2 * Copyright (c) 2014 Cisco 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
9 package org.opendaylight.controller.cluster.raft;
11 import akka.japi.Procedure;
12 import akka.persistence.SnapshotSelectionCriteria;
13 import com.google.protobuf.ByteString;
14 import org.opendaylight.controller.cluster.DataPersistenceProvider;
15 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
16 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
17 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
18 import org.slf4j.Logger;
20 public class SnapshotManager implements SnapshotState {
23 private final SnapshotState IDLE = new Idle();
24 private final SnapshotState CAPTURING = new Capturing();
25 private final SnapshotState PERSISTING = new Persisting();
26 private final SnapshotState CREATING = new Creating();
28 private final Logger LOG;
29 private final RaftActorContext context;
30 private final LastAppliedTermInformationReader lastAppliedTermInformationReader =
31 new LastAppliedTermInformationReader();
32 private final ReplicatedToAllTermInformationReader replicatedToAllTermInformationReader =
33 new ReplicatedToAllTermInformationReader();
36 private SnapshotState currentState = IDLE;
37 private CaptureSnapshot captureSnapshot;
39 public SnapshotManager(RaftActorContext context, Logger logger) {
40 this.context = context;
45 public boolean isCapturing() {
46 return currentState.isCapturing();
50 public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
51 return currentState.captureToInstall(lastLogEntry, replicatedToAllIndex, targetFollower);
55 public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
56 return currentState.capture(lastLogEntry, replicatedToAllIndex);
60 public void create(Procedure<Void> callback) {
61 currentState.create(callback);
65 public void persist(DataPersistenceProvider persistenceProvider, byte[] snapshotBytes,
66 RaftActorBehavior currentBehavior, long totalMemory) {
67 currentState.persist(persistenceProvider, snapshotBytes, currentBehavior, totalMemory);
71 public void commit(DataPersistenceProvider persistenceProvider, long sequenceNumber) {
72 currentState.commit(persistenceProvider, sequenceNumber);
76 public void rollback() {
77 currentState.rollback();
81 public long trimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior) {
82 return currentState.trimLog(desiredTrimIndex, currentBehavior);
85 private boolean hasFollowers(){
86 return context.getPeerAddresses().keySet().size() > 0;
89 private String persistenceId(){
90 return context.getId();
93 private class AbstractSnapshotState implements SnapshotState {
96 public boolean isCapturing() {
101 public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
102 LOG.debug("capture should not be called in state {}", this);
107 public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
108 LOG.debug("captureToInstall should not be called in state {}", this);
113 public void create(Procedure<Void> callback) {
114 LOG.debug("create should not be called in state {}", this);
118 public void persist(DataPersistenceProvider persistenceProvider, byte[] snapshotBytes,
119 RaftActorBehavior currentBehavior, long totalMemory) {
120 LOG.debug("persist should not be called in state {}", this);
124 public void commit(DataPersistenceProvider persistenceProvider, long sequenceNumber) {
125 LOG.debug("commit should not be called in state {}", this);
129 public void rollback() {
130 LOG.debug("rollback should not be called in state {}", this);
134 public long trimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior) {
135 LOG.debug("trimLog should not be called in state {}", this);
139 protected long doTrimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior){
140 // we would want to keep the lastApplied as its used while capturing snapshots
141 long lastApplied = context.getLastApplied();
142 long tempMin = Math.min(desiredTrimIndex, (lastApplied > -1 ? lastApplied - 1 : -1));
144 if(LOG.isTraceEnabled()) {
145 LOG.trace("{}: performSnapshotWithoutCapture: desiredTrimIndex: {}, lastApplied: {}, tempMin: {}",
146 persistenceId(), desiredTrimIndex, lastApplied, tempMin);
149 if (tempMin > -1 && context.getReplicatedLog().isPresent(tempMin)) {
150 LOG.debug("{}: fakeSnapshot purging log to {} for term {}", persistenceId(), tempMin,
151 context.getTermInformation().getCurrentTerm());
153 //use the term of the temp-min, since we check for isPresent, entry will not be null
154 ReplicatedLogEntry entry = context.getReplicatedLog().get(tempMin);
155 context.getReplicatedLog().snapshotPreCommit(tempMin, entry.getTerm());
156 context.getReplicatedLog().snapshotCommit();
158 } else if(tempMin > currentBehavior.getReplicatedToAllIndex()) {
159 // It's possible a follower was lagging and an install snapshot advanced its match index past
160 // the current replicatedToAllIndex. Since the follower is now caught up we should advance the
161 // replicatedToAllIndex (to tempMin). The fact that tempMin wasn't found in the log is likely
162 // due to a previous snapshot triggered by the memory threshold exceeded, in that case we
163 // trim the log to the last applied index even if previous entries weren't replicated to all followers.
164 currentBehavior.setReplicatedToAllIndex(tempMin);
170 private class Idle extends AbstractSnapshotState {
172 private boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
173 TermInformationReader lastAppliedTermInfoReader =
174 lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
175 lastLogEntry, hasFollowers());
177 long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
178 long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
180 TermInformationReader replicatedToAllTermInfoReader =
181 replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
183 long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
184 long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
186 // send a CaptureSnapshot to self to make the expensive operation async.
187 captureSnapshot = new CaptureSnapshot(lastLogEntry.getIndex(),
188 lastLogEntry.getTerm(), lastAppliedIndex, lastAppliedTerm,
189 newReplicatedToAllIndex, newReplicatedToAllTerm, targetFollower!=null);
191 SnapshotManager.this.currentState = CAPTURING;
193 if(targetFollower != null){
194 LOG.info("{}: Initiating snapshot capture {}", persistenceId(), captureSnapshot);
196 LOG.info("{}: Initiating snapshot capture {} to install on {}",
197 persistenceId(), captureSnapshot, targetFollower);
200 context.getActor().tell(captureSnapshot, context.getActor());
206 public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
207 return capture(lastLogEntry, replicatedToAllIndex, null);
211 public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
212 return capture(lastLogEntry, replicatedToAllIndex, targetFollower);
216 public String toString() {
221 public long trimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior) {
222 return doTrimLog(desiredTrimIndex, currentBehavior);
226 private class Capturing extends AbstractSnapshotState {
229 public boolean isCapturing() {
234 public void create(Procedure<Void> callback) {
236 callback.apply(null);
237 SnapshotManager.this.currentState = CREATING;
238 } catch (Exception e) {
239 LOG.error("Unexpected error occurred", e);
244 public String toString() {
250 private class Creating extends AbstractSnapshotState {
253 public boolean isCapturing() {
258 public void persist(DataPersistenceProvider persistenceProvider, byte[] snapshotBytes,
259 RaftActorBehavior currentBehavior, long totalMemory) {
260 // create a snapshot object from the state provided and save it
261 // when snapshot is saved async, SaveSnapshotSuccess is raised.
263 Snapshot sn = Snapshot.create(snapshotBytes,
264 context.getReplicatedLog().getFrom(captureSnapshot.getLastAppliedIndex() + 1),
265 captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
266 captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
268 persistenceProvider.saveSnapshot(sn);
270 LOG.info("{}: Persisting of snapshot done:{}", persistenceId(), sn.getLogMessage());
272 long dataThreshold = totalMemory *
273 context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
274 if (context.getReplicatedLog().dataSize() > dataThreshold) {
276 if(LOG.isDebugEnabled()) {
277 LOG.debug("{}: dataSize {} exceeds dataThreshold {} - doing snapshotPreCommit with index {}",
278 persistenceId(), context.getReplicatedLog().dataSize(), dataThreshold,
279 captureSnapshot.getLastAppliedIndex());
282 // if memory is less, clear the log based on lastApplied.
283 // this could/should only happen if one of the followers is down
284 // as normally we keep removing from the log when its replicated to all.
285 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
286 captureSnapshot.getLastAppliedTerm());
288 // Don't reset replicatedToAllIndex to -1 as this may prevent us from trimming the log after an
289 // install snapshot to a follower.
290 if(captureSnapshot.getReplicatedToAllIndex() >= 0) {
291 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
294 } else if(captureSnapshot.getReplicatedToAllIndex() != -1){
295 // clear the log based on replicatedToAllIndex
296 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(),
297 captureSnapshot.getReplicatedToAllTerm());
299 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
301 // The replicatedToAllIndex was not found in the log
302 // This means that replicatedToAllIndex never moved beyond -1 or that it is already in the snapshot.
303 // In this scenario we may need to save the snapshot to the akka persistence
304 // snapshot for recovery but we do not need to do the replicated log trimming.
305 context.getReplicatedLog().snapshotPreCommit(context.getReplicatedLog().getSnapshotIndex(),
306 context.getReplicatedLog().getSnapshotTerm());
309 LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
310 "and term:{}", persistenceId(), captureSnapshot.getLastAppliedIndex(),
311 captureSnapshot.getLastAppliedTerm());
313 if (context.getId().equals(currentBehavior.getLeaderId())
314 && captureSnapshot.isInstallSnapshotInitiated()) {
315 // this would be call straight to the leader and won't initiate in serialization
316 currentBehavior.handleMessage(context.getActor(), new SendInstallSnapshot(
317 ByteString.copyFrom(snapshotBytes)));
320 captureSnapshot = null;
321 SnapshotManager.this.currentState = PERSISTING;
325 public String toString() {
331 private class Persisting extends AbstractSnapshotState {
334 public void commit(DataPersistenceProvider persistenceProvider, long sequenceNumber) {
335 context.getReplicatedLog().snapshotCommit();
336 persistenceProvider.deleteSnapshots(new SnapshotSelectionCriteria(
337 sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), 43200000));
339 persistenceProvider.deleteMessages(sequenceNumber);
341 SnapshotManager.this.currentState = IDLE;
345 public void rollback() {
346 context.getReplicatedLog().snapshotRollback();
348 LOG.info("{}: Replicated Log rolled back. Snapshot will be attempted in the next cycle." +
349 "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
350 context.getReplicatedLog().getSnapshotIndex(),
351 context.getReplicatedLog().getSnapshotTerm(),
352 context.getReplicatedLog().size());
354 SnapshotManager.this.currentState = IDLE;
358 public String toString() {
364 private static interface TermInformationReader {
369 static class LastAppliedTermInformationReader implements TermInformationReader{
373 public LastAppliedTermInformationReader init(ReplicatedLog log, long originalIndex,
374 ReplicatedLogEntry lastLogEntry, boolean hasFollowers){
375 ReplicatedLogEntry entry = log.get(originalIndex);
379 if(lastLogEntry != null) {
380 index = lastLogEntry.getIndex();
381 term = lastLogEntry.getTerm();
383 } else if (entry != null) {
384 index = entry.getIndex();
385 term = entry.getTerm();
386 } else if(log.getSnapshotIndex() > -1){
387 index = log.getSnapshotIndex();
388 term = log.getSnapshotTerm();
394 public long getIndex(){
399 public long getTerm(){
404 private static class ReplicatedToAllTermInformationReader implements TermInformationReader{
408 ReplicatedToAllTermInformationReader init(ReplicatedLog log, long originalIndex){
409 ReplicatedLogEntry entry = log.get(originalIndex);
414 index = entry.getIndex();
415 term = entry.getTerm();
422 public long getIndex(){
427 public long getTerm(){