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 2018/11/09 14:53:06 UTC
[10/47] hbase-site git commit: Published site at
fe2265fa4a1e828b2e68ff8e42639c5942dccb1b.
http://git-wip-us.apache.org/repos/asf/hbase-site/blob/9592fdb5/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.ReplicationQueueOperation.html
----------------------------------------------------------------------
diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.ReplicationQueueOperation.html b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.ReplicationQueueOperation.html
index 083ab07..be28dfa 100644
--- a/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.ReplicationQueueOperation.html
+++ b/devapidocs/src-html/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceManager.ReplicationQueueOperation.html
@@ -31,54 +31,54 @@
<span class="sourceLineNo">023</span>import java.util.Collection;<a name="line.23"></a>
<span class="sourceLineNo">024</span>import java.util.Collections;<a name="line.24"></a>
<span class="sourceLineNo">025</span>import java.util.HashMap;<a name="line.25"></a>
-<span class="sourceLineNo">026</span>import java.util.HashSet;<a name="line.26"></a>
-<span class="sourceLineNo">027</span>import java.util.Iterator;<a name="line.27"></a>
-<span class="sourceLineNo">028</span>import java.util.List;<a name="line.28"></a>
-<span class="sourceLineNo">029</span>import java.util.Map;<a name="line.29"></a>
-<span class="sourceLineNo">030</span>import java.util.NavigableSet;<a name="line.30"></a>
-<span class="sourceLineNo">031</span>import java.util.Set;<a name="line.31"></a>
-<span class="sourceLineNo">032</span>import java.util.SortedSet;<a name="line.32"></a>
-<span class="sourceLineNo">033</span>import java.util.TreeSet;<a name="line.33"></a>
-<span class="sourceLineNo">034</span>import java.util.UUID;<a name="line.34"></a>
-<span class="sourceLineNo">035</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.35"></a>
-<span class="sourceLineNo">036</span>import java.util.concurrent.ConcurrentMap;<a name="line.36"></a>
-<span class="sourceLineNo">037</span>import java.util.concurrent.Future;<a name="line.37"></a>
-<span class="sourceLineNo">038</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.38"></a>
-<span class="sourceLineNo">039</span>import java.util.concurrent.RejectedExecutionException;<a name="line.39"></a>
-<span class="sourceLineNo">040</span>import java.util.concurrent.ThreadLocalRandom;<a name="line.40"></a>
-<span class="sourceLineNo">041</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.41"></a>
-<span class="sourceLineNo">042</span>import java.util.concurrent.TimeUnit;<a name="line.42"></a>
-<span class="sourceLineNo">043</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.43"></a>
-<span class="sourceLineNo">044</span>import java.util.stream.Collectors;<a name="line.44"></a>
-<span class="sourceLineNo">045</span>import org.apache.hadoop.conf.Configuration;<a name="line.45"></a>
-<span class="sourceLineNo">046</span>import org.apache.hadoop.fs.FileSystem;<a name="line.46"></a>
-<span class="sourceLineNo">047</span>import org.apache.hadoop.fs.Path;<a name="line.47"></a>
-<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.CompatibilitySingletonFactory;<a name="line.48"></a>
-<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.HConstants;<a name="line.49"></a>
-<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.Server;<a name="line.50"></a>
-<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.ServerName;<a name="line.51"></a>
-<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.TableName;<a name="line.52"></a>
-<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.53"></a>
-<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.54"></a>
-<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.55"></a>
-<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.56"></a>
-<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.ReplicationPeerConfig;<a name="line.57"></a>
-<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.replication.ReplicationPeerImpl;<a name="line.58"></a>
-<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.59"></a>
-<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.60"></a>
-<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.61"></a>
-<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.62"></a>
-<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.replication.ReplicationUtils;<a name="line.63"></a>
-<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.replication.SyncReplicationState;<a name="line.64"></a>
-<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.65"></a>
-<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.66"></a>
-<span class="sourceLineNo">067</span>import org.apache.hadoop.hbase.wal.SyncReplicationWALProvider;<a name="line.67"></a>
-<span class="sourceLineNo">068</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.68"></a>
-<span class="sourceLineNo">069</span>import org.apache.zookeeper.KeeperException;<a name="line.69"></a>
-<span class="sourceLineNo">070</span>import org.slf4j.Logger;<a name="line.70"></a>
-<span class="sourceLineNo">071</span>import org.slf4j.LoggerFactory;<a name="line.71"></a>
-<span class="sourceLineNo">072</span><a name="line.72"></a>
-<span class="sourceLineNo">073</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.73"></a>
+<span class="sourceLineNo">026</span>import java.util.Iterator;<a name="line.26"></a>
+<span class="sourceLineNo">027</span>import java.util.List;<a name="line.27"></a>
+<span class="sourceLineNo">028</span>import java.util.Map;<a name="line.28"></a>
+<span class="sourceLineNo">029</span>import java.util.NavigableSet;<a name="line.29"></a>
+<span class="sourceLineNo">030</span>import java.util.Set;<a name="line.30"></a>
+<span class="sourceLineNo">031</span>import java.util.SortedSet;<a name="line.31"></a>
+<span class="sourceLineNo">032</span>import java.util.TreeSet;<a name="line.32"></a>
+<span class="sourceLineNo">033</span>import java.util.UUID;<a name="line.33"></a>
+<span class="sourceLineNo">034</span>import java.util.concurrent.ConcurrentHashMap;<a name="line.34"></a>
+<span class="sourceLineNo">035</span>import java.util.concurrent.ConcurrentMap;<a name="line.35"></a>
+<span class="sourceLineNo">036</span>import java.util.concurrent.Future;<a name="line.36"></a>
+<span class="sourceLineNo">037</span>import java.util.concurrent.LinkedBlockingQueue;<a name="line.37"></a>
+<span class="sourceLineNo">038</span>import java.util.concurrent.RejectedExecutionException;<a name="line.38"></a>
+<span class="sourceLineNo">039</span>import java.util.concurrent.ThreadLocalRandom;<a name="line.39"></a>
+<span class="sourceLineNo">040</span>import java.util.concurrent.ThreadPoolExecutor;<a name="line.40"></a>
+<span class="sourceLineNo">041</span>import java.util.concurrent.TimeUnit;<a name="line.41"></a>
+<span class="sourceLineNo">042</span>import java.util.concurrent.atomic.AtomicLong;<a name="line.42"></a>
+<span class="sourceLineNo">043</span>import java.util.stream.Collectors;<a name="line.43"></a>
+<span class="sourceLineNo">044</span>import org.apache.hadoop.conf.Configuration;<a name="line.44"></a>
+<span class="sourceLineNo">045</span>import org.apache.hadoop.fs.FileSystem;<a name="line.45"></a>
+<span class="sourceLineNo">046</span>import org.apache.hadoop.fs.Path;<a name="line.46"></a>
+<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.CompatibilitySingletonFactory;<a name="line.47"></a>
+<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.HConstants;<a name="line.48"></a>
+<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.Server;<a name="line.49"></a>
+<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.ServerName;<a name="line.50"></a>
+<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.TableName;<a name="line.51"></a>
+<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.replication.ReplicationException;<a name="line.52"></a>
+<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.replication.ReplicationListener;<a name="line.53"></a>
+<span class="sourceLineNo">054</span>import org.apache.hadoop.hbase.replication.ReplicationPeer;<a name="line.54"></a>
+<span class="sourceLineNo">055</span>import org.apache.hadoop.hbase.replication.ReplicationPeer.PeerState;<a name="line.55"></a>
+<span class="sourceLineNo">056</span>import org.apache.hadoop.hbase.replication.ReplicationPeerConfig;<a name="line.56"></a>
+<span class="sourceLineNo">057</span>import org.apache.hadoop.hbase.replication.ReplicationPeerImpl;<a name="line.57"></a>
+<span class="sourceLineNo">058</span>import org.apache.hadoop.hbase.replication.ReplicationPeers;<a name="line.58"></a>
+<span class="sourceLineNo">059</span>import org.apache.hadoop.hbase.replication.ReplicationQueueInfo;<a name="line.59"></a>
+<span class="sourceLineNo">060</span>import org.apache.hadoop.hbase.replication.ReplicationQueueStorage;<a name="line.60"></a>
+<span class="sourceLineNo">061</span>import org.apache.hadoop.hbase.replication.ReplicationTracker;<a name="line.61"></a>
+<span class="sourceLineNo">062</span>import org.apache.hadoop.hbase.replication.ReplicationUtils;<a name="line.62"></a>
+<span class="sourceLineNo">063</span>import org.apache.hadoop.hbase.replication.SyncReplicationState;<a name="line.63"></a>
+<span class="sourceLineNo">064</span>import org.apache.hadoop.hbase.util.Pair;<a name="line.64"></a>
+<span class="sourceLineNo">065</span>import org.apache.hadoop.hbase.wal.AbstractFSWALProvider;<a name="line.65"></a>
+<span class="sourceLineNo">066</span>import org.apache.hadoop.hbase.wal.SyncReplicationWALProvider;<a name="line.66"></a>
+<span class="sourceLineNo">067</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.67"></a>
+<span class="sourceLineNo">068</span>import org.apache.zookeeper.KeeperException;<a name="line.68"></a>
+<span class="sourceLineNo">069</span>import org.slf4j.Logger;<a name="line.69"></a>
+<span class="sourceLineNo">070</span>import org.slf4j.LoggerFactory;<a name="line.70"></a>
+<span class="sourceLineNo">071</span><a name="line.71"></a>
+<span class="sourceLineNo">072</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.72"></a>
+<span class="sourceLineNo">073</span>import org.apache.hbase.thirdparty.com.google.common.collect.Sets;<a name="line.73"></a>
<span class="sourceLineNo">074</span>import org.apache.hbase.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;<a name="line.74"></a>
<span class="sourceLineNo">075</span><a name="line.75"></a>
<span class="sourceLineNo">076</span>/**<a name="line.76"></a>
@@ -156,7 +156,7 @@
<span class="sourceLineNo">148</span> private final Configuration conf;<a name="line.148"></a>
<span class="sourceLineNo">149</span> private final FileSystem fs;<a name="line.149"></a>
<span class="sourceLineNo">150</span> // The paths to the latest log of each wal group, for new coming peers<a name="line.150"></a>
-<span class="sourceLineNo">151</span> private final Set<Path> latestPaths;<a name="line.151"></a>
+<span class="sourceLineNo">151</span> private final Map<String, Path> latestPaths;<a name="line.151"></a>
<span class="sourceLineNo">152</span> // Path to the wals directories<a name="line.152"></a>
<span class="sourceLineNo">153</span> private final Path logDir;<a name="line.153"></a>
<span class="sourceLineNo">154</span> // Path to the wal archive<a name="line.154"></a>
@@ -224,7 +224,7 @@
<span class="sourceLineNo">216</span> tfb.setNameFormat("ReplicationExecutor-%d");<a name="line.216"></a>
<span class="sourceLineNo">217</span> tfb.setDaemon(true);<a name="line.217"></a>
<span class="sourceLineNo">218</span> this.executor.setThreadFactory(tfb.build());<a name="line.218"></a>
-<span class="sourceLineNo">219</span> this.latestPaths = new HashSet<Path>();<a name="line.219"></a>
+<span class="sourceLineNo">219</span> this.latestPaths = new HashMap<>();<a name="line.219"></a>
<span class="sourceLineNo">220</span> this.replicationForBulkLoadDataEnabled = conf.getBoolean(<a name="line.220"></a>
<span class="sourceLineNo">221</span> HConstants.REPLICATION_BULKLOAD_ENABLE_KEY, HConstants.REPLICATION_BULKLOAD_ENABLE_DEFAULT);<a name="line.221"></a>
<span class="sourceLineNo">222</span> this.sleepForRetries = this.conf.getLong("replication.source.sync.sleepforretries", 1000);<a name="line.222"></a>
@@ -379,756 +379,754 @@
<span class="sourceLineNo">371</span> Map<String, NavigableSet<String>> walsByGroup = new HashMap<>();<a name="line.371"></a>
<span class="sourceLineNo">372</span> this.walsById.put(peerId, walsByGroup);<a name="line.372"></a>
<span class="sourceLineNo">373</span> // Add the latest wal to that source's queue<a name="line.373"></a>
-<span class="sourceLineNo">374</span> if (this.latestPaths.size() > 0) {<a name="line.374"></a>
-<span class="sourceLineNo">375</span> for (Path logPath : latestPaths) {<a name="line.375"></a>
-<span class="sourceLineNo">376</span> String name = logPath.getName();<a name="line.376"></a>
-<span class="sourceLineNo">377</span> String walPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(name);<a name="line.377"></a>
-<span class="sourceLineNo">378</span> NavigableSet<String> logs = new TreeSet<>();<a name="line.378"></a>
-<span class="sourceLineNo">379</span> logs.add(name);<a name="line.379"></a>
-<span class="sourceLineNo">380</span> walsByGroup.put(walPrefix, logs);<a name="line.380"></a>
-<span class="sourceLineNo">381</span> // Abort RS and throw exception to make add peer failed<a name="line.381"></a>
-<span class="sourceLineNo">382</span> abortAndThrowIOExceptionWhenFail(<a name="line.382"></a>
-<span class="sourceLineNo">383</span> () -> this.queueStorage.addWAL(server.getServerName(), peerId, name));<a name="line.383"></a>
-<span class="sourceLineNo">384</span> src.enqueueLog(logPath);<a name="line.384"></a>
-<span class="sourceLineNo">385</span> }<a name="line.385"></a>
-<span class="sourceLineNo">386</span> }<a name="line.386"></a>
-<span class="sourceLineNo">387</span> }<a name="line.387"></a>
-<span class="sourceLineNo">388</span> ReplicationPeerConfig peerConfig = peer.getPeerConfig();<a name="line.388"></a>
-<span class="sourceLineNo">389</span> if (peerConfig.isSyncReplication()) {<a name="line.389"></a>
-<span class="sourceLineNo">390</span> syncReplicationPeerMappingManager.add(peer.getId(), peerConfig);<a name="line.390"></a>
-<span class="sourceLineNo">391</span> }<a name="line.391"></a>
-<span class="sourceLineNo">392</span> src.startup();<a name="line.392"></a>
-<span class="sourceLineNo">393</span> return src;<a name="line.393"></a>
-<span class="sourceLineNo">394</span> }<a name="line.394"></a>
-<span class="sourceLineNo">395</span><a name="line.395"></a>
-<span class="sourceLineNo">396</span> /**<a name="line.396"></a>
-<span class="sourceLineNo">397</span> * <p><a name="line.397"></a>
-<span class="sourceLineNo">398</span> * This is used when we transit a sync replication peer to {@link SyncReplicationState#STANDBY}.<a name="line.398"></a>
-<span class="sourceLineNo">399</span> * </p><a name="line.399"></a>
-<span class="sourceLineNo">400</span> * <p><a name="line.400"></a>
-<span class="sourceLineNo">401</span> * When transiting to {@link SyncReplicationState#STANDBY}, we can remove all the pending wal<a name="line.401"></a>
-<span class="sourceLineNo">402</span> * files for a replication peer as we do not need to replicate them any more. And this is<a name="line.402"></a>
-<span class="sourceLineNo">403</span> * necessary, otherwise when we transit back to {@link SyncReplicationState#DOWNGRADE_ACTIVE}<a name="line.403"></a>
-<span class="sourceLineNo">404</span> * later, the stale data will be replicated again and cause inconsistency.<a name="line.404"></a>
-<span class="sourceLineNo">405</span> * </p><a name="line.405"></a>
-<span class="sourceLineNo">406</span> * <p><a name="line.406"></a>
-<span class="sourceLineNo">407</span> * See HBASE-20426 for more details.<a name="line.407"></a>
-<span class="sourceLineNo">408</span> * </p><a name="line.408"></a>
-<span class="sourceLineNo">409</span> * @param peerId the id of the sync replication peer<a name="line.409"></a>
-<span class="sourceLineNo">410</span> */<a name="line.410"></a>
-<span class="sourceLineNo">411</span> public void drainSources(String peerId) throws IOException, ReplicationException {<a name="line.411"></a>
-<span class="sourceLineNo">412</span> String terminateMessage = "Sync replication peer " + peerId +<a name="line.412"></a>
-<span class="sourceLineNo">413</span> " is transiting to STANDBY. Will close the previous replication source and open a new one";<a name="line.413"></a>
-<span class="sourceLineNo">414</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.414"></a>
-<span class="sourceLineNo">415</span> assert peer.getPeerConfig().isSyncReplication();<a name="line.415"></a>
-<span class="sourceLineNo">416</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.416"></a>
-<span class="sourceLineNo">417</span> // synchronized here to avoid race with preLogRoll where we add new log to source and also<a name="line.417"></a>
-<span class="sourceLineNo">418</span> // walsById.<a name="line.418"></a>
-<span class="sourceLineNo">419</span> ReplicationSourceInterface toRemove;<a name="line.419"></a>
-<span class="sourceLineNo">420</span> Map<String, NavigableSet<String>> wals = new HashMap<>();<a name="line.420"></a>
-<span class="sourceLineNo">421</span> synchronized (latestPaths) {<a name="line.421"></a>
-<span class="sourceLineNo">422</span> toRemove = sources.put(peerId, src);<a name="line.422"></a>
-<span class="sourceLineNo">423</span> if (toRemove != null) {<a name="line.423"></a>
-<span class="sourceLineNo">424</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.424"></a>
-<span class="sourceLineNo">425</span> toRemove.terminate(terminateMessage);<a name="line.425"></a>
-<span class="sourceLineNo">426</span> toRemove.getSourceMetrics().clear();<a name="line.426"></a>
-<span class="sourceLineNo">427</span> }<a name="line.427"></a>
-<span class="sourceLineNo">428</span> // Here we make a copy of all the remaining wal files and then delete them from the<a name="line.428"></a>
-<span class="sourceLineNo">429</span> // replication queue storage after releasing the lock. It is not safe to just remove the old<a name="line.429"></a>
-<span class="sourceLineNo">430</span> // map from walsById since later we may fail to delete them from the replication queue<a name="line.430"></a>
-<span class="sourceLineNo">431</span> // storage, and when we retry next time, we can not know the wal files that need to be deleted<a name="line.431"></a>
-<span class="sourceLineNo">432</span> // from the replication queue storage.<a name="line.432"></a>
-<span class="sourceLineNo">433</span> walsById.get(peerId).forEach((k, v) -> wals.put(k, new TreeSet<>(v)));<a name="line.433"></a>
-<span class="sourceLineNo">434</span> }<a name="line.434"></a>
-<span class="sourceLineNo">435</span> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.435"></a>
-<span class="sourceLineNo">436</span> src.startup();<a name="line.436"></a>
-<span class="sourceLineNo">437</span> for (NavigableSet<String> walsByGroup : wals.values()) {<a name="line.437"></a>
-<span class="sourceLineNo">438</span> for (String wal : walsByGroup) {<a name="line.438"></a>
-<span class="sourceLineNo">439</span> queueStorage.removeWAL(server.getServerName(), peerId, wal);<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> synchronized (walsById) {<a name="line.442"></a>
-<span class="sourceLineNo">443</span> Map<String, NavigableSet<String>> oldWals = walsById.get(peerId);<a name="line.443"></a>
-<span class="sourceLineNo">444</span> wals.forEach((k, v) -> {<a name="line.444"></a>
-<span class="sourceLineNo">445</span> NavigableSet<String> walsByGroup = oldWals.get(k);<a name="line.445"></a>
-<span class="sourceLineNo">446</span> if (walsByGroup != null) {<a name="line.446"></a>
-<span class="sourceLineNo">447</span> walsByGroup.removeAll(v);<a name="line.447"></a>
-<span class="sourceLineNo">448</span> }<a name="line.448"></a>
-<span class="sourceLineNo">449</span> });<a name="line.449"></a>
-<span class="sourceLineNo">450</span> }<a name="line.450"></a>
-<span class="sourceLineNo">451</span> // synchronized on oldsources to avoid race with NodeFailoverWorker. Since NodeFailoverWorker is<a name="line.451"></a>
-<span class="sourceLineNo">452</span> // a background task, we will delete the file from replication queue storage under the lock to<a name="line.452"></a>
-<span class="sourceLineNo">453</span> // simplify the logic.<a name="line.453"></a>
-<span class="sourceLineNo">454</span> synchronized (this.oldsources) {<a name="line.454"></a>
-<span class="sourceLineNo">455</span> for (Iterator<ReplicationSourceInterface> iter = oldsources.iterator(); iter.hasNext();) {<a name="line.455"></a>
-<span class="sourceLineNo">456</span> ReplicationSourceInterface oldSource = iter.next();<a name="line.456"></a>
-<span class="sourceLineNo">457</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.457"></a>
-<span class="sourceLineNo">458</span> String queueId = oldSource.getQueueId();<a name="line.458"></a>
-<span class="sourceLineNo">459</span> oldSource.terminate(terminateMessage);<a name="line.459"></a>
-<span class="sourceLineNo">460</span> oldSource.getSourceMetrics().clear();<a name="line.460"></a>
-<span class="sourceLineNo">461</span> queueStorage.removeQueue(server.getServerName(), queueId);<a name="line.461"></a>
-<span class="sourceLineNo">462</span> walsByIdRecoveredQueues.remove(queueId);<a name="line.462"></a>
-<span class="sourceLineNo">463</span> iter.remove();<a name="line.463"></a>
-<span class="sourceLineNo">464</span> }<a name="line.464"></a>
-<span class="sourceLineNo">465</span> }<a name="line.465"></a>
-<span class="sourceLineNo">466</span> }<a name="line.466"></a>
-<span class="sourceLineNo">467</span> }<a name="line.467"></a>
-<span class="sourceLineNo">468</span><a name="line.468"></a>
-<span class="sourceLineNo">469</span> /**<a name="line.469"></a>
-<span class="sourceLineNo">470</span> * Close the previous replication sources of this peer id and open new sources to trigger the new<a name="line.470"></a>
-<span class="sourceLineNo">471</span> * replication state changes or new replication config changes. Here we don't need to change<a name="line.471"></a>
-<span class="sourceLineNo">472</span> * replication queue storage and only to enqueue all logs to the new replication source<a name="line.472"></a>
-<span class="sourceLineNo">473</span> * @param peerId the id of the replication peer<a name="line.473"></a>
-<span class="sourceLineNo">474</span> */<a name="line.474"></a>
-<span class="sourceLineNo">475</span> public void refreshSources(String peerId) throws IOException {<a name="line.475"></a>
-<span class="sourceLineNo">476</span> String terminateMessage = "Peer " + peerId +<a name="line.476"></a>
-<span class="sourceLineNo">477</span> " state or config changed. Will close the previous replication source and open a new one";<a name="line.477"></a>
-<span class="sourceLineNo">478</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.478"></a>
-<span class="sourceLineNo">479</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.479"></a>
-<span class="sourceLineNo">480</span> // synchronized on latestPaths to avoid missing the new log<a name="line.480"></a>
-<span class="sourceLineNo">481</span> synchronized (this.latestPaths) {<a name="line.481"></a>
-<span class="sourceLineNo">482</span> ReplicationSourceInterface toRemove = this.sources.put(peerId, src);<a name="line.482"></a>
-<span class="sourceLineNo">483</span> if (toRemove != null) {<a name="line.483"></a>
-<span class="sourceLineNo">484</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.484"></a>
-<span class="sourceLineNo">485</span> toRemove.terminate(terminateMessage);<a name="line.485"></a>
-<span class="sourceLineNo">486</span> }<a name="line.486"></a>
-<span class="sourceLineNo">487</span> for (NavigableSet<String> walsByGroup : walsById.get(peerId).values()) {<a name="line.487"></a>
-<span class="sourceLineNo">488</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(this.logDir, wal)));<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> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.491"></a>
-<span class="sourceLineNo">492</span> src.startup();<a name="line.492"></a>
-<span class="sourceLineNo">493</span><a name="line.493"></a>
-<span class="sourceLineNo">494</span> List<ReplicationSourceInterface> toStartup = new ArrayList<>();<a name="line.494"></a>
-<span class="sourceLineNo">495</span> // synchronized on oldsources to avoid race with NodeFailoverWorker<a name="line.495"></a>
-<span class="sourceLineNo">496</span> synchronized (this.oldsources) {<a name="line.496"></a>
-<span class="sourceLineNo">497</span> List<String> previousQueueIds = new ArrayList<>();<a name="line.497"></a>
-<span class="sourceLineNo">498</span> for (ReplicationSourceInterface oldSource : this.oldsources) {<a name="line.498"></a>
-<span class="sourceLineNo">499</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.499"></a>
-<span class="sourceLineNo">500</span> previousQueueIds.add(oldSource.getQueueId());<a name="line.500"></a>
-<span class="sourceLineNo">501</span> oldSource.terminate(terminateMessage);<a name="line.501"></a>
-<span class="sourceLineNo">502</span> this.oldsources.remove(oldSource);<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> for (String queueId : previousQueueIds) {<a name="line.505"></a>
-<span class="sourceLineNo">506</span> ReplicationSourceInterface replicationSource = createSource(queueId, peer);<a name="line.506"></a>
-<span class="sourceLineNo">507</span> this.oldsources.add(replicationSource);<a name="line.507"></a>
-<span class="sourceLineNo">508</span> for (SortedSet<String> walsByGroup : walsByIdRecoveredQueues.get(queueId).values()) {<a name="line.508"></a>
-<span class="sourceLineNo">509</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(wal)));<a name="line.509"></a>
-<span class="sourceLineNo">510</span> }<a name="line.510"></a>
-<span class="sourceLineNo">511</span> toStartup.add(replicationSource);<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> for (ReplicationSourceInterface replicationSource : toStartup) {<a name="line.514"></a>
-<span class="sourceLineNo">515</span> replicationSource.startup();<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><a name="line.518"></a>
-<span class="sourceLineNo">519</span> /**<a name="line.519"></a>
-<span class="sourceLineNo">520</span> * Clear the metrics and related replication queue of the specified old source<a name="line.520"></a>
-<span class="sourceLineNo">521</span> * @param src source to clear<a name="line.521"></a>
-<span class="sourceLineNo">522</span> */<a name="line.522"></a>
-<span class="sourceLineNo">523</span> private boolean removeRecoveredSource(ReplicationSourceInterface src) {<a name="line.523"></a>
-<span class="sourceLineNo">524</span> if (!this.oldsources.remove(src)) {<a name="line.524"></a>
-<span class="sourceLineNo">525</span> return false;<a name="line.525"></a>
-<span class="sourceLineNo">526</span> }<a name="line.526"></a>
-<span class="sourceLineNo">527</span> LOG.info("Done with the recovered queue {}", src.getQueueId());<a name="line.527"></a>
-<span class="sourceLineNo">528</span> // Delete queue from storage and memory<a name="line.528"></a>
-<span class="sourceLineNo">529</span> deleteQueue(src.getQueueId());<a name="line.529"></a>
-<span class="sourceLineNo">530</span> this.walsByIdRecoveredQueues.remove(src.getQueueId());<a name="line.530"></a>
-<span class="sourceLineNo">531</span> return true;<a name="line.531"></a>
-<span class="sourceLineNo">532</span> }<a name="line.532"></a>
-<span class="sourceLineNo">533</span><a name="line.533"></a>
-<span class="sourceLineNo">534</span> void finishRecoveredSource(ReplicationSourceInterface src) {<a name="line.534"></a>
-<span class="sourceLineNo">535</span> synchronized (oldsources) {<a name="line.535"></a>
-<span class="sourceLineNo">536</span> if (!removeRecoveredSource(src)) {<a name="line.536"></a>
-<span class="sourceLineNo">537</span> return;<a name="line.537"></a>
-<span class="sourceLineNo">538</span> }<a name="line.538"></a>
-<span class="sourceLineNo">539</span> }<a name="line.539"></a>
-<span class="sourceLineNo">540</span> LOG.info("Finished recovering queue {} with the following stats: {}", src.getQueueId(),<a name="line.540"></a>
-<span class="sourceLineNo">541</span> src.getStats());<a name="line.541"></a>
-<span class="sourceLineNo">542</span> }<a name="line.542"></a>
-<span class="sourceLineNo">543</span><a name="line.543"></a>
-<span class="sourceLineNo">544</span> /**<a name="line.544"></a>
-<span class="sourceLineNo">545</span> * Clear the metrics and related replication queue of the specified old source<a name="line.545"></a>
-<span class="sourceLineNo">546</span> * @param src source to clear<a name="line.546"></a>
-<span class="sourceLineNo">547</span> */<a name="line.547"></a>
-<span class="sourceLineNo">548</span> void removeSource(ReplicationSourceInterface src) {<a name="line.548"></a>
-<span class="sourceLineNo">549</span> LOG.info("Done with the queue " + src.getQueueId());<a name="line.549"></a>
-<span class="sourceLineNo">550</span> this.sources.remove(src.getPeerId());<a name="line.550"></a>
-<span class="sourceLineNo">551</span> // Delete queue from storage and memory<a name="line.551"></a>
-<span class="sourceLineNo">552</span> deleteQueue(src.getQueueId());<a name="line.552"></a>
-<span class="sourceLineNo">553</span> this.walsById.remove(src.getQueueId());<a name="line.553"></a>
-<span class="sourceLineNo">554</span><a name="line.554"></a>
-<span class="sourceLineNo">555</span> }<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> * Delete a complete queue of wals associated with a replication source<a name="line.558"></a>
-<span class="sourceLineNo">559</span> * @param queueId the id of replication queue to delete<a name="line.559"></a>
-<span class="sourceLineNo">560</span> */<a name="line.560"></a>
-<span class="sourceLineNo">561</span> private void deleteQueue(String queueId) {<a name="line.561"></a>
-<span class="sourceLineNo">562</span> abortWhenFail(() -> this.queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.562"></a>
-<span class="sourceLineNo">563</span> }<a name="line.563"></a>
-<span class="sourceLineNo">564</span><a name="line.564"></a>
-<span class="sourceLineNo">565</span> @FunctionalInterface<a name="line.565"></a>
-<span class="sourceLineNo">566</span> private interface ReplicationQueueOperation {<a name="line.566"></a>
-<span class="sourceLineNo">567</span> void exec() throws ReplicationException;<a name="line.567"></a>
-<span class="sourceLineNo">568</span> }<a name="line.568"></a>
-<span class="sourceLineNo">569</span><a name="line.569"></a>
-<span class="sourceLineNo">570</span> /**<a name="line.570"></a>
-<span class="sourceLineNo">571</span> * Refresh replication source will terminate the old source first, then the source thread will be<a name="line.571"></a>
-<span class="sourceLineNo">572</span> * interrupted. Need to handle it instead of abort the region server.<a name="line.572"></a>
-<span class="sourceLineNo">573</span> */<a name="line.573"></a>
-<span class="sourceLineNo">574</span> private void interruptOrAbortWhenFail(ReplicationQueueOperation op) {<a name="line.574"></a>
-<span class="sourceLineNo">575</span> try {<a name="line.575"></a>
-<span class="sourceLineNo">576</span> op.exec();<a name="line.576"></a>
-<span class="sourceLineNo">577</span> } catch (ReplicationException e) {<a name="line.577"></a>
-<span class="sourceLineNo">578</span> if (e.getCause() != null && e.getCause() instanceof KeeperException.SystemErrorException<a name="line.578"></a>
-<span class="sourceLineNo">579</span> && e.getCause().getCause() != null && e.getCause()<a name="line.579"></a>
-<span class="sourceLineNo">580</span> .getCause() instanceof InterruptedException) {<a name="line.580"></a>
-<span class="sourceLineNo">581</span> throw new RuntimeException(<a name="line.581"></a>
-<span class="sourceLineNo">582</span> "Thread is interrupted, the replication source may be terminated");<a name="line.582"></a>
-<span class="sourceLineNo">583</span> }<a name="line.583"></a>
-<span class="sourceLineNo">584</span> server.abort("Failed to operate on replication queue", e);<a name="line.584"></a>
-<span class="sourceLineNo">585</span> }<a name="line.585"></a>
-<span class="sourceLineNo">586</span> }<a name="line.586"></a>
-<span class="sourceLineNo">587</span><a name="line.587"></a>
-<span class="sourceLineNo">588</span> private void abortWhenFail(ReplicationQueueOperation op) {<a name="line.588"></a>
-<span class="sourceLineNo">589</span> try {<a name="line.589"></a>
-<span class="sourceLineNo">590</span> op.exec();<a name="line.590"></a>
-<span class="sourceLineNo">591</span> } catch (ReplicationException e) {<a name="line.591"></a>
-<span class="sourceLineNo">592</span> server.abort("Failed to operate on replication queue", e);<a name="line.592"></a>
-<span class="sourceLineNo">593</span> }<a name="line.593"></a>
-<span class="sourceLineNo">594</span> }<a name="line.594"></a>
-<span class="sourceLineNo">595</span><a name="line.595"></a>
-<span class="sourceLineNo">596</span> private void throwIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.596"></a>
-<span class="sourceLineNo">597</span> try {<a name="line.597"></a>
-<span class="sourceLineNo">598</span> op.exec();<a name="line.598"></a>
-<span class="sourceLineNo">599</span> } catch (ReplicationException e) {<a name="line.599"></a>
-<span class="sourceLineNo">600</span> throw new IOException(e);<a name="line.600"></a>
-<span class="sourceLineNo">601</span> }<a name="line.601"></a>
-<span class="sourceLineNo">602</span> }<a name="line.602"></a>
-<span class="sourceLineNo">603</span><a name="line.603"></a>
-<span class="sourceLineNo">604</span> private void abortAndThrowIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.604"></a>
-<span class="sourceLineNo">605</span> try {<a name="line.605"></a>
-<span class="sourceLineNo">606</span> op.exec();<a name="line.606"></a>
-<span class="sourceLineNo">607</span> } catch (ReplicationException e) {<a name="line.607"></a>
-<span class="sourceLineNo">608</span> server.abort("Failed to operate on replication queue", e);<a name="line.608"></a>
-<span class="sourceLineNo">609</span> throw new IOException(e);<a name="line.609"></a>
-<span class="sourceLineNo">610</span> }<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> /**<a name="line.613"></a>
-<span class="sourceLineNo">614</span> * This method will log the current position to storage. And also clean old logs from the<a name="line.614"></a>
-<span class="sourceLineNo">615</span> * replication queue.<a name="line.615"></a>
-<span class="sourceLineNo">616</span> * @param source the replication source<a name="line.616"></a>
-<span class="sourceLineNo">617</span> * @param entryBatch the wal entry batch we just shipped<a name="line.617"></a>
-<span class="sourceLineNo">618</span> */<a name="line.618"></a>
-<span class="sourceLineNo">619</span> public void logPositionAndCleanOldLogs(ReplicationSourceInterface source,<a name="line.619"></a>
-<span class="sourceLineNo">620</span> WALEntryBatch entryBatch) {<a name="line.620"></a>
-<span class="sourceLineNo">621</span> String fileName = entryBatch.getLastWalPath().getName();<a name="line.621"></a>
-<span class="sourceLineNo">622</span> interruptOrAbortWhenFail(() -> this.queueStorage.setWALPosition(server.getServerName(),<a name="line.622"></a>
-<span class="sourceLineNo">623</span> source.getQueueId(), fileName, entryBatch.getLastWalPosition(), entryBatch.getLastSeqIds()));<a name="line.623"></a>
-<span class="sourceLineNo">624</span> cleanOldLogs(fileName, entryBatch.isEndOfFile(), source);<a name="line.624"></a>
-<span class="sourceLineNo">625</span> }<a name="line.625"></a>
-<span class="sourceLineNo">626</span><a name="line.626"></a>
-<span class="sourceLineNo">627</span> /**<a name="line.627"></a>
-<span class="sourceLineNo">628</span> * Cleans a log file and all older logs from replication queue. Called when we are sure that a log<a name="line.628"></a>
-<span class="sourceLineNo">629</span> * file is closed and has no more entries.<a name="line.629"></a>
-<span class="sourceLineNo">630</span> * @param log Path to the log<a name="line.630"></a>
-<span class="sourceLineNo">631</span> * @param inclusive whether we should also remove the given log file<a name="line.631"></a>
-<span class="sourceLineNo">632</span> * @param source the replication source<a name="line.632"></a>
-<span class="sourceLineNo">633</span> */<a name="line.633"></a>
-<span class="sourceLineNo">634</span> @VisibleForTesting<a name="line.634"></a>
-<span class="sourceLineNo">635</span> void cleanOldLogs(String log, boolean inclusive, ReplicationSourceInterface source) {<a name="line.635"></a>
-<span class="sourceLineNo">636</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log);<a name="line.636"></a>
-<span class="sourceLineNo">637</span> if (source.isRecovered()) {<a name="line.637"></a>
-<span class="sourceLineNo">638</span> NavigableSet<String> wals = walsByIdRecoveredQueues.get(source.getQueueId()).get(logPrefix);<a name="line.638"></a>
-<span class="sourceLineNo">639</span> if (wals != null) {<a name="line.639"></a>
-<span class="sourceLineNo">640</span> NavigableSet<String> walsToRemove = wals.headSet(log, inclusive);<a name="line.640"></a>
-<span class="sourceLineNo">641</span> if (walsToRemove.isEmpty()) {<a name="line.641"></a>
-<span class="sourceLineNo">642</span> return;<a name="line.642"></a>
-<span class="sourceLineNo">643</span> }<a name="line.643"></a>
-<span class="sourceLineNo">644</span> cleanOldLogs(walsToRemove, source);<a name="line.644"></a>
-<span class="sourceLineNo">645</span> walsToRemove.clear();<a name="line.645"></a>
-<span class="sourceLineNo">646</span> }<a name="line.646"></a>
-<span class="sourceLineNo">647</span> } else {<a name="line.647"></a>
-<span class="sourceLineNo">648</span> NavigableSet<String> wals;<a name="line.648"></a>
-<span class="sourceLineNo">649</span> NavigableSet<String> walsToRemove;<a name="line.649"></a>
-<span class="sourceLineNo">650</span> // synchronized on walsById to avoid race with preLogRoll<a name="line.650"></a>
-<span class="sourceLineNo">651</span> synchronized (this.walsById) {<a name="line.651"></a>
-<span class="sourceLineNo">652</span> wals = walsById.get(source.getQueueId()).get(logPrefix);<a name="line.652"></a>
-<span class="sourceLineNo">653</span> if (wals == null) {<a name="line.653"></a>
-<span class="sourceLineNo">654</span> return;<a name="line.654"></a>
-<span class="sourceLineNo">655</span> }<a name="line.655"></a>
-<span class="sourceLineNo">656</span> walsToRemove = wals.headSet(log, inclusive);<a name="line.656"></a>
-<span class="sourceLineNo">657</span> if (walsToRemove.isEmpty()) {<a name="line.657"></a>
-<span class="sourceLineNo">658</span> return;<a name="line.658"></a>
-<span class="sourceLineNo">659</span> }<a name="line.659"></a>
-<span class="sourceLineNo">660</span> walsToRemove = new TreeSet<>(walsToRemove);<a name="line.660"></a>
-<span class="sourceLineNo">661</span> }<a name="line.661"></a>
-<span class="sourceLineNo">662</span> // cleanOldLogs may spend some time, especially for sync replication where we may want to<a name="line.662"></a>
-<span class="sourceLineNo">663</span> // remove remote wals as the remote cluster may have already been down, so we do it outside<a name="line.663"></a>
-<span class="sourceLineNo">664</span> // the lock to avoid block preLogRoll<a name="line.664"></a>
-<span class="sourceLineNo">665</span> cleanOldLogs(walsToRemove, source);<a name="line.665"></a>
-<span class="sourceLineNo">666</span> // now let's remove the files in the set<a name="line.666"></a>
-<span class="sourceLineNo">667</span> synchronized (this.walsById) {<a name="line.667"></a>
-<span class="sourceLineNo">668</span> wals.removeAll(walsToRemove);<a name="line.668"></a>
-<span class="sourceLineNo">669</span> }<a name="line.669"></a>
-<span class="sourceLineNo">670</span> }<a name="line.670"></a>
-<span class="sourceLineNo">671</span> }<a name="line.671"></a>
-<span class="sourceLineNo">672</span><a name="line.672"></a>
-<span class="sourceLineNo">673</span> private void removeRemoteWALs(String peerId, String remoteWALDir, Collection<String> wals)<a name="line.673"></a>
-<span class="sourceLineNo">674</span> throws IOException {<a name="line.674"></a>
-<span class="sourceLineNo">675</span> Path remoteWALDirForPeer = ReplicationUtils.getPeerRemoteWALDir(remoteWALDir, peerId);<a name="line.675"></a>
-<span class="sourceLineNo">676</span> FileSystem fs = ReplicationUtils.getRemoteWALFileSystem(conf, remoteWALDir);<a name="line.676"></a>
-<span class="sourceLineNo">677</span> for (String wal : wals) {<a name="line.677"></a>
-<span class="sourceLineNo">678</span> Path walFile = new Path(remoteWALDirForPeer, wal);<a name="line.678"></a>
-<span class="sourceLineNo">679</span> try {<a name="line.679"></a>
-<span class="sourceLineNo">680</span> if (!fs.delete(walFile, false) && fs.exists(walFile)) {<a name="line.680"></a>
-<span class="sourceLineNo">681</span> throw new IOException("Can not delete " + walFile);<a name="line.681"></a>
-<span class="sourceLineNo">682</span> }<a name="line.682"></a>
-<span class="sourceLineNo">683</span> } catch (FileNotFoundException e) {<a name="line.683"></a>
-<span class="sourceLineNo">684</span> // Just ignore since this means the file has already been deleted.<a name="line.684"></a>
-<span class="sourceLineNo">685</span> // The javadoc of the FileSystem.delete methods does not specify the behavior of deleting an<a name="line.685"></a>
-<span class="sourceLineNo">686</span> // inexistent file, so here we deal with both, i.e, check the return value of the<a name="line.686"></a>
-<span class="sourceLineNo">687</span> // FileSystem.delete, and also catch FNFE.<a name="line.687"></a>
-<span class="sourceLineNo">688</span> LOG.debug("The remote wal {} has already been deleted?", walFile, e);<a name="line.688"></a>
-<span class="sourceLineNo">689</span> }<a name="line.689"></a>
-<span class="sourceLineNo">690</span> }<a name="line.690"></a>
-<span class="sourceLineNo">691</span> }<a name="line.691"></a>
-<span class="sourceLineNo">692</span><a name="line.692"></a>
-<span class="sourceLineNo">693</span> private void cleanOldLogs(NavigableSet<String> wals, ReplicationSourceInterface source) {<a name="line.693"></a>
-<span class="sourceLineNo">694</span> LOG.debug("Removing {} logs in the list: {}", wals.size(), wals);<a name="line.694"></a>
-<span class="sourceLineNo">695</span> // The intention here is that, we want to delete the remote wal files ASAP as it may effect the<a name="line.695"></a>
-<span class="sourceLineNo">696</span> // failover time if you want to transit the remote cluster from S to A. And the infinite retry<a name="line.696"></a>
-<span class="sourceLineNo">697</span> // is not a problem, as if we can not contact with the remote HDFS cluster, then usually we can<a name="line.697"></a>
-<span class="sourceLineNo">698</span> // not contact with the HBase cluster either, so the replication will be blocked either.<a name="line.698"></a>
-<span class="sourceLineNo">699</span> if (source.isSyncReplication()) {<a name="line.699"></a>
-<span class="sourceLineNo">700</span> String peerId = source.getPeerId();<a name="line.700"></a>
-<span class="sourceLineNo">701</span> String remoteWALDir = source.getPeer().getPeerConfig().getRemoteWALDir();<a name="line.701"></a>
-<span class="sourceLineNo">702</span> // Filter out the wals need to be removed from the remote directory. Its name should be the<a name="line.702"></a>
-<span class="sourceLineNo">703</span> // special format, and also, the peer id in its name should match the peer id for the<a name="line.703"></a>
-<span class="sourceLineNo">704</span> // replication source.<a name="line.704"></a>
-<span class="sourceLineNo">705</span> List<String> remoteWals = wals.stream().filter(w -> SyncReplicationWALProvider<a name="line.705"></a>
-<span class="sourceLineNo">706</span> .getSyncReplicationPeerIdFromWALName(w).map(peerId::equals).orElse(false))<a name="line.706"></a>
-<span class="sourceLineNo">707</span> .collect(Collectors.toList());<a name="line.707"></a>
-<span class="sourceLineNo">708</span> LOG.debug("Removing {} logs from remote dir {} in the list: {}", remoteWals.size(),<a name="line.708"></a>
-<span class="sourceLineNo">709</span> remoteWALDir, remoteWals);<a name="line.709"></a>
-<span class="sourceLineNo">710</span> if (!remoteWals.isEmpty()) {<a name="line.710"></a>
-<span class="sourceLineNo">711</span> for (int sleepMultiplier = 0;;) {<a name="line.711"></a>
-<span class="sourceLineNo">712</span> try {<a name="line.712"></a>
-<span class="sourceLineNo">713</span> removeRemoteWALs(peerId, remoteWALDir, remoteWals);<a name="line.713"></a>
-<span class="sourceLineNo">714</span> break;<a name="line.714"></a>
-<span class="sourceLineNo">715</span> } catch (IOException e) {<a name="line.715"></a>
-<span class="sourceLineNo">716</span> LOG.warn("Failed to delete remote wals from remote dir {} for peer {}", remoteWALDir,<a name="line.716"></a>
-<span class="sourceLineNo">717</span> peerId);<a name="line.717"></a>
-<span class="sourceLineNo">718</span> }<a name="line.718"></a>
-<span class="sourceLineNo">719</span> if (!source.isSourceActive()) {<a name="line.719"></a>
-<span class="sourceLineNo">720</span> // skip the following operations<a name="line.720"></a>
-<span class="sourceLineNo">721</span> return;<a name="line.721"></a>
-<span class="sourceLineNo">722</span> }<a name="line.722"></a>
-<span class="sourceLineNo">723</span> if (ReplicationUtils.sleepForRetries("Failed to delete remote wals", sleepForRetries,<a name="line.723"></a>
-<span class="sourceLineNo">724</span> sleepMultiplier, maxRetriesMultiplier)) {<a name="line.724"></a>
-<span class="sourceLineNo">725</span> sleepMultiplier++;<a name="line.725"></a>
-<span class="sourceLineNo">726</span> }<a name="line.726"></a>
-<span class="sourceLineNo">727</span> }<a name="line.727"></a>
-<span class="sourceLineNo">728</span> }<a name="line.728"></a>
-<span class="sourceLineNo">729</span> }<a name="line.729"></a>
-<span class="sourceLineNo">730</span> String queueId = source.getQueueId();<a name="line.730"></a>
-<span class="sourceLineNo">731</span> for (String wal : wals) {<a name="line.731"></a>
-<span class="sourceLineNo">732</span> interruptOrAbortWhenFail(<a name="line.732"></a>
-<span class="sourceLineNo">733</span> () -> this.queueStorage.removeWAL(server.getServerName(), queueId, wal));<a name="line.733"></a>
-<span class="sourceLineNo">734</span> }<a name="line.734"></a>
-<span class="sourceLineNo">735</span> }<a name="line.735"></a>
-<span class="sourceLineNo">736</span><a name="line.736"></a>
-<span class="sourceLineNo">737</span> // public because of we call it in TestReplicationEmptyWALRecovery<a name="line.737"></a>
-<span class="sourceLineNo">738</span> @VisibleForTesting<a name="line.738"></a>
-<span class="sourceLineNo">739</span> public void preLogRoll(Path newLog) throws IOException {<a name="line.739"></a>
-<span class="sourceLineNo">740</span> String logName = newLog.getName();<a name="line.740"></a>
-<span class="sourceLineNo">741</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(logName);<a name="line.741"></a>
-<span class="sourceLineNo">742</span> // synchronized on latestPaths to avoid the new open source miss the new log<a name="line.742"></a>
-<span class="sourceLineNo">743</span> synchronized (this.latestPaths) {<a name="line.743"></a>
-<span class="sourceLineNo">744</span> // Add log to queue storage<a name="line.744"></a>
-<span class="sourceLineNo">745</span> for (ReplicationSourceInterface source : this.sources.values()) {<a name="line.745"></a>
-<span class="sourceLineNo">746</span> // If record log to queue storage failed, abort RS and throw exception to make log roll<a name="line.746"></a>
-<span class="sourceLineNo">747</span> // failed<a name="line.747"></a>
-<span class="sourceLineNo">748</span> abortAndThrowIOExceptionWhenFail(<a name="line.748"></a>
-<span class="sourceLineNo">749</span> () -> this.queueStorage.addWAL(server.getServerName(), source.getQueueId(), logName));<a name="line.749"></a>
-<span class="sourceLineNo">750</span> }<a name="line.750"></a>
-<span class="sourceLineNo">751</span><a name="line.751"></a>
-<span class="sourceLineNo">752</span> // synchronized on walsById to avoid race with cleanOldLogs<a name="line.752"></a>
-<span class="sourceLineNo">753</span> synchronized (this.walsById) {<a name="line.753"></a>
-<span class="sourceLineNo">754</span> // Update walsById map<a name="line.754"></a>
-<span class="sourceLineNo">755</span> for (Map.Entry<String, Map<String, NavigableSet<String>>> entry : this.walsById<a name="line.755"></a>
-<span class="sourceLineNo">756</span> .entrySet()) {<a name="line.756"></a>
-<span class="sourceLineNo">757</span> String peerId = entry.getKey();<a name="line.757"></a>
-<span class="sourceLineNo">758</span> Map<String, NavigableSet<String>> walsByPrefix = entry.getValue();<a name="line.758"></a>
-<span class="sourceLineNo">759</span> boolean existingPrefix = false;<a name="line.759"></a>
-<span class="sourceLineNo">760</span> for (Map.Entry<String, NavigableSet<String>> walsEntry : walsByPrefix.entrySet()) {<a name="line.760"></a>
-<span class="sourceLineNo">761</span> SortedSet<String> wals = walsEntry.getValue();<a name="line.761"></a>
-<span class="sourceLineNo">762</span> if (this.sources.isEmpty()) {<a name="line.762"></a>
-<span class="sourceLineNo">763</span> // If there's no slaves, don't need to keep the old wals since<a name="line.763"></a>
-<span class="sourceLineNo">764</span> // we only consider the last one when a new slave comes in<a name="line.764"></a>
-<span class="sourceLineNo">765</span> wals.clear();<a name="line.765"></a>
-<span class="sourceLineNo">766</span> }<a name="line.766"></a>
-<span class="sourceLineNo">767</span> if (logPrefix.equals(walsEntry.getKey())) {<a name="line.767"></a>
-<span class="sourceLineNo">768</span> wals.add(logName);<a name="line.768"></a>
-<span class="sourceLineNo">769</span> existingPrefix = true;<a name="line.769"></a>
-<span class="sourceLineNo">770</span> }<a name="line.770"></a>
-<span class="sourceLineNo">771</span> }<a name="line.771"></a>
-<span class="sourceLineNo">772</span> if (!existingPrefix) {<a name="line.772"></a>
-<span class="sourceLineNo">773</span> // The new log belongs to a new group, add it into this peer<a name="line.773"></a>
-<span class="sourceLineNo">774</span> LOG.debug("Start tracking logs for wal group {} for peer {}", logPrefix, peerId);<a name="line.774"></a>
-<span class="sourceLineNo">775</span> NavigableSet<String> wals = new TreeSet<>();<a name="line.775"></a>
-<span class="sourceLineNo">776</span> wals.add(logName);<a name="line.776"></a>
-<span class="sourceLineNo">777</span> walsByPrefix.put(logPrefix, wals);<a name="line.777"></a>
-<span class="sourceLineNo">778</span> }<a name="line.778"></a>
-<span class="sourceLineNo">779</span> }<a name="line.779"></a>
-<span class="sourceLineNo">780</span> }<a name="line.780"></a>
-<span class="sourceLineNo">781</span><a name="line.781"></a>
-<span class="sourceLineNo">782</span> // Add to latestPaths<a name="line.782"></a>
-<span class="sourceLineNo">783</span> Iterator<Path> iterator = latestPaths.iterator();<a name="line.783"></a>
-<span class="sourceLineNo">784</span> while (iterator.hasNext()) {<a name="line.784"></a>
-<span class="sourceLineNo">785</span> Path path = iterator.next();<a name="line.785"></a>
-<span class="sourceLineNo">786</span> if (path.getName().contains(logPrefix)) {<a name="line.786"></a>
-<span class="sourceLineNo">787</span> iterator.remove();<a name="line.787"></a>
-<span class="sourceLineNo">788</span> break;<a name="line.788"></a>
-<span class="sourceLineNo">789</span> }<a name="line.789"></a>
-<span class="sourceLineNo">790</span> }<a name="line.790"></a>
-<span class="sourceLineNo">791</span> this.latestPaths.add(newLog);<a name="line.791"></a>
+<span class="sourceLineNo">374</span> if (!latestPaths.isEmpty()) {<a name="line.374"></a>
+<span class="sourceLineNo">375</span> for (Map.Entry<String, Path> walPrefixAndPath : latestPaths.entrySet()) {<a name="line.375"></a>
+<span class="sourceLineNo">376</span> Path walPath = walPrefixAndPath.getValue();<a name="line.376"></a>
+<span class="sourceLineNo">377</span> NavigableSet<String> wals = new TreeSet<>();<a name="line.377"></a>
+<span class="sourceLineNo">378</span> wals.add(walPath.getName());<a name="line.378"></a>
+<span class="sourceLineNo">379</span> walsByGroup.put(walPrefixAndPath.getKey(), wals);<a name="line.379"></a>
+<span class="sourceLineNo">380</span> // Abort RS and throw exception to make add peer failed<a name="line.380"></a>
+<span class="sourceLineNo">381</span> abortAndThrowIOExceptionWhenFail(<a name="line.381"></a>
+<span class="sourceLineNo">382</span> () -> this.queueStorage.addWAL(server.getServerName(), peerId, walPath.getName()));<a name="line.382"></a>
+<span class="sourceLineNo">383</span> src.enqueueLog(walPath);<a name="line.383"></a>
+<span class="sourceLineNo">384</span> }<a name="line.384"></a>
+<span class="sourceLineNo">385</span> }<a name="line.385"></a>
+<span class="sourceLineNo">386</span> }<a name="line.386"></a>
+<span class="sourceLineNo">387</span> ReplicationPeerConfig peerConfig = peer.getPeerConfig();<a name="line.387"></a>
+<span class="sourceLineNo">388</span> if (peerConfig.isSyncReplication()) {<a name="line.388"></a>
+<span class="sourceLineNo">389</span> syncReplicationPeerMappingManager.add(peer.getId(), peerConfig);<a name="line.389"></a>
+<span class="sourceLineNo">390</span> }<a name="line.390"></a>
+<span class="sourceLineNo">391</span> src.startup();<a name="line.391"></a>
+<span class="sourceLineNo">392</span> return src;<a name="line.392"></a>
+<span class="sourceLineNo">393</span> }<a name="line.393"></a>
+<span class="sourceLineNo">394</span><a name="line.394"></a>
+<span class="sourceLineNo">395</span> /**<a name="line.395"></a>
+<span class="sourceLineNo">396</span> * <p><a name="line.396"></a>
+<span class="sourceLineNo">397</span> * This is used when we transit a sync replication peer to {@link SyncReplicationState#STANDBY}.<a name="line.397"></a>
+<span class="sourceLineNo">398</span> * </p><a name="line.398"></a>
+<span class="sourceLineNo">399</span> * <p><a name="line.399"></a>
+<span class="sourceLineNo">400</span> * When transiting to {@link SyncReplicationState#STANDBY}, we can remove all the pending wal<a name="line.400"></a>
+<span class="sourceLineNo">401</span> * files for a replication peer as we do not need to replicate them any more. And this is<a name="line.401"></a>
+<span class="sourceLineNo">402</span> * necessary, otherwise when we transit back to {@link SyncReplicationState#DOWNGRADE_ACTIVE}<a name="line.402"></a>
+<span class="sourceLineNo">403</span> * later, the stale data will be replicated again and cause inconsistency.<a name="line.403"></a>
+<span class="sourceLineNo">404</span> * </p><a name="line.404"></a>
+<span class="sourceLineNo">405</span> * <p><a name="line.405"></a>
+<span class="sourceLineNo">406</span> * See HBASE-20426 for more details.<a name="line.406"></a>
+<span class="sourceLineNo">407</span> * </p><a name="line.407"></a>
+<span class="sourceLineNo">408</span> * @param peerId the id of the sync replication peer<a name="line.408"></a>
+<span class="sourceLineNo">409</span> */<a name="line.409"></a>
+<span class="sourceLineNo">410</span> public void drainSources(String peerId) throws IOException, ReplicationException {<a name="line.410"></a>
+<span class="sourceLineNo">411</span> String terminateMessage = "Sync replication peer " + peerId +<a name="line.411"></a>
+<span class="sourceLineNo">412</span> " is transiting to STANDBY. Will close the previous replication source and open a new one";<a name="line.412"></a>
+<span class="sourceLineNo">413</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.413"></a>
+<span class="sourceLineNo">414</span> assert peer.getPeerConfig().isSyncReplication();<a name="line.414"></a>
+<span class="sourceLineNo">415</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.415"></a>
+<span class="sourceLineNo">416</span> // synchronized here to avoid race with preLogRoll where we add new log to source and also<a name="line.416"></a>
+<span class="sourceLineNo">417</span> // walsById.<a name="line.417"></a>
+<span class="sourceLineNo">418</span> ReplicationSourceInterface toRemove;<a name="line.418"></a>
+<span class="sourceLineNo">419</span> Map<String, NavigableSet<String>> wals = new HashMap<>();<a name="line.419"></a>
+<span class="sourceLineNo">420</span> synchronized (latestPaths) {<a name="line.420"></a>
+<span class="sourceLineNo">421</span> toRemove = sources.put(peerId, src);<a name="line.421"></a>
+<span class="sourceLineNo">422</span> if (toRemove != null) {<a name="line.422"></a>
+<span class="sourceLineNo">423</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.423"></a>
+<span class="sourceLineNo">424</span> toRemove.terminate(terminateMessage);<a name="line.424"></a>
+<span class="sourceLineNo">425</span> toRemove.getSourceMetrics().clear();<a name="line.425"></a>
+<span class="sourceLineNo">426</span> }<a name="line.426"></a>
+<span class="sourceLineNo">427</span> // Here we make a copy of all the remaining wal files and then delete them from the<a name="line.427"></a>
+<span class="sourceLineNo">428</span> // replication queue storage after releasing the lock. It is not safe to just remove the old<a name="line.428"></a>
+<span class="sourceLineNo">429</span> // map from walsById since later we may fail to delete them from the replication queue<a name="line.429"></a>
+<span class="sourceLineNo">430</span> // storage, and when we retry next time, we can not know the wal files that need to be deleted<a name="line.430"></a>
+<span class="sourceLineNo">431</span> // from the replication queue storage.<a name="line.431"></a>
+<span class="sourceLineNo">432</span> walsById.get(peerId).forEach((k, v) -> wals.put(k, new TreeSet<>(v)));<a name="line.432"></a>
+<span class="sourceLineNo">433</span> }<a name="line.433"></a>
+<span class="sourceLineNo">434</span> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.434"></a>
+<span class="sourceLineNo">435</span> src.startup();<a name="line.435"></a>
+<span class="sourceLineNo">436</span> for (NavigableSet<String> walsByGroup : wals.values()) {<a name="line.436"></a>
+<span class="sourceLineNo">437</span> for (String wal : walsByGroup) {<a name="line.437"></a>
+<span class="sourceLineNo">438</span> queueStorage.removeWAL(server.getServerName(), peerId, wal);<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> synchronized (walsById) {<a name="line.441"></a>
+<span class="sourceLineNo">442</span> Map<String, NavigableSet<String>> oldWals = walsById.get(peerId);<a name="line.442"></a>
+<span class="sourceLineNo">443</span> wals.forEach((k, v) -> {<a name="line.443"></a>
+<span class="sourceLineNo">444</span> NavigableSet<String> walsByGroup = oldWals.get(k);<a name="line.444"></a>
+<span class="sourceLineNo">445</span> if (walsByGroup != null) {<a name="line.445"></a>
+<span class="sourceLineNo">446</span> walsByGroup.removeAll(v);<a name="line.446"></a>
+<span class="sourceLineNo">447</span> }<a name="line.447"></a>
+<span class="sourceLineNo">448</span> });<a name="line.448"></a>
+<span class="sourceLineNo">449</span> }<a name="line.449"></a>
+<span class="sourceLineNo">450</span> // synchronized on oldsources to avoid race with NodeFailoverWorker. Since NodeFailoverWorker is<a name="line.450"></a>
+<span class="sourceLineNo">451</span> // a background task, we will delete the file from replication queue storage under the lock to<a name="line.451"></a>
+<span class="sourceLineNo">452</span> // simplify the logic.<a name="line.452"></a>
+<span class="sourceLineNo">453</span> synchronized (this.oldsources) {<a name="line.453"></a>
+<span class="sourceLineNo">454</span> for (Iterator<ReplicationSourceInterface> iter = oldsources.iterator(); iter.hasNext();) {<a name="line.454"></a>
+<span class="sourceLineNo">455</span> ReplicationSourceInterface oldSource = iter.next();<a name="line.455"></a>
+<span class="sourceLineNo">456</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.456"></a>
+<span class="sourceLineNo">457</span> String queueId = oldSource.getQueueId();<a name="line.457"></a>
+<span class="sourceLineNo">458</span> oldSource.terminate(terminateMessage);<a name="line.458"></a>
+<span class="sourceLineNo">459</span> oldSource.getSourceMetrics().clear();<a name="line.459"></a>
+<span class="sourceLineNo">460</span> queueStorage.removeQueue(server.getServerName(), queueId);<a name="line.460"></a>
+<span class="sourceLineNo">461</span> walsByIdRecoveredQueues.remove(queueId);<a name="line.461"></a>
+<span class="sourceLineNo">462</span> iter.remove();<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> }<a name="line.465"></a>
+<span class="sourceLineNo">466</span> }<a name="line.466"></a>
+<span class="sourceLineNo">467</span><a name="line.467"></a>
+<span class="sourceLineNo">468</span> /**<a name="line.468"></a>
+<span class="sourceLineNo">469</span> * Close the previous replication sources of this peer id and open new sources to trigger the new<a name="line.469"></a>
+<span class="sourceLineNo">470</span> * replication state changes or new replication config changes. Here we don't need to change<a name="line.470"></a>
+<span class="sourceLineNo">471</span> * replication queue storage and only to enqueue all logs to the new replication source<a name="line.471"></a>
+<span class="sourceLineNo">472</span> * @param peerId the id of the replication peer<a name="line.472"></a>
+<span class="sourceLineNo">473</span> */<a name="line.473"></a>
+<span class="sourceLineNo">474</span> public void refreshSources(String peerId) throws IOException {<a name="line.474"></a>
+<span class="sourceLineNo">475</span> String terminateMessage = "Peer " + peerId +<a name="line.475"></a>
+<span class="sourceLineNo">476</span> " state or config changed. Will close the previous replication source and open a new one";<a name="line.476"></a>
+<span class="sourceLineNo">477</span> ReplicationPeer peer = replicationPeers.getPeer(peerId);<a name="line.477"></a>
+<span class="sourceLineNo">478</span> ReplicationSourceInterface src = createSource(peerId, peer);<a name="line.478"></a>
+<span class="sourceLineNo">479</span> // synchronized on latestPaths to avoid missing the new log<a name="line.479"></a>
+<span class="sourceLineNo">480</span> synchronized (this.latestPaths) {<a name="line.480"></a>
+<span class="sourceLineNo">481</span> ReplicationSourceInterface toRemove = this.sources.put(peerId, src);<a name="line.481"></a>
+<span class="sourceLineNo">482</span> if (toRemove != null) {<a name="line.482"></a>
+<span class="sourceLineNo">483</span> LOG.info("Terminate replication source for " + toRemove.getPeerId());<a name="line.483"></a>
+<span class="sourceLineNo">484</span> toRemove.terminate(terminateMessage);<a name="line.484"></a>
+<span class="sourceLineNo">485</span> }<a name="line.485"></a>
+<span class="sourceLineNo">486</span> for (NavigableSet<String> walsByGroup : walsById.get(peerId).values()) {<a name="line.486"></a>
+<span class="sourceLineNo">487</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(this.logDir, wal)));<a name="line.487"></a>
+<span class="sourceLineNo">488</span> }<a name="line.488"></a>
+<span class="sourceLineNo">489</span> }<a name="line.489"></a>
+<span class="sourceLineNo">490</span> LOG.info("Startup replication source for " + src.getPeerId());<a name="line.490"></a>
+<span class="sourceLineNo">491</span> src.startup();<a name="line.491"></a>
+<span class="sourceLineNo">492</span><a name="line.492"></a>
+<span class="sourceLineNo">493</span> List<ReplicationSourceInterface> toStartup = new ArrayList<>();<a name="line.493"></a>
+<span class="sourceLineNo">494</span> // synchronized on oldsources to avoid race with NodeFailoverWorker<a name="line.494"></a>
+<span class="sourceLineNo">495</span> synchronized (this.oldsources) {<a name="line.495"></a>
+<span class="sourceLineNo">496</span> List<String> previousQueueIds = new ArrayList<>();<a name="line.496"></a>
+<span class="sourceLineNo">497</span> for (ReplicationSourceInterface oldSource : this.oldsources) {<a name="line.497"></a>
+<span class="sourceLineNo">498</span> if (oldSource.getPeerId().equals(peerId)) {<a name="line.498"></a>
+<span class="sourceLineNo">499</span> previousQueueIds.add(oldSource.getQueueId());<a name="line.499"></a>
+<span class="sourceLineNo">500</span> oldSource.terminate(terminateMessage);<a name="line.500"></a>
+<span class="sourceLineNo">501</span> this.oldsources.remove(oldSource);<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> for (String queueId : previousQueueIds) {<a name="line.504"></a>
+<span class="sourceLineNo">505</span> ReplicationSourceInterface replicationSource = createSource(queueId, peer);<a name="line.505"></a>
+<span class="sourceLineNo">506</span> this.oldsources.add(replicationSource);<a name="line.506"></a>
+<span class="sourceLineNo">507</span> for (SortedSet<String> walsByGroup : walsByIdRecoveredQueues.get(queueId).values()) {<a name="line.507"></a>
+<span class="sourceLineNo">508</span> walsByGroup.forEach(wal -> src.enqueueLog(new Path(wal)));<a name="line.508"></a>
+<span class="sourceLineNo">509</span> }<a name="line.509"></a>
+<span class="sourceLineNo">510</span> toStartup.add(replicationSource);<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> for (ReplicationSourceInterface replicationSource : toStartup) {<a name="line.513"></a>
+<span class="sourceLineNo">514</span> replicationSource.startup();<a name="line.514"></a>
+<span class="sourceLineNo">515</span> }<a name="line.515"></a>
+<span class="sourceLineNo">516</span> }<a name="line.516"></a>
+<span class="sourceLineNo">517</span><a name="line.517"></a>
+<span class="sourceLineNo">518</span> /**<a name="line.518"></a>
+<span class="sourceLineNo">519</span> * Clear the metrics and related replication queue of the specified old source<a name="line.519"></a>
+<span class="sourceLineNo">520</span> * @param src source to clear<a name="line.520"></a>
+<span class="sourceLineNo">521</span> */<a name="line.521"></a>
+<span class="sourceLineNo">522</span> private boolean removeRecoveredSource(ReplicationSourceInterface src) {<a name="line.522"></a>
+<span class="sourceLineNo">523</span> if (!this.oldsources.remove(src)) {<a name="line.523"></a>
+<span class="sourceLineNo">524</span> return false;<a name="line.524"></a>
+<span class="sourceLineNo">525</span> }<a name="line.525"></a>
+<span class="sourceLineNo">526</span> LOG.info("Done with the recovered queue {}", src.getQueueId());<a name="line.526"></a>
+<span class="sourceLineNo">527</span> // Delete queue from storage and memory<a name="line.527"></a>
+<span class="sourceLineNo">528</span> deleteQueue(src.getQueueId());<a name="line.528"></a>
+<span class="sourceLineNo">529</span> this.walsByIdRecoveredQueues.remove(src.getQueueId());<a name="line.529"></a>
+<span class="sourceLineNo">530</span> return true;<a name="line.530"></a>
+<span class="sourceLineNo">531</span> }<a name="line.531"></a>
+<span class="sourceLineNo">532</span><a name="line.532"></a>
+<span class="sourceLineNo">533</span> void finishRecoveredSource(ReplicationSourceInterface src) {<a name="line.533"></a>
+<span class="sourceLineNo">534</span> synchronized (oldsources) {<a name="line.534"></a>
+<span class="sourceLineNo">535</span> if (!removeRecoveredSource(src)) {<a name="line.535"></a>
+<span class="sourceLineNo">536</span> return;<a name="line.536"></a>
+<span class="sourceLineNo">537</span> }<a name="line.537"></a>
+<span class="sourceLineNo">538</span> }<a name="line.538"></a>
+<span class="sourceLineNo">539</span> LOG.info("Finished recovering queue {} with the following stats: {}", src.getQueueId(),<a name="line.539"></a>
+<span class="sourceLineNo">540</span> src.getStats());<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> /**<a name="line.543"></a>
+<span class="sourceLineNo">544</span> * Clear the metrics and related replication queue of the specified old source<a name="line.544"></a>
+<span class="sourceLineNo">545</span> * @param src source to clear<a name="line.545"></a>
+<span class="sourceLineNo">546</span> */<a name="line.546"></a>
+<span class="sourceLineNo">547</span> void removeSource(ReplicationSourceInterface src) {<a name="line.547"></a>
+<span class="sourceLineNo">548</span> LOG.info("Done with the queue " + src.getQueueId());<a name="line.548"></a>
+<span class="sourceLineNo">549</span> this.sources.remove(src.getPeerId());<a name="line.549"></a>
+<span class="sourceLineNo">550</span> // Delete queue from storage and memory<a name="line.550"></a>
+<span class="sourceLineNo">551</span> deleteQueue(src.getQueueId());<a name="line.551"></a>
+<span class="sourceLineNo">552</span> this.walsById.remove(src.getQueueId());<a name="line.552"></a>
+<span class="sourceLineNo">553</span><a name="line.553"></a>
+<span class="sourceLineNo">554</span> }<a name="line.554"></a>
+<span class="sourceLineNo">555</span><a name="line.555"></a>
+<span class="sourceLineNo">556</span> /**<a name="line.556"></a>
+<span class="sourceLineNo">557</span> * Delete a complete queue of wals associated with a replication source<a name="line.557"></a>
+<span class="sourceLineNo">558</span> * @param queueId the id of replication queue to delete<a name="line.558"></a>
+<span class="sourceLineNo">559</span> */<a name="line.559"></a>
+<span class="sourceLineNo">560</span> private void deleteQueue(String queueId) {<a name="line.560"></a>
+<span class="sourceLineNo">561</span> abortWhenFail(() -> this.queueStorage.removeQueue(server.getServerName(), queueId));<a name="line.561"></a>
+<span class="sourceLineNo">562</span> }<a name="line.562"></a>
+<span class="sourceLineNo">563</span><a name="line.563"></a>
+<span class="sourceLineNo">564</span> @FunctionalInterface<a name="line.564"></a>
+<span class="sourceLineNo">565</span> private interface ReplicationQueueOperation {<a name="line.565"></a>
+<span class="sourceLineNo">566</span> void exec() throws ReplicationException;<a name="line.566"></a>
+<span class="sourceLineNo">567</span> }<a name="line.567"></a>
+<span class="sourceLineNo">568</span><a name="line.568"></a>
+<span class="sourceLineNo">569</span> /**<a name="line.569"></a>
+<span class="sourceLineNo">570</span> * Refresh replication source will terminate the old source first, then the source thread will be<a name="line.570"></a>
+<span class="sourceLineNo">571</span> * interrupted. Need to handle it instead of abort the region server.<a name="line.571"></a>
+<span class="sourceLineNo">572</span> */<a name="line.572"></a>
+<span class="sourceLineNo">573</span> private void interruptOrAbortWhenFail(ReplicationQueueOperation op) {<a name="line.573"></a>
+<span class="sourceLineNo">574</span> try {<a name="line.574"></a>
+<span class="sourceLineNo">575</span> op.exec();<a name="line.575"></a>
+<span class="sourceLineNo">576</span> } catch (ReplicationException e) {<a name="line.576"></a>
+<span class="sourceLineNo">577</span> if (e.getCause() != null && e.getCause() instanceof KeeperException.SystemErrorException<a name="line.577"></a>
+<span class="sourceLineNo">578</span> && e.getCause().getCause() != null && e.getCause()<a name="line.578"></a>
+<span class="sourceLineNo">579</span> .getCause() instanceof InterruptedException) {<a name="line.579"></a>
+<span class="sourceLineNo">580</span> throw new RuntimeException(<a name="line.580"></a>
+<span class="sourceLineNo">581</span> "Thread is interrupted, the replication source may be terminated");<a name="line.581"></a>
+<span class="sourceLineNo">582</span> }<a name="line.582"></a>
+<span class="sourceLineNo">583</span> server.abort("Failed to operate on replication queue", e);<a name="line.583"></a>
+<span class="sourceLineNo">584</span> }<a name="line.584"></a>
+<span class="sourceLineNo">585</span> }<a name="line.585"></a>
+<span class="sourceLineNo">586</span><a name="line.586"></a>
+<span class="sourceLineNo">587</span> private void abortWhenFail(ReplicationQueueOperation op) {<a name="line.587"></a>
+<span class="sourceLineNo">588</span> try {<a name="line.588"></a>
+<span class="sourceLineNo">589</span> op.exec();<a name="line.589"></a>
+<span class="sourceLineNo">590</span> } catch (ReplicationException e) {<a name="line.590"></a>
+<span class="sourceLineNo">591</span> server.abort("Failed to operate on replication queue", e);<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><a name="line.594"></a>
+<span class="sourceLineNo">595</span> private void throwIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.595"></a>
+<span class="sourceLineNo">596</span> try {<a name="line.596"></a>
+<span class="sourceLineNo">597</span> op.exec();<a name="line.597"></a>
+<span class="sourceLineNo">598</span> } catch (ReplicationException e) {<a name="line.598"></a>
+<span class="sourceLineNo">599</span> throw new IOException(e);<a name="line.599"></a>
+<span class="sourceLineNo">600</span> }<a name="line.600"></a>
+<span class="sourceLineNo">601</span> }<a name="line.601"></a>
+<span class="sourceLineNo">602</span><a name="line.602"></a>
+<span class="sourceLineNo">603</span> private void abortAndThrowIOExceptionWhenFail(ReplicationQueueOperation op) throws IOException {<a name="line.603"></a>
+<span class="sourceLineNo">604</span> try {<a name="line.604"></a>
+<span class="sourceLineNo">605</span> op.exec();<a name="line.605"></a>
+<span class="sourceLineNo">606</span> } catch (ReplicationException e) {<a name="line.606"></a>
+<span class="sourceLineNo">607</span> server.abort("Failed to operate on replication queue", e);<a name="line.607"></a>
+<span class="sourceLineNo">608</span> throw new IOException(e);<a name="line.608"></a>
+<span class="sourceLineNo">609</span> }<a name="line.609"></a>
+<span class="sourceLineNo">610</span> }<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> * This method will log the current position to storage. And also clean old logs from the<a name="line.613"></a>
+<span class="sourceLineNo">614</span> * replication queue.<a name="line.614"></a>
+<span class="sourceLineNo">615</span> * @param source the replication source<a name="line.615"></a>
+<span class="sourceLineNo">616</span> * @param entryBatch the wal entry batch we just shipped<a name="line.616"></a>
+<span class="sourceLineNo">617</span> */<a name="line.617"></a>
+<span class="sourceLineNo">618</span> public void logPositionAndCleanOldLogs(ReplicationSourceInterface source,<a name="line.618"></a>
+<span class="sourceLineNo">619</span> WALEntryBatch entryBatch) {<a name="line.619"></a>
+<span class="sourceLineNo">620</span> String fileName = entryBatch.getLastWalPath().getName();<a name="line.620"></a>
+<span class="sourceLineNo">621</span> interruptOrAbortWhenFail(() -> this.queueStorage.setWALPosition(server.getServerName(),<a name="line.621"></a>
+<span class="sourceLineNo">622</span> source.getQueueId(), fileName, entryBatch.getLastWalPosition(), entryBatch.getLastSeqIds()));<a name="line.622"></a>
+<span class="sourceLineNo">623</span> cleanOldLogs(fileName, entryBatch.isEndOfFile(), source);<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">626</span> /**<a name="line.626"></a>
+<span class="sourceLineNo">627</span> * Cleans a log file and all older logs from replication queue. Called when we are sure that a log<a name="line.627"></a>
+<span class="sourceLineNo">628</span> * file is closed and has no more entries.<a name="line.628"></a>
+<span class="sourceLineNo">629</span> * @param log Path to the log<a name="line.629"></a>
+<span class="sourceLineNo">630</span> * @param inclusive whether we should also remove the given log file<a name="line.630"></a>
+<span class="sourceLineNo">631</span> * @param source the replication source<a name="line.631"></a>
+<span class="sourceLineNo">632</span> */<a name="line.632"></a>
+<span class="sourceLineNo">633</span> @VisibleForTesting<a name="line.633"></a>
+<span class="sourceLineNo">634</span> void cleanOldLogs(String log, boolean inclusive, ReplicationSourceInterface source) {<a name="line.634"></a>
+<span class="sourceLineNo">635</span> String logPrefix = AbstractFSWALProvider.getWALPrefixFromWALName(log);<a name="line.635"></a>
+<span class="sourceLineNo">636</span> if (source.isRecovered()) {<a name="line.636"></a>
+<span class="sourceLineNo">637</span> NavigableSet<String> wals = walsByIdRecoveredQueues.get(source.getQueueId()).get(logPrefix);<a name="line.637"></a>
+<span class="sourceLineNo">638</span> if (wals != null) {<a name="line.638"></a>
+<span class="sourceLineNo">639</span> NavigableSet<String> walsToRemove = wals.headSet(log, inclusive);<a name="line.639"></a>
+<span class="sourceLineNo">640</span> if (walsToRemove.isEmpty()) {<a name="line.640"></a>
+<span class="sourceLineNo">641</span> return;<a name="line.641"></a>
+<span class="sourceLineNo">642</span> }<a name="line.642"></a>
+<span class="sourceLineNo">643</span> cleanOldLogs(walsToRemove, source);<a name="line.643"></a>
+<span class="sourceLineNo">644</span> walsToRemove.clear();<a name="line.644"></a>
+<span class="sourceLineNo">645</span> }<a name="line.645"></a>
+<span class="sourceLineNo">646</span> } else {<a name="line.646"></a>
+<span class="sourceLineNo">647</span> NavigableSet<String> wals;<a name="line.647"></a>
+<span class="sourceLineNo">648</span> NavigableSet<String> walsToRemove;<a name="line.648"></a>
+<span class="sourceLineNo">649</span> // synchronized on walsById to avoid race with preLogRoll<a name="line.649"></a>
+<span class="sourceLineNo">650</span> synchronized (this.walsById) {<a name="line.650"></a>
+<span class="sourceLineNo">651</span> wals = walsById.get(source.getQueueId()).get(logPrefix);<a name="line.651"></a>
+<span class="sourceLineNo">652</span> if (wals == null) {<a name="line.652"></a>
+<span class="sourceLineNo">653</span> return;<a name="line.653"></a>
+<span class="sourceLineNo">654</span> }<a name="line.654"></a>
+<span class="sourceLineNo">655</span> walsToRemove = wals.headSet(log, inclusive);<a name="line.655"></a>
+<span class="sourceLineNo">656</span> if (walsToRemove.isEmpty()) {<a name="line.656"></a>
+<span class="sourceLineNo">657</span> return;<a name="line.657"></a>
+<span class="sourceLineNo">658</span> }<a name="line.658"></a>
+<span class="sourceLineNo">659</span> walsToRemove = new TreeSet<>(walsToRemove);<a name="line.659"></a>
+<span class="sourceLineNo">660</span> }<a name="line.660"></a>
+<span class="sourceLineNo">661</span> // cleanOldLogs may spend some time, especially for sync replication where we may want to<a name="line.661"></a>
+<span class="sourceLineNo">662</span> // remove remote wals as the remote cluster may have already been down, so we do it outside<a name="line.662"></a>
+<span class="sourceLineNo">663</span> // the lock to avoid block preLogRoll<a name="line.663"></a>
+<span class="sourceLineNo">664</span> cleanOldLogs(walsToRemove, source);<a name="line.664"></a>
+<span class="sourceLineNo">665</span> // now let's remove the files in the set<a name="line.665"></a>
+<span class="sourceLineNo">666</span> synchronized (this.walsById) {<a name="line.666"></a>
+<span class="sourceLineNo">667</span> wals.removeAll(walsToRemove);<a name="line.667"></a>
+<span class="sourceLineNo">668</span> }<a name="line.668"></a>
+<span class="sourceLineNo">669</span> }<a name="line.669"></a>
+<span class="sourceLineNo">670</span> }<a name="line.670"></a>
+<span class="sourceLineNo">671</span><a name="line.671"></a>
+<span class="sourceLineNo">672</span> private void removeRemoteWALs(String peerId, String remoteWALDir, Collection<String> wals)<a name="line.672"></a>
+<span class="sourceLineNo">673</span> throws IOException {<a name="line.673"></a>
+<span class="sourceLineNo">674</span> Path remoteWALDirForPeer = ReplicationUtils.getPeerRemoteWALDir(remoteWALDir, peerId);<a name="line.674"></a>
+<span class="sourceLineNo">675</span> FileSystem fs = ReplicationUtils.getRemoteWALFileSystem(conf, remoteWALDir);<a name="line.675"></a>
+<span class="sourceLineNo">676</span> for (String wal : wals) {<a name="line.676"></a>
+<span class="sourceLineNo">677</span> Path walFile = new Path(remoteWALDirForPeer, wal);<a name="line.677"></a>
+<span class="sourceLineNo">678</span> try {<a name="line.678"></a>
+<span class="sourceLineNo">679</span> if (!fs.delete(walFile, false) && fs.exists(walFile)) {<a name="line.679"></a>
+<span class="sourceLineNo">680</span> throw new IOException("Can not delete " + walFile);<a name="line.680"></a>
+<span class="sourceLineNo">681</span> }<a name="line.681"></a>
+<span class="sourceLineNo">682</span> } catch (FileNotFoundException e) {<a name="line.682"></a>
+<span class="sourceLineNo">683</span> // Just ignore since this means the file has already been deleted.<a name="line.683"></a>
+<span class="sourceLineNo">684</span> // The javadoc of the FileSystem.delete methods does not specify the behavior of deleting an<a name="line.684"></a>
+<span class="sourceLineNo">685</span> // inexistent file, so here we deal with both, i.e, check the return value of the<a name="line.685"></a>
+<span class="sourceLineNo">686</span> // FileSystem.delete, and also catch FNFE.<a name="line.686"></a>
+<span class="sourceLineNo">687</span> LOG.debug("The remote wal {} has already been deleted?", walFile, e);<a name="line.687"></a>
+<span class="sourceLineNo">688</span> }<a name="line.688"
<TRUNCATED>