opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActorRecoverySupport.java

   1 /*
   2  * Copyright (c) 2015 Brocade Communications Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8 package org.opendaylight.controller.cluster.raft;
   9
  10 import akka.persistence.RecoveryCompleted;
  11 import akka.persistence.SnapshotOffer;
  12 import akka.persistence.SnapshotSelectionCriteria;
  13 import com.google.common.base.Stopwatch;
  14 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  15 import org.opendaylight.controller.cluster.PersistentDataProvider;
  16 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  17 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  18 import org.opendaylight.controller.cluster.raft.base.messages.DeleteEntries;
  19 import org.opendaylight.controller.cluster.raft.base.messages.UpdateElectionTerm;
  20 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  21 import org.slf4j.Logger;
  22
  23 /**
  24  * Support class that handles persistence recovery for a RaftActor.
  25  *
  26  * @author Thomas Pantelis
  27  */
  28 class RaftActorRecoverySupport {
  29     private final RaftActorContext context;
  30     private final RaftActorBehavior currentBehavior;
  31     private final RaftActorRecoveryCohort cohort;
  32
  33     private int currentRecoveryBatchCount;
  34     private boolean dataRecoveredWithPersistenceDisabled;
  35
  36     private Stopwatch recoveryTimer;
  37     private final Logger log;
  38
  39     RaftActorRecoverySupport(RaftActorContext context, RaftActorBehavior currentBehavior,
  40             RaftActorRecoveryCohort cohort) {
  41         this.context = context;
  42         this.currentBehavior = currentBehavior;
  43         this.cohort = cohort;
  44         this.log = context.getLogger();
  45     }
  46
  47     boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) {
  48         log.trace("handleRecoveryMessage: {}", message);
  49
  50         boolean recoveryComplete = false;
  51         DataPersistenceProvider persistence = context.getPersistenceProvider();
  52         if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm) {
  53             // Handle this message for backwards compatibility with pre-Lithium versions.
  54             org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm update =
  55                     (org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm)message;
  56             context.getTermInformation().update(update.getCurrentTerm(), update.getVotedFor());
  57         } else if (message instanceof UpdateElectionTerm) {
  58             context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
  59                     ((UpdateElectionTerm) message).getVotedFor());
  60         } else if(persistence.isRecoveryApplicable()) {
  61             if (message instanceof SnapshotOffer) {
  62                 onRecoveredSnapshot((SnapshotOffer) message);
  63             } else if (message instanceof ReplicatedLogEntry) {
  64                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
  65             } else if (message instanceof ApplyLogEntries) {
  66                 // Handle this message for backwards compatibility with pre-Lithium versions.
  67                 onRecoveredApplyLogEntries(((ApplyLogEntries) message).getToIndex());
  68             } else if (message instanceof ApplyJournalEntries) {
  69                 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
  70             } else if (message instanceof DeleteEntries) {
  71                 replicatedLog().removeFrom(((DeleteEntries) message).getFromIndex());
  72             } else if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) {
  73                 // Handle this message for backwards compatibility with pre-Lithium versions.
  74                 replicatedLog().removeFrom(((org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) message).getFromIndex());
  75             } else if (message instanceof RecoveryCompleted) {
  76                 onRecoveryCompletedMessage();
  77                 recoveryComplete = true;
  78             }
  79         } else if (message instanceof RecoveryCompleted) {
  80             recoveryComplete = true;
  81
  82             if(dataRecoveredWithPersistenceDisabled) {
  83                 // Data persistence is disabled but we recovered some data entries so we must have just
  84                 // transitioned to disabled or a persistence backup was restored. Either way, delete all the
  85                 // messages from the akka journal for efficiency and so that we do not end up with consistency
  86                 // issues in case persistence is -re-enabled.
  87                 persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
  88
  89                 // Delete all the akka snapshots as they will not be needed
  90                 persistentProvider.deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(),
  91                         scala.Long.MaxValue()));
  92
  93                 // Since we cleaned out the journal, we need to re-write the current election info.
  94                 context.getTermInformation().updateAndPersist(context.getTermInformation().getCurrentTerm(),
  95                         context.getTermInformation().getVotedFor());
  96             }
  97         } else {
  98             dataRecoveredWithPersistenceDisabled = true;
  99         }
 100
 101         return recoveryComplete;
 102     }
 103
 104     private ReplicatedLog replicatedLog() {
 105         return context.getReplicatedLog();
 106     }
 107
 108     private void initRecoveryTimer() {
 109         if(recoveryTimer == null) {
 110             recoveryTimer = Stopwatch.createStarted();
 111         }
 112     }
 113
 114     private void onRecoveredSnapshot(SnapshotOffer offer) {
 115         if(log.isDebugEnabled()) {
 116             log.debug("{}: SnapshotOffer called..", context.getId());
 117         }
 118
 119         initRecoveryTimer();
 120
 121         Snapshot snapshot = (Snapshot) offer.snapshot();
 122
 123         // Create a replicated log with the snapshot information
 124         // The replicated log can be used later on to retrieve this snapshot
 125         // when we need to install it on a peer
 126
 127         context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, currentBehavior));
 128         context.setLastApplied(snapshot.getLastAppliedIndex());
 129         context.setCommitIndex(snapshot.getLastAppliedIndex());
 130
 131         Stopwatch timer = Stopwatch.createStarted();
 132
 133         // Apply the snapshot to the actors state
 134         cohort.applyRecoverySnapshot(snapshot.getState());
 135
 136         timer.stop();
 137         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 138                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 139                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 140     }
 141
 142     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 143         if(log.isDebugEnabled()) {
 144             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 145                     logEntry.getIndex(), logEntry.size());
 146         }
 147
 148         replicatedLog().append(logEntry);
 149     }
 150
 151     private void onRecoveredApplyLogEntries(long toIndex) {
 152         long lastUnappliedIndex = context.getLastApplied() + 1;
 153
 154         if(log.isDebugEnabled()) {
 155             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 156             // but the entry itself has made it to that state and recovered via the snapshot
 157             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 158                     context.getId(), lastUnappliedIndex, toIndex);
 159         }
 160
 161         long lastApplied = lastUnappliedIndex - 1;
 162         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 163             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 164             if(logEntry != null) {
 165                 lastApplied++;
 166                 batchRecoveredLogEntry(logEntry);
 167             } else {
 168                 // Shouldn't happen but cover it anyway.
 169                 log.error("Log entry not found for index {}", i);
 170                 break;
 171             }
 172         }
 173
 174         context.setLastApplied(lastApplied);
 175         context.setCommitIndex(lastApplied);
 176     }
 177
 178     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 179         initRecoveryTimer();
 180
 181         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 182         if(currentRecoveryBatchCount == 0) {
 183             cohort.startLogRecoveryBatch(batchSize);
 184         }
 185
 186         cohort.appendRecoveredLogEntry(logEntry.getData());
 187
 188         if(++currentRecoveryBatchCount >= batchSize) {
 189             endCurrentLogRecoveryBatch();
 190         }
 191     }
 192
 193     private void endCurrentLogRecoveryBatch() {
 194         cohort.applyCurrentLogRecoveryBatch();
 195         currentRecoveryBatchCount = 0;
 196     }
 197
 198     private void onRecoveryCompletedMessage() {
 199         if(currentRecoveryBatchCount > 0) {
 200             endCurrentLogRecoveryBatch();
 201         }
 202
 203         String recoveryTime = "";
 204         if(recoveryTimer != null) {
 205             recoveryTimer.stop();
 206             recoveryTime = " in " + recoveryTimer.toString();
 207             recoveryTimer = null;
 208         }
 209
 210         log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 211                  "Persistence Id =  " + context.getId() +
 212                  " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, " +
 213                  "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 214                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 215     }
 216 }