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.actor.ActorRef;
12 import akka.actor.ActorSelection;
13 import akka.event.Logging;
14 import akka.event.LoggingAdapter;
15 import akka.japi.Procedure;
16 import akka.persistence.RecoveryCompleted;
17 import akka.persistence.SaveSnapshotFailure;
18 import akka.persistence.SaveSnapshotSuccess;
19 import akka.persistence.SnapshotOffer;
20 import akka.persistence.SnapshotSelectionCriteria;
21 import akka.persistence.UntypedPersistentActor;
22 import com.google.common.base.Optional;
23 import com.google.protobuf.ByteString;
24 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
25 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
26 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
27 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
28 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
29 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
30 import org.opendaylight.controller.cluster.raft.behaviors.Candidate;
31 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
32 import org.opendaylight.controller.cluster.raft.behaviors.Leader;
33 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
34 import org.opendaylight.controller.cluster.raft.client.messages.AddRaftPeer;
35 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
36 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
37 import org.opendaylight.controller.cluster.raft.client.messages.RemoveRaftPeer;
38 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
39 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
40 import org.opendaylight.controller.protobuff.messages.cluster.raft.AppendEntriesMessages;
42 import java.io.Serializable;
46 * RaftActor encapsulates a state machine that needs to be kept synchronized
47 * in a cluster. It implements the RAFT algorithm as described in the paper
48 * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
49 * In Search of an Understandable Consensus Algorithm</a>
51 * RaftActor has 3 states and each state has a certain behavior associated
52 * with it. A Raft actor can behave as,
55 * <li> A Follower (or) </li>
56 * <li> A Candidate </li>
60 * A RaftActor MUST be a Leader in order to accept requests from clients to
61 * change the state of it's encapsulated state machine. Once a RaftActor becomes
62 * a Leader it is also responsible for ensuring that all followers ultimately
63 * have the same log and therefore the same state machine as itself.
66 * The current behavior of a RaftActor determines how election for leadership
67 * is initiated and how peer RaftActors react to request for votes.
70 * Each RaftActor also needs to know the current election term. It uses this
71 * information for a couple of things. One is to simply figure out who it
72 * voted for in the last election. Another is to figure out if the message
73 * it received to update it's state is stale.
76 * The RaftActor uses akka-persistence to store it's replicated log.
77 * Furthermore through it's behaviors a Raft Actor determines
80 * <li> when a log entry should be persisted </li>
81 * <li> when a log entry should be applied to the state machine (and) </li>
82 * <li> when a snapshot should be saved </li>
85 public abstract class RaftActor extends UntypedPersistentActor {
86 protected final LoggingAdapter LOG =
87 Logging.getLogger(getContext().system(), this);
90 * The current state determines the current behavior of a RaftActor
91 * A Raft Actor always starts off in the Follower State
93 private RaftActorBehavior currentBehavior;
96 * This context should NOT be passed directly to any other actor it is
97 * only to be consumed by the RaftActorBehaviors
99 protected RaftActorContext context;
102 * The in-memory journal
104 private ReplicatedLogImpl replicatedLog = new ReplicatedLogImpl();
106 private CaptureSnapshot captureSnapshot = null;
108 private volatile boolean hasSnapshotCaptureInitiated = false;
110 public RaftActor(String id, Map<String, String> peerAddresses) {
111 this(id, peerAddresses, Optional.<ConfigParams>absent());
114 public RaftActor(String id, Map<String, String> peerAddresses,
115 Optional<ConfigParams> configParams) {
117 context = new RaftActorContextImpl(this.getSelf(),
118 this.getContext(), id, new ElectionTermImpl(),
119 -1, -1, replicatedLog, peerAddresses,
120 (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
124 @Override public void onReceiveRecover(Object message) {
125 if (message instanceof SnapshotOffer) {
126 LOG.info("SnapshotOffer called..");
127 SnapshotOffer offer = (SnapshotOffer) message;
128 Snapshot snapshot = (Snapshot) offer.snapshot();
130 // Create a replicated log with the snapshot information
131 // The replicated log can be used later on to retrieve this snapshot
132 // when we need to install it on a peer
133 replicatedLog = new ReplicatedLogImpl(snapshot);
135 context.setReplicatedLog(replicatedLog);
136 context.setLastApplied(snapshot.getLastAppliedIndex());
137 context.setCommitIndex(snapshot.getLastAppliedIndex());
139 LOG.info("Applied snapshot to replicatedLog. " +
140 "snapshotIndex={}, snapshotTerm={}, journal-size={}",
141 replicatedLog.snapshotIndex, replicatedLog.snapshotTerm,
145 // Apply the snapshot to the actors state
146 applySnapshot(ByteString.copyFrom(snapshot.getState()));
148 } else if (message instanceof ReplicatedLogEntry) {
149 ReplicatedLogEntry logEntry = (ReplicatedLogEntry) message;
151 // Apply State immediately
152 replicatedLog.append(logEntry);
153 applyState(null, "recovery", logEntry.getData());
154 context.setLastApplied(logEntry.getIndex());
155 context.setCommitIndex(logEntry.getIndex());
157 } else if (message instanceof DeleteEntries) {
158 replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
160 } else if (message instanceof UpdateElectionTerm) {
161 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
162 ((UpdateElectionTerm) message).getVotedFor());
164 } else if (message instanceof RecoveryCompleted) {
166 "RecoveryCompleted - Switching actor to Follower - " +
167 "Persistence Id = " + persistenceId() +
168 " Last index in log:{}, snapshotIndex={}, snapshotTerm={}, " +
170 replicatedLog.lastIndex(), replicatedLog.snapshotIndex,
171 replicatedLog.snapshotTerm, replicatedLog.size());
172 currentBehavior = switchBehavior(RaftState.Follower);
177 @Override public void onReceiveCommand(Object message) {
178 if (message instanceof ApplyState){
179 ApplyState applyState = (ApplyState) message;
181 if(LOG.isDebugEnabled()) {
182 LOG.debug("Applying state for log index {} data {}",
183 applyState.getReplicatedLogEntry().getIndex(),
184 applyState.getReplicatedLogEntry().getData());
187 applyState(applyState.getClientActor(), applyState.getIdentifier(),
188 applyState.getReplicatedLogEntry().getData());
190 } else if(message instanceof ApplySnapshot ) {
191 Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
193 if(LOG.isDebugEnabled()) {
194 LOG.debug("ApplySnapshot called on Follower Actor " +
195 "snapshotIndex:{}, snapshotTerm:{}", snapshot.getLastAppliedIndex(),
196 snapshot.getLastAppliedTerm()
199 applySnapshot(ByteString.copyFrom(snapshot.getState()));
201 //clears the followers log, sets the snapshot index to ensure adjusted-index works
202 replicatedLog = new ReplicatedLogImpl(snapshot);
203 context.setReplicatedLog(replicatedLog);
204 context.setLastApplied(snapshot.getLastAppliedIndex());
206 } else if (message instanceof FindLeader) {
208 new FindLeaderReply(getLeaderAddress()),
212 } else if (message instanceof SaveSnapshotSuccess) {
213 SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
214 LOG.info("SaveSnapshotSuccess received for snapshot");
216 context.getReplicatedLog().snapshotCommit();
218 // TODO: Not sure if we want to be this aggressive with trimming stuff
219 trimPersistentData(success.metadata().sequenceNr());
221 } else if (message instanceof SaveSnapshotFailure) {
222 SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message;
224 LOG.info("saveSnapshotFailure.metadata():{}", saveSnapshotFailure.metadata().toString());
225 LOG.error(saveSnapshotFailure.cause(), "SaveSnapshotFailure received for snapshot Cause:");
227 context.getReplicatedLog().snapshotRollback();
229 LOG.info("Replicated Log rollbacked. Snapshot will be attempted in the next cycle." +
230 "snapshotIndex:{}, snapshotTerm:{}, log-size:{}",
231 context.getReplicatedLog().getSnapshotIndex(),
232 context.getReplicatedLog().getSnapshotTerm(),
233 context.getReplicatedLog().size());
235 } else if (message instanceof AddRaftPeer){
237 // FIXME : Do not add raft peers like this.
238 // When adding a new Peer we have to ensure that the a majority of
239 // the peers know about the new Peer. Doing it this way may cause
240 // a situation where multiple Leaders may emerge
241 AddRaftPeer arp = (AddRaftPeer)message;
242 context.addToPeers(arp.getName(), arp.getAddress());
244 } else if (message instanceof RemoveRaftPeer){
246 RemoveRaftPeer rrp = (RemoveRaftPeer)message;
247 context.removePeer(rrp.getName());
249 } else if (message instanceof CaptureSnapshot) {
250 LOG.info("CaptureSnapshot received by actor");
251 CaptureSnapshot cs = (CaptureSnapshot)message;
252 captureSnapshot = cs;
255 } else if (message instanceof CaptureSnapshotReply){
256 LOG.info("CaptureSnapshotReply received by actor");
257 CaptureSnapshotReply csr = (CaptureSnapshotReply) message;
259 ByteString stateInBytes = csr.getSnapshot();
260 LOG.info("CaptureSnapshotReply stateInBytes size:{}", stateInBytes.size());
261 handleCaptureSnapshotReply(stateInBytes);
264 if (!(message instanceof AppendEntriesMessages.AppendEntries)
265 && !(message instanceof AppendEntriesReply) && !(message instanceof SendHeartBeat)) {
266 if(LOG.isDebugEnabled()) {
267 LOG.debug("onReceiveCommand: message:" + message.getClass());
272 currentBehavior.handleMessage(getSender(), message);
273 RaftActorBehavior oldBehavior = currentBehavior;
274 currentBehavior = switchBehavior(state);
275 if(oldBehavior != currentBehavior){
279 onLeaderChanged(oldBehavior.getLeaderId(), currentBehavior.getLeaderId());
283 public java.util.Set<String> getPeers() {
284 return context.getPeerAddresses().keySet();
287 protected String getReplicatedLogState() {
288 return "snapshotIndex=" + context.getReplicatedLog().getSnapshotIndex()
289 + ", snapshotTerm=" + context.getReplicatedLog().getSnapshotTerm()
290 + ", im-mem journal size=" + context.getReplicatedLog().size();
295 * When a derived RaftActor needs to persist something it must call
302 protected void persistData(ActorRef clientActor, String identifier,
305 ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
306 context.getReplicatedLog().lastIndex() + 1,
307 context.getTermInformation().getCurrentTerm(), data);
309 if(LOG.isDebugEnabled()) {
310 LOG.debug("Persist data {}", replicatedLogEntry);
314 .appendAndPersist(clientActor, identifier, replicatedLogEntry);
317 protected String getId() {
318 return context.getId();
322 * Derived actors can call the isLeader method to check if the current
323 * RaftActor is the Leader or not
325 * @return true it this RaftActor is a Leader false otherwise
327 protected boolean isLeader() {
328 return context.getId().equals(currentBehavior.getLeaderId());
332 * Derived actor can call getLeader if they need a reference to the Leader.
333 * This would be useful for example in forwarding a request to an actor
334 * which is the leader
336 * @return A reference to the leader if known, null otherwise
338 protected ActorSelection getLeader(){
339 String leaderAddress = getLeaderAddress();
341 if(leaderAddress == null){
345 return context.actorSelection(leaderAddress);
350 * @return the current leader's id
352 protected String getLeaderId(){
353 return currentBehavior.getLeaderId();
356 protected RaftState getRaftState() {
357 return currentBehavior.state();
360 protected ReplicatedLogEntry getLastLogEntry() {
361 return replicatedLog.last();
364 protected Long getCurrentTerm(){
365 return context.getTermInformation().getCurrentTerm();
368 protected Long getCommitIndex(){
369 return context.getCommitIndex();
372 protected Long getLastApplied(){
373 return context.getLastApplied();
377 * setPeerAddress sets the address of a known peer at a later time.
379 * This is to account for situations where a we know that a peer
380 * exists but we do not know an address up-front. This may also be used in
381 * situations where a known peer starts off in a different location and we
382 * need to change it's address
384 * Note that if the peerId does not match the list of peers passed to
385 * this actor during construction an IllegalStateException will be thrown.
390 protected void setPeerAddress(String peerId, String peerAddress){
391 context.setPeerAddress(peerId, peerAddress);
397 * The applyState method will be called by the RaftActor when some data
398 * needs to be applied to the actor's state
400 * @param clientActor A reference to the client who sent this message. This
401 * is the same reference that was passed to persistData
402 * by the derived actor. clientActor may be null when
403 * the RaftActor is behaving as a follower or during
405 * @param identifier The identifier of the persisted data. This is also
406 * the same identifier that was passed to persistData by
407 * the derived actor. identifier may be null when
408 * the RaftActor is behaving as a follower or during
410 * @param data A piece of data that was persisted by the persistData call.
411 * This should NEVER be null.
413 protected abstract void applyState(ActorRef clientActor, String identifier,
417 * This method will be called by the RaftActor when a snapshot needs to be
418 * created. The derived actor should respond with its current state.
420 * During recovery the state that is returned by the derived actor will
421 * be passed back to it by calling the applySnapshot method
423 * @return The current state of the actor
425 protected abstract void createSnapshot();
428 * This method will be called by the RaftActor during recovery to
429 * reconstruct the state of the actor.
431 * This method may also be called at any other point during normal
432 * operations when the derived actor is out of sync with it's peers
433 * and the only way to bring it in sync is by applying a snapshot
435 * @param snapshot A snapshot of the state of the actor
437 protected abstract void applySnapshot(ByteString snapshot);
440 * This method will be called by the RaftActor when the state of the
441 * RaftActor changes. The derived actor can then use methods like
442 * isLeader or getLeader to do something useful
444 protected abstract void onStateChanged();
446 protected void onLeaderChanged(String oldLeader, String newLeader){};
448 private RaftActorBehavior switchBehavior(RaftState state) {
449 if (currentBehavior != null) {
450 if (currentBehavior.state() == state) {
451 return currentBehavior;
453 LOG.info("Switching from state " + currentBehavior.state() + " to "
457 currentBehavior.close();
458 } catch (Exception e) {
460 "Failed to close behavior : " + currentBehavior.state());
464 LOG.info("Switching behavior to " + state);
466 RaftActorBehavior behavior = null;
467 if (state == RaftState.Candidate) {
468 behavior = new Candidate(context);
469 } else if (state == RaftState.Follower) {
470 behavior = new Follower(context);
472 behavior = new Leader(context);
480 private void trimPersistentData(long sequenceNumber) {
481 // Trim akka snapshots
482 // FIXME : Not sure how exactly the SnapshotSelectionCriteria is applied
483 // For now guessing that it is ANDed.
484 deleteSnapshots(new SnapshotSelectionCriteria(
485 sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), 43200000));
488 deleteMessages(sequenceNumber);
491 private String getLeaderAddress(){
493 return getSelf().path().toString();
495 String leaderId = currentBehavior.getLeaderId();
496 if (leaderId == null) {
499 String peerAddress = context.getPeerAddress(leaderId);
500 if(LOG.isDebugEnabled()) {
501 LOG.debug("getLeaderAddress leaderId = " + leaderId + " peerAddress = "
508 private void handleCaptureSnapshotReply(ByteString stateInBytes) {
509 // create a snapshot object from the state provided and save it
510 // when snapshot is saved async, SaveSnapshotSuccess is raised.
512 Snapshot sn = Snapshot.create(stateInBytes.toByteArray(),
513 context.getReplicatedLog().getFrom(captureSnapshot.getLastAppliedIndex() + 1),
514 captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
515 captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
519 LOG.info("Persisting of snapshot done:{}", sn.getLogMessage());
521 //be greedy and remove entries from in-mem journal which are in the snapshot
522 // and update snapshotIndex and snapshotTerm without waiting for the success,
524 context.getReplicatedLog().snapshotPreCommit(stateInBytes,
525 captureSnapshot.getLastAppliedIndex(),
526 captureSnapshot.getLastAppliedTerm());
528 LOG.info("Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
529 "and term:{}", captureSnapshot.getLastAppliedIndex(),
530 captureSnapshot.getLastAppliedTerm());
532 captureSnapshot = null;
533 hasSnapshotCaptureInitiated = false;
537 private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
539 public ReplicatedLogImpl(Snapshot snapshot) {
540 super(ByteString.copyFrom(snapshot.getState()),
541 snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
542 snapshot.getUnAppliedEntries());
545 public ReplicatedLogImpl() {
549 @Override public void removeFromAndPersist(long logEntryIndex) {
550 int adjustedIndex = adjustedIndex(logEntryIndex);
552 if (adjustedIndex < 0) {
556 // FIXME: Maybe this should be done after the command is saved
557 journal.subList(adjustedIndex , journal.size()).clear();
559 persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>(){
561 @Override public void apply(DeleteEntries param)
563 //FIXME : Doing nothing for now
568 @Override public void appendAndPersist(
569 final ReplicatedLogEntry replicatedLogEntry) {
570 appendAndPersist(null, null, replicatedLogEntry);
573 public void appendAndPersist(final ActorRef clientActor,
574 final String identifier,
575 final ReplicatedLogEntry replicatedLogEntry) {
576 context.getLogger().debug(
577 "Append log entry and persist {} ", replicatedLogEntry);
578 // FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
579 journal.add(replicatedLogEntry);
581 // When persisting events with persist it is guaranteed that the
582 // persistent actor will not receive further commands between the
583 // persist call and the execution(s) of the associated event
584 // handler. This also holds for multiple persist calls in context
585 // of a single command.
586 persist(replicatedLogEntry,
587 new Procedure<ReplicatedLogEntry>() {
588 public void apply(ReplicatedLogEntry evt) throws Exception {
589 // when a snaphsot is being taken, captureSnapshot != null
590 if (hasSnapshotCaptureInitiated == false &&
591 journal.size() % context.getConfigParams().getSnapshotBatchCount() == 0) {
593 LOG.info("Initiating Snapshot Capture..");
594 long lastAppliedIndex = -1;
595 long lastAppliedTerm = -1;
597 ReplicatedLogEntry lastAppliedEntry = get(context.getLastApplied());
598 if (lastAppliedEntry != null) {
599 lastAppliedIndex = lastAppliedEntry.getIndex();
600 lastAppliedTerm = lastAppliedEntry.getTerm();
603 if(LOG.isDebugEnabled()) {
604 LOG.debug("Snapshot Capture logSize: {}", journal.size());
605 LOG.debug("Snapshot Capture lastApplied:{} ",
606 context.getLastApplied());
607 LOG.debug("Snapshot Capture lastAppliedIndex:{}", lastAppliedIndex);
608 LOG.debug("Snapshot Capture lastAppliedTerm:{}", lastAppliedTerm);
611 // send a CaptureSnapshot to self to make the expensive operation async.
612 getSelf().tell(new CaptureSnapshot(
613 lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm),
615 hasSnapshotCaptureInitiated = true;
617 // Send message for replication
618 if (clientActor != null) {
619 currentBehavior.handleMessage(getSelf(),
620 new Replicate(clientActor, identifier,
631 private static class DeleteEntries implements Serializable {
632 private final int fromIndex;
635 public DeleteEntries(int fromIndex) {
636 this.fromIndex = fromIndex;
639 public int getFromIndex() {
645 private class ElectionTermImpl implements ElectionTerm {
647 * Identifier of the actor whose election term information this is
649 private long currentTerm = 0;
650 private String votedFor = null;
652 public long getCurrentTerm() {
656 public String getVotedFor() {
660 @Override public void update(long currentTerm, String votedFor) {
661 if(LOG.isDebugEnabled()) {
662 LOG.debug("Set currentTerm={}, votedFor={}", currentTerm, votedFor);
664 this.currentTerm = currentTerm;
665 this.votedFor = votedFor;
669 public void updateAndPersist(long currentTerm, String votedFor){
670 update(currentTerm, votedFor);
671 // FIXME : Maybe first persist then update the state
672 persist(new UpdateElectionTerm(this.currentTerm, this.votedFor), new Procedure<UpdateElectionTerm>(){
674 @Override public void apply(UpdateElectionTerm param)
682 private static class UpdateElectionTerm implements Serializable {
683 private final long currentTerm;
684 private final String votedFor;
686 public UpdateElectionTerm(long currentTerm, String votedFor) {
687 this.currentTerm = currentTerm;
688 this.votedFor = votedFor;
691 public long getCurrentTerm() {
695 public String getVotedFor() {