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%2FRaftActorRecoverySupport.java;h=ae7fdeadbc12a657c6e6c33189a1b7e36c16d687;hp=5dc8361cc43470b72165dfd2c7b51d6372ad3438;hb=b8c6400766f7324dd57d059bd48e435569fe1a27;hpb=edb47bc56dc36013c1dbb9567212ed0119d14914 diff --git a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActorRecoverySupport.java b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActorRecoverySupport.java index 5dc8361cc4..ae7fdeadbc 100644 --- a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActorRecoverySupport.java +++ b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActorRecoverySupport.java @@ -9,13 +9,17 @@ package org.opendaylight.controller.cluster.raft; import akka.persistence.RecoveryCompleted; import akka.persistence.SnapshotOffer; +import akka.persistence.SnapshotSelectionCriteria; import com.google.common.base.Stopwatch; +import java.io.ByteArrayInputStream; +import java.io.ObjectInputStream; import org.opendaylight.controller.cluster.DataPersistenceProvider; -import org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm; +import org.opendaylight.controller.cluster.PersistentDataProvider; import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries; import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries; +import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot; import org.opendaylight.controller.cluster.raft.base.messages.DeleteEntries; -import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior; +import org.opendaylight.controller.cluster.raft.base.messages.UpdateElectionTerm; import org.slf4j.Logger; /** @@ -24,28 +28,38 @@ import org.slf4j.Logger; * @author Thomas Pantelis */ class RaftActorRecoverySupport { - private final DataPersistenceProvider persistence; private final RaftActorContext context; - private final RaftActorBehavior currentBehavior; private final RaftActorRecoveryCohort cohort; private int currentRecoveryBatchCount; + private boolean dataRecoveredWithPersistenceDisabled; + private boolean anyDataRecovered; private Stopwatch recoveryTimer; private final Logger log; - RaftActorRecoverySupport(DataPersistenceProvider persistence, RaftActorContext context, - RaftActorBehavior currentBehavior, RaftActorRecoveryCohort cohort) { - this.persistence = persistence; + RaftActorRecoverySupport(final RaftActorContext context, final RaftActorRecoveryCohort cohort) { this.context = context; - this.currentBehavior = currentBehavior; this.cohort = cohort; this.log = context.getLogger(); } - boolean handleRecoveryMessage(Object message) { + boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) { + log.trace("{}: handleRecoveryMessage: {}", context.getId(), message); + + anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted); + boolean recoveryComplete = false; - if(persistence.isRecoveryApplicable()) { + DataPersistenceProvider persistence = context.getPersistenceProvider(); + if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm) { + // Handle this message for backwards compatibility with pre-Lithium versions. + org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm update = + (org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm)message; + context.getTermInformation().update(update.getCurrentTerm(), update.getVotedFor()); + } else if (message instanceof UpdateElectionTerm) { + context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(), + ((UpdateElectionTerm) message).getVotedFor()); + } else if(persistence.isRecoveryApplicable()) { if (message instanceof SnapshotOffer) { onRecoveredSnapshot((SnapshotOffer) message); } else if (message instanceof ReplicatedLogEntry) { @@ -60,20 +74,72 @@ class RaftActorRecoverySupport { } else if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) { // Handle this message for backwards compatibility with pre-Lithium versions. replicatedLog().removeFrom(((org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) message).getFromIndex()); - } else if (message instanceof UpdateElectionTerm) { - context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(), - ((UpdateElectionTerm) message).getVotedFor()); } else if (message instanceof RecoveryCompleted) { onRecoveryCompletedMessage(); + possiblyRestoreFromSnapshot(); recoveryComplete = true; } } else if (message instanceof RecoveryCompleted) { recoveryComplete = true; + + if(dataRecoveredWithPersistenceDisabled) { + // Data persistence is disabled but we recovered some data entries so we must have just + // transitioned to disabled or a persistence backup was restored. Either way, delete all the + // messages from the akka journal for efficiency and so that we do not end up with consistency + // issues in case persistence is -re-enabled. + persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber()); + + // Delete all the akka snapshots as they will not be needed + persistentProvider.deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), + scala.Long.MaxValue(), 0L, 0L)); + + // Since we cleaned out the journal, we need to re-write the current election info. + context.getTermInformation().updateAndPersist(context.getTermInformation().getCurrentTerm(), + context.getTermInformation().getVotedFor()); + } + + possiblyRestoreFromSnapshot(); + } else { + boolean isServerConfigPayload = false; + if(message instanceof ReplicatedLogEntry){ + ReplicatedLogEntry repLogEntry = (ReplicatedLogEntry)message; + if(isServerConfigurationPayload(repLogEntry)){ + isServerConfigPayload = true; + context.updatePeerIds((ServerConfigurationPayload)repLogEntry.getData()); + } + } + + if(!isServerConfigPayload){ + dataRecoveredWithPersistenceDisabled = true; + } } return recoveryComplete; } + private void possiblyRestoreFromSnapshot() { + byte[] restoreFromSnapshot = cohort.getRestoreFromSnapshot(); + if(restoreFromSnapshot == null) { + return; + } + + if(anyDataRecovered) { + log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty", + context.getId()); + return; + } + + try(ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(restoreFromSnapshot))) { + Snapshot snapshot = (Snapshot) ois.readObject(); + + log.debug("{}: Deserialized restore snapshot: {}", context.getId(), snapshot); + + context.getSnapshotManager().apply(new ApplySnapshot(snapshot)); + } catch(Exception e) { + log.error("{}: Error deserializing snapshot restore", context.getId(), e); + } + } + private ReplicatedLog replicatedLog() { return context.getReplicatedLog(); } @@ -97,15 +163,20 @@ class RaftActorRecoverySupport { // The replicated log can be used later on to retrieve this snapshot // when we need to install it on a peer - context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, persistence, currentBehavior)); + context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context)); context.setLastApplied(snapshot.getLastAppliedIndex()); context.setCommitIndex(snapshot.getLastAppliedIndex()); + context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor()); Stopwatch timer = Stopwatch.createStarted(); // Apply the snapshot to the actors state cohort.applyRecoverySnapshot(snapshot.getState()); + if (snapshot.getServerConfiguration() != null) { + context.updatePeerIds(snapshot.getServerConfiguration()); + } + timer.stop(); log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}", context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(), @@ -118,35 +189,53 @@ class RaftActorRecoverySupport { logEntry.getIndex(), logEntry.size()); } + if(isServerConfigurationPayload(logEntry)){ + context.updatePeerIds((ServerConfigurationPayload)logEntry.getData()); + } replicatedLog().append(logEntry); } private void onRecoveredApplyLogEntries(long toIndex) { + long lastUnappliedIndex = context.getLastApplied() + 1; + if(log.isDebugEnabled()) { - log.debug("{}: Received ApplyLogEntries for recovery, applying to state: {} to {}", - context.getId(), context.getLastApplied() + 1, toIndex); + // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal + // but the entry itself has made it to that state and recovered via the snapshot + log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}", + context.getId(), lastUnappliedIndex, toIndex); } - for (long i = context.getLastApplied() + 1; i <= toIndex; i++) { - batchRecoveredLogEntry(replicatedLog().get(i)); + long lastApplied = lastUnappliedIndex - 1; + for (long i = lastUnappliedIndex; i <= toIndex; i++) { + ReplicatedLogEntry logEntry = replicatedLog().get(i); + if(logEntry != null) { + lastApplied++; + batchRecoveredLogEntry(logEntry); + } else { + // Shouldn't happen but cover it anyway. + log.error("{}: Log entry not found for index {}", context.getId(), i); + break; + } } - context.setLastApplied(toIndex); - context.setCommitIndex(toIndex); + context.setLastApplied(lastApplied); + context.setCommitIndex(lastApplied); } private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) { initRecoveryTimer(); int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize(); - if(currentRecoveryBatchCount == 0) { - cohort.startLogRecoveryBatch(batchSize); - } + if(!isServerConfigurationPayload(logEntry)){ + if(currentRecoveryBatchCount == 0) { + cohort.startLogRecoveryBatch(batchSize); + } - cohort.appendRecoveredLogEntry(logEntry.getData()); + cohort.appendRecoveredLogEntry(logEntry.getData()); - if(++currentRecoveryBatchCount >= batchSize) { - endCurrentLogRecoveryBatch(); + if(++currentRecoveryBatchCount >= batchSize) { + endCurrentLogRecoveryBatch(); + } } } @@ -173,4 +262,8 @@ class RaftActorRecoverySupport { "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(), replicatedLog().getSnapshotTerm(), replicatedLog().size()); } + + private static boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry){ + return (repLogEntry.getData() instanceof ServerConfigurationPayload); + } }