X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?p=controller.git;a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-akka-raft%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fraft%2Fbehaviors%2FAbstractLeader.java;h=31464c5aff818c7e44bcc2a42527db1af048c061;hp=da1627b98e7e4e8204385914795300ce073a71ad;hb=726ee824671781b5031c0108794c22bd0d96eaad;hpb=58c031c8fbec9f8d7e05eac60c1bea168d3471f6 diff --git a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/AbstractLeader.java b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/AbstractLeader.java index da1627b98e..31464c5aff 100644 --- a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/AbstractLeader.java +++ b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/AbstractLeader.java @@ -93,6 +93,8 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { private Optional snapshot; + private long replicatedToAllIndex = -1; + public AbstractLeader(RaftActorContext context) { super(context); @@ -226,9 +228,27 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { applyLogToStateMachine(context.getCommitIndex()); } + if (!context.isSnapshotCaptureInitiated()) { + purgeInMemoryLog(); + } + + //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event + sendUpdatesToFollower(followerId, followerLogInformation, false); return this; } + private void purgeInMemoryLog() { + //find the lowest index across followers which has been replicated to all. -1 if there are no followers. + // we would delete the in-mem log from that index on, in-order to minimize mem usage + // we would also share this info thru AE with the followers so that they can delete their log entries as well. + long minReplicatedToAllIndex = followerToLog.isEmpty() ? -1 : Long.MAX_VALUE; + for (FollowerLogInformation info : followerToLog.values()) { + minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex()); + } + + replicatedToAllIndex = fakeSnapshot(minReplicatedToAllIndex, replicatedToAllIndex); + } + @Override protected ClientRequestTracker removeClientRequestTracker(long logIndex) { final Iterator it = trackerList.iterator(); @@ -276,6 +296,9 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { // set currentTerm = T, convert to follower (§5.1) // This applies to all RPC messages and responses if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) { + LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {}", context.getId(), + rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm()); + context.getTermInformation().updateAndPersist(rpc.getTerm(), null); return switchBehavior(new Follower(context)); @@ -312,12 +335,18 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { private void handleInstallSnapshotReply(InstallSnapshotReply reply) { String followerId = reply.getFollowerId(); FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId); + + if (followerToSnapshot == null) { + LOG.error("{}: FollowerId {} in InstallSnapshotReply not known to Leader", + context.getId(), followerId); + return; + } + FollowerLogInformation followerLogInformation = followerToLog.get(followerId); followerLogInformation.markFollowerActive(); - if (followerToSnapshot != null && - followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) { - + if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) { + boolean wasLastChunk = false; if (reply.isSuccess()) { if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) { //this was the last chunk reply @@ -345,6 +374,7 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { // we can remove snapshot from the memory setSnapshot(Optional.absent()); } + wasLastChunk = true; } else { followerToSnapshot.markSendStatus(true); @@ -356,11 +386,17 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { followerToSnapshot.markSendStatus(false); } + if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) { + ActorSelection followerActor = context.getPeerActorSelection(followerId); + if(followerActor != null) { + sendSnapshotChunk(followerActor, followerId); + } + } + } else { - LOG.error("{}: FollowerId in InstallSnapshotReply not known to Leader" + - " or Chunk Index in InstallSnapshotReply not matching {} != {}", - context.getId(), followerToSnapshot.getChunkIndex(), reply.getChunkIndex() - ); + LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}", + context.getId(), reply.getChunkIndex(), followerId, + followerToSnapshot.getChunkIndex()); if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){ // Since the Follower did not find this index to be valid we should reset the follower snapshot @@ -395,74 +431,94 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { private void sendAppendEntries() { // Send an AppendEntries to all followers + long heartbeatInterval = context.getConfigParams().getHeartBeatInterval().toMillis(); for (Entry e : followerToLog.entrySet()) { final String followerId = e.getKey(); - ActorSelection followerActor = context.getPeerActorSelection(followerId); + final FollowerLogInformation followerLogInformation = e.getValue(); + // This checks helps not to send a repeat message to the follower + if(followerLogInformation.timeSinceLastActivity() >= heartbeatInterval) { + sendUpdatesToFollower(followerId, followerLogInformation, true); + } + } + } - if (followerActor != null) { - FollowerLogInformation followerLogInformation = followerToLog.get(followerId); - long followerNextIndex = followerLogInformation.getNextIndex(); - boolean isFollowerActive = followerLogInformation.isFollowerActive(); - - if (mapFollowerToSnapshot.get(followerId) != null) { - // if install snapshot is in process , then sent next chunk if possible - if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) { - sendSnapshotChunk(followerActor, followerId); - } else { - // we send a heartbeat even if we have not received a reply for the last chunk - sendAppendEntriesToFollower(followerActor, followerNextIndex, - Collections.emptyList()); - } + /** + * + * This method checks if any update needs to be sent to the given follower. This includes append log entries, + * sending next snapshot chunk, and initiating a snapshot. + * @return true if any update is sent, false otherwise + */ - } else { - long leaderLastIndex = context.getReplicatedLog().lastIndex(); - long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex(); - final List entries; - - if (isFollowerActive && - context.getReplicatedLog().isPresent(followerNextIndex)) { - // FIXME : Sending one entry at a time - entries = context.getReplicatedLog().getFrom(followerNextIndex, 1); - - } else if (isFollowerActive && followerNextIndex >= 0 && - leaderLastIndex >= followerNextIndex ) { - // if the followers next index is not present in the leaders log, and - // if the follower is just not starting and if leader's index is more than followers index - // then snapshot should be sent - - if(LOG.isDebugEnabled()) { - LOG.debug(String.format("%s: InitiateInstallSnapshot to follower: %s," + - "follower-nextIndex: %s, leader-snapshot-index: %s, " + - "leader-last-index: %s", context.getId(), followerId, - followerNextIndex, leaderSnapShotIndex, leaderLastIndex)); - } - actor().tell(new InitiateInstallSnapshot(), actor()); - - // we would want to sent AE as the capture snapshot might take time - entries = Collections.emptyList(); - - } else { - //we send an AppendEntries, even if the follower is inactive - // in-order to update the followers timestamp, in case it becomes active again - entries = Collections.emptyList(); + private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation, + boolean sendHeartbeat) { + + ActorSelection followerActor = context.getPeerActorSelection(followerId); + if (followerActor != null) { + long followerNextIndex = followerLogInformation.getNextIndex(); + boolean isFollowerActive = followerLogInformation.isFollowerActive(); + + if (mapFollowerToSnapshot.get(followerId) != null) { + // if install snapshot is in process , then sent next chunk if possible + if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) { + 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.emptyList(), followerId); + } + } else { + long leaderLastIndex = context.getReplicatedLog().lastIndex(); + long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex(); + if (isFollowerActive && + context.getReplicatedLog().isPresent(followerNextIndex)) { + // FIXME : Sending one entry at a time + final List entries = context.getReplicatedLog().getFrom(followerNextIndex, 1); + + sendAppendEntriesToFollower(followerActor, followerNextIndex, entries, followerId); + + } else if (isFollowerActive && followerNextIndex >= 0 && + leaderLastIndex >= followerNextIndex) { + // if the followers next index is not present in the leaders log, and + // if the follower is just not starting and if leader's index is more than followers index + // then snapshot should be sent + + if (LOG.isDebugEnabled()) { + LOG.debug("InitiateInstallSnapshot to follower:{}," + + "follower-nextIndex:{}, leader-snapshot-index:{}, " + + "leader-last-index:{}", followerId, + followerNextIndex, leaderSnapShotIndex, leaderLastIndex + ); } + actor().tell(new InitiateInstallSnapshot(), actor()); - sendAppendEntriesToFollower(followerActor, followerNextIndex, entries); + // Send heartbeat to follower whenever install snapshot is initiated. + sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(), + Collections.emptyList(), followerId); + } else if(sendHeartbeat) { + //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.emptyList(), followerId); } + } } } private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex, - List entries) { - followerActor.tell( - new AppendEntries(currentTerm(), context.getId(), - prevLogIndex(followerNextIndex), - prevLogTerm(followerNextIndex), entries, - context.getCommitIndex()).toSerializable(), - actor() - ); + List entries, String followerId) { + AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(), + prevLogIndex(followerNextIndex), + prevLogTerm(followerNextIndex), entries, + context.getCommitIndex(), replicatedToAllIndex); + + if(!entries.isEmpty()) { + LOG.debug("{}: Sending AppendEntries to follower {}: {}", context.getId(), followerId, + appendEntries); + } + + followerActor.tell(appendEntries.toSerializable(), actor()); } /** @@ -482,6 +538,10 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { * */ private void installSnapshotIfNeeded() { + if(LOG.isDebugEnabled()) { + LOG.debug("{}: installSnapshotIfNeeded, followers {}", context.getId(), followerToLog.keySet()); + } + for (Entry e : followerToLog.entrySet()) { final ActorSelection followerActor = context.getPeerActorSelection(e.getKey()); @@ -489,14 +549,14 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { long nextIndex = e.getValue().getNextIndex(); if (!context.getReplicatedLog().isPresent(nextIndex) && - context.getReplicatedLog().isInSnapshot(nextIndex)) { + context.getReplicatedLog().isInSnapshot(nextIndex)) { LOG.info("{}: {} follower needs a snapshot install", context.getId(), e.getKey()); if (snapshot.isPresent()) { // if a snapshot is present in the memory, most likely another install is in progress // no need to capture snapshot sendSnapshotChunk(followerActor, e.getKey()); - } else { + } else if (!context.isSnapshotCaptureInitiated()) { initiateCaptureSnapshot(); //we just need 1 follower who would need snapshot to be installed. // when we have the snapshot captured, we would again check (in SendInstallSnapshot) @@ -529,6 +589,7 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm, isInstallSnapshotInitiated), actor()); + context.setSnapshotCaptureInitiated(true); } @@ -554,24 +615,30 @@ public abstract class AbstractLeader extends AbstractRaftActorBehavior { private void sendSnapshotChunk(ActorSelection followerActor, String followerId) { try { if (snapshot.isPresent()) { + ByteString nextSnapshotChunk = getNextSnapshotChunk(followerId,snapshot.get()); + + // Note: the previous call to getNextSnapshotChunk has the side-effect of adding + // followerId to the followerToSnapshot map. + FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId); + followerActor.tell( new InstallSnapshot(currentTerm(), context.getId(), context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm(), - getNextSnapshotChunk(followerId,snapshot.get()), - mapFollowerToSnapshot.get(followerId).incrementChunkIndex(), - mapFollowerToSnapshot.get(followerId).getTotalChunks(), - Optional.of(mapFollowerToSnapshot.get(followerId).getLastChunkHashCode()) + nextSnapshotChunk, + followerToSnapshot.incrementChunkIndex(), + followerToSnapshot.getTotalChunks(), + Optional.of(followerToSnapshot.getLastChunkHashCode()) ).toSerializable(), actor() ); LOG.info("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}", context.getId(), followerActor.path(), - mapFollowerToSnapshot.get(followerId).getChunkIndex(), - mapFollowerToSnapshot.get(followerId).getTotalChunks()); + followerToSnapshot.getChunkIndex(), + followerToSnapshot.getTotalChunks()); } } catch (IOException e) { - LOG.error(e, "{}: InstallSnapshot failed for Leader.", context.getId()); + LOG.error("{}: InstallSnapshot failed for Leader.", context.getId(), e); } }