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 java.io.ByteArrayInputStream;
  15 import java.io.ObjectInputStream;
  16 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  17 import org.opendaylight.controller.cluster.PersistentDataProvider;
  18 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  19 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  20 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  21 import org.opendaylight.controller.cluster.raft.base.messages.DeleteEntries;
  22 import org.opendaylight.controller.cluster.raft.base.messages.UpdateElectionTerm;
  23 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  24 import org.slf4j.Logger;
  25
  26 /**
  27  * Support class that handles persistence recovery for a RaftActor.
  28  *
  29  * @author Thomas Pantelis
  30  */
  31 class RaftActorRecoverySupport {
  32     private final RaftActorContext context;
  33     private final RaftActorBehavior currentBehavior;
  34     private final RaftActorRecoveryCohort cohort;
  35
  36     private int currentRecoveryBatchCount;
  37     private boolean dataRecoveredWithPersistenceDisabled;
  38     private boolean anyDataRecovered;
  39
  40     private Stopwatch recoveryTimer;
  41     private final Logger log;
  42
  43     RaftActorRecoverySupport(RaftActorContext context, RaftActorBehavior currentBehavior,
  44             RaftActorRecoveryCohort cohort) {
  45         this.context = context;
  46         this.currentBehavior = currentBehavior;
  47         this.cohort = cohort;
  48         this.log = context.getLogger();
  49     }
  50
  51     boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) {
  52         log.trace("{}: handleRecoveryMessage: {}", context.getId(), message);
  53
  54         anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted);
  55
  56         boolean recoveryComplete = false;
  57         DataPersistenceProvider persistence = context.getPersistenceProvider();
  58         if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm) {
  59             // Handle this message for backwards compatibility with pre-Lithium versions.
  60             org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm update =
  61                     (org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm)message;
  62             context.getTermInformation().update(update.getCurrentTerm(), update.getVotedFor());
  63         } else if (message instanceof UpdateElectionTerm) {
  64             context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
  65                     ((UpdateElectionTerm) message).getVotedFor());
  66         } else if(persistence.isRecoveryApplicable()) {
  67             if (message instanceof SnapshotOffer) {
  68                 onRecoveredSnapshot((SnapshotOffer) message);
  69             } else if (message instanceof ReplicatedLogEntry) {
  70                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
  71             } else if (message instanceof ApplyLogEntries) {
  72                 // Handle this message for backwards compatibility with pre-Lithium versions.
  73                 onRecoveredApplyLogEntries(((ApplyLogEntries) message).getToIndex());
  74             } else if (message instanceof ApplyJournalEntries) {
  75                 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
  76             } else if (message instanceof DeleteEntries) {
  77                 replicatedLog().removeFrom(((DeleteEntries) message).getFromIndex());
  78             } else if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) {
  79                 // Handle this message for backwards compatibility with pre-Lithium versions.
  80                 replicatedLog().removeFrom(((org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) message).getFromIndex());
  81             } else if (message instanceof RecoveryCompleted) {
  82                 onRecoveryCompletedMessage();
  83                 possiblyRestoreFromSnapshot();
  84                 recoveryComplete = true;
  85             }
  86         } else if (message instanceof RecoveryCompleted) {
  87             recoveryComplete = true;
  88
  89             if(dataRecoveredWithPersistenceDisabled) {
  90                 // Data persistence is disabled but we recovered some data entries so we must have just
  91                 // transitioned to disabled or a persistence backup was restored. Either way, delete all the
  92                 // messages from the akka journal for efficiency and so that we do not end up with consistency
  93                 // issues in case persistence is -re-enabled.
  94                 persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
  95
  96                 // Delete all the akka snapshots as they will not be needed
  97                 persistentProvider.deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(),
  98                         scala.Long.MaxValue()));
  99
 100                 // Since we cleaned out the journal, we need to re-write the current election info.
 101                 context.getTermInformation().updateAndPersist(context.getTermInformation().getCurrentTerm(),
 102                         context.getTermInformation().getVotedFor());
 103             }
 104
 105             possiblyRestoreFromSnapshot();
 106         } else {
 107             boolean isServerConfigPayload = false;
 108             if(message instanceof ReplicatedLogEntry){
 109                 ReplicatedLogEntry repLogEntry = (ReplicatedLogEntry)message;
 110                 if(isServerConfigurationPayload(repLogEntry)){
 111                     isServerConfigPayload = true;
 112                     context.updatePeerIds((ServerConfigurationPayload)repLogEntry.getData());
 113                 }
 114             }
 115
 116             if(!isServerConfigPayload){
 117                 dataRecoveredWithPersistenceDisabled = true;
 118             }
 119         }
 120
 121         return recoveryComplete;
 122     }
 123
 124     private void possiblyRestoreFromSnapshot() {
 125         byte[] restoreFromSnapshot = cohort.getRestoreFromSnapshot();
 126         if(restoreFromSnapshot == null) {
 127             return;
 128         }
 129
 130         if(anyDataRecovered) {
 131             log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty",
 132                     context.getId());
 133             return;
 134         }
 135
 136         try(ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(restoreFromSnapshot))) {
 137             Snapshot snapshot = (Snapshot) ois.readObject();
 138
 139             log.debug("{}: Deserialized restore snapshot: {}", context.getId(), snapshot);
 140
 141             context.getSnapshotManager().apply(new ApplySnapshot(snapshot));
 142         } catch(Exception e) {
 143             log.error("{}: Error deserializing snapshot restore", context.getId(), e);
 144         }
 145     }
 146
 147     private ReplicatedLog replicatedLog() {
 148         return context.getReplicatedLog();
 149     }
 150
 151     private void initRecoveryTimer() {
 152         if(recoveryTimer == null) {
 153             recoveryTimer = Stopwatch.createStarted();
 154         }
 155     }
 156
 157     private void onRecoveredSnapshot(SnapshotOffer offer) {
 158         if(log.isDebugEnabled()) {
 159             log.debug("{}: SnapshotOffer called..", context.getId());
 160         }
 161
 162         initRecoveryTimer();
 163
 164         Snapshot snapshot = (Snapshot) offer.snapshot();
 165
 166         // Create a replicated log with the snapshot information
 167         // The replicated log can be used later on to retrieve this snapshot
 168         // when we need to install it on a peer
 169
 170         context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, currentBehavior));
 171         context.setLastApplied(snapshot.getLastAppliedIndex());
 172         context.setCommitIndex(snapshot.getLastAppliedIndex());
 173         context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 174
 175         Stopwatch timer = Stopwatch.createStarted();
 176
 177         // Apply the snapshot to the actors state
 178         cohort.applyRecoverySnapshot(snapshot.getState());
 179
 180         if (snapshot.getServerConfiguration() != null) {
 181             context.updatePeerIds(snapshot.getServerConfiguration());
 182         }
 183
 184         timer.stop();
 185         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 186                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 187                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 188     }
 189
 190     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 191         if(log.isDebugEnabled()) {
 192             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 193                     logEntry.getIndex(), logEntry.size());
 194         }
 195
 196         if(isServerConfigurationPayload(logEntry)){
 197             context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
 198         }
 199         replicatedLog().append(logEntry);
 200     }
 201
 202     private void onRecoveredApplyLogEntries(long toIndex) {
 203         long lastUnappliedIndex = context.getLastApplied() + 1;
 204
 205         if(log.isDebugEnabled()) {
 206             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 207             // but the entry itself has made it to that state and recovered via the snapshot
 208             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 209                     context.getId(), lastUnappliedIndex, toIndex);
 210         }
 211
 212         long lastApplied = lastUnappliedIndex - 1;
 213         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 214             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 215             if(logEntry != null) {
 216                 lastApplied++;
 217                 batchRecoveredLogEntry(logEntry);
 218             } else {
 219                 // Shouldn't happen but cover it anyway.
 220                 log.error("{}: Log entry not found for index {}", context.getId(), i);
 221                 break;
 222             }
 223         }
 224
 225         context.setLastApplied(lastApplied);
 226         context.setCommitIndex(lastApplied);
 227     }
 228
 229     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 230         initRecoveryTimer();
 231
 232         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 233         if(!isServerConfigurationPayload(logEntry)){
 234             if(currentRecoveryBatchCount == 0) {
 235                 cohort.startLogRecoveryBatch(batchSize);
 236             }
 237
 238             cohort.appendRecoveredLogEntry(logEntry.getData());
 239
 240             if(++currentRecoveryBatchCount >= batchSize) {
 241                 endCurrentLogRecoveryBatch();
 242             }
 243         }
 244     }
 245
 246     private void endCurrentLogRecoveryBatch() {
 247         cohort.applyCurrentLogRecoveryBatch();
 248         currentRecoveryBatchCount = 0;
 249     }
 250
 251     private void onRecoveryCompletedMessage() {
 252         if(currentRecoveryBatchCount > 0) {
 253             endCurrentLogRecoveryBatch();
 254         }
 255
 256         String recoveryTime = "";
 257         if(recoveryTimer != null) {
 258             recoveryTimer.stop();
 259             recoveryTime = " in " + recoveryTimer.toString();
 260             recoveryTimer = null;
 261         }
 262
 263         log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 264                  "Persistence Id =  " + context.getId() +
 265                  " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, " +
 266                  "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 267                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 268     }
 269
 270     private boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry){
 271         return (repLogEntry.getData() instanceof ServerConfigurationPayload);
 272     }
 273 }