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
 162         timer.stop();
 163         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 164                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 165                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 166     }
 167
 168     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 169         if (log.isDebugEnabled()) {
 170             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 171                     logEntry.getIndex(), logEntry.size());
 172         }
 173
 174         if (isServerConfigurationPayload(logEntry)) {
 175             context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
 176         }
 177
 178         if (isMigratedPayload(logEntry)) {
 179             hasMigratedDataRecovered = true;
 180         }
 181
 182         if (context.getPersistenceProvider().isRecoveryApplicable()) {
 183             replicatedLog().append(logEntry);
 184         } else if (!isPersistentPayload(logEntry)) {
 185             dataRecoveredWithPersistenceDisabled = true;
 186         }
 187     }
 188
 189     private void onRecoveredApplyLogEntries(long toIndex) {
 190         if (!context.getPersistenceProvider().isRecoveryApplicable()) {
 191             dataRecoveredWithPersistenceDisabled = true;
 192             return;
 193         }
 194
 195         long lastUnappliedIndex = context.getLastApplied() + 1;
 196
 197         if (log.isDebugEnabled()) {
 198             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 199             // but the entry itself has made it to that state and recovered via the snapshot
 200             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 201                     context.getId(), lastUnappliedIndex, toIndex);
 202         }
 203
 204         long lastApplied = lastUnappliedIndex - 1;
 205         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 206             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 207             if (logEntry != null) {
 208                 lastApplied++;
 209                 batchRecoveredLogEntry(logEntry);
 210             } else {
 211                 // Shouldn't happen but cover it anyway.
 212                 log.error("{}: Log entry not found for index {}", context.getId(), i);
 213                 break;
 214             }
 215         }
 216
 217         context.setLastApplied(lastApplied);
 218         context.setCommitIndex(lastApplied);
 219     }
 220
 221     private void onDeleteEntries(DeleteEntries deleteEntries) {
 222         if (context.getPersistenceProvider().isRecoveryApplicable()) {
 223             replicatedLog().removeFrom(deleteEntries.getFromIndex());
 224         } else {
 225             dataRecoveredWithPersistenceDisabled = true;
 226         }
 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(PersistentDataProvider persistentProvider) {
 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 - " + "Persistence Id =  "
 264                   + context.getId() + " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, "
 265                   + "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 266                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 267
 268         if (dataRecoveredWithPersistenceDisabled
 269                 || hasMigratedDataRecovered && !context.getPersistenceProvider().isRecoveryApplicable()) {
 270             if (hasMigratedDataRecovered) {
 271                 log.info("{}: Saving snapshot after recovery due to migrated messages", context.getId());
 272             } else {
 273                 log.info("{}: Saving snapshot after recovery due to data persistence disabled", context.getId());
 274             }
 275
 276             // Either data persistence is disabled and we recovered some data entries (ie we must have just
 277             // transitioned to disabled or a persistence backup was restored) or we recovered migrated
 278             // messages. Either way, we persist a snapshot and delete all the messages from the akka journal
 279             // to clean out unwanted messages.
 280
 281             Snapshot snapshot = Snapshot.create(
 282                     EmptyState.INSTANCE, Collections.<ReplicatedLogEntry>emptyList(),
 283                     -1, -1, -1, -1,
 284                     context.getTermInformation().getCurrentTerm(), context.getTermInformation().getVotedFor(),
 285                     context.getPeerServerInfo(true));
 286
 287             persistentProvider.saveSnapshot(snapshot);
 288
 289             persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
 290         } else if (hasMigratedDataRecovered) {
 291             log.info("{}: Snapshot capture initiated after recovery due to migrated messages", context.getId());
 292
 293             context.getSnapshotManager().capture(replicatedLog().last(), -1);
 294         } else {
 295             possiblyRestoreFromSnapshot();
 296         }
 297     }
 298
 299     private static boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry) {
 300         return repLogEntry.getData() instanceof ServerConfigurationPayload;
 301     }
 302
 303     private static boolean isPersistentPayload(ReplicatedLogEntry repLogEntry) {
 304         return repLogEntry.getData() instanceof PersistentPayload;
 305     }
 306
 307     private static boolean isMigratedPayload(ReplicatedLogEntry repLogEntry) {
 308         return isMigratedSerializable(repLogEntry.getData());
 309     }
 310
 311     private static boolean isMigratedSerializable(Object message) {
 312         return message instanceof MigratedSerializable && ((MigratedSerializable)message).isMigrated();
 313     }
 314 }