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.common.annotations.VisibleForTesting;
14 import java.util.List;
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 {
22 private final SnapshotState IDLE = new Idle();
23 private final SnapshotState PERSISTING = new Persisting();
24 private final SnapshotState CREATING = new Creating();
26 private final Logger LOG;
27 private final RaftActorContext context;
28 private final LastAppliedTermInformationReader lastAppliedTermInformationReader =
29 new LastAppliedTermInformationReader();
30 private final ReplicatedToAllTermInformationReader replicatedToAllTermInformationReader =
31 new ReplicatedToAllTermInformationReader();
34 private SnapshotState currentState = IDLE;
35 private CaptureSnapshot captureSnapshot;
36 private long lastSequenceNumber = -1;
38 private Procedure<Void> createSnapshotProcedure;
40 private Snapshot applySnapshot;
41 private Procedure<byte[]> applySnapshotProcedure;
43 public SnapshotManager(RaftActorContext context, Logger logger) {
44 this.context = context;
49 public boolean isCapturing() {
50 return currentState.isCapturing();
54 public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
55 return currentState.captureToInstall(lastLogEntry, replicatedToAllIndex, targetFollower);
59 public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
60 return currentState.capture(lastLogEntry, replicatedToAllIndex);
64 public void apply(Snapshot snapshot) {
65 currentState.apply(snapshot);
69 public void persist(byte[] snapshotBytes, RaftActorBehavior currentBehavior, long totalMemory) {
70 currentState.persist(snapshotBytes, currentBehavior, totalMemory);
74 public void commit(long sequenceNumber, RaftActorBehavior currentBehavior) {
75 currentState.commit(sequenceNumber, currentBehavior);
79 public void rollback() {
80 currentState.rollback();
84 public long trimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior) {
85 return currentState.trimLog(desiredTrimIndex, currentBehavior);
88 public void setCreateSnapshotCallable(Procedure<Void> createSnapshotProcedure) {
89 this.createSnapshotProcedure = createSnapshotProcedure;
92 public void setApplySnapshotProcedure(Procedure<byte[]> applySnapshotProcedure) {
93 this.applySnapshotProcedure = applySnapshotProcedure;
96 public long getLastSequenceNumber() {
97 return lastSequenceNumber;
101 public CaptureSnapshot getCaptureSnapshot() {
102 return captureSnapshot;
105 private boolean hasFollowers(){
106 return context.getPeerAddresses().keySet().size() > 0;
109 private String persistenceId(){
110 return context.getId();
113 private class AbstractSnapshotState implements SnapshotState {
116 public boolean isCapturing() {
121 public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
122 LOG.debug("capture should not be called in state {}", this);
127 public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
128 LOG.debug("captureToInstall should not be called in state {}", this);
133 public void apply(Snapshot snapshot) {
134 LOG.debug("apply should not be called in state {}", this);
138 public void persist(byte[] snapshotBytes, RaftActorBehavior currentBehavior, long totalMemory) {
139 LOG.debug("persist should not be called in state {}", this);
143 public void commit(long sequenceNumber, RaftActorBehavior currentBehavior) {
144 LOG.debug("commit should not be called in state {}", this);
148 public void rollback() {
149 LOG.debug("rollback should not be called in state {}", this);
153 public long trimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior) {
154 LOG.debug("trimLog should not be called in state {}", this);
158 protected long doTrimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior){
159 // we would want to keep the lastApplied as its used while capturing snapshots
160 long lastApplied = context.getLastApplied();
161 long tempMin = Math.min(desiredTrimIndex, (lastApplied > -1 ? lastApplied - 1 : -1));
163 if(LOG.isTraceEnabled()) {
164 LOG.trace("{}: performSnapshotWithoutCapture: desiredTrimIndex: {}, lastApplied: {}, tempMin: {}",
165 persistenceId(), desiredTrimIndex, lastApplied, tempMin);
168 if (tempMin > -1 && context.getReplicatedLog().isPresent(tempMin)) {
169 LOG.debug("{}: fakeSnapshot purging log to {} for term {}", persistenceId(), tempMin,
170 context.getTermInformation().getCurrentTerm());
172 //use the term of the temp-min, since we check for isPresent, entry will not be null
173 ReplicatedLogEntry entry = context.getReplicatedLog().get(tempMin);
174 context.getReplicatedLog().snapshotPreCommit(tempMin, entry.getTerm());
175 context.getReplicatedLog().snapshotCommit();
177 } else if(tempMin > currentBehavior.getReplicatedToAllIndex()) {
178 // It's possible a follower was lagging and an install snapshot advanced its match index past
179 // the current replicatedToAllIndex. Since the follower is now caught up we should advance the
180 // replicatedToAllIndex (to tempMin). The fact that tempMin wasn't found in the log is likely
181 // due to a previous snapshot triggered by the memory threshold exceeded, in that case we
182 // trim the log to the last applied index even if previous entries weren't replicated to all followers.
183 currentBehavior.setReplicatedToAllIndex(tempMin);
189 private class Idle extends AbstractSnapshotState {
192 public boolean isCapturing() {
196 private boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
197 TermInformationReader lastAppliedTermInfoReader =
198 lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
199 lastLogEntry, hasFollowers());
201 long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
202 long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
204 TermInformationReader replicatedToAllTermInfoReader =
205 replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
207 long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
208 long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
210 // send a CaptureSnapshot to self to make the expensive operation async.
212 List<ReplicatedLogEntry> unAppliedEntries = context.getReplicatedLog().getFrom(lastAppliedIndex + 1);
214 long lastLogEntryIndex = lastAppliedIndex;
215 long lastLogEntryTerm = lastAppliedTerm;
216 if(lastLogEntry != null) {
217 lastLogEntryIndex = lastLogEntry.getIndex();
218 lastLogEntryTerm = lastLogEntry.getTerm();
220 LOG.warn("Capturing Snapshot : lastLogEntry is null. Using lastAppliedIndex {} and lastAppliedTerm {} instead.",
221 lastAppliedIndex, lastAppliedTerm);
224 captureSnapshot = new CaptureSnapshot(lastLogEntryIndex,
225 lastLogEntryTerm, lastAppliedIndex, lastAppliedTerm,
226 newReplicatedToAllIndex, newReplicatedToAllTerm, unAppliedEntries, targetFollower != null);
228 if(captureSnapshot.isInstallSnapshotInitiated()) {
229 LOG.info("{}: Initiating snapshot capture {} to install on {}",
230 persistenceId(), captureSnapshot, targetFollower);
232 LOG.info("{}: Initiating snapshot capture {}", persistenceId(), captureSnapshot);
235 lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
237 LOG.debug("lastSequenceNumber prior to capture: {}", lastSequenceNumber);
239 SnapshotManager.this.currentState = CREATING;
242 createSnapshotProcedure.apply(null);
243 } catch (Exception e) {
244 SnapshotManager.this.currentState = IDLE;
245 LOG.error("Error creating snapshot", e);
253 public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
254 return capture(lastLogEntry, replicatedToAllIndex, null);
258 public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
259 return capture(lastLogEntry, replicatedToAllIndex, targetFollower);
263 public void apply(Snapshot snapshot) {
264 applySnapshot = snapshot;
266 lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
268 LOG.debug("lastSequenceNumber prior to persisting applied snapshot: {}", lastSequenceNumber);
270 context.getPersistenceProvider().saveSnapshot(snapshot);
272 SnapshotManager.this.currentState = PERSISTING;
276 public String toString() {
281 public long trimLog(long desiredTrimIndex, RaftActorBehavior currentBehavior) {
282 return doTrimLog(desiredTrimIndex, currentBehavior);
286 private class Creating extends AbstractSnapshotState {
289 public void persist(byte[] snapshotBytes, RaftActorBehavior currentBehavior, long totalMemory) {
290 // create a snapshot object from the state provided and save it
291 // when snapshot is saved async, SaveSnapshotSuccess is raised.
293 Snapshot snapshot = Snapshot.create(snapshotBytes,
294 captureSnapshot.getUnAppliedEntries(),
295 captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
296 captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm(),
297 context.getTermInformation().getCurrentTerm(),
298 context.getTermInformation().getVotedFor());
300 context.getPersistenceProvider().saveSnapshot(snapshot);
302 LOG.info("{}: Persisting of snapshot done: {}", persistenceId(), snapshot);
304 long dataThreshold = totalMemory *
305 context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
306 boolean dataSizeThresholdExceeded = context.getReplicatedLog().dataSize() > dataThreshold;
308 boolean logSizeExceededSnapshotBatchCount =
309 context.getReplicatedLog().size() >= context.getConfigParams().getSnapshotBatchCount();
311 if (dataSizeThresholdExceeded || logSizeExceededSnapshotBatchCount) {
312 if(LOG.isDebugEnabled()) {
313 if(dataSizeThresholdExceeded) {
314 LOG.debug("{}: log data size {} exceeds the memory threshold {} - doing snapshotPreCommit with index {}",
315 context.getId(), context.getReplicatedLog().dataSize(), dataThreshold,
316 captureSnapshot.getLastAppliedIndex());
318 LOG.debug("{}: log size {} exceeds the snapshot batch count {} - doing snapshotPreCommit with index {}",
319 context.getId(), context.getReplicatedLog().size(),
320 context.getConfigParams().getSnapshotBatchCount(), captureSnapshot.getLastAppliedIndex());
324 // We either exceeded the memory threshold or the log size exceeded the snapshot batch
325 // count so, to keep the log memory footprint in check, clear the log based on lastApplied.
326 // This could/should only happen if one of the followers is down as normally we keep
327 // removing from the log as entries are replicated to all.
328 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
329 captureSnapshot.getLastAppliedTerm());
331 // Don't reset replicatedToAllIndex to -1 as this may prevent us from trimming the log after an
332 // install snapshot to a follower.
333 if(captureSnapshot.getReplicatedToAllIndex() >= 0) {
334 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
337 } else if(captureSnapshot.getReplicatedToAllIndex() != -1){
338 // clear the log based on replicatedToAllIndex
339 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(),
340 captureSnapshot.getReplicatedToAllTerm());
342 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
344 // The replicatedToAllIndex was not found in the log
345 // This means that replicatedToAllIndex never moved beyond -1 or that it is already in the snapshot.
346 // In this scenario we may need to save the snapshot to the akka persistence
347 // snapshot for recovery but we do not need to do the replicated log trimming.
348 context.getReplicatedLog().snapshotPreCommit(context.getReplicatedLog().getSnapshotIndex(),
349 context.getReplicatedLog().getSnapshotTerm());
352 LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex: {} " +
353 "and term: {}", context.getId(), context.getReplicatedLog().getSnapshotIndex(),
354 context.getReplicatedLog().getSnapshotTerm());
356 if (context.getId().equals(currentBehavior.getLeaderId())
357 && captureSnapshot.isInstallSnapshotInitiated()) {
358 // this would be call straight to the leader and won't initiate in serialization
359 currentBehavior.handleMessage(context.getActor(), new SendInstallSnapshot(snapshot));
362 captureSnapshot = null;
363 SnapshotManager.this.currentState = PERSISTING;
367 public String toString() {
373 private class Persisting extends AbstractSnapshotState {
376 public void commit(long sequenceNumber, RaftActorBehavior currentBehavior) {
377 LOG.debug("Snapshot success sequence number:", sequenceNumber);
379 if(applySnapshot != null) {
381 applySnapshotProcedure.apply(applySnapshot.getState());
383 //clears the followers log, sets the snapshot index to ensure adjusted-index works
384 context.setReplicatedLog(ReplicatedLogImpl.newInstance(applySnapshot, context, currentBehavior));
385 context.setLastApplied(applySnapshot.getLastAppliedIndex());
386 context.setCommitIndex(applySnapshot.getLastAppliedIndex());
387 } catch (Exception e) {
388 LOG.error("Error applying snapshot", e);
391 context.getReplicatedLog().snapshotCommit();
394 context.getPersistenceProvider().deleteSnapshots(new SnapshotSelectionCriteria(
395 sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), 43200000));
397 context.getPersistenceProvider().deleteMessages(lastSequenceNumber);
399 lastSequenceNumber = -1;
400 applySnapshot = null;
401 SnapshotManager.this.currentState = IDLE;
405 public void rollback() {
406 // Nothing to rollback if we're applying a snapshot from the leader.
407 if(applySnapshot == null) {
408 context.getReplicatedLog().snapshotRollback();
410 LOG.info("{}: Replicated Log rolled back. Snapshot will be attempted in the next cycle." +
411 "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
412 context.getReplicatedLog().getSnapshotIndex(),
413 context.getReplicatedLog().getSnapshotTerm(),
414 context.getReplicatedLog().size());
417 lastSequenceNumber = -1;
418 applySnapshot = null;
419 SnapshotManager.this.currentState = IDLE;
423 public String toString() {
429 private static interface TermInformationReader {
434 static class LastAppliedTermInformationReader implements TermInformationReader{
438 public LastAppliedTermInformationReader init(ReplicatedLog log, long originalIndex,
439 ReplicatedLogEntry lastLogEntry, boolean hasFollowers){
440 ReplicatedLogEntry entry = log.get(originalIndex);
444 if(lastLogEntry != null) {
445 // since we have persisted the last-log-entry to persistent journal before the capture,
446 // we would want to snapshot from this entry.
447 index = lastLogEntry.getIndex();
448 term = lastLogEntry.getTerm();
450 } else if (entry != null) {
451 index = entry.getIndex();
452 term = entry.getTerm();
453 } else if(log.getSnapshotIndex() > -1){
454 index = log.getSnapshotIndex();
455 term = log.getSnapshotTerm();
461 public long getIndex(){
466 public long getTerm(){
471 private static class ReplicatedToAllTermInformationReader implements TermInformationReader{
475 ReplicatedToAllTermInformationReader init(ReplicatedLog log, long originalIndex){
476 ReplicatedLogEntry entry = log.get(originalIndex);
481 index = entry.getIndex();
482 term = entry.getTerm();
489 public long getIndex(){
494 public long getTerm(){