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.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.Snapshot.State;
23 import org.opendaylight.controller.cluster.raft.persisted.UpdateElectionTerm;
24 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.PersistentPayload;
25 import org.slf4j.Logger;
28 * Support class that handles persistence recovery for a RaftActor.
30 * @author Thomas Pantelis
32 class RaftActorRecoverySupport {
33 private final RaftActorContext context;
34 private final RaftActorRecoveryCohort cohort;
36 private int currentRecoveryBatchCount;
37 private boolean dataRecoveredWithPersistenceDisabled;
38 private boolean anyDataRecovered;
39 private boolean hasMigratedDataRecovered;
41 private Stopwatch recoveryTimer;
42 private final Logger log;
44 RaftActorRecoverySupport(final RaftActorContext context, final RaftActorRecoveryCohort cohort) {
45 this.context = context;
47 this.log = context.getLogger();
50 boolean handleRecoveryMessage(final Object message, final PersistentDataProvider persistentProvider) {
51 log.trace("{}: handleRecoveryMessage: {}", context.getId(), message);
53 anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted);
55 if (isMigratedSerializable(message)) {
56 hasMigratedDataRecovered = true;
59 boolean recoveryComplete = false;
60 if (message instanceof UpdateElectionTerm) {
61 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
62 ((UpdateElectionTerm) message).getVotedFor());
63 } else if (message instanceof SnapshotOffer) {
64 onRecoveredSnapshot((SnapshotOffer) message);
65 } else if (message instanceof ReplicatedLogEntry) {
66 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
67 } else if (message instanceof ApplyJournalEntries) {
68 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
69 } else if (message instanceof DeleteEntries) {
70 onDeleteEntries((DeleteEntries) message);
71 } else if (message instanceof ServerConfigurationPayload) {
72 context.updatePeerIds((ServerConfigurationPayload)message);
73 } else if (message instanceof RecoveryCompleted) {
74 recoveryComplete = true;
75 onRecoveryCompletedMessage(persistentProvider);
78 return recoveryComplete;
81 @SuppressWarnings("checkstyle:IllegalCatch")
82 private void possiblyRestoreFromSnapshot() {
83 Snapshot restoreFromSnapshot = cohort.getRestoreFromSnapshot();
84 if (restoreFromSnapshot == null) {
88 if (anyDataRecovered) {
89 log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty",
94 log.debug("{}: Restore snapshot: {}", context.getId(), restoreFromSnapshot);
96 context.getSnapshotManager().apply(new ApplySnapshot(restoreFromSnapshot));
99 private ReplicatedLog replicatedLog() {
100 return context.getReplicatedLog();
103 private void initRecoveryTimer() {
104 if (recoveryTimer == null) {
105 recoveryTimer = Stopwatch.createStarted();
109 private void onRecoveredSnapshot(final SnapshotOffer offer) {
110 log.debug("{}: SnapshotOffer called.", context.getId());
114 Snapshot snapshot = (Snapshot) offer.snapshot();
116 for (ReplicatedLogEntry entry: snapshot.getUnAppliedEntries()) {
117 if (isMigratedPayload(entry)) {
118 hasMigratedDataRecovered = true;
122 if (!context.getPersistenceProvider().isRecoveryApplicable()) {
123 // We may have just transitioned to disabled and have a snapshot containing state data and/or log
124 // entries - we don't want to preserve these, only the server config and election term info.
126 snapshot = Snapshot.create(
127 EmptyState.INSTANCE, Collections.emptyList(), -1, -1, -1, -1,
128 snapshot.getElectionTerm(), snapshot.getElectionVotedFor(), snapshot.getServerConfiguration());
131 // Create a replicated log with the snapshot information
132 // The replicated log can be used later on to retrieve this snapshot
133 // when we need to install it on a peer
135 context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
136 context.setLastApplied(snapshot.getLastAppliedIndex());
137 context.setCommitIndex(snapshot.getLastAppliedIndex());
138 context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
140 final Stopwatch timer = Stopwatch.createStarted();
142 // Apply the snapshot to the actors state
143 final State snapshotState = snapshot.getState();
144 if (snapshotState.needsMigration()) {
145 hasMigratedDataRecovered = true;
147 if (!(snapshotState instanceof EmptyState)) {
148 cohort.applyRecoverySnapshot(snapshotState);
151 if (snapshot.getServerConfiguration() != null) {
152 context.updatePeerIds(snapshot.getServerConfiguration());
156 log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
157 context.getId(), timer, replicatedLog().getSnapshotIndex(), replicatedLog().getSnapshotTerm(),
158 replicatedLog().size());
161 private void onRecoveredJournalLogEntry(final ReplicatedLogEntry logEntry) {
162 if (log.isDebugEnabled()) {
163 log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
164 logEntry.getIndex(), logEntry.size());
167 if (isServerConfigurationPayload(logEntry)) {
168 context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
171 if (isMigratedPayload(logEntry)) {
172 hasMigratedDataRecovered = true;
175 if (context.getPersistenceProvider().isRecoveryApplicable()) {
176 replicatedLog().append(logEntry);
177 } else if (!isPersistentPayload(logEntry)) {
178 dataRecoveredWithPersistenceDisabled = true;
182 private void onRecoveredApplyLogEntries(final long toIndex) {
183 if (!context.getPersistenceProvider().isRecoveryApplicable()) {
184 dataRecoveredWithPersistenceDisabled = true;
188 long lastUnappliedIndex = context.getLastApplied() + 1;
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);
197 long lastApplied = lastUnappliedIndex - 1;
198 for (long i = lastUnappliedIndex; i <= toIndex; i++) {
199 ReplicatedLogEntry logEntry = replicatedLog().get(i);
200 if (logEntry != null) {
202 batchRecoveredLogEntry(logEntry);
204 // Shouldn't happen but cover it anyway.
205 log.error("{}: Log entry not found for index {}", context.getId(), i);
210 context.setLastApplied(lastApplied);
211 context.setCommitIndex(lastApplied);
214 private void onDeleteEntries(final DeleteEntries deleteEntries) {
215 if (context.getPersistenceProvider().isRecoveryApplicable()) {
216 replicatedLog().removeFrom(deleteEntries.getFromIndex());
218 dataRecoveredWithPersistenceDisabled = true;
222 private void batchRecoveredLogEntry(final ReplicatedLogEntry logEntry) {
225 int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
226 if (!isServerConfigurationPayload(logEntry)) {
227 if (currentRecoveryBatchCount == 0) {
228 cohort.startLogRecoveryBatch(batchSize);
231 cohort.appendRecoveredLogEntry(logEntry.getData());
233 if (++currentRecoveryBatchCount >= batchSize) {
234 endCurrentLogRecoveryBatch();
239 private void endCurrentLogRecoveryBatch() {
240 cohort.applyCurrentLogRecoveryBatch();
241 currentRecoveryBatchCount = 0;
244 private void onRecoveryCompletedMessage(final PersistentDataProvider persistentProvider) {
245 if (currentRecoveryBatchCount > 0) {
246 endCurrentLogRecoveryBatch();
249 final String recoveryTime;
250 if (recoveryTimer != null) {
251 recoveryTime = " in " + recoveryTimer.stop();
252 recoveryTimer = null;
257 log.info("{}: Recovery completed {} - Switching actor to Follower - last log index = {}, last log term = {}, "
258 + "snapshot index = {}, snapshot term = {}, journal size = {}", context.getId(), recoveryTime,
259 replicatedLog().lastIndex(), replicatedLog().lastTerm(), replicatedLog().getSnapshotIndex(),
260 replicatedLog().getSnapshotTerm(), replicatedLog().size());
262 if (dataRecoveredWithPersistenceDisabled
263 || hasMigratedDataRecovered && !context.getPersistenceProvider().isRecoveryApplicable()) {
264 if (hasMigratedDataRecovered) {
265 log.info("{}: Saving snapshot after recovery due to migrated messages", context.getId());
267 log.info("{}: Saving snapshot after recovery due to data persistence disabled", context.getId());
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.
275 Snapshot snapshot = Snapshot.create(
276 EmptyState.INSTANCE, 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(final ReplicatedLogEntry repLogEntry) {
294 return repLogEntry.getData() instanceof ServerConfigurationPayload;
297 private static boolean isPersistentPayload(final ReplicatedLogEntry repLogEntry) {
298 return repLogEntry.getData() instanceof PersistentPayload;
301 private static boolean isMigratedPayload(final ReplicatedLogEntry repLogEntry) {
302 return isMigratedSerializable(repLogEntry.getData());
305 private static boolean isMigratedSerializable(final Object message) {
306 return message instanceof MigratedSerializable && ((MigratedSerializable)message).isMigrated();