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 com.google.common.base.Stopwatch;
  13 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  14 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  15 import org.opendaylight.controller.cluster.raft.base.messages.DeleteEntries;
  16 import org.opendaylight.controller.cluster.raft.base.messages.UpdateElectionTerm;
  17 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  18 import org.slf4j.Logger;
  19
  20 /**
  21  * Support class that handles persistence recovery for a RaftActor.
  22  *
  23  * @author Thomas Pantelis
  24  */
  25 class RaftActorRecoverySupport {
  26     private final RaftActorContext context;
  27     private final RaftActorBehavior currentBehavior;
  28     private final RaftActorRecoveryCohort cohort;
  29
  30     private int currentRecoveryBatchCount;
  31
  32     private Stopwatch recoveryTimer;
  33     private final Logger log;
  34
  35     RaftActorRecoverySupport(RaftActorContext context, RaftActorBehavior currentBehavior,
  36             RaftActorRecoveryCohort cohort) {
  37         this.context = context;
  38         this.currentBehavior = currentBehavior;
  39         this.cohort = cohort;
  40         this.log = context.getLogger();
  41     }
  42
  43     boolean handleRecoveryMessage(Object message) {
  44         boolean recoveryComplete = false;
  45         if(context.getPersistenceProvider().isRecoveryApplicable()) {
  46             if (message instanceof SnapshotOffer) {
  47                 onRecoveredSnapshot((SnapshotOffer) message);
  48             } else if (message instanceof ReplicatedLogEntry) {
  49                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
  50             } else if (message instanceof ApplyLogEntries) {
  51                 // Handle this message for backwards compatibility with pre-Lithium versions.
  52                 onRecoveredApplyLogEntries(((ApplyLogEntries) message).getToIndex());
  53             } else if (message instanceof ApplyJournalEntries) {
  54                 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
  55             } else if (message instanceof DeleteEntries) {
  56                 replicatedLog().removeFrom(((DeleteEntries) message).getFromIndex());
  57             } else if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) {
  58                 // Handle this message for backwards compatibility with pre-Lithium versions.
  59                 replicatedLog().removeFrom(((org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) message).getFromIndex());
  60             } else if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm) {
  61                 // Handle this message for backwards compatibility with pre-Lithium versions.
  62                 org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm update =
  63                         (org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm)message;
  64                 context.getTermInformation().update(update.getCurrentTerm(), update.getVotedFor());
  65             } else if (message instanceof UpdateElectionTerm) {
  66                 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
  67                         ((UpdateElectionTerm) message).getVotedFor());
  68             } else if (message instanceof RecoveryCompleted) {
  69                 onRecoveryCompletedMessage();
  70                 recoveryComplete = true;
  71             }
  72         } else if (message instanceof RecoveryCompleted) {
  73             recoveryComplete = true;
  74         }
  75
  76         return recoveryComplete;
  77     }
  78
  79     private ReplicatedLog replicatedLog() {
  80         return context.getReplicatedLog();
  81     }
  82
  83     private void initRecoveryTimer() {
  84         if(recoveryTimer == null) {
  85             recoveryTimer = Stopwatch.createStarted();
  86         }
  87     }
  88
  89     private void onRecoveredSnapshot(SnapshotOffer offer) {
  90         if(log.isDebugEnabled()) {
  91             log.debug("{}: SnapshotOffer called..", context.getId());
  92         }
  93
  94         initRecoveryTimer();
  95
  96         Snapshot snapshot = (Snapshot) offer.snapshot();
  97
  98         // Create a replicated log with the snapshot information
  99         // The replicated log can be used later on to retrieve this snapshot
 100         // when we need to install it on a peer
 101
 102         context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, currentBehavior));
 103         context.setLastApplied(snapshot.getLastAppliedIndex());
 104         context.setCommitIndex(snapshot.getLastAppliedIndex());
 105
 106         Stopwatch timer = Stopwatch.createStarted();
 107
 108         // Apply the snapshot to the actors state
 109         cohort.applyRecoverySnapshot(snapshot.getState());
 110
 111         timer.stop();
 112         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 113                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 114                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 115     }
 116
 117     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 118         if(log.isDebugEnabled()) {
 119             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 120                     logEntry.getIndex(), logEntry.size());
 121         }
 122
 123         replicatedLog().append(logEntry);
 124     }
 125
 126     private void onRecoveredApplyLogEntries(long toIndex) {
 127         long lastUnappliedIndex = context.getLastApplied() + 1;
 128
 129         if(log.isDebugEnabled()) {
 130             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 131             // but the entry itself has made it to that state and recovered via the snapshot
 132             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 133                     context.getId(), lastUnappliedIndex, toIndex);
 134         }
 135
 136         long lastApplied = lastUnappliedIndex - 1;
 137         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 138             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 139             if(logEntry != null) {
 140                 lastApplied++;
 141                 batchRecoveredLogEntry(logEntry);
 142             } else {
 143                 // Shouldn't happen but cover it anyway.
 144                 log.error("Log entry not found for index {}", i);
 145                 break;
 146             }
 147         }
 148
 149         context.setLastApplied(lastApplied);
 150         context.setCommitIndex(lastApplied);
 151     }
 152
 153     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 154         initRecoveryTimer();
 155
 156         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 157         if(currentRecoveryBatchCount == 0) {
 158             cohort.startLogRecoveryBatch(batchSize);
 159         }
 160
 161         cohort.appendRecoveredLogEntry(logEntry.getData());
 162
 163         if(++currentRecoveryBatchCount >= batchSize) {
 164             endCurrentLogRecoveryBatch();
 165         }
 166     }
 167
 168     private void endCurrentLogRecoveryBatch() {
 169         cohort.applyCurrentLogRecoveryBatch();
 170         currentRecoveryBatchCount = 0;
 171     }
 172
 173     private void onRecoveryCompletedMessage() {
 174         if(currentRecoveryBatchCount > 0) {
 175             endCurrentLogRecoveryBatch();
 176         }
 177
 178         String recoveryTime = "";
 179         if(recoveryTimer != null) {
 180             recoveryTimer.stop();
 181             recoveryTime = " in " + recoveryTimer.toString();
 182             recoveryTimer = null;
 183         }
 184
 185         log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 186                  "Persistence Id =  " + context.getId() +
 187                  " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, " +
 188                  "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 189                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 190     }
 191 }