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             boolean isServerConfigPayload = false;
  99             if(message instanceof ReplicatedLogEntry){
 100                 ReplicatedLogEntry repLogEntry = (ReplicatedLogEntry)message;
 101                 if(isServerConfigurationPayload(repLogEntry)){
 102                     isServerConfigPayload = true;
 103                     context.updatePeerIds((ServerConfigurationPayload)repLogEntry.getData());
 104                 }
 105             }
 106
 107             if(!isServerConfigPayload){
 108                 dataRecoveredWithPersistenceDisabled = true;
 109             }
 110         }
 111
 112         return recoveryComplete;
 113     }
 114
 115     private ReplicatedLog replicatedLog() {
 116         return context.getReplicatedLog();
 117     }
 118
 119     private void initRecoveryTimer() {
 120         if(recoveryTimer == null) {
 121             recoveryTimer = Stopwatch.createStarted();
 122         }
 123     }
 124
 125     private void onRecoveredSnapshot(SnapshotOffer offer) {
 126         if(log.isDebugEnabled()) {
 127             log.debug("{}: SnapshotOffer called..", context.getId());
 128         }
 129
 130         initRecoveryTimer();
 131
 132         Snapshot snapshot = (Snapshot) offer.snapshot();
 133
 134         // Create a replicated log with the snapshot information
 135         // The replicated log can be used later on to retrieve this snapshot
 136         // when we need to install it on a peer
 137
 138         context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, currentBehavior));
 139         context.setLastApplied(snapshot.getLastAppliedIndex());
 140         context.setCommitIndex(snapshot.getLastAppliedIndex());
 141         context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 142
 143         Stopwatch timer = Stopwatch.createStarted();
 144
 145         // Apply the snapshot to the actors state
 146         cohort.applyRecoverySnapshot(snapshot.getState());
 147
 148         timer.stop();
 149         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 150                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 151                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 152     }
 153
 154     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 155         if(log.isDebugEnabled()) {
 156             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 157                     logEntry.getIndex(), logEntry.size());
 158         }
 159
 160         if(isServerConfigurationPayload(logEntry)){
 161             context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
 162         }
 163         replicatedLog().append(logEntry);
 164     }
 165
 166     private void onRecoveredApplyLogEntries(long toIndex) {
 167         long lastUnappliedIndex = context.getLastApplied() + 1;
 168
 169         if(log.isDebugEnabled()) {
 170             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 171             // but the entry itself has made it to that state and recovered via the snapshot
 172             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 173                     context.getId(), lastUnappliedIndex, toIndex);
 174         }
 175
 176         long lastApplied = lastUnappliedIndex - 1;
 177         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 178             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 179             if(logEntry != null) {
 180                 lastApplied++;
 181                 batchRecoveredLogEntry(logEntry);
 182             } else {
 183                 // Shouldn't happen but cover it anyway.
 184                 log.error("Log entry not found for index {}", i);
 185                 break;
 186             }
 187         }
 188
 189         context.setLastApplied(lastApplied);
 190         context.setCommitIndex(lastApplied);
 191     }
 192
 193     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 194         initRecoveryTimer();
 195
 196         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 197         if(!isServerConfigurationPayload(logEntry)){
 198             if(currentRecoveryBatchCount == 0) {
 199                 cohort.startLogRecoveryBatch(batchSize);
 200             }
 201
 202             cohort.appendRecoveredLogEntry(logEntry.getData());
 203
 204             if(++currentRecoveryBatchCount >= batchSize) {
 205                 endCurrentLogRecoveryBatch();
 206             }
 207         }
 208     }
 209
 210     private void endCurrentLogRecoveryBatch() {
 211         cohort.applyCurrentLogRecoveryBatch();
 212         currentRecoveryBatchCount = 0;
 213     }
 214
 215     private void onRecoveryCompletedMessage() {
 216         if(currentRecoveryBatchCount > 0) {
 217             endCurrentLogRecoveryBatch();
 218         }
 219
 220         String recoveryTime = "";
 221         if(recoveryTimer != null) {
 222             recoveryTimer.stop();
 223             recoveryTime = " in " + recoveryTimer.toString();
 224             recoveryTimer = null;
 225         }
 226
 227         log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 228                  "Persistence Id =  " + context.getId() +
 229                  " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, " +
 230                  "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 231                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 232     }
 233
 234     private boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry){
 235         return (repLogEntry.getData() instanceof ServerConfigurationPayload);
 236     }
 237 }