You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hbase.apache.org by gi...@apache.org on 2018/03/26 14:48:07 UTC

[13/19] hbase-site git commit: Published site at 7a1d00c7a0719ccb8d97b42965835f124ef27804.

http://git-wip-us.apache.org/repos/asf/hbase-site/blob/b4aaf75c/devapidocs/src-html/org/apache/hadoop/hbase/replication/ZKReplicationQueueStorage.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/ZKReplicationQueueStorage.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/ZKReplicationQueueStorage.html
index 477cdb1..72e0a46 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/ZKReplicationQueueStorage.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/ZKReplicationQueueStorage.html
@@ -212,379 +212,419 @@
 <span class="sourceLineNo">204</span>  }<a name="line.204"></a>
 <span class="sourceLineNo">205</span><a name="line.205"></a>
 <span class="sourceLineNo">206</span>  private void addLastSeqIdsToOps(String queueId, Map&lt;String, Long&gt; lastSeqIds,<a name="line.206"></a>
-<span class="sourceLineNo">207</span>      List&lt;ZKUtilOp&gt; listOfOps) throws KeeperException {<a name="line.207"></a>
-<span class="sourceLineNo">208</span>    for (Entry&lt;String, Long&gt; lastSeqEntry : lastSeqIds.entrySet()) {<a name="line.208"></a>
-<span class="sourceLineNo">209</span>      String peerId = new ReplicationQueueInfo(queueId).getPeerId();<a name="line.209"></a>
+<span class="sourceLineNo">207</span>      List&lt;ZKUtilOp&gt; listOfOps) throws KeeperException, ReplicationException {<a name="line.207"></a>
+<span class="sourceLineNo">208</span>    String peerId = new ReplicationQueueInfo(queueId).getPeerId();<a name="line.208"></a>
+<span class="sourceLineNo">209</span>    for (Entry&lt;String, Long&gt; lastSeqEntry : lastSeqIds.entrySet()) {<a name="line.209"></a>
 <span class="sourceLineNo">210</span>      String path = getSerialReplicationRegionPeerNode(lastSeqEntry.getKey(), peerId);<a name="line.210"></a>
-<span class="sourceLineNo">211</span>      /*<a name="line.211"></a>
-<span class="sourceLineNo">212</span>       * Make sure the existence of path<a name="line.212"></a>
-<span class="sourceLineNo">213</span>       * /hbase/replication/regions/&lt;hash&gt;/&lt;encoded-region-name&gt;-&lt;peer-id&gt;. As the javadoc in<a name="line.213"></a>
-<span class="sourceLineNo">214</span>       * multiOrSequential() method said, if received a NodeExistsException, all operations will<a name="line.214"></a>
-<span class="sourceLineNo">215</span>       * fail. So create the path here, and in fact, no need to add this operation to listOfOps,<a name="line.215"></a>
-<span class="sourceLineNo">216</span>       * because only need to make sure that update file position and sequence id atomically.<a name="line.216"></a>
-<span class="sourceLineNo">217</span>       */<a name="line.217"></a>
-<span class="sourceLineNo">218</span>      ZKUtil.createWithParents(zookeeper, path);<a name="line.218"></a>
-<span class="sourceLineNo">219</span>      // Persist the max sequence id of region to zookeeper.<a name="line.219"></a>
-<span class="sourceLineNo">220</span>      listOfOps.add(ZKUtilOp.setData(path, ZKUtil.positionToByteArray(lastSeqEntry.getValue())));<a name="line.220"></a>
-<span class="sourceLineNo">221</span>    }<a name="line.221"></a>
-<span class="sourceLineNo">222</span>  }<a name="line.222"></a>
-<span class="sourceLineNo">223</span><a name="line.223"></a>
-<span class="sourceLineNo">224</span>  @Override<a name="line.224"></a>
-<span class="sourceLineNo">225</span>  public void setWALPosition(ServerName serverName, String queueId, String fileName, long position,<a name="line.225"></a>
-<span class="sourceLineNo">226</span>      Map&lt;String, Long&gt; lastSeqIds) throws ReplicationException {<a name="line.226"></a>
-<span class="sourceLineNo">227</span>    try {<a name="line.227"></a>
-<span class="sourceLineNo">228</span>      List&lt;ZKUtilOp&gt; listOfOps = new ArrayList&lt;&gt;();<a name="line.228"></a>
-<span class="sourceLineNo">229</span>      if (position &gt; 0) {<a name="line.229"></a>
-<span class="sourceLineNo">230</span>        listOfOps.add(ZKUtilOp.setData(getFileNode(serverName, queueId, fileName),<a name="line.230"></a>
-<span class="sourceLineNo">231</span>          ZKUtil.positionToByteArray(position)));<a name="line.231"></a>
-<span class="sourceLineNo">232</span>      }<a name="line.232"></a>
-<span class="sourceLineNo">233</span>      // Persist the max sequence id(s) of regions for serial replication atomically.<a name="line.233"></a>
-<span class="sourceLineNo">234</span>      addLastSeqIdsToOps(queueId, lastSeqIds, listOfOps);<a name="line.234"></a>
-<span class="sourceLineNo">235</span>      ZKUtil.multiOrSequential(zookeeper, listOfOps, false);<a name="line.235"></a>
-<span class="sourceLineNo">236</span>    } catch (KeeperException e) {<a name="line.236"></a>
-<span class="sourceLineNo">237</span>      throw new ReplicationException("Failed to set log position (serverName=" + serverName<a name="line.237"></a>
-<span class="sourceLineNo">238</span>          + ", queueId=" + queueId + ", fileName=" + fileName + ", position=" + position + ")", e);<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>  @Override<a name="line.242"></a>
-<span class="sourceLineNo">243</span>  public long getLastSequenceId(String encodedRegionName, String peerId)<a name="line.243"></a>
-<span class="sourceLineNo">244</span>      throws ReplicationException {<a name="line.244"></a>
-<span class="sourceLineNo">245</span>    byte[] data;<a name="line.245"></a>
-<span class="sourceLineNo">246</span>    try {<a name="line.246"></a>
-<span class="sourceLineNo">247</span>      data =<a name="line.247"></a>
-<span class="sourceLineNo">248</span>          ZKUtil.getData(zookeeper, getSerialReplicationRegionPeerNode(encodedRegionName, peerId));<a name="line.248"></a>
-<span class="sourceLineNo">249</span>    } catch (KeeperException | InterruptedException e) {<a name="line.249"></a>
-<span class="sourceLineNo">250</span>      throw new ReplicationException("Failed to get the last sequence id(region="<a name="line.250"></a>
-<span class="sourceLineNo">251</span>          + encodedRegionName + ", peerId=" + peerId + ")");<a name="line.251"></a>
-<span class="sourceLineNo">252</span>    }<a name="line.252"></a>
-<span class="sourceLineNo">253</span>    try {<a name="line.253"></a>
-<span class="sourceLineNo">254</span>      return ZKUtil.parseWALPositionFrom(data);<a name="line.254"></a>
-<span class="sourceLineNo">255</span>    } catch (DeserializationException de) {<a name="line.255"></a>
-<span class="sourceLineNo">256</span>      LOG.warn("Failed to parse log position (region=" + encodedRegionName + ", peerId=" + peerId<a name="line.256"></a>
-<span class="sourceLineNo">257</span>          + "), data=" + Bytes.toStringBinary(data));<a name="line.257"></a>
-<span class="sourceLineNo">258</span>    }<a name="line.258"></a>
-<span class="sourceLineNo">259</span>    return HConstants.NO_SEQNUM;<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>  @Override<a name="line.262"></a>
-<span class="sourceLineNo">263</span>  public void setLastSequenceIds(String peerId, Map&lt;String, Long&gt; lastSeqIds)<a name="line.263"></a>
-<span class="sourceLineNo">264</span>      throws ReplicationException {<a name="line.264"></a>
-<span class="sourceLineNo">265</span>    try {<a name="line.265"></a>
-<span class="sourceLineNo">266</span>      List&lt;ZKUtilOp&gt; listOfOps = new ArrayList&lt;&gt;();<a name="line.266"></a>
-<span class="sourceLineNo">267</span>      addLastSeqIdsToOps(peerId, lastSeqIds, listOfOps);<a name="line.267"></a>
-<span class="sourceLineNo">268</span>      ZKUtil.multiOrSequential(zookeeper, listOfOps, false);<a name="line.268"></a>
-<span class="sourceLineNo">269</span>    } catch (KeeperException e) {<a name="line.269"></a>
-<span class="sourceLineNo">270</span>      throw new ReplicationException("Failed to set last sequence ids, peerId=" + peerId +<a name="line.270"></a>
-<span class="sourceLineNo">271</span>        ", lastSeqIds.size=" + lastSeqIds.size(), e);<a name="line.271"></a>
-<span class="sourceLineNo">272</span>    }<a name="line.272"></a>
-<span class="sourceLineNo">273</span>  }<a name="line.273"></a>
-<span class="sourceLineNo">274</span><a name="line.274"></a>
-<span class="sourceLineNo">275</span>  @Override<a name="line.275"></a>
-<span class="sourceLineNo">276</span>  public long getWALPosition(ServerName serverName, String queueId, String fileName)<a name="line.276"></a>
-<span class="sourceLineNo">277</span>      throws ReplicationException {<a name="line.277"></a>
-<span class="sourceLineNo">278</span>    byte[] bytes;<a name="line.278"></a>
-<span class="sourceLineNo">279</span>    try {<a name="line.279"></a>
-<span class="sourceLineNo">280</span>      bytes = ZKUtil.getData(zookeeper, getFileNode(serverName, queueId, fileName));<a name="line.280"></a>
-<span class="sourceLineNo">281</span>    } catch (KeeperException | InterruptedException e) {<a name="line.281"></a>
-<span class="sourceLineNo">282</span>      throw new ReplicationException("Failed to get log position (serverName=" + serverName +<a name="line.282"></a>
-<span class="sourceLineNo">283</span>        ", queueId=" + queueId + ", fileName=" + fileName + ")", e);<a name="line.283"></a>
-<span class="sourceLineNo">284</span>    }<a name="line.284"></a>
-<span class="sourceLineNo">285</span>    try {<a name="line.285"></a>
-<span class="sourceLineNo">286</span>      return ZKUtil.parseWALPositionFrom(bytes);<a name="line.286"></a>
-<span class="sourceLineNo">287</span>    } catch (DeserializationException de) {<a name="line.287"></a>
-<span class="sourceLineNo">288</span>      LOG.warn("Failed parse log position (serverName={}, queueId={}, fileName={})",<a name="line.288"></a>
-<span class="sourceLineNo">289</span>          serverName, queueId, fileName);<a name="line.289"></a>
-<span class="sourceLineNo">290</span>    }<a name="line.290"></a>
-<span class="sourceLineNo">291</span>    // if we can not parse the position, start at the beginning of the wal file again<a name="line.291"></a>
-<span class="sourceLineNo">292</span>    return 0;<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>  @Override<a name="line.295"></a>
-<span class="sourceLineNo">296</span>  public Pair&lt;String, SortedSet&lt;String&gt;&gt; claimQueue(ServerName sourceServerName, String queueId,<a name="line.296"></a>
-<span class="sourceLineNo">297</span>      ServerName destServerName) throws ReplicationException {<a name="line.297"></a>
-<span class="sourceLineNo">298</span>    LOG.info("Atomically moving {}/{}'s WALs to {}", sourceServerName, queueId, destServerName);<a name="line.298"></a>
-<span class="sourceLineNo">299</span>    try {<a name="line.299"></a>
-<span class="sourceLineNo">300</span>      ZKUtil.createWithParents(zookeeper, getRsNode(destServerName));<a name="line.300"></a>
-<span class="sourceLineNo">301</span>    } catch (KeeperException e) {<a name="line.301"></a>
-<span class="sourceLineNo">302</span>      throw new ReplicationException(<a name="line.302"></a>
-<span class="sourceLineNo">303</span>          "Claim queue queueId=" + queueId + " from " + sourceServerName + " to " + destServerName +<a name="line.303"></a>
-<span class="sourceLineNo">304</span>            " failed when creating the node for " + destServerName,<a name="line.304"></a>
-<span class="sourceLineNo">305</span>          e);<a name="line.305"></a>
-<span class="sourceLineNo">306</span>    }<a name="line.306"></a>
-<span class="sourceLineNo">307</span>    try {<a name="line.307"></a>
-<span class="sourceLineNo">308</span>      String oldQueueNode = getQueueNode(sourceServerName, queueId);<a name="line.308"></a>
-<span class="sourceLineNo">309</span>      List&lt;String&gt; wals = ZKUtil.listChildrenNoWatch(zookeeper, oldQueueNode);<a name="line.309"></a>
-<span class="sourceLineNo">310</span>      String newQueueId = queueId + "-" + sourceServerName;<a name="line.310"></a>
-<span class="sourceLineNo">311</span>      if (CollectionUtils.isEmpty(wals)) {<a name="line.311"></a>
-<span class="sourceLineNo">312</span>        ZKUtil.deleteNodeFailSilent(zookeeper, oldQueueNode);<a name="line.312"></a>
-<span class="sourceLineNo">313</span>        LOG.info("Removed empty {}/{}", sourceServerName, queueId);<a name="line.313"></a>
-<span class="sourceLineNo">314</span>        return new Pair&lt;&gt;(newQueueId, Collections.emptySortedSet());<a name="line.314"></a>
-<span class="sourceLineNo">315</span>      }<a name="line.315"></a>
-<span class="sourceLineNo">316</span>      String newQueueNode = getQueueNode(destServerName, newQueueId);<a name="line.316"></a>
-<span class="sourceLineNo">317</span>      List&lt;ZKUtilOp&gt; listOfOps = new ArrayList&lt;&gt;();<a name="line.317"></a>
-<span class="sourceLineNo">318</span>      SortedSet&lt;String&gt; logQueue = new TreeSet&lt;&gt;();<a name="line.318"></a>
-<span class="sourceLineNo">319</span>      // create the new cluster znode<a name="line.319"></a>
-<span class="sourceLineNo">320</span>      listOfOps.add(ZKUtilOp.createAndFailSilent(newQueueNode, HConstants.EMPTY_BYTE_ARRAY));<a name="line.320"></a>
-<span class="sourceLineNo">321</span>      // get the offset of the logs and set it to new znodes<a name="line.321"></a>
-<span class="sourceLineNo">322</span>      for (String wal : wals) {<a name="line.322"></a>
-<span class="sourceLineNo">323</span>        String oldWalNode = getFileNode(oldQueueNode, wal);<a name="line.323"></a>
-<span class="sourceLineNo">324</span>        byte[] logOffset = ZKUtil.getData(this.zookeeper, oldWalNode);<a name="line.324"></a>
-<span class="sourceLineNo">325</span>        LOG.debug("Creating {} with data {}", wal, Bytes.toStringBinary(logOffset));<a name="line.325"></a>
-<span class="sourceLineNo">326</span>        String newWalNode = getFileNode(newQueueNode, wal);<a name="line.326"></a>
-<span class="sourceLineNo">327</span>        listOfOps.add(ZKUtilOp.createAndFailSilent(newWalNode, logOffset));<a name="line.327"></a>
-<span class="sourceLineNo">328</span>        listOfOps.add(ZKUtilOp.deleteNodeFailSilent(oldWalNode));<a name="line.328"></a>
-<span class="sourceLineNo">329</span>        logQueue.add(wal);<a name="line.329"></a>
-<span class="sourceLineNo">330</span>      }<a name="line.330"></a>
-<span class="sourceLineNo">331</span>      // add delete op for peer<a name="line.331"></a>
-<span class="sourceLineNo">332</span>      listOfOps.add(ZKUtilOp.deleteNodeFailSilent(oldQueueNode));<a name="line.332"></a>
-<span class="sourceLineNo">333</span><a name="line.333"></a>
-<span class="sourceLineNo">334</span>      LOG.trace("The multi list size is {}", listOfOps.size());<a name="line.334"></a>
-<span class="sourceLineNo">335</span>      ZKUtil.multiOrSequential(zookeeper, listOfOps, false);<a name="line.335"></a>
-<span class="sourceLineNo">336</span><a name="line.336"></a>
-<span class="sourceLineNo">337</span>      LOG.info("Atomically moved {}/{}'s WALs to {}", sourceServerName, queueId, destServerName);<a name="line.337"></a>
-<span class="sourceLineNo">338</span>      return new Pair&lt;&gt;(newQueueId, logQueue);<a name="line.338"></a>
-<span class="sourceLineNo">339</span>    } catch (NoNodeException | NodeExistsException | NotEmptyException | BadVersionException e) {<a name="line.339"></a>
-<span class="sourceLineNo">340</span>      // Multi call failed; it looks like some other regionserver took away the logs.<a name="line.340"></a>
-<span class="sourceLineNo">341</span>      // These exceptions mean that zk tells us the request can not be execute so it is safe to just<a name="line.341"></a>
-<span class="sourceLineNo">342</span>      // return a null. For other types of exception should be thrown out to notify the upper layer.<a name="line.342"></a>
-<span class="sourceLineNo">343</span>      LOG.info("Claim queue queueId={} from {} to {} failed with {}, someone else took the log?",<a name="line.343"></a>
-<span class="sourceLineNo">344</span>          queueId,sourceServerName, destServerName, e.toString());<a name="line.344"></a>
-<span class="sourceLineNo">345</span>      return null;<a name="line.345"></a>
-<span class="sourceLineNo">346</span>    } catch (KeeperException | InterruptedException e) {<a name="line.346"></a>
-<span class="sourceLineNo">347</span>      throw new ReplicationException("Claim queue queueId=" + queueId + " from " +<a name="line.347"></a>
-<span class="sourceLineNo">348</span>        sourceServerName + " to " + destServerName + " failed", e);<a name="line.348"></a>
-<span class="sourceLineNo">349</span>    }<a name="line.349"></a>
-<span class="sourceLineNo">350</span>  }<a name="line.350"></a>
-<span class="sourceLineNo">351</span><a name="line.351"></a>
-<span class="sourceLineNo">352</span>  @Override<a name="line.352"></a>
-<span class="sourceLineNo">353</span>  public void removeReplicatorIfQueueIsEmpty(ServerName serverName) throws ReplicationException {<a name="line.353"></a>
-<span class="sourceLineNo">354</span>    try {<a name="line.354"></a>
-<span class="sourceLineNo">355</span>      ZKUtil.deleteNodeFailSilent(zookeeper, getRsNode(serverName));<a name="line.355"></a>
-<span class="sourceLineNo">356</span>    } catch (NotEmptyException e) {<a name="line.356"></a>
-<span class="sourceLineNo">357</span>      // keep silence to avoid logging too much.<a name="line.357"></a>
-<span class="sourceLineNo">358</span>    } catch (KeeperException e) {<a name="line.358"></a>
-<span class="sourceLineNo">359</span>      throw new ReplicationException("Failed to remove replicator for " + serverName, e);<a name="line.359"></a>
-<span class="sourceLineNo">360</span>    }<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>  private List&lt;ServerName&gt; getListOfReplicators0() throws KeeperException {<a name="line.363"></a>
-<span class="sourceLineNo">364</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, queuesZNode)).stream()<a name="line.364"></a>
-<span class="sourceLineNo">365</span>        .map(ServerName::parseServerName).collect(toList());<a name="line.365"></a>
-<span class="sourceLineNo">366</span>  }<a name="line.366"></a>
-<span class="sourceLineNo">367</span><a name="line.367"></a>
-<span class="sourceLineNo">368</span>  @Override<a name="line.368"></a>
-<span class="sourceLineNo">369</span>  public List&lt;ServerName&gt; getListOfReplicators() throws ReplicationException {<a name="line.369"></a>
-<span class="sourceLineNo">370</span>    try {<a name="line.370"></a>
-<span class="sourceLineNo">371</span>      return getListOfReplicators0();<a name="line.371"></a>
-<span class="sourceLineNo">372</span>    } catch (KeeperException e) {<a name="line.372"></a>
-<span class="sourceLineNo">373</span>      throw new ReplicationException("Failed to get list of replicators", e);<a name="line.373"></a>
-<span class="sourceLineNo">374</span>    }<a name="line.374"></a>
-<span class="sourceLineNo">375</span>  }<a name="line.375"></a>
+<span class="sourceLineNo">211</span>      Pair&lt;Long, Integer&gt; p = getLastSequenceIdWithVersion(lastSeqEntry.getKey(), peerId);<a name="line.211"></a>
+<span class="sourceLineNo">212</span>      byte[] data = ZKUtil.positionToByteArray(lastSeqEntry.getValue());<a name="line.212"></a>
+<span class="sourceLineNo">213</span>      if (p.getSecond() &lt; 0) { // ZNode does not exist.<a name="line.213"></a>
+<span class="sourceLineNo">214</span>        ZKUtil.createWithParents(zookeeper,<a name="line.214"></a>
+<span class="sourceLineNo">215</span>          path.substring(0, path.lastIndexOf(ZNodePaths.ZNODE_PATH_SEPARATOR)));<a name="line.215"></a>
+<span class="sourceLineNo">216</span>        listOfOps.add(ZKUtilOp.createAndFailSilent(path, data));<a name="line.216"></a>
+<span class="sourceLineNo">217</span>        continue;<a name="line.217"></a>
+<span class="sourceLineNo">218</span>      }<a name="line.218"></a>
+<span class="sourceLineNo">219</span>      // Perform CAS in a specific version v0 (HBASE-20138)<a name="line.219"></a>
+<span class="sourceLineNo">220</span>      int v0 = p.getSecond();<a name="line.220"></a>
+<span class="sourceLineNo">221</span>      long lastPushedSeqId = p.getFirst();<a name="line.221"></a>
+<span class="sourceLineNo">222</span>      if (lastSeqEntry.getValue() &lt;= lastPushedSeqId) {<a name="line.222"></a>
+<span class="sourceLineNo">223</span>        continue;<a name="line.223"></a>
+<span class="sourceLineNo">224</span>      }<a name="line.224"></a>
+<span class="sourceLineNo">225</span>      listOfOps.add(ZKUtilOp.setData(path, data, v0));<a name="line.225"></a>
+<span class="sourceLineNo">226</span>    }<a name="line.226"></a>
+<span class="sourceLineNo">227</span>  }<a name="line.227"></a>
+<span class="sourceLineNo">228</span><a name="line.228"></a>
+<span class="sourceLineNo">229</span>  @Override<a name="line.229"></a>
+<span class="sourceLineNo">230</span>  public void setWALPosition(ServerName serverName, String queueId, String fileName, long position,<a name="line.230"></a>
+<span class="sourceLineNo">231</span>      Map&lt;String, Long&gt; lastSeqIds) throws ReplicationException {<a name="line.231"></a>
+<span class="sourceLineNo">232</span>    try {<a name="line.232"></a>
+<span class="sourceLineNo">233</span>      for (int retry = 0;; retry++) {<a name="line.233"></a>
+<span class="sourceLineNo">234</span>        List&lt;ZKUtilOp&gt; listOfOps = new ArrayList&lt;&gt;();<a name="line.234"></a>
+<span class="sourceLineNo">235</span>        if (position &gt; 0) {<a name="line.235"></a>
+<span class="sourceLineNo">236</span>          listOfOps.add(ZKUtilOp.setData(getFileNode(serverName, queueId, fileName),<a name="line.236"></a>
+<span class="sourceLineNo">237</span>            ZKUtil.positionToByteArray(position)));<a name="line.237"></a>
+<span class="sourceLineNo">238</span>        }<a name="line.238"></a>
+<span class="sourceLineNo">239</span>        // Persist the max sequence id(s) of regions for serial replication atomically.<a name="line.239"></a>
+<span class="sourceLineNo">240</span>        addLastSeqIdsToOps(queueId, lastSeqIds, listOfOps);<a name="line.240"></a>
+<span class="sourceLineNo">241</span>        if (listOfOps.isEmpty()) {<a name="line.241"></a>
+<span class="sourceLineNo">242</span>          return;<a name="line.242"></a>
+<span class="sourceLineNo">243</span>        }<a name="line.243"></a>
+<span class="sourceLineNo">244</span>        try {<a name="line.244"></a>
+<span class="sourceLineNo">245</span>          ZKUtil.multiOrSequential(zookeeper, listOfOps, false);<a name="line.245"></a>
+<span class="sourceLineNo">246</span>          return;<a name="line.246"></a>
+<span class="sourceLineNo">247</span>        } catch (KeeperException.BadVersionException | KeeperException.NodeExistsException e) {<a name="line.247"></a>
+<span class="sourceLineNo">248</span>          LOG.warn(<a name="line.248"></a>
+<span class="sourceLineNo">249</span>            "Bad version(or node exist) when persist the last pushed sequence id to zookeeper storage, "<a name="line.249"></a>
+<span class="sourceLineNo">250</span>                + "Retry = " + retry + ", serverName=" + serverName + ", queueId=" + queueId<a name="line.250"></a>
+<span class="sourceLineNo">251</span>                + ", fileName=" + fileName);<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>    } catch (KeeperException e) {<a name="line.254"></a>
+<span class="sourceLineNo">255</span>      throw new ReplicationException("Failed to set log position (serverName=" + serverName<a name="line.255"></a>
+<span class="sourceLineNo">256</span>          + ", queueId=" + queueId + ", fileName=" + fileName + ", position=" + position + ")", 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><a name="line.259"></a>
+<span class="sourceLineNo">260</span>  /**<a name="line.260"></a>
+<span class="sourceLineNo">261</span>   * Return the {lastPushedSequenceId, ZNodeDataVersion} pair. if ZNodeDataVersion is -1, it means<a name="line.261"></a>
+<span class="sourceLineNo">262</span>   * that the ZNode does not exist.<a name="line.262"></a>
+<span class="sourceLineNo">263</span>   */<a name="line.263"></a>
+<span class="sourceLineNo">264</span>  @VisibleForTesting<a name="line.264"></a>
+<span class="sourceLineNo">265</span>  protected Pair&lt;Long, Integer&gt; getLastSequenceIdWithVersion(String encodedRegionName,<a name="line.265"></a>
+<span class="sourceLineNo">266</span>      String peerId) throws KeeperException {<a name="line.266"></a>
+<span class="sourceLineNo">267</span>    Stat stat = new Stat();<a name="line.267"></a>
+<span class="sourceLineNo">268</span>    String path = getSerialReplicationRegionPeerNode(encodedRegionName, peerId);<a name="line.268"></a>
+<span class="sourceLineNo">269</span>    byte[] data = ZKUtil.getDataNoWatch(zookeeper, path, stat);<a name="line.269"></a>
+<span class="sourceLineNo">270</span>    if (data == null) {<a name="line.270"></a>
+<span class="sourceLineNo">271</span>      // ZNode does not exist, so just return version -1 to indicate that no node exist.<a name="line.271"></a>
+<span class="sourceLineNo">272</span>      return Pair.newPair(HConstants.NO_SEQNUM, -1);<a name="line.272"></a>
+<span class="sourceLineNo">273</span>    }<a name="line.273"></a>
+<span class="sourceLineNo">274</span>    try {<a name="line.274"></a>
+<span class="sourceLineNo">275</span>      return Pair.newPair(ZKUtil.parseWALPositionFrom(data), stat.getVersion());<a name="line.275"></a>
+<span class="sourceLineNo">276</span>    } catch (DeserializationException de) {<a name="line.276"></a>
+<span class="sourceLineNo">277</span>      LOG.warn("Failed to parse log position (region=" + encodedRegionName + ", peerId=" + peerId<a name="line.277"></a>
+<span class="sourceLineNo">278</span>          + "), data=" + Bytes.toStringBinary(data));<a name="line.278"></a>
+<span class="sourceLineNo">279</span>    }<a name="line.279"></a>
+<span class="sourceLineNo">280</span>    return Pair.newPair(HConstants.NO_SEQNUM, stat.getVersion());<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>  @Override<a name="line.283"></a>
+<span class="sourceLineNo">284</span>  public long getLastSequenceId(String encodedRegionName, String peerId)<a name="line.284"></a>
+<span class="sourceLineNo">285</span>      throws ReplicationException {<a name="line.285"></a>
+<span class="sourceLineNo">286</span>    try {<a name="line.286"></a>
+<span class="sourceLineNo">287</span>      return getLastSequenceIdWithVersion(encodedRegionName, peerId).getFirst();<a name="line.287"></a>
+<span class="sourceLineNo">288</span>    } catch (KeeperException e) {<a name="line.288"></a>
+<span class="sourceLineNo">289</span>      throw new ReplicationException("Failed to get last pushed sequence id (encodedRegionName="<a name="line.289"></a>
+<span class="sourceLineNo">290</span>          + encodedRegionName + ", peerId=" + peerId + ")", e);<a name="line.290"></a>
+<span class="sourceLineNo">291</span>    }<a name="line.291"></a>
+<span class="sourceLineNo">292</span>  }<a name="line.292"></a>
+<span class="sourceLineNo">293</span><a name="line.293"></a>
+<span class="sourceLineNo">294</span>  @Override<a name="line.294"></a>
+<span class="sourceLineNo">295</span>  public void setLastSequenceIds(String peerId, Map&lt;String, Long&gt; lastSeqIds)<a name="line.295"></a>
+<span class="sourceLineNo">296</span>      throws ReplicationException {<a name="line.296"></a>
+<span class="sourceLineNo">297</span>    try {<a name="line.297"></a>
+<span class="sourceLineNo">298</span>      // No need CAS and retry here, because it'll call setLastSequenceIds() for disabled peers<a name="line.298"></a>
+<span class="sourceLineNo">299</span>      // only, so no conflict happen.<a name="line.299"></a>
+<span class="sourceLineNo">300</span>      List&lt;ZKUtilOp&gt; listOfOps = new ArrayList&lt;&gt;();<a name="line.300"></a>
+<span class="sourceLineNo">301</span>      for (Entry&lt;String, Long&gt; lastSeqEntry : lastSeqIds.entrySet()) {<a name="line.301"></a>
+<span class="sourceLineNo">302</span>        String path = getSerialReplicationRegionPeerNode(lastSeqEntry.getKey(), peerId);<a name="line.302"></a>
+<span class="sourceLineNo">303</span>        ZKUtil.createWithParents(zookeeper, path);<a name="line.303"></a>
+<span class="sourceLineNo">304</span>        listOfOps.add(ZKUtilOp.setData(path, ZKUtil.positionToByteArray(lastSeqEntry.getValue())));<a name="line.304"></a>
+<span class="sourceLineNo">305</span>      }<a name="line.305"></a>
+<span class="sourceLineNo">306</span>      if (!listOfOps.isEmpty()) {<a name="line.306"></a>
+<span class="sourceLineNo">307</span>        ZKUtil.multiOrSequential(zookeeper, listOfOps, true);<a name="line.307"></a>
+<span class="sourceLineNo">308</span>      }<a name="line.308"></a>
+<span class="sourceLineNo">309</span>    } catch (KeeperException e) {<a name="line.309"></a>
+<span class="sourceLineNo">310</span>      throw new ReplicationException("Failed to set last sequence ids, peerId=" + peerId<a name="line.310"></a>
+<span class="sourceLineNo">311</span>          + ", size of lastSeqIds=" + lastSeqIds.size(), e);<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>  @Override<a name="line.315"></a>
+<span class="sourceLineNo">316</span>  public long getWALPosition(ServerName serverName, String queueId, String fileName)<a name="line.316"></a>
+<span class="sourceLineNo">317</span>      throws ReplicationException {<a name="line.317"></a>
+<span class="sourceLineNo">318</span>    byte[] bytes;<a name="line.318"></a>
+<span class="sourceLineNo">319</span>    try {<a name="line.319"></a>
+<span class="sourceLineNo">320</span>      bytes = ZKUtil.getData(zookeeper, getFileNode(serverName, queueId, fileName));<a name="line.320"></a>
+<span class="sourceLineNo">321</span>    } catch (KeeperException | InterruptedException e) {<a name="line.321"></a>
+<span class="sourceLineNo">322</span>      throw new ReplicationException("Failed to get log position (serverName=" + serverName +<a name="line.322"></a>
+<span class="sourceLineNo">323</span>        ", queueId=" + queueId + ", fileName=" + fileName + ")", e);<a name="line.323"></a>
+<span class="sourceLineNo">324</span>    }<a name="line.324"></a>
+<span class="sourceLineNo">325</span>    try {<a name="line.325"></a>
+<span class="sourceLineNo">326</span>      return ZKUtil.parseWALPositionFrom(bytes);<a name="line.326"></a>
+<span class="sourceLineNo">327</span>    } catch (DeserializationException de) {<a name="line.327"></a>
+<span class="sourceLineNo">328</span>      LOG.warn("Failed parse log position (serverName={}, queueId={}, fileName={})",<a name="line.328"></a>
+<span class="sourceLineNo">329</span>          serverName, queueId, fileName);<a name="line.329"></a>
+<span class="sourceLineNo">330</span>    }<a name="line.330"></a>
+<span class="sourceLineNo">331</span>    // if we can not parse the position, start at the beginning of the wal file again<a name="line.331"></a>
+<span class="sourceLineNo">332</span>    return 0;<a name="line.332"></a>
+<span class="sourceLineNo">333</span>  }<a name="line.333"></a>
+<span class="sourceLineNo">334</span><a name="line.334"></a>
+<span class="sourceLineNo">335</span>  @Override<a name="line.335"></a>
+<span class="sourceLineNo">336</span>  public Pair&lt;String, SortedSet&lt;String&gt;&gt; claimQueue(ServerName sourceServerName, String queueId,<a name="line.336"></a>
+<span class="sourceLineNo">337</span>      ServerName destServerName) throws ReplicationException {<a name="line.337"></a>
+<span class="sourceLineNo">338</span>    LOG.info("Atomically moving {}/{}'s WALs to {}", sourceServerName, queueId, destServerName);<a name="line.338"></a>
+<span class="sourceLineNo">339</span>    try {<a name="line.339"></a>
+<span class="sourceLineNo">340</span>      ZKUtil.createWithParents(zookeeper, getRsNode(destServerName));<a name="line.340"></a>
+<span class="sourceLineNo">341</span>    } catch (KeeperException e) {<a name="line.341"></a>
+<span class="sourceLineNo">342</span>      throw new ReplicationException(<a name="line.342"></a>
+<span class="sourceLineNo">343</span>          "Claim queue queueId=" + queueId + " from " + sourceServerName + " to " + destServerName +<a name="line.343"></a>
+<span class="sourceLineNo">344</span>            " failed when creating the node for " + destServerName,<a name="line.344"></a>
+<span class="sourceLineNo">345</span>          e);<a name="line.345"></a>
+<span class="sourceLineNo">346</span>    }<a name="line.346"></a>
+<span class="sourceLineNo">347</span>    try {<a name="line.347"></a>
+<span class="sourceLineNo">348</span>      String oldQueueNode = getQueueNode(sourceServerName, queueId);<a name="line.348"></a>
+<span class="sourceLineNo">349</span>      List&lt;String&gt; wals = ZKUtil.listChildrenNoWatch(zookeeper, oldQueueNode);<a name="line.349"></a>
+<span class="sourceLineNo">350</span>      String newQueueId = queueId + "-" + sourceServerName;<a name="line.350"></a>
+<span class="sourceLineNo">351</span>      if (CollectionUtils.isEmpty(wals)) {<a name="line.351"></a>
+<span class="sourceLineNo">352</span>        ZKUtil.deleteNodeFailSilent(zookeeper, oldQueueNode);<a name="line.352"></a>
+<span class="sourceLineNo">353</span>        LOG.info("Removed empty {}/{}", sourceServerName, queueId);<a name="line.353"></a>
+<span class="sourceLineNo">354</span>        return new Pair&lt;&gt;(newQueueId, Collections.emptySortedSet());<a name="line.354"></a>
+<span class="sourceLineNo">355</span>      }<a name="line.355"></a>
+<span class="sourceLineNo">356</span>      String newQueueNode = getQueueNode(destServerName, newQueueId);<a name="line.356"></a>
+<span class="sourceLineNo">357</span>      List&lt;ZKUtilOp&gt; listOfOps = new ArrayList&lt;&gt;();<a name="line.357"></a>
+<span class="sourceLineNo">358</span>      SortedSet&lt;String&gt; logQueue = new TreeSet&lt;&gt;();<a name="line.358"></a>
+<span class="sourceLineNo">359</span>      // create the new cluster znode<a name="line.359"></a>
+<span class="sourceLineNo">360</span>      listOfOps.add(ZKUtilOp.createAndFailSilent(newQueueNode, HConstants.EMPTY_BYTE_ARRAY));<a name="line.360"></a>
+<span class="sourceLineNo">361</span>      // get the offset of the logs and set it to new znodes<a name="line.361"></a>
+<span class="sourceLineNo">362</span>      for (String wal : wals) {<a name="line.362"></a>
+<span class="sourceLineNo">363</span>        String oldWalNode = getFileNode(oldQueueNode, wal);<a name="line.363"></a>
+<span class="sourceLineNo">364</span>        byte[] logOffset = ZKUtil.getData(this.zookeeper, oldWalNode);<a name="line.364"></a>
+<span class="sourceLineNo">365</span>        LOG.debug("Creating {} with data {}", wal, Bytes.toStringBinary(logOffset));<a name="line.365"></a>
+<span class="sourceLineNo">366</span>        String newWalNode = getFileNode(newQueueNode, wal);<a name="line.366"></a>
+<span class="sourceLineNo">367</span>        listOfOps.add(ZKUtilOp.createAndFailSilent(newWalNode, logOffset));<a name="line.367"></a>
+<span class="sourceLineNo">368</span>        listOfOps.add(ZKUtilOp.deleteNodeFailSilent(oldWalNode));<a name="line.368"></a>
+<span class="sourceLineNo">369</span>        logQueue.add(wal);<a name="line.369"></a>
+<span class="sourceLineNo">370</span>      }<a name="line.370"></a>
+<span class="sourceLineNo">371</span>      // add delete op for peer<a name="line.371"></a>
+<span class="sourceLineNo">372</span>      listOfOps.add(ZKUtilOp.deleteNodeFailSilent(oldQueueNode));<a name="line.372"></a>
+<span class="sourceLineNo">373</span><a name="line.373"></a>
+<span class="sourceLineNo">374</span>      LOG.trace("The multi list size is {}", listOfOps.size());<a name="line.374"></a>
+<span class="sourceLineNo">375</span>      ZKUtil.multiOrSequential(zookeeper, listOfOps, false);<a name="line.375"></a>
 <span class="sourceLineNo">376</span><a name="line.376"></a>
-<span class="sourceLineNo">377</span>  private List&lt;String&gt; getWALsInQueue0(ServerName serverName, String queueId)<a name="line.377"></a>
-<span class="sourceLineNo">378</span>      throws KeeperException {<a name="line.378"></a>
-<span class="sourceLineNo">379</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, getQueueNode(serverName, queueId)));<a name="line.379"></a>
-<span class="sourceLineNo">380</span>  }<a name="line.380"></a>
-<span class="sourceLineNo">381</span><a name="line.381"></a>
-<span class="sourceLineNo">382</span>  @Override<a name="line.382"></a>
-<span class="sourceLineNo">383</span>  public List&lt;String&gt; getWALsInQueue(ServerName serverName, String queueId)<a name="line.383"></a>
-<span class="sourceLineNo">384</span>      throws ReplicationException {<a name="line.384"></a>
-<span class="sourceLineNo">385</span>    try {<a name="line.385"></a>
-<span class="sourceLineNo">386</span>      return getWALsInQueue0(serverName, queueId);<a name="line.386"></a>
-<span class="sourceLineNo">387</span>    } catch (KeeperException e) {<a name="line.387"></a>
-<span class="sourceLineNo">388</span>      throw new ReplicationException(<a name="line.388"></a>
-<span class="sourceLineNo">389</span>          "Failed to get wals in queue (serverName=" + serverName + ", queueId=" + queueId + ")",<a name="line.389"></a>
-<span class="sourceLineNo">390</span>          e);<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>  private List&lt;String&gt; getAllQueues0(ServerName serverName) throws KeeperException {<a name="line.394"></a>
-<span class="sourceLineNo">395</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, getRsNode(serverName)));<a name="line.395"></a>
-<span class="sourceLineNo">396</span>  }<a name="line.396"></a>
-<span class="sourceLineNo">397</span><a name="line.397"></a>
-<span class="sourceLineNo">398</span>  @Override<a name="line.398"></a>
-<span class="sourceLineNo">399</span>  public List&lt;String&gt; getAllQueues(ServerName serverName) throws ReplicationException {<a name="line.399"></a>
-<span class="sourceLineNo">400</span>    try {<a name="line.400"></a>
-<span class="sourceLineNo">401</span>      return getAllQueues0(serverName);<a name="line.401"></a>
-<span class="sourceLineNo">402</span>    } catch (KeeperException e) {<a name="line.402"></a>
-<span class="sourceLineNo">403</span>      throw new ReplicationException("Failed to get all queues (serverName=" + serverName + ")", e);<a name="line.403"></a>
-<span class="sourceLineNo">404</span>    }<a name="line.404"></a>
-<span class="sourceLineNo">405</span>  }<a name="line.405"></a>
-<span class="sourceLineNo">406</span><a name="line.406"></a>
-<span class="sourceLineNo">407</span>  // will be overridden in UTs<a name="line.407"></a>
-<span class="sourceLineNo">408</span>  @VisibleForTesting<a name="line.408"></a>
-<span class="sourceLineNo">409</span>  protected int getQueuesZNodeCversion() throws KeeperException {<a name="line.409"></a>
-<span class="sourceLineNo">410</span>    Stat stat = new Stat();<a name="line.410"></a>
-<span class="sourceLineNo">411</span>    ZKUtil.getDataNoWatch(this.zookeeper, this.queuesZNode, stat);<a name="line.411"></a>
-<span class="sourceLineNo">412</span>    return stat.getCversion();<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>  @Override<a name="line.415"></a>
-<span class="sourceLineNo">416</span>  public Set&lt;String&gt; getAllWALs() throws ReplicationException {<a name="line.416"></a>
-<span class="sourceLineNo">417</span>    try {<a name="line.417"></a>
-<span class="sourceLineNo">418</span>      for (int retry = 0;; retry++) {<a name="line.418"></a>
-<span class="sourceLineNo">419</span>        int v0 = getQueuesZNodeCversion();<a name="line.419"></a>
-<span class="sourceLineNo">420</span>        List&lt;ServerName&gt; rss = getListOfReplicators0();<a name="line.420"></a>
-<span class="sourceLineNo">421</span>        if (rss.isEmpty()) {<a name="line.421"></a>
-<span class="sourceLineNo">422</span>          LOG.debug("Didn't find a RegionServer that replicates, won't prevent deletions.");<a name="line.422"></a>
-<span class="sourceLineNo">423</span>          return Collections.emptySet();<a name="line.423"></a>
-<span class="sourceLineNo">424</span>        }<a name="line.424"></a>
-<span class="sourceLineNo">425</span>        Set&lt;String&gt; wals = new HashSet&lt;&gt;();<a name="line.425"></a>
-<span class="sourceLineNo">426</span>        for (ServerName rs : rss) {<a name="line.426"></a>
-<span class="sourceLineNo">427</span>          for (String queueId : getAllQueues0(rs)) {<a name="line.427"></a>
-<span class="sourceLineNo">428</span>            wals.addAll(getWALsInQueue0(rs, queueId));<a name="line.428"></a>
-<span class="sourceLineNo">429</span>          }<a name="line.429"></a>
-<span class="sourceLineNo">430</span>        }<a name="line.430"></a>
-<span class="sourceLineNo">431</span>        int v1 = getQueuesZNodeCversion();<a name="line.431"></a>
-<span class="sourceLineNo">432</span>        if (v0 == v1) {<a name="line.432"></a>
-<span class="sourceLineNo">433</span>          return wals;<a name="line.433"></a>
-<span class="sourceLineNo">434</span>        }<a name="line.434"></a>
-<span class="sourceLineNo">435</span>        LOG.info("Replication queue node cversion changed from %d to %d, retry = %d",<a name="line.435"></a>
-<span class="sourceLineNo">436</span>            v0, v1, retry);<a name="line.436"></a>
-<span class="sourceLineNo">437</span>      }<a name="line.437"></a>
-<span class="sourceLineNo">438</span>    } catch (KeeperException e) {<a name="line.438"></a>
-<span class="sourceLineNo">439</span>      throw new ReplicationException("Failed to get all wals", e);<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>  private String getHFileRefsPeerNode(String peerId) {<a name="line.443"></a>
-<span class="sourceLineNo">444</span>    return ZNodePaths.joinZNode(hfileRefsZNode, peerId);<a name="line.444"></a>
+<span class="sourceLineNo">377</span>      LOG.info("Atomically moved {}/{}'s WALs to {}", sourceServerName, queueId, destServerName);<a name="line.377"></a>
+<span class="sourceLineNo">378</span>      return new Pair&lt;&gt;(newQueueId, logQueue);<a name="line.378"></a>
+<span class="sourceLineNo">379</span>    } catch (NoNodeException | NodeExistsException | NotEmptyException | BadVersionException e) {<a name="line.379"></a>
+<span class="sourceLineNo">380</span>      // Multi call failed; it looks like some other regionserver took away the logs.<a name="line.380"></a>
+<span class="sourceLineNo">381</span>      // These exceptions mean that zk tells us the request can not be execute so it is safe to just<a name="line.381"></a>
+<span class="sourceLineNo">382</span>      // return a null. For other types of exception should be thrown out to notify the upper layer.<a name="line.382"></a>
+<span class="sourceLineNo">383</span>      LOG.info("Claim queue queueId={} from {} to {} failed with {}, someone else took the log?",<a name="line.383"></a>
+<span class="sourceLineNo">384</span>          queueId,sourceServerName, destServerName, e.toString());<a name="line.384"></a>
+<span class="sourceLineNo">385</span>      return null;<a name="line.385"></a>
+<span class="sourceLineNo">386</span>    } catch (KeeperException | InterruptedException e) {<a name="line.386"></a>
+<span class="sourceLineNo">387</span>      throw new ReplicationException("Claim queue queueId=" + queueId + " from " +<a name="line.387"></a>
+<span class="sourceLineNo">388</span>        sourceServerName + " to " + destServerName + " failed", e);<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><a name="line.391"></a>
+<span class="sourceLineNo">392</span>  @Override<a name="line.392"></a>
+<span class="sourceLineNo">393</span>  public void removeReplicatorIfQueueIsEmpty(ServerName serverName) throws ReplicationException {<a name="line.393"></a>
+<span class="sourceLineNo">394</span>    try {<a name="line.394"></a>
+<span class="sourceLineNo">395</span>      ZKUtil.deleteNodeFailSilent(zookeeper, getRsNode(serverName));<a name="line.395"></a>
+<span class="sourceLineNo">396</span>    } catch (NotEmptyException e) {<a name="line.396"></a>
+<span class="sourceLineNo">397</span>      // keep silence to avoid logging too much.<a name="line.397"></a>
+<span class="sourceLineNo">398</span>    } catch (KeeperException e) {<a name="line.398"></a>
+<span class="sourceLineNo">399</span>      throw new ReplicationException("Failed to remove replicator for " + serverName, e);<a name="line.399"></a>
+<span class="sourceLineNo">400</span>    }<a name="line.400"></a>
+<span class="sourceLineNo">401</span>  }<a name="line.401"></a>
+<span class="sourceLineNo">402</span><a name="line.402"></a>
+<span class="sourceLineNo">403</span>  private List&lt;ServerName&gt; getListOfReplicators0() throws KeeperException {<a name="line.403"></a>
+<span class="sourceLineNo">404</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, queuesZNode)).stream()<a name="line.404"></a>
+<span class="sourceLineNo">405</span>        .map(ServerName::parseServerName).collect(toList());<a name="line.405"></a>
+<span class="sourceLineNo">406</span>  }<a name="line.406"></a>
+<span class="sourceLineNo">407</span><a name="line.407"></a>
+<span class="sourceLineNo">408</span>  @Override<a name="line.408"></a>
+<span class="sourceLineNo">409</span>  public List&lt;ServerName&gt; getListOfReplicators() throws ReplicationException {<a name="line.409"></a>
+<span class="sourceLineNo">410</span>    try {<a name="line.410"></a>
+<span class="sourceLineNo">411</span>      return getListOfReplicators0();<a name="line.411"></a>
+<span class="sourceLineNo">412</span>    } catch (KeeperException e) {<a name="line.412"></a>
+<span class="sourceLineNo">413</span>      throw new ReplicationException("Failed to get list of replicators", e);<a name="line.413"></a>
+<span class="sourceLineNo">414</span>    }<a name="line.414"></a>
+<span class="sourceLineNo">415</span>  }<a name="line.415"></a>
+<span class="sourceLineNo">416</span><a name="line.416"></a>
+<span class="sourceLineNo">417</span>  private List&lt;String&gt; getWALsInQueue0(ServerName serverName, String queueId)<a name="line.417"></a>
+<span class="sourceLineNo">418</span>      throws KeeperException {<a name="line.418"></a>
+<span class="sourceLineNo">419</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, getQueueNode(serverName, queueId)));<a name="line.419"></a>
+<span class="sourceLineNo">420</span>  }<a name="line.420"></a>
+<span class="sourceLineNo">421</span><a name="line.421"></a>
+<span class="sourceLineNo">422</span>  @Override<a name="line.422"></a>
+<span class="sourceLineNo">423</span>  public List&lt;String&gt; getWALsInQueue(ServerName serverName, String queueId)<a name="line.423"></a>
+<span class="sourceLineNo">424</span>      throws ReplicationException {<a name="line.424"></a>
+<span class="sourceLineNo">425</span>    try {<a name="line.425"></a>
+<span class="sourceLineNo">426</span>      return getWALsInQueue0(serverName, queueId);<a name="line.426"></a>
+<span class="sourceLineNo">427</span>    } catch (KeeperException e) {<a name="line.427"></a>
+<span class="sourceLineNo">428</span>      throw new ReplicationException(<a name="line.428"></a>
+<span class="sourceLineNo">429</span>          "Failed to get wals in queue (serverName=" + serverName + ", queueId=" + queueId + ")",<a name="line.429"></a>
+<span class="sourceLineNo">430</span>          e);<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><a name="line.433"></a>
+<span class="sourceLineNo">434</span>  private List&lt;String&gt; getAllQueues0(ServerName serverName) throws KeeperException {<a name="line.434"></a>
+<span class="sourceLineNo">435</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, getRsNode(serverName)));<a name="line.435"></a>
+<span class="sourceLineNo">436</span>  }<a name="line.436"></a>
+<span class="sourceLineNo">437</span><a name="line.437"></a>
+<span class="sourceLineNo">438</span>  @Override<a name="line.438"></a>
+<span class="sourceLineNo">439</span>  public List&lt;String&gt; getAllQueues(ServerName serverName) throws ReplicationException {<a name="line.439"></a>
+<span class="sourceLineNo">440</span>    try {<a name="line.440"></a>
+<span class="sourceLineNo">441</span>      return getAllQueues0(serverName);<a name="line.441"></a>
+<span class="sourceLineNo">442</span>    } catch (KeeperException e) {<a name="line.442"></a>
+<span class="sourceLineNo">443</span>      throw new ReplicationException("Failed to get all queues (serverName=" + serverName + ")", e);<a name="line.443"></a>
+<span class="sourceLineNo">444</span>    }<a name="line.444"></a>
 <span class="sourceLineNo">445</span>  }<a name="line.445"></a>
 <span class="sourceLineNo">446</span><a name="line.446"></a>
-<span class="sourceLineNo">447</span>  private String getHFileNode(String peerNode, String fileName) {<a name="line.447"></a>
-<span class="sourceLineNo">448</span>    return ZNodePaths.joinZNode(peerNode, fileName);<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>  @Override<a name="line.451"></a>
-<span class="sourceLineNo">452</span>  public void addPeerToHFileRefs(String peerId) throws ReplicationException {<a name="line.452"></a>
-<span class="sourceLineNo">453</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.453"></a>
-<span class="sourceLineNo">454</span>    try {<a name="line.454"></a>
-<span class="sourceLineNo">455</span>      if (ZKUtil.checkExists(zookeeper, peerNode) == -1) {<a name="line.455"></a>
-<span class="sourceLineNo">456</span>        LOG.info("Adding peer {} to hfile reference queue.", peerId);<a name="line.456"></a>
-<span class="sourceLineNo">457</span>        ZKUtil.createWithParents(zookeeper, peerNode);<a name="line.457"></a>
-<span class="sourceLineNo">458</span>      }<a name="line.458"></a>
-<span class="sourceLineNo">459</span>    } catch (KeeperException e) {<a name="line.459"></a>
-<span class="sourceLineNo">460</span>      throw new ReplicationException("Failed to add peer " + peerId + " to hfile reference queue.",<a name="line.460"></a>
-<span class="sourceLineNo">461</span>          e);<a name="line.461"></a>
-<span class="sourceLineNo">462</span>    }<a name="line.462"></a>
-<span class="sourceLineNo">463</span>  }<a name="line.463"></a>
-<span class="sourceLineNo">464</span><a name="line.464"></a>
-<span class="sourceLineNo">465</span>  @Override<a name="line.465"></a>
-<span class="sourceLineNo">466</span>  public void removePeerFromHFileRefs(String peerId) throws ReplicationException {<a name="line.466"></a>
-<span class="sourceLineNo">467</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.467"></a>
-<span class="sourceLineNo">468</span>    try {<a name="line.468"></a>
-<span class="sourceLineNo">469</span>      if (ZKUtil.checkExists(zookeeper, peerNode) == -1) {<a name="line.469"></a>
-<span class="sourceLineNo">470</span>          LOG.debug("Peer {} not found in hfile reference queue.", peerNode);<a name="line.470"></a>
-<span class="sourceLineNo">471</span>      } else {<a name="line.471"></a>
-<span class="sourceLineNo">472</span>        LOG.info("Removing peer {} from hfile reference queue.", peerNode);<a name="line.472"></a>
-<span class="sourceLineNo">473</span>        ZKUtil.deleteNodeRecursively(zookeeper, peerNode);<a name="line.473"></a>
-<span class="sourceLineNo">474</span>      }<a name="line.474"></a>
-<span class="sourceLineNo">475</span>    } catch (KeeperException e) {<a name="line.475"></a>
-<span class="sourceLineNo">476</span>      throw new ReplicationException(<a name="line.476"></a>
-<span class="sourceLineNo">477</span>          "Failed to remove peer " + peerId + " from hfile reference queue.", e);<a name="line.477"></a>
-<span class="sourceLineNo">478</span>    }<a name="line.478"></a>
-<span class="sourceLineNo">479</span>  }<a name="line.479"></a>
-<span class="sourceLineNo">480</span><a name="line.480"></a>
-<span class="sourceLineNo">481</span>  @Override<a name="line.481"></a>
-<span class="sourceLineNo">482</span>  public void addHFileRefs(String peerId, List&lt;Pair&lt;Path, Path&gt;&gt; pairs)<a name="line.482"></a>
-<span class="sourceLineNo">483</span>      throws ReplicationException {<a name="line.483"></a>
-<span class="sourceLineNo">484</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.484"></a>
-<span class="sourceLineNo">485</span>    LOG.debug("Adding hfile references {} in queue {}", pairs, peerNode);<a name="line.485"></a>
-<span class="sourceLineNo">486</span>    List&lt;ZKUtilOp&gt; listOfOps = pairs.stream().map(p -&gt; p.getSecond().getName())<a name="line.486"></a>
-<span class="sourceLineNo">487</span>        .map(n -&gt; getHFileNode(peerNode, n))<a name="line.487"></a>
-<span class="sourceLineNo">488</span>        .map(f -&gt; ZKUtilOp.createAndFailSilent(f, HConstants.EMPTY_BYTE_ARRAY)).collect(toList());<a name="line.488"></a>
-<span class="sourceLineNo">489</span>      LOG.debug("The multi list size for adding hfile references in zk for node {} is {}",<a name="line.489"></a>
-<span class="sourceLineNo">490</span>          peerNode, listOfOps.size());<a name="line.490"></a>
-<span class="sourceLineNo">491</span>    try {<a name="line.491"></a>
-<span class="sourceLineNo">492</span>      ZKUtil.multiOrSequential(this.zookeeper, listOfOps, true);<a name="line.492"></a>
-<span class="sourceLineNo">493</span>    } catch (KeeperException e) {<a name="line.493"></a>
-<span class="sourceLineNo">494</span>      throw new ReplicationException("Failed to add hfile reference to peer " + peerId, e);<a name="line.494"></a>
-<span class="sourceLineNo">495</span>    }<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>  @Override<a name="line.498"></a>
-<span class="sourceLineNo">499</span>  public void removeHFileRefs(String peerId, List&lt;String&gt; files) throws ReplicationException {<a name="line.499"></a>
-<span class="sourceLineNo">500</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.500"></a>
-<span class="sourceLineNo">501</span>    LOG.debug("Removing hfile references {} from queue {}", files, peerNode);<a name="line.501"></a>
-<span class="sourceLineNo">502</span><a name="line.502"></a>
-<span class="sourceLineNo">503</span>    List&lt;ZKUtilOp&gt; listOfOps = files.stream().map(n -&gt; getHFileNode(peerNode, n))<a name="line.503"></a>
-<span class="sourceLineNo">504</span>        .map(ZKUtilOp::deleteNodeFailSilent).collect(toList());<a name="line.504"></a>
-<span class="sourceLineNo">505</span>    LOG.debug("The multi list size for removing hfile references in zk for node {} is {}",<a name="line.505"></a>
-<span class="sourceLineNo">506</span>        peerNode, listOfOps.size());<a name="line.506"></a>
-<span class="sourceLineNo">507</span>    try {<a name="line.507"></a>
-<span class="sourceLineNo">508</span>      ZKUtil.multiOrSequential(this.zookeeper, listOfOps, true);<a name="line.508"></a>
-<span class="sourceLineNo">509</span>    } catch (KeeperException e) {<a name="line.509"></a>
-<span class="sourceLineNo">510</span>      throw new ReplicationException("Failed to remove hfile reference from peer " + peerId, e);<a name="line.510"></a>
-<span class="sourceLineNo">511</span>    }<a name="line.511"></a>
-<span class="sourceLineNo">512</span>  }<a name="line.512"></a>
-<span class="sourceLineNo">513</span><a name="line.513"></a>
-<span class="sourceLineNo">514</span>  private List&lt;String&gt; getAllPeersFromHFileRefsQueue0() throws KeeperException {<a name="line.514"></a>
-<span class="sourceLineNo">515</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, hfileRefsZNode));<a name="line.515"></a>
-<span class="sourceLineNo">516</span>  }<a name="line.516"></a>
-<span class="sourceLineNo">517</span><a name="line.517"></a>
-<span class="sourceLineNo">518</span>  @Override<a name="line.518"></a>
-<span class="sourceLineNo">519</span>  public List&lt;String&gt; getAllPeersFromHFileRefsQueue() throws ReplicationException {<a name="line.519"></a>
-<span class="sourceLineNo">520</span>    try {<a name="line.520"></a>
-<span class="sourceLineNo">521</span>      return getAllPeersFromHFileRefsQueue0();<a name="line.521"></a>
-<span class="sourceLineNo">522</span>    } catch (KeeperException e) {<a name="line.522"></a>
-<span class="sourceLineNo">523</span>      throw new ReplicationException("Failed to get list of all peers in hfile references node.",<a name="line.523"></a>
-<span class="sourceLineNo">524</span>          e);<a name="line.524"></a>
-<span class="sourceLineNo">525</span>    }<a name="line.525"></a>
-<span class="sourceLineNo">526</span>  }<a name="line.526"></a>
-<span class="sourceLineNo">527</span><a name="line.527"></a>
-<span class="sourceLineNo">528</span>  private List&lt;String&gt; getReplicableHFiles0(String peerId) throws KeeperException {<a name="line.528"></a>
-<span class="sourceLineNo">529</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(this.zookeeper, getHFileRefsPeerNode(peerId)));<a name="line.529"></a>
-<span class="sourceLineNo">530</span>  }<a name="line.530"></a>
-<span class="sourceLineNo">531</span><a name="line.531"></a>
-<span class="sourceLineNo">532</span>  @Override<a name="line.532"></a>
-<span class="sourceLineNo">533</span>  public List&lt;String&gt; getReplicableHFiles(String peerId) throws ReplicationException {<a name="line.533"></a>
-<span class="sourceLineNo">534</span>    try {<a name="line.534"></a>
-<span class="sourceLineNo">535</span>      return getReplicableHFiles0(peerId);<a name="line.535"></a>
-<span class="sourceLineNo">536</span>    } catch (KeeperException e) {<a name="line.536"></a>
-<span class="sourceLineNo">537</span>      throw new ReplicationException("Failed to get list of hfile references for peer " + peerId,<a name="line.537"></a>
-<span class="sourceLineNo">538</span>          e);<a name="line.538"></a>
-<span class="sourceLineNo">539</span>    }<a name="line.539"></a>
-<span class="sourceLineNo">540</span>  }<a name="line.540"></a>
-<span class="sourceLineNo">541</span><a name="line.541"></a>
-<span class="sourceLineNo">542</span>  // will be overridden in UTs<a name="line.542"></a>
-<span class="sourceLineNo">543</span>  @VisibleForTesting<a name="line.543"></a>
-<span class="sourceLineNo">544</span>  protected int getHFileRefsZNodeCversion() throws ReplicationException {<a name="line.544"></a>
-<span class="sourceLineNo">545</span>    Stat stat = new Stat();<a name="line.545"></a>
-<span class="sourceLineNo">546</span>    try {<a name="line.546"></a>
-<span class="sourceLineNo">547</span>      ZKUtil.getDataNoWatch(zookeeper, hfileRefsZNode, stat);<a name="line.547"></a>
-<span class="sourceLineNo">548</span>    } catch (KeeperException e) {<a name="line.548"></a>
-<span class="sourceLineNo">549</span>      throw new ReplicationException("Failed to get stat of replication hfile references node.", e);<a name="line.549"></a>
-<span class="sourceLineNo">550</span>    }<a name="line.550"></a>
-<span class="sourceLineNo">551</span>    return stat.getCversion();<a name="line.551"></a>
+<span class="sourceLineNo">447</span>  // will be overridden in UTs<a name="line.447"></a>
+<span class="sourceLineNo">448</span>  @VisibleForTesting<a name="line.448"></a>
+<span class="sourceLineNo">449</span>  protected int getQueuesZNodeCversion() throws KeeperException {<a name="line.449"></a>
+<span class="sourceLineNo">450</span>    Stat stat = new Stat();<a name="line.450"></a>
+<span class="sourceLineNo">451</span>    ZKUtil.getDataNoWatch(this.zookeeper, this.queuesZNode, stat);<a name="line.451"></a>
+<span class="sourceLineNo">452</span>    return stat.getCversion();<a name="line.452"></a>
+<span class="sourceLineNo">453</span>  }<a name="line.453"></a>
+<span class="sourceLineNo">454</span><a name="line.454"></a>
+<span class="sourceLineNo">455</span>  @Override<a name="line.455"></a>
+<span class="sourceLineNo">456</span>  public Set&lt;String&gt; getAllWALs() throws ReplicationException {<a name="line.456"></a>
+<span class="sourceLineNo">457</span>    try {<a name="line.457"></a>
+<span class="sourceLineNo">458</span>      for (int retry = 0;; retry++) {<a name="line.458"></a>
+<span class="sourceLineNo">459</span>        int v0 = getQueuesZNodeCversion();<a name="line.459"></a>
+<span class="sourceLineNo">460</span>        List&lt;ServerName&gt; rss = getListOfReplicators0();<a name="line.460"></a>
+<span class="sourceLineNo">461</span>        if (rss.isEmpty()) {<a name="line.461"></a>
+<span class="sourceLineNo">462</span>          LOG.debug("Didn't find a RegionServer that replicates, won't prevent deletions.");<a name="line.462"></a>
+<span class="sourceLineNo">463</span>          return Collections.emptySet();<a name="line.463"></a>
+<span class="sourceLineNo">464</span>        }<a name="line.464"></a>
+<span class="sourceLineNo">465</span>        Set&lt;String&gt; wals = new HashSet&lt;&gt;();<a name="line.465"></a>
+<span class="sourceLineNo">466</span>        for (ServerName rs : rss) {<a name="line.466"></a>
+<span class="sourceLineNo">467</span>          for (String queueId : getAllQueues0(rs)) {<a name="line.467"></a>
+<span class="sourceLineNo">468</span>            wals.addAll(getWALsInQueue0(rs, queueId));<a name="line.468"></a>
+<span class="sourceLineNo">469</span>          }<a name="line.469"></a>
+<span class="sourceLineNo">470</span>        }<a name="line.470"></a>
+<span class="sourceLineNo">471</span>        int v1 = getQueuesZNodeCversion();<a name="line.471"></a>
+<span class="sourceLineNo">472</span>        if (v0 == v1) {<a name="line.472"></a>
+<span class="sourceLineNo">473</span>          return wals;<a name="line.473"></a>
+<span class="sourceLineNo">474</span>        }<a name="line.474"></a>
+<span class="sourceLineNo">475</span>        LOG.info("Replication queue node cversion changed from %d to %d, retry = %d",<a name="line.475"></a>
+<span class="sourceLineNo">476</span>            v0, v1, retry);<a name="line.476"></a>
+<span class="sourceLineNo">477</span>      }<a name="line.477"></a>
+<span class="sourceLineNo">478</span>    } catch (KeeperException e) {<a name="line.478"></a>
+<span class="sourceLineNo">479</span>      throw new ReplicationException("Failed to get all wals", e);<a name="line.479"></a>
+<span class="sourceLineNo">480</span>    }<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>  private String getHFileRefsPeerNode(String peerId) {<a name="line.483"></a>
+<span class="sourceLineNo">484</span>    return ZNodePaths.joinZNode(hfileRefsZNode, peerId);<a name="line.484"></a>
+<span class="sourceLineNo">485</span>  }<a name="line.485"></a>
+<span class="sourceLineNo">486</span><a name="line.486"></a>
+<span class="sourceLineNo">487</span>  private String getHFileNode(String peerNode, String fileName) {<a name="line.487"></a>
+<span class="sourceLineNo">488</span>    return ZNodePaths.joinZNode(peerNode, fileName);<a name="line.488"></a>
+<span class="sourceLineNo">489</span>  }<a name="line.489"></a>
+<span class="sourceLineNo">490</span><a name="line.490"></a>
+<span class="sourceLineNo">491</span>  @Override<a name="line.491"></a>
+<span class="sourceLineNo">492</span>  public void addPeerToHFileRefs(String peerId) throws ReplicationException {<a name="line.492"></a>
+<span class="sourceLineNo">493</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.493"></a>
+<span class="sourceLineNo">494</span>    try {<a name="line.494"></a>
+<span class="sourceLineNo">495</span>      if (ZKUtil.checkExists(zookeeper, peerNode) == -1) {<a name="line.495"></a>
+<span class="sourceLineNo">496</span>        LOG.info("Adding peer {} to hfile reference queue.", peerId);<a name="line.496"></a>
+<span class="sourceLineNo">497</span>        ZKUtil.createWithParents(zookeeper, peerNode);<a name="line.497"></a>
+<span class="sourceLineNo">498</span>      }<a name="line.498"></a>
+<span class="sourceLineNo">499</span>    } catch (KeeperException e) {<a name="line.499"></a>
+<span class="sourceLineNo">500</span>      throw new ReplicationException("Failed to add peer " + peerId + " to hfile reference queue.",<a name="line.500"></a>
+<span class="sourceLineNo">501</span>          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><a name="line.504"></a>
+<span class="sourceLineNo">505</span>  @Override<a name="line.505"></a>
+<span class="sourceLineNo">506</span>  public void removePeerFromHFileRefs(String peerId) throws ReplicationException {<a name="line.506"></a>
+<span class="sourceLineNo">507</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.507"></a>
+<span class="sourceLineNo">508</span>    try {<a name="line.508"></a>
+<span class="sourceLineNo">509</span>      if (ZKUtil.checkExists(zookeeper, peerNode) == -1) {<a name="line.509"></a>
+<span class="sourceLineNo">510</span>          LOG.debug("Peer {} not found in hfile reference queue.", peerNode);<a name="line.510"></a>
+<span class="sourceLineNo">511</span>      } else {<a name="line.511"></a>
+<span class="sourceLineNo">512</span>        LOG.info("Removing peer {} from hfile reference queue.", peerNode);<a name="line.512"></a>
+<span class="sourceLineNo">513</span>        ZKUtil.deleteNodeRecursively(zookeeper, peerNode);<a name="line.513"></a>
+<span class="sourceLineNo">514</span>      }<a name="line.514"></a>
+<span class="sourceLineNo">515</span>    } catch (KeeperException e) {<a name="line.515"></a>
+<span class="sourceLineNo">516</span>      throw new ReplicationException(<a name="line.516"></a>
+<span class="sourceLineNo">517</span>          "Failed to remove peer " + peerId + " from hfile reference queue.", e);<a name="line.517"></a>
+<span class="sourceLineNo">518</span>    }<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>  @Override<a name="line.521"></a>
+<span class="sourceLineNo">522</span>  public void addHFileRefs(String peerId, List&lt;Pair&lt;Path, Path&gt;&gt; pairs)<a name="line.522"></a>
+<span class="sourceLineNo">523</span>      throws ReplicationException {<a name="line.523"></a>
+<span class="sourceLineNo">524</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.524"></a>
+<span class="sourceLineNo">525</span>    LOG.debug("Adding hfile references {} in queue {}", pairs, peerNode);<a name="line.525"></a>
+<span class="sourceLineNo">526</span>    List&lt;ZKUtilOp&gt; listOfOps = pairs.stream().map(p -&gt; p.getSecond().getName())<a name="line.526"></a>
+<span class="sourceLineNo">527</span>        .map(n -&gt; getHFileNode(peerNode, n))<a name="line.527"></a>
+<span class="sourceLineNo">528</span>        .map(f -&gt; ZKUtilOp.createAndFailSilent(f, HConstants.EMPTY_BYTE_ARRAY)).collect(toList());<a name="line.528"></a>
+<span class="sourceLineNo">529</span>      LOG.debug("The multi list size for adding hfile references in zk for node {} is {}",<a name="line.529"></a>
+<span class="sourceLineNo">530</span>          peerNode, listOfOps.size());<a name="line.530"></a>
+<span class="sourceLineNo">531</span>    try {<a name="line.531"></a>
+<span class="sourceLineNo">532</span>      ZKUtil.multiOrSequential(this.zookeeper, listOfOps, true);<a name="line.532"></a>
+<span class="sourceLineNo">533</span>    } catch (KeeperException e) {<a name="line.533"></a>
+<span class="sourceLineNo">534</span>      throw new ReplicationException("Failed to add hfile reference to peer " + peerId, e);<a name="line.534"></a>
+<span class="sourceLineNo">535</span>    }<a name="line.535"></a>
+<span class="sourceLineNo">536</span>  }<a name="line.536"></a>
+<span class="sourceLineNo">537</span><a name="line.537"></a>
+<span class="sourceLineNo">538</span>  @Override<a name="line.538"></a>
+<span class="sourceLineNo">539</span>  public void removeHFileRefs(String peerId, List&lt;String&gt; files) throws ReplicationException {<a name="line.539"></a>
+<span class="sourceLineNo">540</span>    String peerNode = getHFileRefsPeerNode(peerId);<a name="line.540"></a>
+<span class="sourceLineNo">541</span>    LOG.debug("Removing hfile references {} from queue {}", files, peerNode);<a name="line.541"></a>
+<span class="sourceLineNo">542</span><a name="line.542"></a>
+<span class="sourceLineNo">543</span>    List&lt;ZKUtilOp&gt; listOfOps = files.stream().map(n -&gt; getHFileNode(peerNode, n))<a name="line.543"></a>
+<span class="sourceLineNo">544</span>        .map(ZKUtilOp::deleteNodeFailSilent).collect(toList());<a name="line.544"></a>
+<span class="sourceLineNo">545</span>    LOG.debug("The multi list size for removing hfile references in zk for node {} is {}",<a name="line.545"></a>
+<span class="sourceLineNo">546</span>        peerNode, listOfOps.size());<a name="line.546"></a>
+<span class="sourceLineNo">547</span>    try {<a name="line.547"></a>
+<span class="sourceLineNo">548</span>      ZKUtil.multiOrSequential(this.zookeeper, listOfOps, true);<a name="line.548"></a>
+<span class="sourceLineNo">549</span>    } catch (KeeperException e) {<a name="line.549"></a>
+<span class="sourceLineNo">550</span>      throw new ReplicationException("Failed to remove hfile reference from peer " + peerId, e);<a name="line.550"></a>
+<span class="sourceLineNo">551</span>    }<a name="line.551"></a>
 <span class="sourceLineNo">552</span>  }<a name="line.552"></a>
 <span class="sourceLineNo">553</span><a name="line.553"></a>
-<span class="sourceLineNo">554</span>  @Override<a name="line.554"></a>
-<span class="sourceLineNo">555</span>  public Set&lt;String&gt; getAllHFileRefs() throws ReplicationException {<a name="line.555"></a>
-<span class="sourceLineNo">556</span>    try {<a name="line.556"></a>
-<span class="sourceLineNo">557</span>      for (int retry = 0;; retry++) {<a name="line.557"></a>
-<span class="sourceLineNo">558</span>        int v0 = getHFileRefsZNodeCversion();<a name="line.558"></a>
-<span class="sourceLineNo">559</span>        List&lt;String&gt; peers = getAllPeersFromHFileRefsQueue();<a name="line.559"></a>
-<span class="sourceLineNo">560</span>        if (peers.isEmpty()) {<a name="line.560"></a>
-<span class="sourceLineNo">561</span>          LOG.debug("Didn't find any peers with hfile references, won't prevent deletions.");<a name="line.561"></a>
-<span class="sourceLineNo">562</span>          return Collections.emptySet();<a name="line.562"></a>
-<span class="sourceLineNo">563</span>        }<a name="line.563"></a>
-<span class="sourceLineNo">564</span>        Set&lt;String&gt; hfileRefs = new HashSet&lt;&gt;();<a name="line.564"></a>
-<span class="sourceLineNo">565</span>        for (String peer : peers) {<a name="line.565"></a>
-<span class="sourceLineNo">566</span>          hfileRefs.addAll(getReplicableHFiles0(peer));<a name="line.566"></a>
-<span class="sourceLineNo">567</span>        }<a name="line.567"></a>
-<span class="sourceLineNo">568</span>        int v1 = getHFileRefsZNodeCversion();<a name="line.568"></a>
-<span class="sourceLineNo">569</span>        if (v0 == v1) {<a name="line.569"></a>
-<span class="sourceLineNo">570</span>          return hfileRefs;<a name="line.570"></a>
-<span class="sourceLineNo">571</span>        }<a name="line.571"></a>
-<span class="sourceLineNo">572</span>        LOG.debug("Replication hfile references node cversion changed from %d to %d, retry = %d",<a name="line.572"></a>
-<span class="sourceLineNo">573</span>            v0, v1, retry);<a name="line.573"></a>
-<span class="sourceLineNo">574</span>      }<a name="line.574"></a>
-<span class="sourceLineNo">575</span>    } catch (KeeperException e) {<a name="line.575"></a>
-<span class="sourceLineNo">576</span>      throw new ReplicationException("Failed to get all hfile refs", e);<a name="line.576"></a>
-<span class="sourceLineNo">577</span>    }<a name="line.577"></a>
-<span class="sourceLineNo">578</span>  }<a name="line.578"></a>
-<span class="sourceLineNo">579</span>}<a name="line.579"></a>
+<span class="sourceLineNo">554</span>  private List&lt;String&gt; getAllPeersFromHFileRefsQueue0() throws KeeperException {<a name="line.554"></a>
+<span class="sourceLineNo">555</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(zookeeper, hfileRefsZNode));<a name="line.555"></a>
+<span class="sourceLineNo">556</span>  }<a name="line.556"></a>
+<span class="sourceLineNo">557</span><a name="line.557"></a>
+<span class="sourceLineNo">558</span>  @Override<a name="line.558"></a>
+<span class="sourceLineNo">559</span>  public List&lt;String&gt; getAllPeersFromHFileRefsQueue() throws ReplicationException {<a name="line.559"></a>
+<span class="sourceLineNo">560</span>    try {<a name="line.560"></a>
+<span class="sourceLineNo">561</span>      return getAllPeersFromHFileRefsQueue0();<a name="line.561"></a>
+<span class="sourceLineNo">562</span>    } catch (KeeperException e) {<a name="line.562"></a>
+<span class="sourceLineNo">563</span>      throw new ReplicationException("Failed to get list of all peers in hfile references node.",<a name="line.563"></a>
+<span class="sourceLineNo">564</span>          e);<a name="line.564"></a>
+<span class="sourceLineNo">565</span>    }<a name="line.565"></a>
+<span class="sourceLineNo">566</span>  }<a name="line.566"></a>
+<span class="sourceLineNo">567</span><a name="line.567"></a>
+<span class="sourceLineNo">568</span>  private List&lt;String&gt; getReplicableHFiles0(String peerId) throws KeeperException {<a name="line.568"></a>
+<span class="sourceLineNo">569</span>    return nullToEmpty(ZKUtil.listChildrenNoWatch(this.zookeeper, getHFileRefsPeerNode(peerId)));<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>  @Override<a name="line.572"></a>
+<span class="sourceLineNo">573</span>  public List&lt;String&gt; getReplicableHFiles(String peerId) throws ReplicationException {<a name="line.573"></a>
+<span class="sourceLineNo">574</span>    try {<a name="line.574"></a>
+<span class="sourceLineNo">575</span>      return getReplicableHFiles0(peerId);<a name="line.575"></a>
+<span class="sourceLineNo">576</span>    } catch (KeeperException e) {<a name="line.576"></a>
+<span class="sourceLineNo">577</span>      throw new ReplicationException("Failed to get list of hfile references for peer " + peerId,<a name="line.577"></a>
+<span class="sourceLineNo">578</span>          e);<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><a name="line.581"></a>
+<span class="sourceLineNo">582</span>  // will be overridden in UTs<a name="line.582"></a>
+<span class="sourceLineNo">583</span>  @VisibleForTesting<a name="line.583"></a>
+<span class="sourceLineNo">584</span>  protected int getHFileRefsZNodeCversion() throws ReplicationException {<a name="line.584"></a>
+<span class="sourceLineNo">585</span>    Stat stat = new Stat();<a name="line.585"></a>
+<span class="sourceLineNo">586</span>    try {<a name="line.586"></a>
+<span class="sourceLineNo">587</span>      ZKUtil.getDataNoWatch(zookeeper, hfileRefsZNode, stat);<a name="line.587"></a>
+<span class="sourceLineNo">588</span>    } catch (KeeperException e) {<a name="line.588"></a>
+<span class="sourceLineNo">589</span>      throw new ReplicationException("Failed to get stat of replication hfile references node.", e);<a name="line.589"></a>
+<span class="sourceLineNo">590</span>    }<a name="line.590"></a>
+<span class="sourceLineNo">591</span>    return stat.getCversion();<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 Set&lt;String&gt; getAllHFileRefs() throws ReplicationException {<a name="line.595"></a>
+<span class="sourceLineNo">596</span>    try {<a name="line.596"></a>
+<span class="sourceLineNo">597</span>      for (int retry = 0;; retry++) {<a name="line.597"></a>
+<span class="sourceLineNo">598</span>        int v0 = getHFileRefsZNodeCversion();<a name="line.598"></a>
+<span class="sourceLineNo">599</span>        List&lt;String&gt; peers = getAllPeersFromHFileRefsQueue();<a name="line.599"></a>
+<span class="sourceLineNo">600</span>        if (peers.isEmpty()) {<a name="line.600"></a>
+<span class="sourceLineNo">601</span>          LOG.debug("Didn't find any peers with hfile references, won't prevent deletions.");<a name="line.601"></a>
+<span class="sourceLineNo">602</span>          return Collections.emptySet();<a name="line.602"></a>
+<span class="sourceLineNo">603</span>        }<a name="line.603"></a>
+<span class="sourceLineNo">604</span>        Set&lt;String&gt; hfileRefs = new HashSet&lt;&gt;();<a name="line.604"></a>
+<span class="sourceLineNo">605</span>        for (String peer : peers) {<a name="line.605"></a>
+<span class="sourceLineNo">606</span>          hfileRefs.addAll(getReplicableHFiles0(peer));<a name="line.606"></a>
+<span class="sourceLineNo">607</span>        }<a name="line.607"></a>
+<span class="sourceLineNo">608</span>        int v1 = getHFileRefsZNodeCversion();<a name="line.608"></a>
+<span class="sourceLineNo">609</span>        if (v0 == v1) {<a name="line.609"></a>
+<span class="sourceLineNo">610</span>          return hfileRefs;<a name="line.610"></a>
+<span class="sourceLineNo">611</span>        }<a name="line.611"></a>
+<span class="sourceLineNo">612</span>        LOG.debug("Replication hfile references node cversion changed from %d to %d, retry = %d",<a name="line.612"></a>
+<span class="sourceLineNo">613</span>            v0, v1, retry);<a name="line.613"></a>
+<span class="sourceLineNo">614</span>      }<a name="line.614"></a>
+<span class="sourceLineNo">615</span>    } catch (KeeperException e) {<a name="line.615"></a>
+<span class="sourceLineNo">616</span>      throw new ReplicationException("Failed to get all hfile refs", e);<a name="line.616"></a>
+<span class="sourceLineNo">617</span>    }<a name="line.617"></a>
+<span class="sourceLineNo">618</span>  }<a name="line.618"></a>
+<span class="sourceLineNo">619</span>}<a name="line.619"></a>