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.io.ByteArrayInputStream;
  14 import java.io.ObjectInputStream;
  15 import java.util.Collections;
  16 import org.opendaylight.controller.cluster.PersistentDataProvider;
  17 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  18 import org.opendaylight.controller.cluster.raft.persisted.ApplyJournalEntries;
  19 import org.opendaylight.controller.cluster.raft.persisted.DeleteEntries;
  20 import org.opendaylight.controller.cluster.raft.persisted.MigratedSerializable;
  21 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  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  * Support class that handles persistence recovery for a RaftActor.
  27  *
  28  * @author Thomas Pantelis
  29  */
  30 class RaftActorRecoverySupport {
  31     private final RaftActorContext context;
  32     private final RaftActorRecoveryCohort cohort;
  33
  34     private int currentRecoveryBatchCount;
  35     private boolean dataRecoveredWithPersistenceDisabled;
  36     private boolean anyDataRecovered;
  37     private boolean hasMigratedDataRecovered;
  38
  39     private Stopwatch recoveryTimer;
  40     private final Logger log;
  41
  42     RaftActorRecoverySupport(final RaftActorContext context, final RaftActorRecoveryCohort cohort) {
  43         this.context = context;
  44         this.cohort = cohort;
  45         this.log = context.getLogger();
  46     }
  47
  48     boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) {
  49         log.trace("{}: handleRecoveryMessage: {}", context.getId(), message);
  50
  51         anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted);
  52
  53         if(isMigratedSerializable(message)) {
  54             hasMigratedDataRecovered = true;
  55         }
  56
  57         boolean recoveryComplete = false;
  58         if (message instanceof UpdateElectionTerm) {
  59             context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
  60                     ((UpdateElectionTerm) message).getVotedFor());
  61         } else if (message instanceof SnapshotOffer) {
  62             onRecoveredSnapshot((SnapshotOffer) message);
  63         } else if (message instanceof ReplicatedLogEntry) {
  64             onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
  65         } else if (message instanceof ApplyJournalEntries) {
  66             onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
  67         } else if (message instanceof DeleteEntries) {
  68             onDeleteEntries((DeleteEntries) message);
  69         } else if (message instanceof RecoveryCompleted) {
  70             recoveryComplete = true;
  71             onRecoveryCompletedMessage(persistentProvider);
  72         }
  73
  74         return recoveryComplete;
  75     }
  76
  77     private void possiblyRestoreFromSnapshot() {
  78         byte[] restoreFromSnapshot = cohort.getRestoreFromSnapshot();
  79         if(restoreFromSnapshot == null) {
  80             return;
  81         }
  82
  83         if(anyDataRecovered) {
  84             log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty",
  85                     context.getId());
  86             return;
  87         }
  88
  89         try(ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(restoreFromSnapshot))) {
  90             Snapshot snapshot = (Snapshot) ois.readObject();
  91
  92             log.debug("{}: Deserialized restore snapshot: {}", context.getId(), snapshot);
  93
  94             context.getSnapshotManager().apply(new ApplySnapshot(snapshot));
  95         } catch(Exception e) {
  96             log.error("{}: Error deserializing snapshot restore", context.getId(), e);
  97         }
  98     }
  99
 100     private ReplicatedLog replicatedLog() {
 101         return context.getReplicatedLog();
 102     }
 103
 104     private void initRecoveryTimer() {
 105         if(recoveryTimer == null) {
 106             recoveryTimer = Stopwatch.createStarted();
 107         }
 108     }
 109
 110     private void onRecoveredSnapshot(SnapshotOffer offer) {
 111         if(log.isDebugEnabled()) {
 112             log.debug("{}: SnapshotOffer called..", context.getId());
 113         }
 114
 115         initRecoveryTimer();
 116
 117         Snapshot snapshot = (Snapshot) offer.snapshot();
 118
 119         for(ReplicatedLogEntry entry: snapshot.getUnAppliedEntries()) {
 120             if(isMigratedPayload(entry)) {
 121                 hasMigratedDataRecovered = true;
 122             }
 123         }
 124
 125         if(!context.getPersistenceProvider().isRecoveryApplicable()) {
 126             // We may have just transitioned to disabled and have a snapshot containing state data and/or log
 127             // entries - we don't want to preserve these, only the server config and election term info.
 128
 129             snapshot = Snapshot.create(new byte[0], Collections.emptyList(), -1, -1, -1, -1,
 130                     snapshot.getElectionTerm(), snapshot.getElectionVotedFor(), snapshot.getServerConfiguration());
 131         }
 132
 133         // Create a replicated log with the snapshot information
 134         // The replicated log can be used later on to retrieve this snapshot
 135         // when we need to install it on a peer
 136
 137         context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
 138         context.setLastApplied(snapshot.getLastAppliedIndex());
 139         context.setCommitIndex(snapshot.getLastAppliedIndex());
 140         context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 141
 142         Stopwatch timer = Stopwatch.createStarted();
 143
 144         // Apply the snapshot to the actors state
 145         cohort.applyRecoverySnapshot(snapshot.getState());
 146
 147         if (snapshot.getServerConfiguration() != null) {
 148             context.updatePeerIds(snapshot.getServerConfiguration());
 149
 150             if(isMigratedSerializable(snapshot.getServerConfiguration())) {
 151                 hasMigratedDataRecovered = true;
 152             }
 153         }
 154
 155         timer.stop();
 156         log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
 157                 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
 158                 replicatedLog().getSnapshotTerm(), replicatedLog().size());
 159     }
 160
 161     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 162         if(log.isDebugEnabled()) {
 163             log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
 164                     logEntry.getIndex(), logEntry.size());
 165         }
 166
 167         if(isServerConfigurationPayload(logEntry)){
 168             context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
 169         }
 170
 171         if(isMigratedPayload(logEntry)) {
 172             hasMigratedDataRecovered = true;
 173         }
 174
 175         if(context.getPersistenceProvider().isRecoveryApplicable()) {
 176             replicatedLog().append(logEntry);
 177         } else if(!isPersistentPayload(logEntry)) {
 178             dataRecoveredWithPersistenceDisabled = true;
 179         }
 180     }
 181
 182     private void onRecoveredApplyLogEntries(long toIndex) {
 183         if(!context.getPersistenceProvider().isRecoveryApplicable()) {
 184             dataRecoveredWithPersistenceDisabled = true;
 185             return;
 186         }
 187
 188         long lastUnappliedIndex = context.getLastApplied() + 1;
 189
 190         if(log.isDebugEnabled()) {
 191             // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
 192             // but the entry itself has made it to that state and recovered via the snapshot
 193             log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
 194                     context.getId(), lastUnappliedIndex, toIndex);
 195         }
 196
 197         long lastApplied = lastUnappliedIndex - 1;
 198         for (long i = lastUnappliedIndex; i <= toIndex; i++) {
 199             ReplicatedLogEntry logEntry = replicatedLog().get(i);
 200             if(logEntry != null) {
 201                 lastApplied++;
 202                 batchRecoveredLogEntry(logEntry);
 203             } else {
 204                 // Shouldn't happen but cover it anyway.
 205                 log.error("{}: Log entry not found for index {}", context.getId(), i);
 206                 break;
 207             }
 208         }
 209
 210         context.setLastApplied(lastApplied);
 211         context.setCommitIndex(lastApplied);
 212     }
 213
 214     private void onDeleteEntries(DeleteEntries deleteEntries) {
 215         if(context.getPersistenceProvider().isRecoveryApplicable()) {
 216             replicatedLog().removeFrom(deleteEntries.getFromIndex());
 217         } else {
 218             dataRecoveredWithPersistenceDisabled = true;
 219         }
 220     }
 221
 222     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 223         initRecoveryTimer();
 224
 225         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 226         if(!isServerConfigurationPayload(logEntry)){
 227             if(currentRecoveryBatchCount == 0) {
 228                 cohort.startLogRecoveryBatch(batchSize);
 229             }
 230
 231             cohort.appendRecoveredLogEntry(logEntry.getData());
 232
 233             if(++currentRecoveryBatchCount >= batchSize) {
 234                 endCurrentLogRecoveryBatch();
 235             }
 236         }
 237     }
 238
 239     private void endCurrentLogRecoveryBatch() {
 240         cohort.applyCurrentLogRecoveryBatch();
 241         currentRecoveryBatchCount = 0;
 242     }
 243
 244     private void onRecoveryCompletedMessage(PersistentDataProvider persistentProvider) {
 245         if(currentRecoveryBatchCount > 0) {
 246             endCurrentLogRecoveryBatch();
 247         }
 248
 249         String recoveryTime = "";
 250         if(recoveryTimer != null) {
 251             recoveryTimer.stop();
 252             recoveryTime = " in " + recoveryTimer.toString();
 253             recoveryTimer = null;
 254         }
 255
 256         log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 257                  "Persistence Id =  " + context.getId() +
 258                  " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, " +
 259                  "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
 260                  replicatedLog().getSnapshotTerm(), replicatedLog().size());
 261
 262         if(dataRecoveredWithPersistenceDisabled ||
 263                 (hasMigratedDataRecovered && !context.getPersistenceProvider().isRecoveryApplicable())) {
 264             if(hasMigratedDataRecovered) {
 265                 log.info("{}: Saving snapshot after recovery due to migrated messages", context.getId());
 266             } else {
 267                 log.info("{}: Saving snapshot after recovery due to data persistence disabled", context.getId());
 268             }
 269
 270             // Either data persistence is disabled and we recovered some data entries (ie we must have just
 271             // transitioned to disabled or a persistence backup was restored) or we recovered migrated
 272             // messages. Either way, we persist a snapshot and delete all the messages from the akka journal
 273             // to clean out unwanted messages.
 274
 275             Snapshot snapshot = Snapshot.create(new byte[0], Collections.<ReplicatedLogEntry>emptyList(), -1, -1, -1, -1,
 276                     context.getTermInformation().getCurrentTerm(), context.getTermInformation().getVotedFor(),
 277                     context.getPeerServerInfo(true));
 278
 279             persistentProvider.saveSnapshot(snapshot);
 280
 281             persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
 282         } else if(hasMigratedDataRecovered) {
 283             log.info("{}: Snapshot capture initiated after recovery due to migrated messages", context.getId());
 284
 285             context.getSnapshotManager().capture(replicatedLog().last(), -1);
 286         } else {
 287             possiblyRestoreFromSnapshot();
 288         }
 289     }
 290
 291     private static boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry){
 292         return repLogEntry.getData() instanceof ServerConfigurationPayload;
 293     }
 294
 295     private static boolean isPersistentPayload(ReplicatedLogEntry repLogEntry){
 296         return repLogEntry.getData() instanceof PersistentPayload;
 297     }
 298
 299     private static boolean isMigratedPayload(ReplicatedLogEntry repLogEntry){
 300         return isMigratedSerializable(repLogEntry.getData());
 301     }
 302
 303     private static boolean isMigratedSerializable(Object message){
 304         return message instanceof MigratedSerializable && ((MigratedSerializable)message).isMigrated();
 305     }
 306 }