Merge "Add more info to ShardStats JXM bean"
[controller.git] / opendaylight / md-sal / sal-akka-raft / src / main / java / org / opendaylight / controller / cluster / raft / behaviors / AbstractLeader.java
index be51ba069cc5056636646566d1db00b30154073a..a63c62fa30740b5830676ab6f15f3de9c1988e7b 100644 (file)
@@ -97,9 +97,7 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
         final Builder<String, FollowerLogInformation> ftlBuilder = ImmutableMap.builder();
         for (String followerId : context.getPeerAddresses().keySet()) {
             FollowerLogInformation followerLogInformation =
-                new FollowerLogInformationImpl(followerId,
-                    context.getCommitIndex(), -1,
-                    context.getConfigParams().getElectionTimeOutInterval());
+                new FollowerLogInformationImpl(followerId, -1, context);
 
             ftlBuilder.put(followerId, followerLogInformation);
         }
@@ -136,7 +134,7 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
      *
      * @return Collection of follower IDs
      */
-    protected final Collection<String> getFollowerIds() {
+    public final Collection<String> getFollowerIds() {
         return followerToLog.keySet();
     }
 
@@ -160,8 +158,6 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
 
         if(LOG.isTraceEnabled()) {
             LOG.trace("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
-        } else if(LOG.isDebugEnabled() && !appendEntriesReply.isSuccess()) {
-            LOG.debug("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
         }
 
         // Update the FollowerLogInformation
@@ -176,7 +172,7 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
 
         if(followerLogInformation.timeSinceLastActivity() >
                 context.getConfigParams().getElectionTimeOutInterval().toMillis()) {
-            LOG.error("{} : handleAppendEntriesReply delayed beyond election timeout, " +
+            LOG.warn("{} : handleAppendEntriesReply delayed beyond election timeout, " +
                             "appendEntriesReply : {}, timeSinceLastActivity : {}, lastApplied : {}, commitIndex : {}",
                     logName(), appendEntriesReply, followerLogInformation.timeSinceLastActivity(),
                     context.getLastApplied(), context.getCommitIndex());
@@ -184,12 +180,17 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
 
         followerLogInformation.markFollowerActive();
 
+        boolean updated = false;
         if (appendEntriesReply.isSuccess()) {
-            followerLogInformation
-                .setMatchIndex(appendEntriesReply.getLogLastIndex());
-            followerLogInformation
-                .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
+            updated = followerLogInformation.setMatchIndex(appendEntriesReply.getLogLastIndex());
+            updated = followerLogInformation.setNextIndex(appendEntriesReply.getLogLastIndex() + 1) || updated;
+
+            if(updated && LOG.isDebugEnabled()) {
+                LOG.debug("{}: handleAppendEntriesReply - FollowerLogInformation for {} updated: matchIndex: {}, nextIndex: {}", logName(),
+                        followerId, followerLogInformation.getMatchIndex(), followerLogInformation.getNextIndex());
+            }
         } else {
+            LOG.debug("{}: handleAppendEntriesReply: received unsuccessful reply: {}", logName(), appendEntriesReply);
 
             // TODO: When we find that the follower is out of sync with the
             // Leader we simply decrement that followers next index by 1.
@@ -226,8 +227,10 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
 
         // Apply the change to the state machine
         if (context.getCommitIndex() > context.getLastApplied()) {
-            LOG.debug("{}: handleAppendEntriesReply: applying to log - commitIndex: {}, lastAppliedIndex: {}",
-                    logName(), context.getCommitIndex(), context.getLastApplied());
+            if(LOG.isDebugEnabled()) {
+                LOG.debug("{}: handleAppendEntriesReply from {}: applying to log - commitIndex: {}, lastAppliedIndex: {}",
+                        logName(), followerId, context.getCommitIndex(), context.getLastApplied());
+            }
 
             applyLogToStateMachine(context.getCommitIndex());
         }
@@ -237,7 +240,7 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
         }
 
         //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
-        sendUpdatesToFollower(followerId, followerLogInformation, false, false);
+        sendUpdatesToFollower(followerId, followerLogInformation, false, !updated);
         return this;
     }
 
@@ -385,7 +388,10 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
                 followerToSnapshot.markSendStatus(false);
             }
 
-            if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
+            if (wasLastChunk && !context.isSnapshotCaptureInitiated()) {
+                // Since the follower is now caught up try to purge the log.
+                purgeInMemoryLog();
+            } else if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
                 ActorSelection followerActor = context.getPeerActorSelection(followerId);
                 if(followerActor != null) {
                     sendSnapshotChunk(followerActor, followerId);
@@ -454,6 +460,8 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
         if (followerActor != null) {
             long followerNextIndex = followerLogInformation.getNextIndex();
             boolean isFollowerActive = followerLogInformation.isFollowerActive();
+            boolean sendAppendEntries = false;
+            List<ReplicatedLogEntry> entries = Collections.EMPTY_LIST;
 
             if (mapFollowerToSnapshot.get(followerId) != null) {
                 // if install snapshot is in process , then sent next chunk if possible
@@ -461,16 +469,15 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
                     sendSnapshotChunk(followerActor, followerId);
                 } else if(sendHeartbeat) {
                     // we send a heartbeat even if we have not received a reply for the last chunk
-                    sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
-                        Collections.<ReplicatedLogEntry>emptyList(), followerId);
+                    sendAppendEntries = true;
                 }
             } else {
                 long leaderLastIndex = context.getReplicatedLog().lastIndex();
                 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 
-                if(!isHeartbeat || LOG.isTraceEnabled()) {
-                    LOG.debug("{}: Checking sendAppendEntries for follower {}, leaderLastIndex: {}, leaderSnapShotIndex: {}",
-                            logName(), followerId, leaderLastIndex, leaderSnapShotIndex);
+                if((!isHeartbeat && LOG.isDebugEnabled()) || LOG.isTraceEnabled()) {
+                    LOG.debug("{}: Checking sendAppendEntries for follower {}, followerNextIndex {}, leaderLastIndex: {}, leaderSnapShotIndex: {}",
+                            logName(), followerId, followerNextIndex, leaderLastIndex, leaderSnapShotIndex);
                 }
 
                 if (isFollowerActive && context.getReplicatedLog().isPresent(followerNextIndex)) {
@@ -479,10 +486,10 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
                             followerNextIndex, followerId);
 
                     // FIXME : Sending one entry at a time
-                    final List<ReplicatedLogEntry> entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
-
-                    sendAppendEntriesToFollower(followerActor, followerNextIndex, entries, followerId);
-
+                    if(followerLogInformation.okToReplicate()) {
+                        entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
+                        sendAppendEntries = true;
+                    }
                 } else if (isFollowerActive && followerNextIndex >= 0 &&
                     leaderLastIndex > followerNextIndex && !context.isSnapshotCaptureInitiated()) {
                     // if the followers next index is not present in the leaders log, and
@@ -497,19 +504,21 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
                     }
 
                     // Send heartbeat to follower whenever install snapshot is initiated.
-                    sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
-                            Collections.<ReplicatedLogEntry>emptyList(), followerId);
-
+                    sendAppendEntries = true;
                     initiateCaptureSnapshot(followerId, followerNextIndex);
 
                 } else if(sendHeartbeat) {
-                    //we send an AppendEntries, even if the follower is inactive
+                    // we send an AppendEntries, even if the follower is inactive
                     // in-order to update the followers timestamp, in case it becomes active again
-                    sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
-                        Collections.<ReplicatedLogEntry>emptyList(), followerId);
+                    sendAppendEntries = true;
                 }
 
             }
+
+            if(sendAppendEntries) {
+                sendAppendEntriesToFollower(followerActor, followerNextIndex,
+                        entries, followerId);
+            }
         }
     }
 
@@ -555,7 +564,6 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
 
             } else if (!context.isSnapshotCaptureInitiated()) {
 
-                LOG.info("{}: Initiating Snapshot Capture to Install Snapshot, Leader:{}", logName(), getLeaderId());
                 ReplicatedLogEntry lastAppliedEntry = context.getReplicatedLog().get(context.getLastApplied());
                 long lastAppliedIndex = -1;
                 long lastAppliedTerm = -1;
@@ -571,10 +579,19 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
                 boolean isInstallSnapshotInitiated = true;
                 long replicatedToAllIndex = super.getReplicatedToAllIndex();
                 ReplicatedLogEntry replicatedToAllEntry = context.getReplicatedLog().get(replicatedToAllIndex);
-                actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm,
-                    (replicatedToAllEntry != null ? replicatedToAllEntry.getIndex() : -1),
-                    (replicatedToAllEntry != null ? replicatedToAllEntry.getTerm() : -1),
-                    isInstallSnapshotInitiated), actor());
+
+                CaptureSnapshot captureSnapshot = new CaptureSnapshot(
+                        lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm,
+                        (replicatedToAllEntry != null ? replicatedToAllEntry.getIndex() : -1),
+                        (replicatedToAllEntry != null ? replicatedToAllEntry.getTerm() : -1),
+                        isInstallSnapshotInitiated);
+
+                if(LOG.isDebugEnabled()) {
+                    LOG.debug("{}: Initiating install snapshot to follower {}: {}", logName(), followerId,
+                            captureSnapshot);
+                }
+
+                actor().tell(captureSnapshot, actor());
                 context.setSnapshotCaptureInitiated(true);
             }
         }
@@ -621,10 +638,12 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior {
                     ).toSerializable(),
                     actor()
                 );
-                LOG.info("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
-                        logName(), followerActor.path(),
-                        followerToSnapshot.getChunkIndex(),
-                        followerToSnapshot.getTotalChunks());
+
+                if(LOG.isDebugEnabled()) {
+                    LOG.debug("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
+                            logName(), followerActor.path(), followerToSnapshot.getChunkIndex(),
+                            followerToSnapshot.getTotalChunks());
+                }
             }
         } catch (IOException e) {
             LOG.error("{}: InstallSnapshot failed for Leader.", logName(), e);