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%2FRaftActor.java;h=65254f2d6277c34c9773570e5938d8a005ab3015;hp=766b80e73dd12c890df3ed493e397a7cd144aab4;hb=9a24e798d7f9855be2518db54adcac60f7f9ba54;hpb=fc54ab8853d36fb1d7aebf2a09ef10567e66aa0d diff --git a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActor.java b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActor.java index 766b80e73d..65254f2d62 100644 --- a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActor.java +++ b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActor.java @@ -10,8 +10,6 @@ package org.opendaylight.controller.cluster.raft; import akka.actor.ActorRef; import akka.actor.ActorSelection; -import akka.event.Logging; -import akka.event.LoggingAdapter; import akka.japi.Procedure; import akka.persistence.RecoveryCompleted; import akka.persistence.SaveSnapshotFailure; @@ -33,16 +31,15 @@ import org.opendaylight.controller.cluster.raft.base.messages.ApplyState; import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot; import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply; import org.opendaylight.controller.cluster.raft.base.messages.Replicate; -import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat; import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot; import org.opendaylight.controller.cluster.raft.behaviors.AbstractRaftActorBehavior; import org.opendaylight.controller.cluster.raft.behaviors.Follower; import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior; import org.opendaylight.controller.cluster.raft.client.messages.FindLeader; import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply; -import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload; -import org.opendaylight.controller.protobuff.messages.cluster.raft.AppendEntriesMessages; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; /** * RaftActor encapsulates a state machine that needs to be kept synchronized @@ -85,8 +82,7 @@ import org.opendaylight.controller.protobuff.messages.cluster.raft.AppendEntries * */ public abstract class RaftActor extends AbstractUntypedPersistentActor { - protected final LoggingAdapter LOG = - Logging.getLogger(getContext().system(), this); + protected final Logger LOG = LoggerFactory.getLogger(getClass()); /** * The current state determines the current behavior of a RaftActor @@ -127,8 +123,7 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { private void initRecoveryTimer() { if(recoveryTimer == null) { - recoveryTimer = new Stopwatch(); - recoveryTimer.start(); + recoveryTimer = Stopwatch.createStarted(); } } @@ -191,8 +186,7 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { context.setLastApplied(snapshot.getLastAppliedIndex()); context.setCommitIndex(snapshot.getLastAppliedIndex()); - Stopwatch timer = new Stopwatch(); - timer.start(); + Stopwatch timer = Stopwatch.createStarted(); // Apply the snapshot to the actors state applyRecoverySnapshot(snapshot.getState()); @@ -338,8 +332,8 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { } else if (message instanceof SaveSnapshotFailure) { SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message; - LOG.error(saveSnapshotFailure.cause(), "{}: SaveSnapshotFailure received for snapshot Cause:", - persistenceId()); + LOG.error("{}: SaveSnapshotFailure received for snapshot Cause:", + persistenceId(), saveSnapshotFailure.cause()); context.getReplicatedLog().snapshotRollback(); @@ -361,13 +355,6 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { handleCaptureSnapshotReply(((CaptureSnapshotReply) message).getSnapshot()); } else { - if (!(message instanceof AppendEntriesMessages.AppendEntries) - && !(message instanceof AppendEntriesReply) && !(message instanceof SendHeartBeat)) { - if(LOG.isDebugEnabled()) { - LOG.debug("{}: onReceiveCommand: message: {}", persistenceId(), message.getClass()); - } - } - RaftActorBehavior oldBehavior = currentBehavior; currentBehavior = currentBehavior.handleMessage(getSender(), message); @@ -576,7 +563,7 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { /** * This method is called during recovery to reconstruct the state of the actor. * - * @param snapshot A snapshot of the state of the actor + * @param snapshotBytes A snapshot of the state of the actor */ protected abstract void applyRecoverySnapshot(byte[] snapshotBytes); @@ -671,12 +658,21 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { LOG.info("{}: Persisting of snapshot done:{}", persistenceId(), sn.getLogMessage()); - //be greedy and remove entries from in-mem journal which are in the snapshot - // and update snapshotIndex and snapshotTerm without waiting for the success, + long dataThreshold = Runtime.getRuntime().totalMemory() * + getRaftActorContext().getConfigParams().getSnapshotDataThresholdPercentage() / 100; + if (context.getReplicatedLog().dataSize() > dataThreshold) { + // if memory is less, clear the log based on lastApplied. + // this could/should only happen if one of the followers is down + // as normally we keep removing from the log when its replicated to all. + context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(), + captureSnapshot.getLastAppliedTerm()); - context.getReplicatedLog().snapshotPreCommit( - captureSnapshot.getLastAppliedIndex(), - captureSnapshot.getLastAppliedTerm()); + } else { + // clear the log based on replicatedToAllIndex + context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(), + captureSnapshot.getReplicatedToAllTerm()); + } + getCurrentBehavior().setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex()); LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " + "and term:{}", persistenceId(), captureSnapshot.getLastAppliedIndex(), @@ -720,13 +716,14 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { // FIXME: Maybe this should be done after the command is saved journal.subList(adjustedIndex , journal.size()).clear(); - persistence().persist(new DeleteEntries(adjustedIndex), new Procedure(){ + persistence().persist(new DeleteEntries(adjustedIndex), new Procedure() { - @Override public void apply(DeleteEntries param) - throws Exception { + @Override + public void apply(DeleteEntries param) + throws Exception { //FIXME : Doing nothing for now dataSize = 0; - for(ReplicatedLogEntry entry : journal){ + for (ReplicatedLogEntry entry : journal) { dataSize += entry.size(); } } @@ -738,11 +735,6 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { appendAndPersist(replicatedLogEntry, null); } - @Override - public int dataSize() { - return dataSize; - } - public void appendAndPersist( final ReplicatedLogEntry replicatedLogEntry, final Procedure callback) { @@ -769,7 +761,7 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { long dataSizeForCheck = dataSize; dataSizeSinceLastSnapshot += logEntrySize; - long journalSize = lastIndex()+1; + long journalSize = lastIndex() + 1; if(!hasFollowers()) { // When we do not have followers we do not maintain an in-memory log @@ -796,7 +788,9 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { dataSizeSinceLastSnapshot = 0; - LOG.info("{}: Initiating Snapshot Capture..", persistenceId()); + LOG.info("{}: Initiating Snapshot Capture, journalSize = {}, dataSizeForCheck = {}," + + " dataThreshold = {}", persistenceId(), journalSize, dataSizeForCheck, dataThreshold); + long lastAppliedIndex = -1; long lastAppliedTerm = -1; @@ -820,12 +814,15 @@ public abstract class RaftActor extends AbstractUntypedPersistentActor { } // send a CaptureSnapshot to self to make the expensive operation async. - getSelf().tell(new CaptureSnapshot( - lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm), + long replicatedToAllIndex = getCurrentBehavior().getReplicatedToAllIndex(); + ReplicatedLogEntry replicatedToAllEntry = context.getReplicatedLog().get(replicatedToAllIndex); + getSelf().tell(new CaptureSnapshot(lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm, + (replicatedToAllEntry != null ? replicatedToAllEntry.getIndex() : -1), + (replicatedToAllEntry != null ? replicatedToAllEntry.getTerm() : -1)), null); context.setSnapshotCaptureInitiated(true); } - if(callback != null){ + if (callback != null){ callback.apply(replicatedLogEntry); } }