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 java.util.Collections;
  14 import org.opendaylight.controller.cluster.PersistentDataProvider;
  15 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  16 import org.opendaylight.controller.cluster.raft.persisted.ApplyJournalEntries;
  17 import org.opendaylight.controller.cluster.raft.persisted.DeleteEntries;
  18 import org.opendaylight.controller.cluster.raft.persisted.EmptyState;
  19 import org.opendaylight.controller.cluster.raft.persisted.MigratedSerializable;
  20 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  21 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
  22 import org.opendaylight.controller.cluster.raft.persisted.UpdateElectionTerm;
  23 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.PersistentPayload;
  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 RaftActorRecoveryCohort cohort;
  34
  35     private int currentRecoveryBatchCount;
  36     private boolean dataRecoveredWithPersistenceDisabled;
  37     private boolean anyDataRecovered;
  38     private boolean hasMigratedDataRecovered;
  39
  40     private Stopwatch recoveryTimer;
  41     private final Logger log;
  42
  43     RaftActorRecoverySupport(final RaftActorContext context, final RaftActorRecoveryCohort cohort) {
  44         this.context = context;
  45         this.cohort = cohort;
  46         this.log = context.getLogger();
  47     }
  48
  49     boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) {
  50         log.trace("{}: handleRecoveryMessage: {}", context.getId(), message);
  51
  52         anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted);
  53
  54         if (isMigratedSerializable(message)) {
  55             hasMigratedDataRecovered = true;
  56         }
  57
  58         boolean recoveryComplete = false;
  59         if (message instanceof UpdateElectionTerm) {
  60             context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
  61                     ((UpdateElectionTerm) message).getVotedFor());
  62         } else if (message instanceof SnapshotOffer) {
  63             onRecoveredSnapshot((SnapshotOffer) message);
  64         } else if (message instanceof ReplicatedLogEntry) {
  65             onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
  66         } else if (message instanceof ApplyJournalEntries) {
  67             onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
  68         } else if (message instanceof DeleteEntries) {
  69             onDeleteEntries((DeleteEntries) message);
  70         } else if (message instanceof ServerConfigurationPayload) {
  71             context.updatePeerIds((ServerConfigurationPayload)message);
  72         } else if (message instanceof RecoveryCompleted) {
  73             recoveryComplete = true;
  74             onRecoveryCompletedMessage(persistentProvider);
  75         }
  76
  77         return recoveryComplete;
  78     }
  79
  80     @SuppressWarnings("checkstyle:IllegalCatch")
  81     private void possiblyRestoreFromSnapshot() {
  82         Snapshot restoreFromSnapshot = cohort.getRestoreFromSnapshot();
  83         if (restoreFromSnapshot == null) {
  84             return;
  85         }
  86
  87         if (anyDataRecovered) {
  88             log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty",
  89                     context.getId());
  90             return;
  91         }
  92
  93         log.debug("{}: Restore snapshot: {}", context.getId(), restoreFromSnapshot);
  94
  95         context.getSnapshotManager().apply(new ApplySnapshot(restoreFromSnapshot));
  96     }
  97
  98     private ReplicatedLog replicatedLog() {
  99         return context.getReplicatedLog();
 100     }
 101
 102     private void initRecoveryTimer() {
 103         if (recoveryTimer == null) {
 104             recoveryTimer = Stopwatch.createStarted();
 105         }
 106     }
 107
 108     private void onRecoveredSnapshot(SnapshotOffer offer) {
 109         log.debug("{}: SnapshotOffer called.", context.getId());
 110
 111         initRecoveryTimer();
 112
 113         Object snapshotObj = offer.snapshot();
 114         Snapshot snapshot;
 115         if (snapshotObj instanceof org.opendaylight.controller.cluster.raft.Snapshot) {
 116             org.opendaylight.controller.cluster.raft.Snapshot legacy =
 117                     (org.opendaylight.controller.cluster.raft.Snapshot)snapshotObj;
 118             snapshot = Snapshot.create(cohort.deserializePreCarbonSnapshot(legacy.getState()),
 119                     legacy.getUnAppliedEntries(), legacy.getLastIndex(), legacy.getLastTerm(),
 120                     legacy.getLastAppliedIndex(), legacy.getLastAppliedTerm(),
 121                     legacy.getElectionTerm(), legacy.getElectionVotedFor(), legacy.getServerConfiguration());
 122             hasMigratedDataRecovered = true;
 123         } else {
 124             snapshot = (Snapshot) offer.snapshot();
 125         }
 126
 127         for (ReplicatedLogEntry entry: snapshot.getUnAppliedEntries()) {
 128             if (isMigratedPayload(entry)) {
 129                 hasMigratedDataRecovered = true;
 130             }
 131         }
 132
 133         if (!context.getPersistenceProvider().isRecoveryApplicable()) {
 134             // We may have just transitioned to disabled and have a snapshot containing state data and/or log
 135             // entries - we don't want to preserve these, only the server config and election term info.
 136
 137             snapshot = Snapshot.create(
 138                     EmptyState.INSTANCE, Collections.emptyList(), -1, -1, -1, -1,
 139                     snapshot.getElectionTerm(), snapshot.getElectionVotedFor(), snapshot.getServerConfiguration());
 140         }
 141
 142         // Create a replicated log with the snapshot information
 143         // The replicated log can be used later on to retrieve this snapshot
 144         // when we need to install it on a peer
 145
 146         context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
 147         context.setLastApplied(snapshot.getLastAppliedIndex());
 148         context.setCommitIndex(snapshot.getLastAppliedIndex());
 149         context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 150
 151         Stopwatch timer = Stopwatch.createStarted();
 152
 153         // Apply the snapshot to the actors state
 154         if (!(snapshot.getState() instanceof EmptyState)) {
 155             cohort.applyRecoverySnapshot(snapshot.getState());
 156         }
 157
 158         if (snapshot.getServerConfiguration() != null) {
 159             context.updatePeerIds(snapshot.getServerConfiguration());
 160
 161             if (isMigratedSerializable(snapshot.getServerConfiguration())) {
 162                 hasMigratedDataRecovered = true;
 163             }
 164         }
 165
 166         timer.stop();
 167         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 168                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 169                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 170     }
 171
 172     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 173         if (log.isDebugEnabled()) {
 174             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 175                     logEntry.getIndex(), logEntry.size());
 176         }
 177
 178         if (isServerConfigurationPayload(logEntry)) {
 179             context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
 180         }
 181
 182         if (isMigratedPayload(logEntry)) {
 183             hasMigratedDataRecovered = true;
 184         }
 185
 186         if (context.getPersistenceProvider().isRecoveryApplicable()) {
 187             replicatedLog().append(logEntry);
 188         } else if (!isPersistentPayload(logEntry)) {
 189             dataRecoveredWithPersistenceDisabled = true;
 190         }
 191     }
 192
 193     private void onRecoveredApplyLogEntries(long toIndex) {
 194         if (!context.getPersistenceProvider().isRecoveryApplicable()) {
 195             dataRecoveredWithPersistenceDisabled = true;
 196             return;
 197         }
 198
 199         long lastUnappliedIndex = context.getLastApplied() + 1;
 200
 201         if (log.isDebugEnabled()) {
 202             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 203             // but the entry itself has made it to that state and recovered via the snapshot
 204             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 205                     context.getId(), lastUnappliedIndex, toIndex);
 206         }
 207
 208         long lastApplied = lastUnappliedIndex - 1;
 209         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 210             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 211             if (logEntry != null) {
 212                 lastApplied++;
 213                 batchRecoveredLogEntry(logEntry);
 214             } else {
 215                 // Shouldn't happen but cover it anyway.
 216                 log.error("{}: Log entry not found for index {}", context.getId(), i);
 217                 break;
 218             }
 219         }
 220
 221         context.setLastApplied(lastApplied);
 222         context.setCommitIndex(lastApplied);
 223     }
 224
 225     private void onDeleteEntries(DeleteEntries deleteEntries) {
 226         if (context.getPersistenceProvider().isRecoveryApplicable()) {
 227             replicatedLog().removeFrom(deleteEntries.getFromIndex());
 228         } else {
 229             dataRecoveredWithPersistenceDisabled = true;
 230         }
 231     }
 232
 233     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 234         initRecoveryTimer();
 235
 236         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 237         if (!isServerConfigurationPayload(logEntry)) {
 238             if (currentRecoveryBatchCount == 0) {
 239                 cohort.startLogRecoveryBatch(batchSize);
 240             }
 241
 242             cohort.appendRecoveredLogEntry(logEntry.getData());
 243
 244             if (++currentRecoveryBatchCount >= batchSize) {
 245                 endCurrentLogRecoveryBatch();
 246             }
 247         }
 248     }
 249
 250     private void endCurrentLogRecoveryBatch() {
 251         cohort.applyCurrentLogRecoveryBatch();
 252         currentRecoveryBatchCount = 0;
 253     }
 254
 255     private void onRecoveryCompletedMessage(PersistentDataProvider persistentProvider) {
 256         if (currentRecoveryBatchCount > 0) {
 257             endCurrentLogRecoveryBatch();
 258         }
 259
 260         String recoveryTime = "";
 261         if (recoveryTimer != null) {
 262             recoveryTimer.stop();
 263             recoveryTime = " in " + recoveryTimer.toString();
 264             recoveryTimer = null;
 265         }
 266
 267         log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " + "Persistence Id =  "
 268                   + context.getId() + " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, "
 269                   + "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 270                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 271
 272         if (dataRecoveredWithPersistenceDisabled
 273                 || hasMigratedDataRecovered && !context.getPersistenceProvider().isRecoveryApplicable()) {
 274             if (hasMigratedDataRecovered) {
 275                 log.info("{}: Saving snapshot after recovery due to migrated messages", context.getId());
 276             } else {
 277                 log.info("{}: Saving snapshot after recovery due to data persistence disabled", context.getId());
 278             }
 279
 280             // Either data persistence is disabled and we recovered some data entries (ie we must have just
 281             // transitioned to disabled or a persistence backup was restored) or we recovered migrated
 282             // messages. Either way, we persist a snapshot and delete all the messages from the akka journal
 283             // to clean out unwanted messages.
 284
 285             Snapshot snapshot = Snapshot.create(
 286                     EmptyState.INSTANCE, Collections.<ReplicatedLogEntry>emptyList(),
 287                     -1, -1, -1, -1,
 288                     context.getTermInformation().getCurrentTerm(), context.getTermInformation().getVotedFor(),
 289                     context.getPeerServerInfo(true));
 290
 291             persistentProvider.saveSnapshot(snapshot);
 292
 293             persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
 294         } else if (hasMigratedDataRecovered) {
 295             log.info("{}: Snapshot capture initiated after recovery due to migrated messages", context.getId());
 296
 297             context.getSnapshotManager().capture(replicatedLog().last(), -1);
 298         } else {
 299             possiblyRestoreFromSnapshot();
 300         }
 301     }
 302
 303     private static boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry) {
 304         return repLogEntry.getData() instanceof ServerConfigurationPayload;
 305     }
 306
 307     private static boolean isPersistentPayload(ReplicatedLogEntry repLogEntry) {
 308         return repLogEntry.getData() instanceof PersistentPayload;
 309     }
 310
 311     private static boolean isMigratedPayload(ReplicatedLogEntry repLogEntry) {
 312         return isMigratedSerializable(repLogEntry.getData());
 313     }
 314
 315     private static boolean isMigratedSerializable(Object message) {
 316         return message instanceof MigratedSerializable && ((MigratedSerializable)message).isMigrated();
 317     }
 318 }