2 * Copyright (c) 2015 Brocade Communications Systems, Inc. and others. All rights reserved.
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
8 package org.opendaylight.controller.cluster.raft;
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;
27 * Support class that handles persistence recovery for a RaftActor.
29 * @author Thomas Pantelis
31 class RaftActorRecoverySupport {
32 private final RaftActorContext context;
33 private final RaftActorRecoveryCohort cohort;
35 private int currentRecoveryBatchCount;
36 private boolean dataRecoveredWithPersistenceDisabled;
37 private boolean anyDataRecovered;
38 private boolean hasMigratedDataRecovered;
40 private Stopwatch recoveryTimer;
41 private final Logger log;
43 RaftActorRecoverySupport(final RaftActorContext context, final RaftActorRecoveryCohort cohort) {
44 this.context = context;
46 this.log = context.getLogger();
49 boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) {
50 log.trace("{}: handleRecoveryMessage: {}", context.getId(), message);
52 anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted);
54 if (isMigratedSerializable(message)) {
55 hasMigratedDataRecovered = true;
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);
77 return recoveryComplete;
80 @SuppressWarnings("checkstyle:IllegalCatch")
81 private void possiblyRestoreFromSnapshot() {
82 byte[] restoreFromSnapshot = cohort.getRestoreFromSnapshot();
83 if (restoreFromSnapshot == null) {
87 if (anyDataRecovered) {
88 log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty",
93 try (ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(restoreFromSnapshot))) {
94 Snapshot snapshot = (Snapshot) ois.readObject();
96 log.debug("{}: Deserialized restore snapshot: {}", context.getId(), snapshot);
98 context.getSnapshotManager().apply(new ApplySnapshot(snapshot));
99 } catch (Exception e) {
100 log.error("{}: Error deserializing snapshot restore", context.getId(), e);
104 private ReplicatedLog replicatedLog() {
105 return context.getReplicatedLog();
108 private void initRecoveryTimer() {
109 if (recoveryTimer == null) {
110 recoveryTimer = Stopwatch.createStarted();
114 private void onRecoveredSnapshot(SnapshotOffer offer) {
115 log.debug("{}: SnapshotOffer called..", context.getId());
119 Snapshot snapshot = (Snapshot) offer.snapshot();
121 for (ReplicatedLogEntry entry: snapshot.getUnAppliedEntries()) {
122 if (isMigratedPayload(entry)) {
123 hasMigratedDataRecovered = true;
127 if (!context.getPersistenceProvider().isRecoveryApplicable()) {
128 // We may have just transitioned to disabled and have a snapshot containing state data and/or log
129 // entries - we don't want to preserve these, only the server config and election term info.
131 snapshot = Snapshot.create(new byte[0], Collections.emptyList(), -1, -1, -1, -1,
132 snapshot.getElectionTerm(), snapshot.getElectionVotedFor(), snapshot.getServerConfiguration());
135 // Create a replicated log with the snapshot information
136 // The replicated log can be used later on to retrieve this snapshot
137 // when we need to install it on a peer
139 context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
140 context.setLastApplied(snapshot.getLastAppliedIndex());
141 context.setCommitIndex(snapshot.getLastAppliedIndex());
142 context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
144 Stopwatch timer = Stopwatch.createStarted();
146 // Apply the snapshot to the actors state
147 cohort.applyRecoverySnapshot(snapshot.getState());
149 if (snapshot.getServerConfiguration() != null) {
150 context.updatePeerIds(snapshot.getServerConfiguration());
152 if (isMigratedSerializable(snapshot.getServerConfiguration())) {
153 hasMigratedDataRecovered = true;
158 log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
159 context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
160 replicatedLog().getSnapshotTerm(), replicatedLog().size());
163 private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
164 if (log.isDebugEnabled()) {
165 log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
166 logEntry.getIndex(), logEntry.size());
169 if (isServerConfigurationPayload(logEntry)) {
170 context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
173 if (isMigratedPayload(logEntry)) {
174 hasMigratedDataRecovered = true;
177 if (context.getPersistenceProvider().isRecoveryApplicable()) {
178 replicatedLog().append(logEntry);
179 } else if (!isPersistentPayload(logEntry)) {
180 dataRecoveredWithPersistenceDisabled = true;
184 private void onRecoveredApplyLogEntries(long toIndex) {
185 if (!context.getPersistenceProvider().isRecoveryApplicable()) {
186 dataRecoveredWithPersistenceDisabled = true;
190 long lastUnappliedIndex = context.getLastApplied() + 1;
192 if (log.isDebugEnabled()) {
193 // it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
194 // but the entry itself has made it to that state and recovered via the snapshot
195 log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
196 context.getId(), lastUnappliedIndex, toIndex);
199 long lastApplied = lastUnappliedIndex - 1;
200 for (long i = lastUnappliedIndex; i <= toIndex; i++) {
201 ReplicatedLogEntry logEntry = replicatedLog().get(i);
202 if (logEntry != null) {
204 batchRecoveredLogEntry(logEntry);
206 // Shouldn't happen but cover it anyway.
207 log.error("{}: Log entry not found for index {}", context.getId(), i);
212 context.setLastApplied(lastApplied);
213 context.setCommitIndex(lastApplied);
216 private void onDeleteEntries(DeleteEntries deleteEntries) {
217 if (context.getPersistenceProvider().isRecoveryApplicable()) {
218 replicatedLog().removeFrom(deleteEntries.getFromIndex());
220 dataRecoveredWithPersistenceDisabled = true;
224 private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
227 int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
228 if (!isServerConfigurationPayload(logEntry)) {
229 if (currentRecoveryBatchCount == 0) {
230 cohort.startLogRecoveryBatch(batchSize);
233 cohort.appendRecoveredLogEntry(logEntry.getData());
235 if (++currentRecoveryBatchCount >= batchSize) {
236 endCurrentLogRecoveryBatch();
241 private void endCurrentLogRecoveryBatch() {
242 cohort.applyCurrentLogRecoveryBatch();
243 currentRecoveryBatchCount = 0;
246 private void onRecoveryCompletedMessage(PersistentDataProvider persistentProvider) {
247 if (currentRecoveryBatchCount > 0) {
248 endCurrentLogRecoveryBatch();
251 String recoveryTime = "";
252 if (recoveryTimer != null) {
253 recoveryTimer.stop();
254 recoveryTime = " in " + recoveryTimer.toString();
255 recoveryTimer = null;
258 log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " + "Persistence Id = "
259 + context.getId() + " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, "
260 + "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
261 replicatedLog().getSnapshotTerm(), replicatedLog().size());
263 if (dataRecoveredWithPersistenceDisabled
264 || hasMigratedDataRecovered && !context.getPersistenceProvider().isRecoveryApplicable()) {
265 if (hasMigratedDataRecovered) {
266 log.info("{}: Saving snapshot after recovery due to migrated messages", context.getId());
268 log.info("{}: Saving snapshot after recovery due to data persistence disabled", context.getId());
271 // Either data persistence is disabled and we recovered some data entries (ie we must have just
272 // transitioned to disabled or a persistence backup was restored) or we recovered migrated
273 // messages. Either way, we persist a snapshot and delete all the messages from the akka journal
274 // to clean out unwanted messages.
276 Snapshot snapshot = Snapshot.create(new byte[0], Collections.<ReplicatedLogEntry>emptyList(),
278 context.getTermInformation().getCurrentTerm(), context.getTermInformation().getVotedFor(),
279 context.getPeerServerInfo(true));
281 persistentProvider.saveSnapshot(snapshot);
283 persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
284 } else if (hasMigratedDataRecovered) {
285 log.info("{}: Snapshot capture initiated after recovery due to migrated messages", context.getId());
287 context.getSnapshotManager().capture(replicatedLog().last(), -1);
289 possiblyRestoreFromSnapshot();
293 private static boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry) {
294 return repLogEntry.getData() instanceof ServerConfigurationPayload;
297 private static boolean isPersistentPayload(ReplicatedLogEntry repLogEntry) {
298 return repLogEntry.getData() instanceof PersistentPayload;
301 private static boolean isMigratedPayload(ReplicatedLogEntry repLogEntry) {
302 return isMigratedSerializable(repLogEntry.getData());
305 private static boolean isMigratedSerializable(Object message) {
306 return message instanceof MigratedSerializable && ((MigratedSerializable)message).isMigrated();