a name="line.150"></a> -<span class="sourceLineNo">151</span> ReplicationQueueStorage queueStorage, ReplicationPeer replicationPeer, Server server,<a name="line.151"></a> -<span class="sourceLineNo">152</span> String queueId, UUID clusterId, WALFileLengthProvider walFileLengthProvider,<a name="line.152"></a> -<span class="sourceLineNo">153</span> MetricsSource metrics) throws IOException {<a name="line.153"></a> -<span class="sourceLineNo">154</span> this.server = server;<a name="line.154"></a> -<span class="sourceLineNo">155</span> this.conf = HBaseConfiguration.create(conf);<a name="line.155"></a> -<span class="sourceLineNo">156</span> this.waitOnEndpointSeconds =<a name="line.156"></a> -<span class="sourceLineNo">157</span> this.conf.getInt(WAIT_ON_ENDPOINT_SECONDS, DEFAULT_WAIT_ON_ENDPOINT_SECONDS);<a name="line.157"></a> -<span class="sourceLineNo">158</span> decorateConf();<a name="line.158"></a> -<span class="sourceLineNo">159</span> this.sleepForRetries =<a name="line.159"></a> -<span class="sourceLineNo">160</span> this.conf.getLong("replication.source.sleepforretries", 1000); // 1 second<a name="line.160"></a> -<span class="sourceLineNo">161</span> this.maxRetriesMultiplier =<a name="line.161"></a> -<span class="sourceLineNo">162</span> this.conf.getInt("replication.source.maxretriesmultiplier", 300); // 5 minutes @ 1 sec per<a name="line.162"></a> -<span class="sourceLineNo">163</span> this.queueSizePerGroup = this.conf.getInt("hbase.regionserver.maxlogs", 32);<a name="line.163"></a> -<span class="sourceLineNo">164</span> this.queueStorage = queueStorage;<a name="line.164"></a> -<span class="sourceLineNo">165</span> this.replicationPeer = replicationPeer;<a name="line.165"></a> -<span class="sourceLineNo">166</span> this.manager = manager;<a name="line.166"></a> -<span class="sourceLineNo">167</span> this.fs = fs;<a name="line.167"></a> -<span class="sourceLineNo">168</span> this.metrics = metrics;<a name="line.168"></a> -<span class="sourceLineNo">169</span> this.clusterId = clusterId;<a name="line.169"></a> -<span class="sourceLineNo">170</span><a name="line.170"></a> -<span class="sourceLineNo">171</span> this.queueId = queueId;<a name="line.171"></a> -<span class="sourceLineNo">172</span> this.replicationQueueInfo = new ReplicationQueueInfo(queueId);<a name="line.172"></a> -<span class="sourceLineNo">173</span> // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.173"></a> -<span class="sourceLineNo">174</span> this.peerId = this.replicationQueueInfo.getPeerId();<a name="line.174"></a> -<span class="sourceLineNo">175</span> this.logQueueWarnThreshold = this.conf.getInt("replication.source.log.queue.warn", 2);<a name="line.175"></a> -<span class="sourceLineNo">176</span><a name="line.176"></a> -<span class="sourceLineNo">177</span> defaultBandwidth = this.conf.getLong("replication.source.per.peer.node.bandwidth", 0);<a name="line.177"></a> -<span class="sourceLineNo">178</span> currentBandwidth = getCurrentBandwidth();<a name="line.178"></a> -<span class="sourceLineNo">179</span> this.throttler = new ReplicationThrottler((double) currentBandwidth / 10.0);<a name="line.179"></a> -<span class="sourceLineNo">180</span> this.totalBufferUsed = manager.getTotalBufferUsed();<a name="line.180"></a> -<span class="sourceLineNo">181</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.181"></a> -<span class="sourceLineNo">182</span> LOG.info("queueId=" + queueId + ", ReplicationSource : " + peerId<a name="line.182"></a> -<span class="sourceLineNo">183</span> + ", currentBandwidth=" + this.currentBandwidth);<a name="line.183"></a> -<span class="sourceLineNo">184</span> }<a name="line.184"></a> -<span class="sourceLineNo">185</span><a name="line.185"></a> -<span class="sourceLineNo">186</span> private void decorateConf() {<a name="line.186"></a> -<span class="sourceLineNo">187</span> String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.187"></a> -<span class="sourceLineNo">188</span> if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.188"></a> -<span class="sourceLineNo">189</span> this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.189"></a> -<span class="sourceLineNo">190</span> }<a name="line.190"></a> -<span class="sourceLineNo">191</span> }<a name="line.191"></a> -<span class="sourceLineNo">192</span><a name="line.192"></a> -<span class="sourceLineNo">193</span> @Override<a name="line.193"></a> -<span class="sourceLineNo">194</span> public void enqueueLog(Path log) {<a name="line.194"></a> -<span class="sourceLineNo">195</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log.getName());<a name="line.195"></a> -<span class="sourceLineNo">196</span> PriorityBlockingQueue<Path> queue = queues.get(logPrefix);<a name="line.196"></a> -<span class="sourceLineNo">197</span> if (queue == null) {<a name="line.197"></a> -<span class="sourceLineNo">198</span> queue = new PriorityBlockingQueue<>(queueSizePerGroup, new LogsComparator());<a name="line.198"></a> -<span class="sourceLineNo">199</span> queues.put(logPrefix, queue);<a name="line.199"></a> -<span class="sourceLineNo">200</span> if (this.isSourceActive() && this.replicationEndpoint != null) {<a name="line.200"></a> -<span class="sourceLineNo">201</span> // new wal group observed after source startup, start a new worker thread to track it<a name="line.201"></a> -<span class="sourceLineNo">202</span> // notice: it's possible that log enqueued when this.running is set but worker thread<a name="line.202"></a> -<span class="sourceLineNo">203</span> // still not launched, so it's necessary to check workerThreads before start the worker<a name="line.203"></a> -<span class="sourceLineNo">204</span> tryStartNewShipper(logPrefix, queue);<a name="line.204"></a> -<span class="sourceLineNo">205</span> }<a name="line.205"></a> -<span class="sourceLineNo">206</span> }<a name="line.206"></a> -<span class="sourceLineNo">207</span> queue.put(log);<a name="line.207"></a> -<span class="sourceLineNo">208</span> this.metrics.incrSizeOfLogQueue();<a name="line.208"></a> -<span class="sourceLineNo">209</span> // This will log a warning for each new log that gets created above the warn threshold<a name="line.209"></a> -<span class="sourceLineNo">210</span> int queueSize = queue.size();<a name="line.210"></a> -<span class="sourceLineNo">211</span> if (queueSize > this.logQueueWarnThreshold) {<a name="line.211"></a> -<span class="sourceLineNo">212</span> LOG.warn("WAL group " + logPrefix + " queue size: " + queueSize<a name="line.212"></a> -<span class="sourceLineNo">213</span> + " exceeds value of replication.source.log.queue.warn: " + logQueueWarnThreshold);<a name="line.213"></a> -<span class="sourceLineNo">214</span> }<a name="line.214"></a> -<span class="sourceLineNo">215</span> }<a name="line.215"></a> -<span class="sourceLineNo">216</span><a name="line.216"></a> -<span class="sourceLineNo">217</span> @Override<a name="line.217"></a> -<span class="sourceLineNo">218</span> public void addHFileRefs(TableName tableName, byte[] family, List<Pair<Path, Path>> pairs)<a name="line.218"></a> -<span class="sourceLineNo">219</span> throws ReplicationException {<a name="line.219"></a> -<span class="sourceLineNo">220</span> Map<TableName, List<String>> tableCFMap = replicationPeer.getTableCFs();<a name="line.220"></a> -<span class="sourceLineNo">221</span> if (tableCFMap != null) {<a name="line.221"></a> -<span class="sourceLineNo">222</span> List<String> tableCfs = tableCFMap.get(tableName);<a name="line.222"></a> -<span class="sourceLineNo">223</span> if (tableCFMap.containsKey(tableName)<a name="line.223"></a> -<span class="sourceLineNo">224</span> && (tableCfs == null || tableCfs.contains(Bytes.toString(family)))) {<a name="line.224"></a> -<span class="sourceLineNo">225</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.225"></a> -<span class="sourceLineNo">226</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.226"></a> -<span class="sourceLineNo">227</span> } else {<a name="line.227"></a> -<span class="sourceLineNo">228</span> LOG.debug("HFiles will not be replicated belonging to the table " + tableName + " family "<a name="line.228"></a> -<span class="sourceLineNo">229</span> + Bytes.toString(family) + " to peer id " + peerId);<a name="line.229"></a> -<span class="sourceLineNo">230</span> }<a name="line.230"></a> -<span class="sourceLineNo">231</span> } else {<a name="line.231"></a> -<span class="sourceLineNo">232</span> // user has explicitly not defined any table cfs for replication, means replicate all the<a name="line.232"></a> -<span class="sourceLineNo">233</span> // data<a name="line.233"></a> -<span class="sourceLineNo">234</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.234"></a> -<span class="sourceLineNo">235</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.235"></a> -<span class="sourceLineNo">236</span> }<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> private ReplicationEndpoint createReplicationEndpoint()<a name="line.239"></a> -<span class="sourceLineNo">240</span> throws InstantiationException, IllegalAccessException, ClassNotFoundException, IOException {<a name="line.240"></a> -<span class="sourceLineNo">241</span> RegionServerCoprocessorHost rsServerHost = null;<a name="line.241"></a> -<span class="sourceLineNo">242</span> if (server instanceof HRegionServer) {<a name="line.242"></a> -<span class="sourceLineNo">243</span> rsServerHost = ((HRegionServer) server).getRegionServerCoprocessorHost();<a name="line.243"></a> -<span class="sourceLineNo">244</span> }<a name="line.244"></a> -<span class="sourceLineNo">245</span> String replicationEndpointImpl = replicationPeer.getPeerConfig().getReplicationEndpointImpl();<a name="line.245"></a> -<span class="sourceLineNo">246</span><a name="line.246"></a> -<span class="sourceLineNo">247</span> ReplicationEndpoint replicationEndpoint;<a name="line.247"></a> -<span class="sourceLineNo">248</span> if (replicationEndpointImpl == null) {<a name="line.248"></a> -<span class="sourceLineNo">249</span> // Default to HBase inter-cluster replication endpoint; skip reflection<a name="line.249"></a> -<span class="sourceLineNo">250</span> replicationEndpoint = new HBaseInterClusterReplicationEndpoint();<a name="line.250"></a> -<span class="sourceLineNo">251</span> } else {<a name="line.251"></a> -<span class="sourceLineNo">252</span> try {<a name="line.252"></a> -<span class="sourceLineNo">253</span> replicationEndpoint = Class.forName(replicationEndpointImpl)<a name="line.253"></a> -<span class="sourceLineNo">254</span> .asSubclass(ReplicationEndpoint.class)<a name="line.254"></a> -<span class="sourceLineNo">255</span> .getDeclaredConstructor()<a name="line.255"></a> -<span class="sourceLineNo">256</span> .newInstance();<a name="line.256"></a> -<span class="sourceLineNo">257</span> } catch (NoSuchMethodException | InvocationTargetException e) {<a name="line.257"></a> -<span class="sourceLineNo">258</span> throw new IllegalArgumentException(e);<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> if (rsServerHost != null) {<a name="line.261"></a> -<span class="sourceLineNo">262</span> ReplicationEndpoint newReplicationEndPoint =<a name="line.262"></a> -<span class="sourceLineNo">263</span> rsServerHost.postCreateReplicationEndPoint(replicationEndpoint);<a name="line.263"></a> -<span class="sourceLineNo">264</span> if (newReplicationEndPoint != null) {<a name="line.264"></a> -<span class="sourceLineNo">265</span> // Override the newly created endpoint from the hook with configured end point<a name="line.265"></a> -<span class="sourceLineNo">266</span> replicationEndpoint = newReplicationEndPoint;<a name="line.266"></a> -<span class="sourceLineNo">267</span> }<a name="line.267"></a> -<span class="sourceLineNo">268</span> }<a name="line.268"></a> -<span class="sourceLineNo">269</span> return replicationEndpoint;<a name="line.269"></a> -<span class="sourceLineNo">270</span> }<a name="line.270"></a> -<span class="sourceLineNo">271</span><a name="line.271"></a> -<span class="sourceLineNo">272</span> private void initAndStartReplicationEndpoint(ReplicationEndpoint replicationEndpoint)<a name="line.272"></a> -<span class="sourceLineNo">273</span> throws IOException, TimeoutException {<a name="line.273"></a> -<span class="sourceLineNo">274</span> TableDescriptors tableDescriptors = null;<a name="line.274"></a> -<span class="sourceLineNo">275</span> if (server instanceof HRegionServer) {<a name="line.275"></a> -<span class="sourceLineNo">276</span> tableDescriptors = ((HRegionServer) server).getTableDescriptors();<a name="line.276"></a> -<span class="sourceLineNo">277</span> }<a name="line.277"></a> -<span class="sourceLineNo">278</span> replicationEndpoint<a name="line.278"></a> -<span class="sourceLineNo">279</span> .init(new ReplicationEndpoint.Context(conf, replicationPeer.getConfiguration(), fs, peerId,<a name="line.279"></a> -<span class="sourceLineNo">280</span> clusterId, replicationPeer, metrics, tableDescriptors, server));<a name="line.280"></a> -<span class="sourceLineNo">281</span> replicationEndpoint.start();<a name="line.281"></a> -<span class="sourceLineNo">282</span> replicationEndpoint.awaitRunning(waitOnEndpointSeconds, TimeUnit.SECONDS);<a name="line.282"></a> -<span class="sourceLineNo">283</span> }<a name="line.283"></a> -<span class="sourceLineNo">284</span><a name="line.284"></a> -<span class="sourceLineNo">285</span> private void initializeWALEntryFilter() {<a name="line.285"></a> -<span class="sourceLineNo">286</span> // get the WALEntryFilter from ReplicationEndpoint and add it to default filters<a name="line.286"></a> -<span class="sourceLineNo">287</span> ArrayList<WALEntryFilter> filters =<a name="line.287"></a> -<span class="sourceLineNo">288</span> Lists.<WALEntryFilter> newArrayList(new SystemTableWALEntryFilter());<a name="line.288"></a> -<span class="sourceLineNo">289</span> WALEntryFilter filterFromEndpoint = this.replicationEndpoint.getWALEntryfilter();<a name="line.289"></a> -<span class="sourceLineNo">290</span> if (filterFromEndpoint != null) {<a name="line.290"></a> -<span class="sourceLineNo">291</span> filters.add(filterFromEndpoint);<a name="line.291"></a> -<span class="sourceLineNo">292</span> }<a name="line.292"></a> -<span class="sourceLineNo">293</span> filters.add(new ClusterMarkingEntryFilter(clusterId, peerClusterId, replicationEndpoint));<a name="line.293"></a> -<span class="sourceLineNo">294</span> this.walEntryFilter = new ChainWALEntryFilter(filters);<a name="line.294"></a> -<span class="sourceLineNo">295</span> }<a name="line.295"></a> -<span class="sourceLineNo">296</span><a name="line.296"></a> -<span class="sourceLineNo">297</span> private void tryStartNewShipper(String walGroupId, PriorityBlockingQueue<Path> queue) {<a name="line.297"></a> -<span class="sourceLineNo">298</span> ReplicationSourceShipper worker = createNewShipper(walGroupId, queue);<a name="line.298"></a> -<span class="sourceLineNo">299</span> ReplicationSourceShipper extant = workerThreads.putIfAbsent(walGroupId, worker);<a name="line.299"></a> -<span class="sourceLineNo">300</span> if (extant != null) {<a name="line.300"></a> -<span class="sourceLineNo">301</span> LOG.debug("Someone has beat us to start a worker thread for wal group {}", walGroupId);<a name="line.301"></a> -<span class="sourceLineNo">302</span> } else {<a name="line.302"></a> -<span class="sourceLineNo">303</span> LOG.debug("Starting up worker for wal group {}", walGroupId);<a name="line.303"></a> -<span class="sourceLineNo">304</span> ReplicationSourceWALReader walReader =<a name="line.304"></a> -<span class="sourceLineNo">305</span> createNewWALReader(walGroupId, queue, worker.getStartPosition());<a name="line.305"></a> -<span class="sourceLineNo">306</span> Threads.setDaemonThreadRunning(walReader, Thread.currentThread().getName() +<a name="line.306"></a> -<span class="sourceLineNo">307</span> ".replicationSource.wal-reader." + walGroupId + "," + queueId, this::uncaughtException);<a name="line.307"></a> -<span class="sourceLineNo">308</span> worker.setWALReader(walReader);<a name="line.308"></a> -<span class="sourceLineNo">309</span> worker.startup(this::uncaughtException);<a name="line.309"></a> -<span class="sourceLineNo">310</span> }<a name="line.310"></a> -<span class="sourceLineNo">311</span> }<a name="line.311"></a> -<span class="sourceLineNo">312</span><a name="line.312"></a> -<span class="sourceLineNo">313</span> protected ReplicationSourceShipper createNewShipper(String walGroupId,<a name="line.313"></a> -<span class="sourceLineNo">314</span> PriorityBlockingQueue<Path> queue) {<a name="line.314"></a> -<span class="sourceLineNo">315</span> return new ReplicationSourceShipper(conf, walGroupId, queue, this);<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> private ReplicationSourceWALReader createNewWALReader(String walGroupId,<a name="line.318"></a> -<span class="sourceLineNo">319</span> PriorityBlockingQueue<Path> queue, long startPosition) {<a name="line.319"></a> -<span class="sourceLineNo">320</span> return replicationPeer.getPeerConfig().isSerial()<a name="line.320"></a> -<span class="sourceLineNo">321</span> ? new SerialReplicationSourceWALReader(fs, conf, queue, startPosition, walEntryFilter, this)<a name="line.321"></a> -<span class="sourceLineNo">322</span> : new ReplicationSourceWALReader(fs, conf, queue, startPosition, walEntryFilter, this);<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> protected final void uncaughtException(Thread t, Throwable e) {<a name="line.325"></a> -<span class="sourceLineNo">326</span> RSRpcServices.exitIfOOME(e);<a name="line.326"></a> -<span class="sourceLineNo">327</span> LOG.error("Unexpected exception in " + t.getName() + " currentPath=" + getCurrentPath(), e);<a name="line.327"></a> -<span class="sourceLineNo">328</span> server.abort("Unexpected exception in " + t.getName(), e);<a name="line.328"></a> -<span class="sourceLineNo">329</span> }<a name="line.329"></a> -<span class="sourceLineNo">330</span><a name="line.330"></a> -<span class="sourceLineNo">331</span> @Override<a name="line.331"></a> -<span class="sourceLineNo">332</span> public ReplicationEndpoint getReplicationEndpoint() {<a name="line.332"></a> -<span class="sourceLineNo">333</span> return this.replicationEndpoint;<a name="line.333"></a> -<span class="sourceLineNo">334</span> }<a name="line.334"></a> -<span class="sourceLineNo">335</span><a name="line.335"></a> -<span class="sourceLineNo">336</span> @Override<a name="line.336"></a> -<span class="sourceLineNo">337</span> public ReplicationSourceManager getSourceManager() {<a name="line.337"></a> -<span class="sourceLineNo">338</span> return this.manager;<a name="line.338"></a> -<span class="sourceLineNo">339</span> }<a name="line.339"></a> -<span class="sourceLineNo">340</span><a name="line.340"></a> -<span class="sourceLineNo">341</span> @Override<a name="line.341"></a> -<span class="sourceLineNo">342</span> public void tryThrottle(int batchSize) throws InterruptedException {<a name="line.342"></a> -<span class="sourceLineNo">343</span> checkBandwidthChangeAndResetThrottler();<a name="line.343"></a> -<span class="sourceLineNo">344</span> if (throttler.isEnabled()) {<a name="line.344"></a> -<span class="sourceLineNo">345</span> long sleepTicks = throttler.getNextSleepInterval(batchSize);<a name="line.345"></a> -<span class="sourceLineNo">346</span> if (sleepTicks > 0) {<a name="line.346"></a> -<span class="sourceLineNo">347</span> if (LOG.isTraceEnabled()) {<a name="line.347"></a> -<span class="sourceLineNo">348</span> LOG.trace("To sleep " + sleepTicks + "ms for throttling control");<a name="line.348"></a> -<span class="sourceLineNo">349</span> }<a name="line.349"></a> -<span class="sourceLineNo">350</span> Thread.sleep(sleepTicks);<a name="line.350"></a> -<span class="sourceLineNo">351</span> // reset throttler's cycle start tick when sleep for throttling occurs<a name="line.351"></a> -<span class="sourceLineNo">352</span> throttler.resetStartTick();<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><a name="line.356"></a> -<span class="sourceLineNo">357</span> private void checkBandwidthChangeAndResetThrottler() {<a name="line.357"></a> -<span class="sourceLineNo">358</span> long peerBandwidth = getCurrentBandwidth();<a name="line.358"></a> -<span class="sourceLineNo">359</span> if (peerBandwidth != currentBandwidth) {<a name="line.359"></a> -<span class="sourceLineNo">360</span> currentBandwidth = peerBandwidth;<a name="line.360"></a> -<span class="sourceLineNo">361</span> throttler.setBandwidth((double) currentBandwidth / 10.0);<a name="line.361"></a> -<span class="sourceLineNo">362</span> LOG.info("ReplicationSource : " + peerId<a name="line.362"></a> -<span class="sourceLineNo">363</span> + " bandwidth throttling changed, currentBandWidth=" + currentBandwidth);<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><a name="line.366"></a> -<span class="sourceLineNo">367</span> private long getCurrentBandwidth() {<a name="line.367"></a> -<span class="sourceLineNo">368</span> long peerBandwidth = replicationPeer.getPeerBandwidth();<a name="line.368"></a> -<span class="sourceLineNo">369</span> // user can set peer bandwidth to 0 to use default bandwidth<a name="line.369"></a> -<span class="sourceLineNo">370</span> return peerBandwidth != 0 ? peerBandwidth : defaultBandwidth;<a name="line.370"></a> -<span class="sourceLineNo">371</span> }<a name="line.371"></a> -<span class="sourceLineNo">372</span><a name="line.372"></a> -<span class="sourceLineNo">373</span> /**<a name="line.373"></a> -<span class="sourceLineNo">374</span> * Do the sleeping logic<a name="line.374"></a> -<span class="sourceLineNo">375</span> * @param msg Why we sleep<a name="line.375"></a> -<span class="sourceLineNo">376</span> * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.376"></a> -<span class="sourceLineNo">377</span> * @return True if <code>sleepMultiplier</code> is &lt; <code>maxRetriesMultiplier</code><a name="line.377"></a> -<span class="sourceLineNo">378</span> */<a name="line.378"></a> -<span class="sourceLineNo">379</span> protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.379"></a> -<span class="sourceLineNo">380</span> try {<a name="line.380"></a> -<span class="sourceLineNo">381</span> if (LOG.isTraceEnabled()) {<a name="line.381"></a> -<span class="sourceLineNo">382</span> LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.382"></a> -<span class="sourceLineNo">383</span> }<a name="line.383"></a> -<span class="sourceLineNo">384</span> Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.384"></a> -<span class="sourceLineNo">385</span> } catch (InterruptedException e) {<a name="line.385"></a> -<span class="sourceLineNo">386</span> LOG.debug("Interrupted while sleeping between retries");<a name="line.386"></a> -<span class="sourceLineNo">387</span> Thread.currentThread().interrupt();<a name="line.387"></a> -<span class="sourceLineNo">388</span> }<a name="line.388"></a> -<span class="sourceLineNo">389</span> return sleepMultiplier < maxRetriesMultiplier;<a name="line.389"></a> -<span class="sourceLineNo">390</span> }<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> * check whether the peer is enabled or not<a name="line.393"></a> -<span class="sourceLineNo">394</span> * @return true if the peer is enabled, otherwise false<a name="line.394"></a> -<span class="sourceLineNo">395</span> */<a name="line.395"></a> -<span class="sourceLineNo">396</span> @Override<a name="line.396"></a> -<span class="sourceLineNo">397</span> public boolean isPeerEnabled() {<a name="line.397"></a> -<span class="sourceLineNo">398</span> return replicationPeer.isPeerEnabled();<a name="line.398"></a> -<span class="sourceLineNo">399</span> }<a name="line.399"></a> -<span class="sourceLineNo">400</span><a name="line.400"></a> -<span class="sourceLineNo">401</span> private void initialize() {<a name="line.401"></a> -<span class="sourceLineNo">402</span> int sleepMultiplier = 1;<a name="line.402"></a> -<span class="sourceLineNo">403</span> while (this.isSourceActive()) {<a name="line.403"></a> -<span class="sourceLineNo">404</span> ReplicationEndpoint replicationEndpoint;<a name="line.404"></a> -<span class="sourceLineNo">405</span> try {<a name="line.405"></a> -<span class="sourceLineNo">406</span> replicationEndpoint = createReplicationEndpoint();<a name="line.406"></a> -<span class="sourceLineNo">407</span> } catch (Exception e) {<a name="line.407"></a> -<span class="sourceLineNo">408</span> LOG.warn("error creating ReplicationEndpoint, retry", e);<a name="line.408"></a> -<span class="sourceLineNo">409</span> if (sleepForRetries("Error creating ReplicationEndpoint", sleepMultiplier)) {<a name="line.409"></a> -<span class="sourceLineNo">410</span> sleepMultiplier++;<a name="line.410"></a> -<span class="sourceLineNo">411</span> }<a name="line.411"></a> -<span class="sourceLineNo">412</span> continue;<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> try {<a name="line.415"></a> -<span class="sourceLineNo">416</span> initAndStartReplicationEndpoint(replicationEndpoint);<a name="line.416"></a> -<span class="sourceLineNo">417</span> this.replicationEndpoint = replicationEndpoint;<a name="line.417"></a> -<span class="sourceLineNo">418</span> break;<a name="line.418"></a> -<span class="sourceLineNo">419</span> } catch (Exception e) {<a name="line.419"></a> -<span class="sourceLineNo">420</span> LOG.warn("Error starting ReplicationEndpoint, retry", e);<a name="line.420"></a> -<span class="sourceLineNo">421</span> replicationEndpoint.stop();<a name="line.421"></a> -<span class="sourceLineNo">422</span> if (sleepForRetries("Error starting ReplicationEndpoint", sleepMultiplier)) {<a name="line.422"></a> -<span class="sourceLineNo">423</span> sleepMultiplier++;<a name="line.423"></a> -<span class="sourceLineNo">424</span> }<a name="line.424"></a> -<span class="sourceLineNo">425</span> }<a name="line.425"></a> -<span class="sourceLineNo">426</span> }<a name="line.426"></a> -<span class="sourceLineNo">427</span><a name="line.427"></a> -<span class="sourceLineNo">428</span> if (!this.isSourceActive()) {<a name="line.428"></a> -<span class="sourceLineNo">429</span> return;<a name="line.429"></a> -<span class="sourceLineNo">430</span> }<a name="line.430"></a> -<span class="sourceLineNo">431</span><a name="line.431"></a> -<span class="sourceLineNo">432</span> sleepMultiplier = 1;<a name="line.432"></a> -<span class="sourceLineNo">433</span> // delay this until we are in an asynchronous thread<a name="line.433"></a> -<span class="sourceLineNo">434</span> while (this.isSourceActive() && this.peerClusterId == null) {<a name="line.434"></a> -<span class="sourceLineNo">435</span> this.peerClusterId = replicationEndpoint.getPeerUUID();<a name="line.435"></a> -<span class="sourceLineNo">436</span> if (this.isSourceActive() && this.peerClusterId == null) {<a name="line.436"></a> -<span class="sourceLineNo">437</span> if (sleepForRetries("Cannot contact the peer's zk ensemble", sleepMultiplier)) {<a name="line.437"></a> -<span class="sourceLineNo">438</span> sleepMultiplier++;<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> // In rare case, zookeeper setting may be messed up. That leads to the incorrect<a name="line.443"></a> -<span class="sourceLineNo">444</span> // peerClusterId value, which is the same as the source clusterId<a name="line.444"></a> -<span class="sourceLineNo">445</span> if (clusterId.equals(peerClusterId) && !replicationEndpoint.canReplicateToSameCluster()) {<a name="line.445"></a> -<span class="sourceLineNo">446</span> this.terminate("ClusterId " + clusterId + " is replicating to itself: peerClusterId "<a name="line.446"></a> -<span class="sourceLineNo">447</span> + peerClusterId + " which is not allowed by ReplicationEndpoint:"<a name="line.447"></a> -<span class="sourceLineNo">448</span> + replicationEndpoint.getClass().getName(), null, false);<a name="line.448"></a> -<span class="sourceLineNo">449</span> this.manager.removeSource(this);<a name="line.449"></a> -<span class="sourceLineNo">450</span> return;<a name="line.450"></a> -<span class="sourceLineNo">451</span> }<a name="line.451"></a> -<span class="sourceLineNo">452</span> LOG.info("Replicating " + clusterId + " -> " + peerClusterId);<a name="line.452"></a> -<span class="sourceLineNo">453</span><a name="line.453"></a> -<span class="sourceLineNo">454</span> initializeWALEntryFilter();<a name="line.454"></a> -<span class="sourceLineNo">455</span> // start workers<a name="line.455"></a> -<span class="sourceLineNo">456</span> for (Map.Entry<String, PriorityBlockingQueue<Path>> entry : queues.entrySet()) {<a name="line.456"></a> -<span class="sourceLineNo">457</span> String walGroupId = entry.getKey();<a name="line.457"></a> -<span class="sourceLineNo">458</span> PriorityBlockingQueue<Path> queue = entry.getValue();<a name="line.458"></a> -<span class="sourceLineNo">459</span> tryStartNewShipper(walGroupId, queue);<a name="line.459"></a> -<span class="sourceLineNo">460</span> }<a name="line.460"></a> -<span class="sourceLineNo">461</span> }<a name="line.461"></a> -<span class="sourceLineNo">462</span><a name="line.462"></a> -<span class="sourceLineNo">463</span> @Override<a name="line.463"></a> -<span class="sourceLineNo">464</span> public void startup() {<a name="line.464"></a> -<span class="sourceLineNo">465</span> // mark we are running now<a name="line.465"></a> -<span class="sourceLineNo">466</span> this.sourceRunning = true;<a name="line.466"></a> -<span class="sourceLineNo">467</span> initThread = new Thread(this::initialize);<a name="line.467"></a> -<span class="sourceLineNo">468</span> Threads.setDaemonThreadRunning(initThread,<a name="line.468"></a> -<span class="sourceLineNo">469</span> Thread.currentThread().getName() + ".replicationSource," + this.queueId,<a name="line.469"></a> -<span class="sourceLineNo">470</span> this::uncaughtException);<a name="line.470"></a> -<span class="sourceLineNo">471</span> }<a name="line.471"></a> -<span class="sourceLineNo">472</span><a name="line.472"></a> -<span class="sourceLineNo">473</span> @Override<a name="line.473"></a> -<span class="sourceLineNo">474</span> public void terminate(String reason) {<a name="line.474"></a> -<span class="sourceLineNo">475</span> terminate(reason, null);<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> @Override<a name="line.478"></a> -<span class="sourceLineNo">479</span> public void terminate(String reason, Exception cause) {<a name="line.479"></a> -<span class="sourceLineNo">480</span> terminate(reason, cause, true);<a name="line.480"></a> -<span class="sourceLineNo">481</span> }<a name="line.481"></a> -<span class="sourceLineNo">482</span><a name="line.482"></a> -<span class="sourceLineNo">483</span> public void terminate(String reason, Exception cause, boolean join) {<a name="line.483"></a> -<span class="sourceLineNo">484</span> if (cause == null) {<a name="line.484"></a> -<span class="sourceLineNo">485</span> LOG.info("Closing source " + this.queueId + " because: " + reason);<a name="line.485"></a> -<span class="sourceLineNo">486</span> } else {<a name="line.486"></a> -<span class="sourceLineNo">487</span> LOG.error("Closing source " + this.queueId + " because an error occurred: " + reason,<a name="line.487"></a> -<span class="sourceLineNo">488</span> cause);<a name="line.488"></a> -<span class="sourceLineNo">489</span> }<a name="line.489"></a> -<span class="sourceLineNo">490</span> this.sourceRunning = false;<a name="line.490"></a> -<span class="sourceLineNo">491</span> if (initThread != null && Thread.currentThread() != initThread) {<a name="line.491"></a> -<span class="sourceLineNo">492</span> // This usually won't happen but anyway, let's wait until the initialization thread exits.<a name="line.492"></a> -<span class="sourceLineNo">493</span> // And notice that we may call terminate directly from the initThread so here we need to<a name="line.493"></a> -<span class="sourceLineNo">494</span> // avoid join on ourselves.<a name="line.494"></a> -<span class="sourceLineNo">495</span> initThread.interrupt();<a name="line.495"></a> -<span class="sourceLineNo">496</span> Threads.shutdown(initThread, this.sleepForRetries);<a name="line.496"></a> -<span class="sourceLineNo">497</span> }<a name="line.497"></a> -<span class="sourceLineNo">498</span> Collection<ReplicationSourceShipper> workers = workerThreads.values();<a name="line.498"></a> -<span class="sourceLineNo">499</span> for (ReplicationSourceShipper worker : workers) {<a name="line.499"></a> -<span class="sourceLineNo">500</span> worker.stopWorker();<a name="line.500"></a> -<span class="sourceLineNo">501</span> worker.entryReader.interrupt();<a name="line.501"></a> -<span class="sourceLineNo">502</span> worker.interrupt();<a name="line.502"></a> -<span class="sourceLineNo">503</span> }<a name="line.503"></a> -<span class="sourceLineNo">504</span> if (this.replicationEndpoint != null) {<a name="line.504"></a> -<span class="sourceLineNo">505</span> this.replicationEndpoint.stop();<a name="line.505"></a> -<span class="sourceLineNo">506</span> }<a name="line.506"></a> -<span class="sourceLineNo">507</span> if (join) {<a name="line.507"></a> -<span class="sourceLineNo">508</span> for (ReplicationSourceShipper worker : workers) {<a name="line.508"></a> -<span class="sourceLineNo">509</span> Threads.shutdown(worker, this.sleepForRetries);<a name="line.509"></a> -<span class="sourceLineNo">510</span> LOG.info("ReplicationSourceWorker " + worker.getName() + " terminated");<a name="line.510"></a> -<span class="sourceLineNo">511</span> }<a name="line.511"></a> -<span class="sourceLineNo">512</span> if (this.replicationEndpoint != null) {<a name="line.512"></a> -<span class="sourceLineNo">513</span> try {<a name="line.513"></a> -<span class="sourceLineNo">514</span> this.replicationEndpoint.awaitTerminated(sleepForRetries * maxRetriesMultiplier,<a name="line.514"></a> -<span class="sourceLineNo">515</span> TimeUnit.MILLISECONDS);<a name="line.515"></a> -<span class="sourceLineNo">516</span> } catch (TimeoutException te) {<a name="line.516"></a> -<span class="sourceLineNo">517</span> LOG.warn("Got exception while waiting for endpoint to shutdown for replication source :" +<a name="line.517"></a> -<span class="sourceLineNo">518</span> this.queueId, te);<a name="line.518"></a> -<span class="sourceLineNo">519</span> }<a name="line.519"></a> -<span class="sourceLineNo">520</span> }<a name="line.520"></a> -<span class="sourceLineNo">521</span> }<a name="line.521"></a> -<span class="sourceLineNo">522</span> this.metrics.clear();<a name="line.522"></a> -<span class="sourceLineNo">523</span> }<a name="line.523"></a> -<span class="sourceLineNo">524</span><a name="line.524"></a> -<span class="sourceLineNo">525</span> @Override<a name="line.525"></a> -<span class="sourceLineNo">526</span> public String getQueueId() {<a name="line.526"></a> -<span class="sourceLineNo">527</span> return this.queueId;<a name="line.527"></a> -<span class="sourceLineNo">528</span> }<a name="line.528"></a> -<span class="sourceLineNo">529</span><a name="line.529"></a> -<span class="sourceLineNo">530</span> @Override<a name="line.530"></a> -<span class="sourceLineNo">531</span> public String getPeerId() {<a name="line.531"></a> -<span class="sourceLineNo">532</span> return this.peerId;<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> @Override<a name="line.535"></a> -<span class="sourceLineNo">536</span> public Path getCurrentPath() {<a name="line.536"></a> -<span class="sourceLineNo">537</span> // only for testing<a name="line.537"></a> -<span class="sourceLineNo">538</span> for (ReplicationSourceShipper worker : workerThreads.values()) {<a name="line.538"></a> -<span class="sourceLineNo">539</span> if (worker.getCurrentPath() != null) {<a name="line.539"></a> -<span class="sourceLineNo">540</span> return worker.getCurrentPath();<a name="line.540"></a> -<span class="sourceLineNo">541</span> }<a name="line.541"></a> -<span class="sourceLineNo">542</span> }<a name="line.542"></a> -<span class="sourceLineNo">543</span> return null;<a name="line.543"></a> -<span class="sourceLineNo">544</span> }<a name="line.544"></a> -<span class="sourceLineNo">545</span><a name="line.545"></a> -<span class="sourceLineNo">546</span> @Override<a name="line.546"></a> -<span class="sourceLineNo">547</span> public boolean isSourceActive() {<a name="line.547"></a> -<span class="sourceLineNo">548</span> return !this.server.isStopped() && this.sourceRunning;<a name="line.548"></a> -<span class="sourceLineNo">549</span> }<a name="line.549"></a> -<span class="sourceLineNo">550</span><a name="line.550"></a> -<span class="sourceLineNo">551</span> /**<a name="line.551"></a> -<span class="sourceLineNo">552</span> * Comparator used to compare logs together based on their start time<a name="line.552"></a> -<span class="sourceLineNo">553</span> */<a name="line.553"></a> -<span class="sourceLineNo">554</span> public static class LogsComparator implements Comparator<Path> {<a name="line.554"></a> -<span class="sourceLineNo">555</span><a name="line.555"></a> -<span class="sourceLineNo">556</span> @Override<a name="line.556"></a> -<span class="sourceLineNo">557</span> public int compare(Path o1, Path o2) {<a name="line.557"></a> -<span class="sourceLineNo">558</span> return Long.compare(getTS(o1), getTS(o2));<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> * Split a path to get the start time<a name="line.562"></a> -<span class="sourceLineNo">563</span> * For example: 10.20.20.171%3A60020.1277499063250<a name="line.563"></a> -<span class="sourceLineNo">564</span> * @param p path to split<a name="line.564"></a> -<span class="sourceLineNo">565</span> * @return start time<a name="line.565"></a> -<span class="sourceLineNo">566</span> */<a name="line.566"></a> -<span class="sourceLineNo">567</span> private static long getTS(Path p) {<a name="line.567"></a> -<span class="sourceLineNo">568</span> int tsIndex = p.getName().lastIndexOf('.') + 1;<a name="line.568"></a> -<span class="sourceLineNo">569</span> return Long.parseLong(p.getName().substring(tsIndex));<a name="line.569"></a> -<span class="sourceLineNo">570</span> }<a name="line.570"></a> -<span class="sourceLineNo">571</span> }<a name="line.571"></a> -<span class="sourceLineNo">572</span><a name="line.572"></a> -<span class="sourceLineNo">573</span> @Override<a name="line.573"></a> -<span class="sourceLineNo">574</span> public String getStats() {<a name="line.574"></a> -<span class="sourceLineNo">575</span> StringBuilder sb = new StringBuilder();<a name="line.575"></a> -<span class="sourceLineNo">576</span> sb.append("Total replicated edits: ").append(totalReplicatedEdits)<a name="line.576"></a> -<span class="sourceLineNo">577</span> .append(", current progress: \n");<a name="line.577"></a> -<span class="sourceLineNo">578</span> for (Map.Entry<String, ReplicationSourceShipper> entry : workerThreads.entrySet()) {<a name="line.578"></a> -<span class="sourceLineNo">579</span> String walGroupId = entry.getKey();<a name="line.579"></a> -<span class="sourceLineNo">580</span> ReplicationSourceShipper worker = entry.getValue();<a name="line.580"></a> -<span class="sourceLineNo">581</span> long position = worker.getCurrentPosition();<a name="line.581"></a> -<span class="sourceLineNo">582</span> Path currentPath = worker.getCurrentPath();<a name="line.582"></a> -<span class="sourceLineNo">583</span> sb.append("walGroup [").append(walGroupId).append("]: ");<a name="line.583"></a> -<span class="sourceLineNo">584</span> if (currentPath != null) {<a name="line.584"></a> -<span class="sourceLineNo">585</span> sb.append("currently replicating from: ").append(currentPath).append(" at position: ")<a name="line.585"></a> -<span class="sourceLineNo">586</span> .append(position).append("\n");<a name="line.586"></a> -<span class="sourceLineNo">587</span> } else {<a name="line.587"></a> -<span class="sourceLineNo">588</span> sb.append("no replication ongoing, waiting for new log");<a name="line.588"></a> -<span class="sourceLineNo">589</span> }<a name="line.589"></a> -<span class="sourceLineNo">590</span> }<a name="line.590"></a> -<span class="sourceLineNo">591</span> return sb.toString();<a name="line.591"></a> -<span class="sourceLineNo">592</span> }<a name="line.592"></a> -<span class="sourceLineNo">593</span><a name="line.593"></a> -<span class="sourceLineNo">594</span> @Override<a name="line.594"></a> -<span class="sourceLineNo">595</span> public MetricsSource getSourceMetrics() {<a name="line.595"></a> -<span class="sourceLineNo">596</span> return this.metrics;<a name="line.596"></a> -<span class="sourceLineNo">597</span> }<a name="line.597"></a> -<span class="sourceLineNo">598</span><a name="line.598"></a> -<span class="sourceLineNo">599</span> @Override<a name="line.599"></a> -<span class="sourceLineNo">600</span> public void postShipEdits(List<Entry> entries, int batchSize) {<a name="line.600"></a> -<span class="sourceLineNo">601</span> if (throttler.isEnabled()) {<a name="line.601"></a> -<span class="sourceLineNo">602</span> throttler.addPushSize(batchSize);<a name="line.602"></a> -<span class="sourceLineNo">603</span> }<a name="line.603"></a> -<span class="sourceLineNo">604</span> totalReplicatedEdits.addAndGet(entries.size());<a name="line.604"></a> -<span class="sourceLineNo">605</span> totalBufferUsed.addAndGet(-batchSize);<a name="line.605"></a> -<span class="sourceLineNo">606</span> }<a name="line.606"></a> -<span class="sourceLineNo">607</span><a name="line.607"></a> -<span class="sourceLineNo">608</span> @Override<a name="line.608"></a> -<span class="sourceLineNo">609</span> public WALFileLengthProvider getWALFileLengthProvider() {<a name="line.609"></a> -<span class="sourceLineNo">610</span> return walFileLengthProvider;<a name="line.610"></a> -<span class="sourceLineNo">611</span> }<a name="line.611"></a> -<span class="sourceLineNo">612</span><a name="line.612"></a> -<span class="sourceLineNo">613</span> @Override<a name="line.613"></a> -<span class="sourceLineNo">614</span> public ServerName getServerWALsBelongTo() {<a name="line.614"></a> -<span class="sourceLineNo">615</span> return server.getServerName();<a name="line.615"></a> -<span class="sourceLineNo">616</span> }<a name="line.616"></a> -<span class="sourceLineNo">617</span><a name="line.617"></a> -<span class="sourceLineNo">618</span> Server getServer() {<a name="line.618"></a> -<span class="sourceLineNo">619</span> return server;<a name="line.619"></a> -<span class="sourceLineNo">620</span> }<a name="line.620"></a> -<span class="sourceLineNo">621</span><a name="line.621"></a> -<span class="sourceLineNo">622</span> ReplicationQueueStorage getQueueStorage() {<a name="line.622"></a> -<span class="sourceLineNo">623</span> return queueStorage;<a name="line.623"></a> -<span class="sourceLineNo">624</span> }<a name="line.624"></a> -<span class="sourceLineNo">625</span>}<a name="line.625"></a> +<span class="sourceLineNo">137</span> /**<a name="line.137"></a> +<span class="sourceLineNo">138</span> * Instantiation method used by region servers<a name="line.138"></a> +<span class="sourceLineNo">139</span> * @param conf configuration to use<a name="line.139"></a> +<span class="sourceLineNo">140</span> * @param fs file system to use<a name="line.140"></a> +<span class="sourceLineNo">141</span> * @param manager replication manager to ping to<a name="line.141"></a> +<span class="sourceLineNo">142</span> * @param server the server for this region server<a name="line.142"></a> +<span class="sourceLineNo">143</span> * @param queueId the id of our replication queue<a name="line.143"></a> +<span class="sourceLineNo">144</span> * @param clusterId unique UUID for the cluster<a name="line.144"></a> +<span class="sourceLineNo">145</span> * @param metrics metrics for replication source<a name="line.145"></a> +<span class="sourceLineNo">146</span> */<a name="line.146"></a> +<span class="sourceLineNo">147</span> @Override<a name="line.147"></a> +<span class="sourceLineNo">148</span> public void init(Configuration conf, FileSystem fs, ReplicationSourceManager manager,<a name="line.148"></a> +<span class="sourceLineNo">149</span> ReplicationQueueStorage queueStorage, ReplicationPeer replicationPeer, Server server,<a name="line.149"></a> +<span class="sourceLineNo">150</span> String queueId, UUID clusterId, WALFileLengthProvider walFileLengthProvider,<a name="line.150"></a> +<span class="sourceLineNo">151</span> MetricsSource metrics) throws IOException {<a name="line.151"></a> +<span class="sourceLineNo">152</span> this.server = server;<a name="line.152"></a> +<span class="sourceLineNo">153</span> this.conf = HBaseConfiguration.create(conf);<a name="line.153"></a> +<span class="sourceLineNo">154</span> this.waitOnEndpointSeconds =<a name="line.154"></a> +<span class="sourceLineNo">155</span> this.conf.getInt(WAIT_ON_ENDPOINT_SECONDS, DEFAULT_WAIT_ON_ENDPOINT_SECONDS);<a name="line.155"></a> +<span class="sourceLineNo">156</span> decorateConf();<a name="line.156"></a> +<span class="sourceLineNo">157</span> this.sleepForRetries =<a name="line.157"></a> +<span class="sourceLineNo">158</span> this.conf.getLong("replication.source.sleepforretries", 1000); // 1 second<a name="line.158"></a> +<span class="sourceLineNo">159</span> this.maxRetriesMultiplier =<a name="line.159"></a> +<span class="sourceLineNo">160</span> this.conf.getInt("replication.source.maxretriesmultiplier", 300); // 5 minutes @ 1 sec per<a name="line.160"></a> +<span class="sourceLineNo">161</span> this.queueSizePerGroup = this.conf.getInt("hbase.regionserver.maxlogs", 32);<a name="line.161"></a> +<span class="sourceLineNo">162</span> this.queueStorage = queueStorage;<a name="line.162"></a> +<span class="sourceLineNo">163</span> this.replicationPeer = replicationPeer;<a name="line.163"></a> +<span class="sourceLineNo">164</span> this.manager = manager;<a name="line.164"></a> +<span class="sourceLineNo">165</span> this.fs = fs;<a name="line.165"></a> +<span class="sourceLineNo">166</span> this.metrics = metrics;<a name="line.166"></a> +<span class="sourceLineNo">167</span> this.clusterId = clusterId;<a name="line.167"></a> +<span class="sourceLineNo">168</span><a name="line.168"></a> +<span class="sourceLineNo">169</span> this.queueId = queueId;<a name="line.169"></a> +<span class="sourceLineNo">170</span> this.replicationQueueInfo = new ReplicationQueueInfo(queueId);<a name="line.170"></a> +<span class="sourceLineNo">171</span> // ReplicationQueueInfo parses the peerId out of the znode for us<a name="line.171"></a> +<span class="sourceLineNo">172</span> this.peerId = this.replicationQueueInfo.getPeerId();<a name="line.172"></a> +<span class="sourceLineNo">173</span> this.logQueueWarnThreshold = this.conf.getInt("replication.source.log.queue.warn", 2);<a name="line.173"></a> +<span class="sourceLineNo">174</span><a name="line.174"></a> +<span class="sourceLineNo">175</span> defaultBandwidth = this.conf.getLong("replication.source.per.peer.node.bandwidth", 0);<a name="line.175"></a> +<span class="sourceLineNo">176</span> currentBandwidth = getCurrentBandwidth();<a name="line.176"></a> +<span class="sourceLineNo">177</span> this.throttler = new ReplicationThrottler((double) currentBandwidth / 10.0);<a name="line.177"></a> +<span class="sourceLineNo">178</span> this.totalBufferUsed = manager.getTotalBufferUsed();<a name="line.178"></a> +<span class="sourceLineNo">179</span> this.walFileLengthProvider = walFileLengthProvider;<a name="line.179"></a> +<span class="sourceLineNo">180</span> LOG.info("queueId=" + queueId + ", ReplicationSource : " + peerId<a name="line.180"></a> +<span class="sourceLineNo">181</span> + ", currentBandwidth=" + this.currentBandwidth);<a name="line.181"></a> +<span class="sourceLineNo">182</span> }<a name="line.182"></a> +<span class="sourceLineNo">183</span><a name="line.183"></a> +<span class="sourceLineNo">184</span> private void decorateConf() {<a name="line.184"></a> +<span class="sourceLineNo">185</span> String replicationCodec = this.conf.get(HConstants.REPLICATION_CODEC_CONF_KEY);<a name="line.185"></a> +<span class="sourceLineNo">186</span> if (StringUtils.isNotEmpty(replicationCodec)) {<a name="line.186"></a> +<span class="sourceLineNo">187</span> this.conf.set(HConstants.RPC_CODEC_CONF_KEY, replicationCodec);<a name="line.187"></a> +<span class="sourceLineNo">188</span> }<a name="line.188"></a> +<span class="sourceLineNo">189</span> }<a name="line.189"></a> +<span class="sourceLineNo">190</span><a name="line.190"></a> +<span class="sourceLineNo">191</span> @Override<a name="line.191"></a> +<span class="sourceLineNo">192</span> public void enqueueLog(Path log) {<a name="line.192"></a> +<span class="sourceLineNo">193</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log.getName());<a name="line.193"></a> +<span class="sourceLineNo">194</span> PriorityBlockingQueue<Path> queue = queues.get(logPrefix);<a name="line.194"></a> +<span class="sourceLineNo">195</span> if (queue == null) {<a name="line.195"></a> +<span class="sourceLineNo">196</span> queue = new PriorityBlockingQueue<>(queueSizePerGroup, new LogsComparator());<a name="line.196"></a> +<span class="sourceLineNo">197</span> queues.put(logPrefix, queue);<a name="line.197"></a> +<span class="sourceLineNo">198</span> if (this.isSourceActive() && this.walEntryFilter != null) {<a name="line.198"></a> +<span class="sourceLineNo">199</span> // new wal group observed after source startup, start a new worker thread to track it<a name="line.199"></a> +<span class="sourceLineNo">200</span> // notice: it's possible that log enqueued when this.running is set but worker thread<a name="line.200"></a> +<span class="sourceLineNo">201</span> // still not launched, so it's necessary to check workerThreads before start the worker<a name="line.201"></a> +<span class="sourceLineNo">202</span> tryStartNewShipper(logPrefix, queue);<a name="line.202"></a> +<span class="sourceLineNo">203</span> }<a name="line.203"></a> +<span class="sourceLineNo">204</span> }<a name="line.204"></a> +<span class="sourceLineNo">205</span> queue.put(log);<a name="line.205"></a> +<span class="sourceLineNo">206</span> this.metrics.incrSizeOfLogQueue();<a name="line.206"></a> +<span class="sourceLineNo">207</span> // This will log a warning for each new log that gets created above the warn threshold<a name="line.207"></a> +<span class="sourceLineNo">208</span> int queueSize = queue.size();<a name="line.208"></a> +<span class="sourceLineNo">209</span> if (queueSize > this.logQueueWarnThreshold) {<a name="line.209"></a> +<span class="sourceLineNo">210</span> LOG.warn("WAL group " + logPrefix + " queue size: " + queueSize<a name="line.210"></a> +<span class="sourceLineNo">211</span> + " exceeds value of replication.source.log.queue.warn: " + logQueueWarnThreshold);<a name="line.211"></a> +<span class="sourceLineNo">212</span> }<a name="line.212"></a> +<span class="sourceLineNo">213</span> }<a name="line.213"></a> +<span class="sourceLineNo">214</span><a name="line.214"></a> +<span class="sourceLineNo">215</span> @Override<a name="line.215"></a> +<span class="sourceLineNo">216</span> public void addHFileRefs(TableName tableName, byte[] family, List<Pair<Path, Path>> pairs)<a name="line.216"></a> +<span class="sourceLineNo">217</span> throws ReplicationException {<a name="line.217"></a> +<span class="sourceLineNo">218</span> Map<TableName, List<String>> tableCFMap = replicationPeer.getTableCFs();<a name="line.218"></a> +<span class="sourceLineNo">219</span> if (tableCFMap != null) {<a name="line.219"></a> +<span class="sourceLineNo">220</span> List<String> tableCfs = tableCFMap.get(tableName);<a name="line.220"></a> +<span class="sourceLineNo">221</span> if (tableCFMap.containsKey(tableName)<a name="line.221"></a> +<span class="sourceLineNo">222</span> && (tableCfs == null || tableCfs.contains(Bytes.toString(family)))) {<a name="line.222"></a> +<span class="sourceLineNo">223</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.223"></a> +<span class="sourceLineNo">224</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.224"></a> +<span class="sourceLineNo">225</span> } else {<a name="line.225"></a> +<span class="sourceLineNo">226</span> LOG.debug("HFiles will not be replicated belonging to the table " + tableName + " family "<a name="line.226"></a> +<span class="sourceLineNo">227</span> + Bytes.toString(family) + " to peer id " + peerId);<a name="line.227"></a> +<span class="sourceLineNo">228</span> }<a name="line.228"></a> +<span class="sourceLineNo">229</span> } else {<a name="line.229"></a> +<span class="sourceLineNo">230</span> // user has explicitly not defined any table cfs for replication, means replicate all the<a name="line.230"></a> +<span class="sourceLineNo">231</span> // data<a name="line.231"></a> +<span class="sourceLineNo">232</span> this.queueStorage.addHFileRefs(peerId, pairs);<a name="line.232"></a> +<span class="sourceLineNo">233</span> metrics.incrSizeOfHFileRefsQueue(pairs.size());<a name="line.233"></a> +<span class="sourceLineNo">234</span> }<a name="line.234"></a> +<span class="sourceLineNo">235</span> }<a name="line.235"></a> +<span class="sourceLineNo">236</span><a name="line.236"></a> +<span class="sourceLineNo">237</span> private ReplicationEndpoint createReplicationEndpoint()<a name="line.237"></a> +<span class="sourceLineNo">238</span> throws InstantiationException, IllegalAccessException, ClassNotFoundException, IOException {<a name="line.238"></a> +<span class="sourceLineNo">239</span> RegionServerCoprocessorHost rsServerHost = null;<a name="line.239"></a> +<span class="sourceLineNo">240</span> if (server instanceof HRegionServer) {<a name="line.240"></a> +<span class="sourceLineNo">241</span> rsServerHost = ((HRegionServer) server).getRegionServerCoprocessorHost();<a name="line.241"></a> +<span class="sourceLineNo">242</span> }<a name="line.242"></a> +<span class="sourceLineNo">243</span> String replicationEndpointImpl = replicationPeer.getPeerConfig().getReplicationEndpointImpl();<a name="line.243"></a> +<span class="sourceLineNo">244</span><a name="line.244"></a> +<span class="sourceLineNo">245</span> ReplicationEndpoint replicationEndpoint;<a name="line.245"></a> +<span class="sourceLineNo">246</span> if (replicationEndpointImpl == null) {<a name="line.246"></a> +<span class="sourceLineNo">247</span> // Default to HBase inter-cluster replication endpoint; skip reflection<a name="line.247"></a> +<span class="sourceLineNo">248</span> replicationEndpoint = new HBaseInterClusterReplicationEndpoint();<a name="line.248"></a> +<span class="sourceLineNo">249</span> } else {<a name="line.249"></a> +<span class="sourceLineNo">250</span> try {<a name="line.250"></a> +<span class="sourceLineNo">251</span> replicationEndpoint = Class.forName(replicationEndpointImpl)<a name="line.251"></a> +<span class="sourceLineNo">252</span> .asSubclass(ReplicationEndpoint.class)<a name="line.252"></a> +<span class="sourceLineNo">253</span> .getDeclaredConstructor()<a name="line.253"></a> +<span class="sourceLineNo">254</span> .newInstance();<a name="line.254"></a> +<span class="sourceLineNo">255</span> } catch (NoSuchMethodException | InvocationTargetException e) {<a name="line.255"></a> +<span class="sourceLineNo">256</span> throw new IllegalArgumentException(e);<a name="line.256"></a> +<span class="sourceLineNo">257</span> }<a name="line.257"></a> +<span class="sourceLineNo">258</span> }<a name="line.258"></a> +<span class="sourceLineNo">259</span> if (rsServerHost != null) {<a name="line.259"></a> +<span class="sourceLineNo">260</span> ReplicationEndpoint newReplicationEndPoint =<a name="line.260"></a> +<span class="sourceLineNo">261</span> rsServerHost.postCreateReplicationEndPoint(replicationEndpoint);<a name="line.261"></a> +<span class="sourceLineNo">262</span> if (newReplicationEndPoint != null) {<a name="line.262"></a> +<span class="sourceLineNo">263</span> // Override the newly created endpoint from the hook with configured end point<a name="line.263"></a> +<span class="sourceLineNo">264</span> replicationEndpoint = newReplicationEndPoint;<a name="line.264"></a> +<span class="sourceLineNo">265</span> }<a name="line.265"></a> +<span class="sourceLineNo">266</span> }<a name="line.266"></a> +<span class="sourceLineNo">267</span> return replicationEndpoint;<a name="line.267"></a> +<span class="sourceLineNo">268</span> }<a name="line.268"></a> +<span class="sourceLineNo">269</span><a name="line.269"></a> +<span class="sourceLineNo">270</span> private void initAndStartReplicationEndpoint(ReplicationEndpoint replicationEndpoint)<a name="line.270"></a> +<span class="sourceLineNo">271</span> throws IOException, TimeoutException {<a name="line.271"></a> +<span class="sourceLineNo">272</span> TableDescriptors tableDescriptors = null;<a name="line.272"></a> +<span class="sourceLineNo">273</span> if (server instanceof HRegionServer) {<a name="line.273"></a> +<span class="sourceLineNo">274</span> tableDescriptors = ((HRegionServer) server).getTableDescriptors();<a name="line.274"></a> +<span class="sourceLineNo">275</span> }<a name="line.275"></a> +<span class="sourceLineNo">276</span> replicationEndpoint<a name="line.276"></a> +<span class="sourceLineNo">277</span> .init(new ReplicationEndpoint.Context(conf, replicationPeer.getConfiguration(), fs, peerId,<a name="line.277"></a> +<span class="sourceLineNo">278</span> clusterId, replicationPeer, metrics, tableDescriptors, server));<a name="line.278"></a> +<span class="sourceLineNo">279</span> replicationEndpoint.start();<a name="line.279"></a> +<span class="sourceLineNo">280</span> replicationEndpoint.awaitRunning(waitOnEndpointSeconds, TimeUnit.SECONDS);<a name="line.280"></a> +<span class="sourceLineNo">281</span> }<a name="line.281"></a> +<span class="sourceLineNo">282</span><a name="line.282"></a> +<span class="sourceLineNo">283</span> private void initializeWALEntryFilter(UUID peerClusterId) {<a name="line.283"></a> +<span class="sourceLineNo">284</span> // get the WALEntryFilter from ReplicationEndpoint and add it to default filters<a name="line.284"></a> +<span class="sourceLineNo">285</span> ArrayList<WALEntryFilter> filters =<a name="line.285"></a> +<span class="sourceLineNo">286</span> Lists.<WALEntryFilter> newArrayList(new SystemTableWALEntryFilter());<a name="line.286"></a> +<span class="sourceLineNo">287</span> WALEntryFilter filterFromEndpoint = this.replicationEndpoint.getWALEntryfilter();<a name="line.287"></a> +<span class="sourceLineNo">288</span> if (filterFromEndpoint != null) {<a name="line.288"></a> +<span class="sourceLineNo">289</span> filters.add(filterFromEndpoint);<a name="line.289"></a> +<span class="sourceLineNo">290</span> }<a name="line.290"></a> +<span class="sourceLineNo">291</span> filters.add(new ClusterMarkingEntryFilter(clusterId, peerClusterId, replicationEndpoint));<a name="line.291"></a> +<span class="sourceLineNo">292</span> this.walEntryFilter = new ChainWALEntryFilter(filters);<a name="line.292"></a> +<span class="sourceLineNo">293</span> }<a name="line.293"></a> +<span class="sourceLineNo">294</span><a name="line.294"></a> +<span class="sourceLineNo">295</span> private void tryStartNewShipper(String walGroupId, PriorityBlockingQueue<Path> queue) {<a name="line.295"></a> +<span class="sourceLineNo">296</span> ReplicationSourceShipper worker = createNewShipper(walGroupId, queue);<a name="line.296"></a> +<span class="sourceLineNo">297</span> ReplicationSourceShipper extant = workerThreads.putIfAbsent(walGroupId, worker);<a name="line.297"></a> +<span class="sourceLineNo">298</span> if (extant != null) {<a name="line.298"></a> +<span class="sourceLineNo">299</span> LOG.debug("Someone has beat us to start a worker thread for wal group {}", walGroupId);<a name="line.299"></a> +<span class="sourceLineNo">300</span> } else {<a name="line.300"></a> +<span class="sourceLineNo">301</span> LOG.debug("Starting up worker for wal group {}", walGroupId);<a name="line.301"></a> +<span class="sourceLineNo">302</span> ReplicationSourceWALReader walReader =<a name="line.302"></a> +<span class="sourceLineNo">303</span> createNewWALReader(walGroupId, queue, worker.getStartPosition());<a name="line.303"></a> +<span class="sourceLineNo">304</span> Threads.setDaemonThreadRunning(walReader, Thread.currentThread().getName() +<a name="line.304"></a> +<span class="sourceLineNo">305</span> ".replicationSource.wal-reader." + walGroupId + "," + queueId, this::uncaughtException);<a name="line.305"></a> +<span class="sourceLineNo">306</span> worker.setWALReader(walReader);<a name="line.306"></a> +<span class="sourceLineNo">307</span> worker.startup(this::uncaughtException);<a name="line.307"></a> +<span class="sourceLineNo">308</span> }<a name="line.308"></a> +<span class="sourceLineNo">309</span> }<a name="line.309"></a> +<span class="sourceLineNo">310</span><a name="line.310"></a> +<span class="sourceLineNo">311</span> protected ReplicationSourceShipper createNewShipper(String walGroupId,<a name="line.311"></a> +<span class="sourceLineNo">312</span> PriorityBlockingQueue<Path> queue) {<a name="line.312"></a> +<span class="sourceLineNo">313</span> return new ReplicationSourceShipper(conf, walGroupId, queue, this);<a name="line.313"></a> +<span class="sourceLineNo">314</span> }<a name="line.314"></a> +<span class="sourceLineNo">315</span><a name="line.315"></a> +<span class="sourceLineNo">316</span> private ReplicationSourceWALReader createNewWALReader(String walGroupId,<a name="line.316"></a> +<span class="sourceLineNo">317</span> PriorityBlockingQueue<Path> queue, long startPosition) {<a name="line.317"></a> +<span class="sourceLineNo">318</span> return replicationPeer.getPeerConfig().isSerial()<a name="line.318"></a> +<span class="sourceLineNo">319</span> ? new SerialReplicationSourceWALReader(fs, conf, queue, startPosition, walEntryFilter, this)<a name="line.319"></a> +<span class="sourceLineNo">320</span> : new ReplicationSourceWALReader(fs, conf, queue, startPosition, walEntryFilter, this);<a name="line.320"></a> +<span class="sourceLineNo">321</span> }<a name="line.321"></a> +<span class="sourceLineNo">322</span><a name="line.322"></a> +<span class="sourceLineNo">323</span> protected final void uncaughtException(Thread t, Throwable e) {<a name="line.323"></a> +<span class="sourceLineNo">324</span> RSRpcServices.exitIfOOME(e);<a name="line.324"></a> +<span class="sourceLineNo">325</span> LOG.error("Unexpected exception in " + t.getName() + " currentPath=" + getCurrentPath(), e);<a name="line.325"></a> +<span class="sourceLineNo">326</span> server.abort("Unexpected exception in " + t.getName(), e);<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> @Override<a name="line.329"></a> +<span class="sourceLineNo">330</span> public ReplicationEndpoint getReplicationEndpoint() {<a name="line.330"></a> +<span class="sourceLineNo">331</span> return this.replicationEndpoint;<a name="line.331"></a> +<span class="sourceLineNo">332</span> }<a name="line.332"></a> +<span class="sourceLineNo">333</span><a name="line.333"></a> +<span class="sourceLineNo">334</span> @Override<a name="line.334"></a> +<span class="sourceLineNo">335</span> public ReplicationSourceManager getSourceManager() {<a name="line.335"></a> +<span class="sourceLineNo">336</span> return this.manager;<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> @Override<a name="line.339"></a> +<span class="sourceLineNo">340</span> public void tryThrottle(int batchSize) throws InterruptedException {<a name="line.340"></a> +<span class="sourceLineNo">341</span> checkBandwidthChangeAndResetThrottler();<a name="line.341"></a> +<span class="sourceLineNo">342</span> if (throttler.isEnabled()) {<a name="line.342"></a> +<span class="sourceLineNo">343</span> long sleepTicks = throttler.getNextSleepInterval(batchSize);<a name="line.343"></a> +<span class="sourceLineNo">344</span> if (sleepTicks > 0) {<a name="line.344"></a> +<span class="sourceLineNo">345</span> if (LOG.isTraceEnabled()) {<a name="line.345"></a> +<span class="sourceLineNo">346</span> LOG.trace("To sleep " + sleepTicks + "ms for throttling control");<a name="line.346"></a> +<span class="sourceLineNo">347</span> }<a name="line.347"></a> +<span class="sourceLineNo">348</span> Thread.sleep(sleepTicks);<a name="line.348"></a> +<span class="sourceLineNo">349</span> // reset throttler's cycle start tick when sleep for throttling occurs<a name="line.349"></a> +<span class="sourceLineNo">350</span> throttler.resetStartTick();<a name="line.350"></a> +<span class="sourceLineNo">351</span> }<a name="line.351"></a> +<span class="sourceLineNo">352</span> }<a name="line.352"></a> +<span class="sourceLineNo">353</span> }<a name="line.353"></a> +<span class="sourceLineNo">354</span><a name="line.354"></a> +<span class="sourceLineNo">355</span> private void checkBandwidthChangeAndResetThrottler() {<a name="line.355"></a> +<span class="sourceLineNo">356</span> long peerBandwidth = getCurrentBandwidth();<a name="line.356"></a> +<span class="sourceLineNo">357</span> if (peerBandwidth != currentBandwidth) {<a name="line.357"></a> +<span class="sourceLineNo">358</span> currentBandwidth = peerBandwidth;<a name="line.358"></a> +<span class="sourceLineNo">359</span> throttler.setBandwidth((double) currentBandwidth / 10.0);<a name="line.359"></a> +<span class="sourceLineNo">360</span> LOG.info("ReplicationSource : " + peerId<a name="line.360"></a> +<span class="sourceLineNo">361</span> + " bandwidth throttling changed, currentBandWidth=" + currentBandwidth);<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><a name="line.364"></a> +<span class="sourceLineNo">365</span> private long getCurrentBandwidth() {<a name="line.365"></a> +<span class="sourceLineNo">366</span> long peerBandwidth = replicationPeer.getPeerBandwidth();<a name="line.366"></a> +<span class="sourceLineNo">367</span> // user can set peer bandwidth to 0 to use default bandwidth<a name="line.367"></a> +<span class="sourceLineNo">368</span> return peerBandwidth != 0 ? peerBandwidth : defaultBandwidth;<a name="line.368"></a> +<span class="sourceLineNo">369</span> }<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> * Do the sleeping logic<a name="line.372"></a> +<span class="sourceLineNo">373</span> * @param msg Why we sleep<a name="line.373"></a> +<span class="sourceLineNo">374</span> * @param sleepMultiplier by how many times the default sleeping time is augmented<a name="line.374"></a> +<span class="sourceLineNo">375</span> * @return True if <code>sleepMultiplier</code> is &lt; <code>maxRetriesMultiplier</code><a name="line.375"></a> +<span class="sourceLineNo">376</span> */<a name="line.376"></a> +<span class="sourceLineNo">377</span> protected boolean sleepForRetries(String msg, int sleepMultiplier) {<a name="line.377"></a> +<span class="sourceLineNo">378</span> try {<a name="line.378"></a> +<span class="sourceLineNo">379</span> if (LOG.isTraceEnabled()) {<a name="line.379"></a> +<span class="sourceLineNo">380</span> LOG.trace(msg + ", sleeping " + sleepForRetries + " times " + sleepMultiplier);<a name="line.380"></a> +<span class="sourceLineNo">381</span> }<a name="line.381"></a> +<span class="sourceLineNo">382</span> Thread.sleep(this.sleepForRetries * sleepMultiplier);<a name="line.382"></a> +<span class="sourceLineNo">383</span> } catch (InterruptedException e) {<a name="line.383"></a> +<span class="sourceLineNo">384</span> LOG.debug("Interrupted while sleeping between retries");<a name="line.384"></a> +<span class="sourceLineNo">385</span> Thread.currentThread().interrupt();<a name="line.385"></a> +<span class="sourceLineNo">386</span> }<a name="line.386"></a> +<span class="sourceLineNo">387</span> return sleepMultiplier < maxRetriesMultiplier;<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> /**<a name="line.390"></a> +<span class="sourceLineNo">391</span> * check whether the peer is enabled or not<a name="line.391"></a> +<span class="sourceLineNo">392</span> * @return true if the peer is enabled, otherwise false<a name="line.392"></a> +<span class="sourceLineNo">393</span> */<a name="line.393"></a> +<span class="sourceLineNo">394</span> @Override<a name="line.394"></a> +<span class="sourceLineNo">395</span> public boolean isPeerEnabled() {<a name="line.395"></a> +<span class="sourceLineNo">396</span> return replicationPeer.isPeerEnabled();<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> +<span class="sourceLineNo">399</span> private void initialize() {<a name="line.399"></a> +<span class="sourceLineNo">400</span> int sleepMultiplier = 1;<a name="line.400"></a> +<span class="sourceLineNo">401</span> while (this.isSourceActive()) {<a name="line.401"></a> +<span class="sourceLineNo">402</span> ReplicationEndpoint replicationEndpoint;<a name="line.402"></a> +<span class="sourceLineNo">403</span> try {<a name="line.403"></a> +<span class="sourceLineNo">404</span> replicationEndpoint = createReplicationEndpoint();<a name="line.404"></a> +<span class="sourceLineNo">405</span> } catch (Exception e) {<a name="line.405"></a> +<span class="sourceLineNo">406</span> LOG.warn("error creating ReplicationEndpoint, retry", e);<a name="line.406"></a> +<span class="sourceLineNo">407</span> if (sleepForRetries("Error creating ReplicationEndpoint", sleepMultiplier)) {<a name="line.407"></a> +<span class="sourceLineNo">408</span> sleepMultiplier++;<a name="line.408"></a> +<span class="sourceLineNo">409</span> }<a name="line.409"></a> +<span class="sourceLineNo">410</span> continue;<a name="line.410"></a> +<span class="sourceLineNo">411</span> }<a name="line.411"></a> +<span class="sourceLineNo">412</span><a name="line.412"></a> +<span class="sourceLineNo">413</span> try {<a name="line.413"></a> +<span class="sourceLineNo">414</span> initAndStartReplicationEndpoint(replicationEndpoint);<a name="line.414"></a> +<span class="sourceLineNo">415</span> this.replicationEndpoint = replicationEndpoint;<a name="line.415"></a> +<span class="sourceLineNo">416</span> break;<a name="line.416"></a> +<span class="sourceLineNo">417</span> } catch (Exception e) {<a name="line.417"></a> +<span class="sourceLineNo">418</span> LOG.warn("Error starting ReplicationEndpoint, retry", e);<a name="line.418"></a> +<span class="sourceLineNo">419</span> replicationEndpoint.stop();<a name="line.419"></a> +<span class="sourceLineNo">420</span> if (sleepForRetries("Error starting ReplicationEndpoint", sleepMultiplier)) {<a name="line.420"></a> +<span class="sourceLineNo">421</span> sleepMultiplier++;<a name="line.421"></a> +<span class="sourceLineNo">422</span> }<a name="line.422"></a> +<span class="sourceLineNo">423</span> }<a name="line.423"></a> +<span class="sourceLineNo">424</span> }<a name="line.424"></a> +<span class="sourceLineNo">425</span><a name="line.425"></a> +<span class="sourceLineNo">426</span> if (!this.isSourceActive()) {<a name="line.426"></a> +<span class="sourceLineNo">427</span> return;<a name="line.427"></a> +<span class="sourceLineNo">428</span> }<a name="line.428"></a> +<span class="sourceLineNo">429</span><a name="line.429"></a> +<span class="sourceLineNo">430</span> sleepMultiplier = 1;<a name="line.430"></a> +<span class="sourceLineNo">431</span> UUID peerClusterId;<a name="line.431"></a> +<span class="sourceLineNo">432</span> // delay this until we are in an asynchronous thread<a name="line.432"></a> +<span class="sourceLineNo">433</span> for (;;) {<a name="line.433"></a> +<span class="sourceLineNo">434</span> peerClusterId = replicationEndpoint.getPeerUUID();<a name="line.434"></a> +<span class="sourceLineNo">435</span> if (this.isSourceActive() && peerClusterId == null) {<a name="line.435"></a> +<span class="sourceLineNo">436</span> if (sleepForRetries("Cannot contact the peer's zk ensemble", sleepMultiplier)) {<a name="line.436"></a> +<span class="sourceLineNo">437</span> sleepMultiplier++;<a name="line.437"></a> +<span class="sourceLineNo">438</span> }<a name="line.438"></a> +<span class="sourceLineNo">439</span> } else {<a name="line.439"></a> +<span class="sourceLineNo">440</span> break;<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> // In rare case, zookeeper setting may be messed up. That leads to the incorrect<a name="line.444"></a> +<span class="sourceLineNo">445</span> // peerClusterId value, which is the same as the source clusterId<a name="line.445"></a> +<span class="sourceLineNo">446</span> if (clusterId.equals(peerClusterId) && !replicationEndpoint.canReplicateToSameCluster()) {<a name="line.446"></a> +<span class="sourceLineNo">447</span> this.terminate("ClusterId " + clusterId + " is replicating to itself: peerClusterId "<a name="line.447"></a> +<span class="sourceLineNo">448</span> + peerClusterId + " which is not allowed by ReplicationEndpoint:"<a name="line.448"></a> +<span class="sourceLineNo">449</span> + replicationEndpoint.getClass().getName(), null, false);<a name="line.449"></a> +<span class="sourceLineNo">450</span> this.manager.removeSource(this);<a name="line.450"></a> +<span class="sourceLineNo">451</span> return;<a name="line.451"></a> +<span class="sourceLineNo">452</span> }<a name="line.452"></a> +<span class="sourceLineNo">453</span> LOG.info("Replicating " + clusterId + " -> " + peerClusterId);<a name="line.453"></a> +<span class="sourceLineNo">454</span><a name="line.454"></a> +<span class="sourceLineNo">455</span> initializeWALEntryFilter(peerClusterId);<a name="line.455"></a> +<span class="sourceLineNo">456</span> // start workers<a name="line.456"></a> +<span class="sourceLineNo">457</span> for (Map.Entry<String, PriorityBlockingQueue<Path>> entry : queues.entrySet()) {<a name="line.457"></a> +<span class="sourceLineNo">458</span> String walGroupId = entry.getKey();<a name="line.458"></a> +<span class="sourceLineNo">459</span> PriorityBlockingQueue<Path> queue = entry.getValue();<a name="line.459"></a> +<span class="sourceLineNo">460</span> tryStartNewShipper(walGroupId, queue);<a name="line.460"></a> +<span class="sourceLineNo">461</span> }<a name="line.4
<TRUNCATED>
