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
8 package org.opendaylight.controller.cluster.raft.behaviors;
10 import static java.util.Objects.requireNonNull;
12 import akka.actor.ActorRef;
13 import akka.actor.ActorSelection;
14 import akka.actor.Cancellable;
15 import com.google.common.annotations.VisibleForTesting;
16 import com.google.common.base.Optional;
17 import com.google.common.io.ByteSource;
18 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
19 import java.io.IOException;
20 import java.io.ObjectOutputStream;
21 import java.util.Collection;
22 import java.util.Collections;
23 import java.util.HashMap;
24 import java.util.Iterator;
25 import java.util.LinkedList;
26 import java.util.List;
28 import java.util.Map.Entry;
29 import java.util.Queue;
30 import java.util.concurrent.TimeUnit;
31 import org.eclipse.jdt.annotation.Nullable;
32 import org.opendaylight.controller.cluster.io.SharedFileBackedOutputStream;
33 import org.opendaylight.controller.cluster.messaging.MessageSlicer;
34 import org.opendaylight.controller.cluster.messaging.SliceOptions;
35 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
36 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
37 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
38 import org.opendaylight.controller.cluster.raft.PeerInfo;
39 import org.opendaylight.controller.cluster.raft.RaftActorContext;
40 import org.opendaylight.controller.cluster.raft.RaftState;
41 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
42 import org.opendaylight.controller.cluster.raft.VotingState;
43 import org.opendaylight.controller.cluster.raft.base.messages.CheckConsensusReached;
44 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
45 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
46 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
47 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
48 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
49 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
50 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
51 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
52 import org.opendaylight.controller.cluster.raft.messages.RequestVote;
53 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
54 import org.opendaylight.controller.cluster.raft.messages.UnInitializedFollowerSnapshotReply;
55 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
56 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
57 import scala.concurrent.duration.FiniteDuration;
60 * The behavior of a RaftActor when it is in the Leader state.
65 * <li> Upon election: send initial empty AppendEntries RPCs
66 * (heartbeat) to each server; repeat during idle periods to
67 * prevent election timeouts (§5.2)
68 * <li> If command received from client: append entry to local log,
69 * respond after entry applied to state machine (§5.3)
70 * <li> If last log index ≥ nextIndex for a follower: send
71 * AppendEntries RPC with log entries starting at nextIndex
72 * <li> If successful: update nextIndex and matchIndex for
74 * <li> If AppendEntries fails because of log inconsistency:
75 * decrement nextIndex and retry (§5.3)
76 * <li> If there exists an N such that N > commitIndex, a majority
77 * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
78 * set commitIndex = N (§5.3, §5.4).
81 public abstract class AbstractLeader extends AbstractRaftActorBehavior {
82 private final Map<String, FollowerLogInformation> followerToLog = new HashMap<>();
85 * Lookup table for request contexts based on journal index. We could use a {@link Map} here, but we really
86 * expect the entries to be modified in sequence, hence we open-code the lookup.
87 * TODO: Evaluate the use of ArrayDeque(), as that has lower memory overhead. Non-head removals are more costly,
88 * but we already expect those to be far from frequent.
90 private final Queue<ClientRequestTracker> trackers = new LinkedList<>();
93 * Map of serialized AppendEntries output streams keyed by log index. This is used in conjunction with the
94 * appendEntriesMessageSlicer for slicing single ReplicatedLogEntry payloads that exceed the message size threshold.
95 * This Map allows the SharedFileBackedOutputStreams to be reused for multiple followers.
97 private final Map<Long, SharedFileBackedOutputStream> sharedSerializedAppendEntriesStreams = new HashMap<>();
98 private final MessageSlicer appendEntriesMessageSlicer;
100 private Cancellable heartbeatSchedule = null;
101 private Optional<SnapshotHolder> snapshotHolder = Optional.absent();
102 private int minReplicationCount;
104 protected AbstractLeader(final RaftActorContext context, final RaftState state,
105 final @Nullable AbstractLeader initializeFromLeader) {
106 super(context, state);
108 appendEntriesMessageSlicer = MessageSlicer.builder().logContext(logName())
109 .messageSliceSize(context.getConfigParams().getSnapshotChunkSize())
110 .expireStateAfterInactivity(context.getConfigParams().getElectionTimeOutInterval().toMillis() * 3,
111 TimeUnit.MILLISECONDS).build();
113 if (initializeFromLeader != null) {
114 followerToLog.putAll(initializeFromLeader.followerToLog);
115 snapshotHolder = initializeFromLeader.snapshotHolder;
116 trackers.addAll(initializeFromLeader.trackers);
118 for (PeerInfo peerInfo: context.getPeers()) {
119 FollowerLogInformation followerLogInformation = new FollowerLogInformation(peerInfo, context);
120 followerToLog.put(peerInfo.getId(), followerLogInformation);
124 log.debug("{}: Election: Leader has following peers: {}", logName(), getFollowerIds());
126 updateMinReplicaCount();
128 // Immediately schedule a heartbeat
129 // Upon election: send initial empty AppendEntries RPCs
130 // (heartbeat) to each server; repeat during idle periods to
131 // prevent election timeouts (§5.2)
132 sendAppendEntries(0, false);
134 // It is important to schedule this heartbeat here
135 scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
138 protected AbstractLeader(final RaftActorContext context, final RaftState state) {
139 this(context, state, null);
143 * Return an immutable collection of follower identifiers.
145 * @return Collection of follower IDs
147 public final Collection<String> getFollowerIds() {
148 return followerToLog.keySet();
151 public void addFollower(final String followerId) {
152 FollowerLogInformation followerLogInformation = new FollowerLogInformation(context.getPeerInfo(followerId),
154 followerToLog.put(followerId, followerLogInformation);
156 if (heartbeatSchedule == null) {
157 scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
161 public void removeFollower(final String followerId) {
162 followerToLog.remove(followerId);
165 public void updateMinReplicaCount() {
167 for (PeerInfo peer: context.getPeers()) {
168 if (peer.isVoting()) {
173 minReplicationCount = getMajorityVoteCount(numVoting);
176 protected int getMinIsolatedLeaderPeerCount() {
177 //the isolated Leader peer count will be 1 less than the majority vote count.
178 //this is because the vote count has the self vote counted in it
180 //0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
181 //2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
182 //4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
184 return minReplicationCount > 0 ? minReplicationCount - 1 : 0;
188 void setSnapshotHolder(final @Nullable SnapshotHolder snapshotHolder) {
189 this.snapshotHolder = Optional.fromNullable(snapshotHolder);
193 boolean hasSnapshot() {
194 return snapshotHolder.isPresent();
198 protected RaftActorBehavior handleAppendEntries(final ActorRef sender,
199 final AppendEntries appendEntries) {
201 log.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
207 protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender,
208 final AppendEntriesReply appendEntriesReply) {
209 log.trace("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
211 // Update the FollowerLogInformation
212 String followerId = appendEntriesReply.getFollowerId();
213 FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
215 if (followerLogInformation == null) {
216 log.error("{}: handleAppendEntriesReply - unknown follower {}", logName(), followerId);
220 final long lastActivityNanos = followerLogInformation.nanosSinceLastActivity();
221 if (lastActivityNanos > context.getConfigParams().getElectionTimeOutInterval().toNanos()) {
222 log.warn("{} : handleAppendEntriesReply delayed beyond election timeout, "
223 + "appendEntriesReply : {}, timeSinceLastActivity : {}, lastApplied : {}, commitIndex : {}",
224 logName(), appendEntriesReply, TimeUnit.NANOSECONDS.toMillis(lastActivityNanos),
225 context.getLastApplied(), context.getCommitIndex());
228 followerLogInformation.markFollowerActive();
229 followerLogInformation.setPayloadVersion(appendEntriesReply.getPayloadVersion());
230 followerLogInformation.setRaftVersion(appendEntriesReply.getRaftVersion());
231 followerLogInformation.setNeedsLeaderAddress(appendEntriesReply.isNeedsLeaderAddress());
233 long followerLastLogIndex = appendEntriesReply.getLogLastIndex();
234 boolean updated = false;
235 if (appendEntriesReply.getLogLastIndex() > context.getReplicatedLog().lastIndex()) {
236 // The follower's log is actually ahead of the leader's log. Normally this doesn't happen
237 // in raft as a node cannot become leader if it's log is behind another's. However, the
238 // non-voting semantics deviate a bit from raft. Only voting members participate in
239 // elections and can become leader so it's possible for a non-voting follower to be ahead
240 // of the leader. This can happen if persistence is disabled and all voting members are
241 // restarted. In this case, the voting leader will start out with an empty log however
242 // the non-voting followers still retain the previous data in memory. On the first
243 // AppendEntries, the non-voting follower returns a successful reply b/c the prevLogIndex
244 // sent by the leader is -1 and thus the integrity checks pass. However the follower's returned
245 // lastLogIndex may be higher in which case we want to reset the follower by installing a
246 // snapshot. It's also possible that the follower's last log index is behind the leader's.
247 // However in this case the log terms won't match and the logs will conflict - this is handled
249 log.info("{}: handleAppendEntriesReply: follower {} lastIndex {} is ahead of our lastIndex {} "
250 + "(snapshotIndex {}, snapshotTerm {}) - forcing install snaphot", logName(),
251 followerLogInformation.getId(), appendEntriesReply.getLogLastIndex(),
252 context.getReplicatedLog().lastIndex(), context.getReplicatedLog().getSnapshotIndex(),
253 context.getReplicatedLog().getSnapshotTerm());
255 followerLogInformation.setMatchIndex(-1);
256 followerLogInformation.setNextIndex(-1);
258 initiateCaptureSnapshot(followerId);
261 } else if (appendEntriesReply.isSuccess()) {
262 long followersLastLogTermInLeadersLog = getLogEntryTerm(followerLastLogIndex);
263 if (followerLastLogIndex >= 0 && followersLastLogTermInLeadersLog >= 0
264 && followersLastLogTermInLeadersLog != appendEntriesReply.getLogLastTerm()) {
265 // The follower's last entry is present in the leader's journal but the terms don't match so the
266 // follower has a conflicting entry. Since the follower didn't report that it's out of sync, this means
267 // either the previous leader entry sent didn't conflict or the previous leader entry is in the snapshot
268 // and no longer in the journal. Either way, we set the follower's next index to 1 less than the last
269 // index reported by the follower. For the former case, the leader will send all entries starting with
270 // the previous follower's index and the follower will remove and replace the conflicting entries as
271 // needed. For the latter, the leader will initiate an install snapshot.
273 followerLogInformation.setNextIndex(followerLastLogIndex - 1);
276 log.info("{}: handleAppendEntriesReply: follower {} last log term {} for index {} conflicts with the "
277 + "leader's {} - set the follower's next index to {}", logName(),
278 followerId, appendEntriesReply.getLogLastTerm(), appendEntriesReply.getLogLastIndex(),
279 followersLastLogTermInLeadersLog, followerLogInformation.getNextIndex());
281 updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
284 log.info("{}: handleAppendEntriesReply - received unsuccessful reply: {}, leader snapshotIndex: {}, "
285 + "snapshotTerm: {}, replicatedToAllIndex: {}", logName(), appendEntriesReply,
286 context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm(),
287 getReplicatedToAllIndex());
289 long followersLastLogTermInLeadersLogOrSnapshot = getLogEntryOrSnapshotTerm(followerLastLogIndex);
290 if (appendEntriesReply.isForceInstallSnapshot()) {
291 // Reset the followers match and next index. This is to signal that this follower has nothing
292 // in common with this Leader and so would require a snapshot to be installed
293 followerLogInformation.setMatchIndex(-1);
294 followerLogInformation.setNextIndex(-1);
296 // Force initiate a snapshot capture
297 initiateCaptureSnapshot(followerId);
298 } else if (followerLastLogIndex < 0 || followersLastLogTermInLeadersLogOrSnapshot >= 0
299 && followersLastLogTermInLeadersLogOrSnapshot == appendEntriesReply.getLogLastTerm()) {
300 // The follower's log is empty or the follower's last entry is present in the leader's journal or
301 // snapshot and the terms match so the follower is just behind the leader's journal from the last
302 // snapshot, if any. We'll catch up the follower quickly by starting at the follower's last log index.
304 updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
306 log.info("{}: follower {} appears to be behind the leader from the last snapshot - "
307 + "updated: matchIndex: {}, nextIndex: {}", logName(), followerId,
308 followerLogInformation.getMatchIndex(), followerLogInformation.getNextIndex());
310 // The follower's log conflicts with leader's log so decrement follower's next index
311 // in an attempt to find where the logs match.
312 if (followerLogInformation.decrNextIndex(appendEntriesReply.getLogLastIndex())) {
315 log.info("{}: follower {} last log term {} conflicts with the leader's {} - dec next index to {}",
316 logName(), followerId, appendEntriesReply.getLogLastTerm(),
317 followersLastLogTermInLeadersLogOrSnapshot, followerLogInformation.getNextIndex());
322 if (log.isTraceEnabled()) {
323 log.trace("{}: handleAppendEntriesReply from {}: commitIndex: {}, lastAppliedIndex: {}, currentTerm: {}",
324 logName(), followerId, context.getCommitIndex(), context.getLastApplied(), currentTerm());
327 possiblyUpdateCommitIndex();
329 //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
330 sendUpdatesToFollower(followerId, followerLogInformation, false, !updated);
335 private void possiblyUpdateCommitIndex() {
336 // Figure out if we can update the the commitIndex as follows:
337 // If there exists an index N such that N > commitIndex, a majority of matchIndex[i] ≥ N,
338 // and log[N].term == currentTerm:
339 // set commitIndex = N (§5.3, §5.4).
340 for (long index = context.getCommitIndex() + 1; ; index++) {
341 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(index);
342 if (replicatedLogEntry == null) {
343 log.trace("{}: ReplicatedLogEntry not found for index {} - snapshotIndex: {}, journal size: {}",
344 logName(), index, context.getReplicatedLog().getSnapshotIndex(),
345 context.getReplicatedLog().size());
349 // Count our entry if it has been persisted.
350 int replicatedCount = replicatedLogEntry.isPersistencePending() ? 0 : 1;
352 if (replicatedCount == 0) {
353 // We don't commit and apply a log entry until we've gotten the ack from our local persistence,
354 // even though there *shouldn't* be any issue with updating the commit index if we get a consensus
355 // amongst the followers w/o the local persistence ack.
359 log.trace("{}: checking Nth index {}", logName(), index);
360 for (FollowerLogInformation info : followerToLog.values()) {
361 final PeerInfo peerInfo = context.getPeerInfo(info.getId());
362 if (info.getMatchIndex() >= index && peerInfo != null && peerInfo.isVoting()) {
364 } else if (log.isTraceEnabled()) {
365 log.trace("{}: Not counting follower {} - matchIndex: {}, {}", logName(), info.getId(),
366 info.getMatchIndex(), peerInfo);
370 if (log.isTraceEnabled()) {
371 log.trace("{}: replicatedCount {}, minReplicationCount: {}", logName(), replicatedCount,
372 minReplicationCount);
375 if (replicatedCount >= minReplicationCount) {
376 // Don't update the commit index if the log entry is from a previous term, as per §5.4.1:
377 // "Raft never commits log entries from previous terms by counting replicas".
378 // However we keep looping so we can make progress when new entries in the current term
379 // reach consensus, as per §5.4.1: "once an entry from the current term is committed by
380 // counting replicas, then all prior entries are committed indirectly".
381 if (replicatedLogEntry.getTerm() == currentTerm()) {
382 log.trace("{}: Setting commit index to {}", logName(), index);
383 context.setCommitIndex(index);
385 log.debug("{}: Not updating commit index to {} - retrieved log entry with index {}, "
386 + "term {} does not match the current term {}", logName(), index,
387 replicatedLogEntry.getIndex(), replicatedLogEntry.getTerm(), currentTerm());
390 log.trace("{}: minReplicationCount not reached, actual {} - breaking", logName(), replicatedCount);
395 // Apply the change to the state machine
396 if (context.getCommitIndex() > context.getLastApplied()) {
397 log.debug("{}: Applying to log - commitIndex: {}, lastAppliedIndex: {}", logName(),
398 context.getCommitIndex(), context.getLastApplied());
400 applyLogToStateMachine(context.getCommitIndex());
403 if (!context.getSnapshotManager().isCapturing()) {
408 private boolean updateFollowerLogInformation(final FollowerLogInformation followerLogInformation,
409 final AppendEntriesReply appendEntriesReply) {
410 boolean updated = followerLogInformation.setMatchIndex(appendEntriesReply.getLogLastIndex());
411 updated = followerLogInformation.setNextIndex(appendEntriesReply.getLogLastIndex() + 1) || updated;
413 if (updated && log.isDebugEnabled()) {
415 "{}: handleAppendEntriesReply - FollowerLogInformation for {} updated: matchIndex: {}, nextIndex: {}",
416 logName(), followerLogInformation.getId(), followerLogInformation.getMatchIndex(),
417 followerLogInformation.getNextIndex());
422 private void purgeInMemoryLog() {
423 //find the lowest index across followers which has been replicated to all.
424 // lastApplied if there are no followers, so that we keep clearing the log for single-node
425 // we would delete the in-mem log from that index on, in-order to minimize mem usage
426 // we would also share this info thru AE with the followers so that they can delete their log entries as well.
427 long minReplicatedToAllIndex = followerToLog.isEmpty() ? context.getLastApplied() : Long.MAX_VALUE;
428 for (FollowerLogInformation info : followerToLog.values()) {
429 minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
432 super.performSnapshotWithoutCapture(minReplicatedToAllIndex);
436 protected ClientRequestTracker removeClientRequestTracker(final long logIndex) {
437 final Iterator<ClientRequestTracker> it = trackers.iterator();
438 while (it.hasNext()) {
439 final ClientRequestTracker t = it.next();
440 if (t.getIndex() == logIndex) {
450 protected RaftActorBehavior handleRequestVoteReply(final ActorRef sender,
451 final RequestVoteReply requestVoteReply) {
455 protected void beforeSendHeartbeat(){}
458 public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) {
459 requireNonNull(sender, "sender should not be null");
461 if (appendEntriesMessageSlicer.handleMessage(message)) {
465 if (message instanceof RaftRPC) {
466 RaftRPC rpc = (RaftRPC) message;
467 // If RPC request or response contains term T > currentTerm:
468 // set currentTerm = T, convert to follower (§5.1)
469 // This applies to all RPC messages and responses
470 if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
471 log.info("{}: Term {} in \"{}\" message is greater than leader's term {} - switching to Follower",
472 logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
474 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
476 // This is a special case. Normally when stepping down as leader we don't process and reply to the
477 // RaftRPC as per raft. But if we're in the process of transferring leadership and we get a
478 // RequestVote, process the RequestVote before switching to Follower. This enables the requesting
479 // candidate node to be elected the leader faster and avoids us possibly timing out in the Follower
480 // state and starting a new election and grabbing leadership back before the other candidate node can
481 // start a new election due to lack of responses. This case would only occur if there isn't a majority
482 // of other nodes available that can elect the requesting candidate. Since we're transferring
483 // leadership, we should make every effort to get the requesting node elected.
484 if (message instanceof RequestVote && context.getRaftActorLeadershipTransferCohort() != null) {
485 log.debug("{}: Leadership transfer in progress - processing RequestVote", logName());
486 super.handleMessage(sender, message);
489 return internalSwitchBehavior(RaftState.Follower);
493 if (message instanceof SendHeartBeat) {
494 beforeSendHeartbeat();
496 scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
497 } else if (message instanceof SendInstallSnapshot) {
498 SendInstallSnapshot sendInstallSnapshot = (SendInstallSnapshot) message;
499 setSnapshotHolder(new SnapshotHolder(sendInstallSnapshot.getSnapshot(),
500 sendInstallSnapshot.getSnapshotBytes()));
501 sendInstallSnapshot();
502 } else if (message instanceof Replicate) {
503 replicate((Replicate) message);
504 } else if (message instanceof InstallSnapshotReply) {
505 handleInstallSnapshotReply((InstallSnapshotReply) message);
506 } else if (message instanceof CheckConsensusReached) {
507 possiblyUpdateCommitIndex();
509 return super.handleMessage(sender, message);
515 @SuppressFBWarnings(value = "NP_NULL_PARAM_DEREF_ALL_TARGETS_DANGEROUS",
516 justification = "JDT nullness with SpotBugs at setSnapshotHolder(null)")
517 private void handleInstallSnapshotReply(final InstallSnapshotReply reply) {
518 log.debug("{}: handleInstallSnapshotReply: {}", logName(), reply);
520 String followerId = reply.getFollowerId();
521 FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
522 if (followerLogInformation == null) {
523 // This can happen during AddServer if it times out.
524 log.error("{}: FollowerLogInformation not found for follower {} in InstallSnapshotReply",
525 logName(), followerId);
529 LeaderInstallSnapshotState installSnapshotState = followerLogInformation.getInstallSnapshotState();
530 if (installSnapshotState == null) {
531 log.error("{}: LeaderInstallSnapshotState not found for follower {} in InstallSnapshotReply",
532 logName(), followerId);
536 installSnapshotState.resetChunkTimer();
537 followerLogInformation.markFollowerActive();
539 if (installSnapshotState.getChunkIndex() == reply.getChunkIndex()) {
540 boolean wasLastChunk = false;
541 if (reply.isSuccess()) {
542 if (installSnapshotState.isLastChunk(reply.getChunkIndex())) {
543 //this was the last chunk reply
545 long followerMatchIndex = snapshotHolder.get().getLastIncludedIndex();
546 followerLogInformation.setMatchIndex(followerMatchIndex);
547 followerLogInformation.setNextIndex(followerMatchIndex + 1);
548 followerLogInformation.clearLeaderInstallSnapshotState();
550 log.info("{}: Snapshot successfully installed on follower {} (last chunk {}) - "
551 + "matchIndex set to {}, nextIndex set to {}", logName(), followerId, reply.getChunkIndex(),
552 followerLogInformation.getMatchIndex(), followerLogInformation.getNextIndex());
554 if (!anyFollowersInstallingSnapshot()) {
555 // once there are no pending followers receiving snapshots
556 // we can remove snapshot from the memory
557 setSnapshotHolder(null);
561 if (context.getPeerInfo(followerId).getVotingState() == VotingState.VOTING_NOT_INITIALIZED) {
562 UnInitializedFollowerSnapshotReply unInitFollowerSnapshotSuccess =
563 new UnInitializedFollowerSnapshotReply(followerId);
564 context.getActor().tell(unInitFollowerSnapshotSuccess, context.getActor());
565 log.debug("Sent message UnInitializedFollowerSnapshotReply to self");
568 installSnapshotState.markSendStatus(true);
571 log.warn("{}: Received failed InstallSnapshotReply - will retry: {}", logName(), reply);
573 installSnapshotState.markSendStatus(false);
577 if (!context.getSnapshotManager().isCapturing()) {
578 // Since the follower is now caught up try to purge the log.
582 ActorSelection followerActor = context.getPeerActorSelection(followerId);
583 if (followerActor != null) {
584 sendSnapshotChunk(followerActor, followerLogInformation);
589 log.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
590 logName(), reply.getChunkIndex(), followerId,
591 installSnapshotState.getChunkIndex());
593 if (reply.getChunkIndex() == LeaderInstallSnapshotState.INVALID_CHUNK_INDEX) {
594 // Since the Follower did not find this index to be valid we should reset the follower snapshot
595 // so that Installing the snapshot can resume from the beginning
596 installSnapshotState.reset();
601 private boolean anyFollowersInstallingSnapshot() {
602 for (FollowerLogInformation info: followerToLog.values()) {
603 if (info.getInstallSnapshotState() != null) {
612 private void replicate(final Replicate replicate) {
613 long logIndex = replicate.getReplicatedLogEntry().getIndex();
615 log.debug("{}: Replicate message: identifier: {}, logIndex: {}, payload: {}, isSendImmediate: {}", logName(),
616 replicate.getIdentifier(), logIndex, replicate.getReplicatedLogEntry().getData().getClass(),
617 replicate.isSendImmediate());
619 // Create a tracker entry we will use this later to notify the
621 if (replicate.getClientActor() != null) {
622 trackers.add(new ClientRequestTrackerImpl(replicate.getClientActor(), replicate.getIdentifier(),
626 boolean applyModificationToState = !context.anyVotingPeers()
627 || context.getRaftPolicy().applyModificationToStateBeforeConsensus();
629 if (applyModificationToState) {
630 context.setCommitIndex(logIndex);
631 applyLogToStateMachine(logIndex);
634 if (replicate.isSendImmediate() && !followerToLog.isEmpty()) {
635 sendAppendEntries(0, false);
639 protected void sendAppendEntries(final long timeSinceLastActivityIntervalNanos, final boolean isHeartbeat) {
640 // Send an AppendEntries to all followers
641 for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
642 final String followerId = e.getKey();
643 final FollowerLogInformation followerLogInformation = e.getValue();
644 // This checks helps not to send a repeat message to the follower
645 if (!followerLogInformation.isFollowerActive()
646 || followerLogInformation.nanosSinceLastActivity() >= timeSinceLastActivityIntervalNanos) {
647 sendUpdatesToFollower(followerId, followerLogInformation, true, isHeartbeat);
653 * This method checks if any update needs to be sent to the given follower. This includes append log entries,
654 * sending next snapshot chunk, and initiating a snapshot.
656 private void sendUpdatesToFollower(final String followerId, final FollowerLogInformation followerLogInformation,
657 final boolean sendHeartbeat, final boolean isHeartbeat) {
659 ActorSelection followerActor = context.getPeerActorSelection(followerId);
660 if (followerActor != null) {
661 long followerNextIndex = followerLogInformation.getNextIndex();
662 boolean isFollowerActive = followerLogInformation.isFollowerActive();
663 boolean sendAppendEntries = false;
664 List<ReplicatedLogEntry> entries = Collections.emptyList();
666 LeaderInstallSnapshotState installSnapshotState = followerLogInformation.getInstallSnapshotState();
667 if (installSnapshotState != null) {
669 // if install snapshot is in process , then sent next chunk if possible
670 if (isFollowerActive) {
671 // 30 seconds with default settings, can be modified via heartbeat or election timeout factor
672 FiniteDuration snapshotReplyTimeout = context.getConfigParams().getHeartBeatInterval()
673 .$times(context.getConfigParams().getElectionTimeoutFactor() * 3);
675 if (installSnapshotState.isChunkTimedOut(snapshotReplyTimeout)) {
676 sendAppendEntries = !resendSnapshotChunk(followerActor, followerLogInformation);
677 } else if (installSnapshotState.canSendNextChunk()) {
678 sendSnapshotChunk(followerActor, followerLogInformation);
680 } else if (sendHeartbeat || followerLogInformation.hasStaleCommitIndex(context.getCommitIndex())) {
681 // we send a heartbeat even if we have not received a reply for the last chunk
682 sendAppendEntries = true;
684 } else if (followerLogInformation.isLogEntrySlicingInProgress()) {
685 sendAppendEntries = sendHeartbeat;
687 long leaderLastIndex = context.getReplicatedLog().lastIndex();
688 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
690 if (!isHeartbeat && log.isDebugEnabled() || log.isTraceEnabled()) {
691 log.debug("{}: Checking sendAppendEntries for follower {}: active: {}, followerNextIndex: {}, "
692 + "leaderLastIndex: {}, leaderSnapShotIndex: {}", logName(), followerId, isFollowerActive,
693 followerNextIndex, leaderLastIndex, leaderSnapShotIndex);
696 if (isFollowerActive && context.getReplicatedLog().isPresent(followerNextIndex)) {
698 log.debug("{}: sendAppendEntries: {} is present for follower {}", logName(),
699 followerNextIndex, followerId);
701 if (followerLogInformation.okToReplicate(context.getCommitIndex())) {
702 entries = getEntriesToSend(followerLogInformation, followerActor);
703 sendAppendEntries = true;
705 } else if (isFollowerActive && followerNextIndex >= 0
706 && leaderLastIndex > followerNextIndex && !context.getSnapshotManager().isCapturing()) {
707 // if the followers next index is not present in the leaders log, and
708 // if the follower is just not starting and if leader's index is more than followers index
709 // then snapshot should be sent
711 // Send heartbeat to follower whenever install snapshot is initiated.
712 sendAppendEntries = true;
713 if (canInstallSnapshot(followerNextIndex)) {
714 log.info("{}: Initiating install snapshot to follower {}: follower nextIndex: {}, leader "
715 + "snapshotIndex: {}, leader lastIndex: {}, leader log size: {}", logName(), followerId,
716 followerNextIndex, leaderSnapShotIndex, leaderLastIndex,
717 context.getReplicatedLog().size());
719 initiateCaptureSnapshot(followerId);
721 // It doesn't seem like we should ever reach here - most likely indicates sonething is
723 log.info("{}: Follower {} is behind but cannot install snapshot: follower nextIndex: {}, "
724 + "leader snapshotIndex: {}, leader lastIndex: {}, leader log size: {}", logName(),
725 followerId, followerNextIndex, leaderSnapShotIndex, leaderLastIndex,
726 context.getReplicatedLog().size());
729 } else if (sendHeartbeat || followerLogInformation.hasStaleCommitIndex(context.getCommitIndex())) {
730 // we send an AppendEntries, even if the follower is inactive
731 // in-order to update the followers timestamp, in case it becomes active again
732 sendAppendEntries = true;
737 if (sendAppendEntries) {
738 sendAppendEntriesToFollower(followerActor, entries, followerLogInformation);
743 private List<ReplicatedLogEntry> getEntriesToSend(final FollowerLogInformation followerLogInfo,
744 final ActorSelection followerActor) {
745 // Try to get all the entries in the journal but not exceeding the max data size for a single AppendEntries
747 int maxEntries = (int) context.getReplicatedLog().size();
748 final int maxDataSize = context.getConfigParams().getSnapshotChunkSize();
749 final long followerNextIndex = followerLogInfo.getNextIndex();
750 List<ReplicatedLogEntry> entries = context.getReplicatedLog().getFrom(followerNextIndex,
751 maxEntries, maxDataSize);
753 // If the first entry's size exceeds the max data size threshold, it will be returned from the call above. If
754 // that is the case, then we need to slice it into smaller chunks.
755 if (!(entries.size() == 1 && entries.get(0).getData().size() > maxDataSize)) {
756 // Don't need to slice.
760 log.debug("{}: Log entry size {} exceeds max payload size {}", logName(), entries.get(0).getData().size(),
763 // If an AppendEntries has already been serialized for the log index then reuse the
764 // SharedFileBackedOutputStream.
765 final Long logIndex = entries.get(0).getIndex();
766 SharedFileBackedOutputStream fileBackedStream = sharedSerializedAppendEntriesStreams.get(logIndex);
767 if (fileBackedStream == null) {
768 fileBackedStream = context.getFileBackedOutputStreamFactory().newSharedInstance();
770 final AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
771 getLogEntryIndex(followerNextIndex - 1), getLogEntryTerm(followerNextIndex - 1), entries,
772 context.getCommitIndex(), getReplicatedToAllIndex(), context.getPayloadVersion());
774 log.debug("{}: Serializing {} for slicing for follower {}", logName(), appendEntries,
775 followerLogInfo.getId());
777 try (ObjectOutputStream out = new ObjectOutputStream(fileBackedStream)) {
778 out.writeObject(appendEntries);
779 } catch (IOException e) {
780 log.error("{}: Error serializing {}", logName(), appendEntries, e);
781 fileBackedStream.cleanup();
782 return Collections.emptyList();
785 sharedSerializedAppendEntriesStreams.put(logIndex, fileBackedStream);
787 fileBackedStream.setOnCleanupCallback(index -> {
788 log.debug("{}: On SharedFileBackedOutputStream cleanup for index {}", logName(), index);
789 sharedSerializedAppendEntriesStreams.remove(index);
792 log.debug("{}: Reusing SharedFileBackedOutputStream for follower {}", logName(), followerLogInfo.getId());
793 fileBackedStream.incrementUsageCount();
796 log.debug("{}: Slicing stream for index {}, follower {}", logName(), logIndex, followerLogInfo.getId());
798 // Record that slicing is in progress for the follower.
799 followerLogInfo.setSlicedLogEntryIndex(logIndex);
801 final FollowerIdentifier identifier = new FollowerIdentifier(followerLogInfo.getId());
802 appendEntriesMessageSlicer.slice(SliceOptions.builder().identifier(identifier)
803 .fileBackedOutputStream(fileBackedStream).sendTo(followerActor).replyTo(actor())
804 .onFailureCallback(failure -> {
805 log.error("{}: Error slicing AppendEntries for follower {}", logName(),
806 followerLogInfo.getId(), failure);
807 followerLogInfo.setSlicedLogEntryIndex(FollowerLogInformation.NO_INDEX);
810 return Collections.emptyList();
813 private void sendAppendEntriesToFollower(final ActorSelection followerActor, final List<ReplicatedLogEntry> entries,
814 final FollowerLogInformation followerLogInformation) {
815 // In certain cases outlined below we don't want to send the actual commit index to prevent the follower from
816 // possibly committing and applying conflicting entries (those with same index, different term) from a prior
817 // term that weren't replicated to a majority, which would be a violation of raft.
818 // - if the follower isn't active. In this case we don't know the state of the follower and we send an
819 // empty AppendEntries as a heart beat to prevent election.
820 // - if we're in the process of installing a snapshot. In this case we don't send any new entries but still
821 // need to send AppendEntries to prevent election.
822 // - if we're in the process of slicing an AppendEntries with a large log entry payload. In this case we
823 // need to send an empty AppendEntries to prevent election.
824 boolean isInstallingSnaphot = followerLogInformation.getInstallSnapshotState() != null;
825 long leaderCommitIndex = isInstallingSnaphot || followerLogInformation.isLogEntrySlicingInProgress()
826 || !followerLogInformation.isFollowerActive() ? -1 : context.getCommitIndex();
828 long followerNextIndex = followerLogInformation.getNextIndex();
829 AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
830 getLogEntryIndex(followerNextIndex - 1),
831 getLogEntryTerm(followerNextIndex - 1), entries,
832 leaderCommitIndex, super.getReplicatedToAllIndex(), context.getPayloadVersion(),
833 followerLogInformation.getRaftVersion(), followerLogInformation.needsLeaderAddress(getId()));
835 if (!entries.isEmpty() || log.isTraceEnabled()) {
836 log.debug("{}: Sending AppendEntries to follower {}: {}", logName(), followerLogInformation.getId(),
840 followerLogInformation.setSentCommitIndex(leaderCommitIndex);
841 followerActor.tell(appendEntries, actor());
845 * Initiates a snapshot capture to install on a follower.
848 * Install Snapshot works as follows
849 * 1. Leader initiates the capture snapshot by calling createSnapshot on the RaftActor.
850 * 2. On receipt of the CaptureSnapshotReply message, the RaftActor persists the snapshot and makes a call to
851 * the Leader's handleMessage with a SendInstallSnapshot message.
852 * 3. The Leader obtains and stores the Snapshot from the SendInstallSnapshot message and sends it in chunks to
853 * the Follower via InstallSnapshot messages.
854 * 4. For each chunk, the Follower sends back an InstallSnapshotReply.
855 * 5. On receipt of the InstallSnapshotReply for the last chunk, the Leader marks the install complete for that
857 * 6. If another follower requires a snapshot and a snapshot has been collected (via SendInstallSnapshot)
858 * then send the existing snapshot in chunks to the follower.
860 * @param followerId the id of the follower.
861 * @return true if capture was initiated, false otherwise.
863 public boolean initiateCaptureSnapshot(final String followerId) {
864 FollowerLogInformation followerLogInfo = followerToLog.get(followerId);
865 if (snapshotHolder.isPresent()) {
866 // If a snapshot is present in the memory, most likely another install is in progress no need to capture
867 // snapshot. This could happen if another follower needs an install when one is going on.
868 final ActorSelection followerActor = context.getPeerActorSelection(followerId);
870 // Note: sendSnapshotChunk will set the LeaderInstallSnapshotState.
871 sendSnapshotChunk(followerActor, followerLogInfo);
875 boolean captureInitiated = context.getSnapshotManager().captureToInstall(context.getReplicatedLog().last(),
876 this.getReplicatedToAllIndex(), followerId);
877 if (captureInitiated) {
878 followerLogInfo.setLeaderInstallSnapshotState(new LeaderInstallSnapshotState(
879 context.getConfigParams().getSnapshotChunkSize(), logName()));
882 return captureInitiated;
885 private boolean canInstallSnapshot(final long nextIndex) {
886 // If the follower's nextIndex is -1 then we might as well send it a snapshot
887 // Otherwise send it a snapshot only if the nextIndex is not present in the log but is present
889 return nextIndex == -1 || !context.getReplicatedLog().isPresent(nextIndex)
890 && context.getReplicatedLog().isInSnapshot(nextIndex);
895 private void sendInstallSnapshot() {
896 log.debug("{}: sendInstallSnapshot", logName());
897 for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
898 String followerId = e.getKey();
899 ActorSelection followerActor = context.getPeerActorSelection(followerId);
900 FollowerLogInformation followerLogInfo = e.getValue();
902 if (followerActor != null) {
903 long nextIndex = followerLogInfo.getNextIndex();
904 if (followerLogInfo.getInstallSnapshotState() != null
905 || context.getPeerInfo(followerId).getVotingState() == VotingState.VOTING_NOT_INITIALIZED
906 || canInstallSnapshot(nextIndex)) {
907 sendSnapshotChunk(followerActor, followerLogInfo);
914 * Sends a snapshot chunk to a given follower
915 * InstallSnapshot should qualify as a heartbeat too.
917 private void sendSnapshotChunk(final ActorSelection followerActor, final FollowerLogInformation followerLogInfo) {
918 if (snapshotHolder.isPresent()) {
919 LeaderInstallSnapshotState installSnapshotState = followerLogInfo.getInstallSnapshotState();
920 if (installSnapshotState == null) {
921 installSnapshotState = new LeaderInstallSnapshotState(context.getConfigParams().getSnapshotChunkSize(),
923 followerLogInfo.setLeaderInstallSnapshotState(installSnapshotState);
927 // Ensure the snapshot bytes are set - this is a no-op.
928 installSnapshotState.setSnapshotBytes(snapshotHolder.get().getSnapshotBytes());
930 if (!installSnapshotState.canSendNextChunk()) {
934 byte[] nextSnapshotChunk = installSnapshotState.getNextChunk();
936 log.debug("{}: next snapshot chunk size for follower {}: {}", logName(), followerLogInfo.getId(),
937 nextSnapshotChunk.length);
939 int nextChunkIndex = installSnapshotState.incrementChunkIndex();
940 Optional<ServerConfigurationPayload> serverConfig = Optional.absent();
941 if (installSnapshotState.isLastChunk(nextChunkIndex)) {
942 serverConfig = Optional.fromNullable(context.getPeerServerInfo(true));
945 sendSnapshotChunk(followerActor, followerLogInfo, nextSnapshotChunk, nextChunkIndex, serverConfig);
947 } catch (IOException e) {
948 throw new RuntimeException(e);
951 log.debug("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}", logName(), followerActor.path(),
952 installSnapshotState.getChunkIndex(), installSnapshotState.getTotalChunks());
956 private void sendSnapshotChunk(final ActorSelection followerActor, final FollowerLogInformation followerLogInfo,
957 final byte[] snapshotChunk, final int chunkIndex,
958 final Optional<ServerConfigurationPayload> serverConfig) {
959 LeaderInstallSnapshotState installSnapshotState = followerLogInfo.getInstallSnapshotState();
961 installSnapshotState.startChunkTimer();
963 new InstallSnapshot(currentTerm(), context.getId(),
964 snapshotHolder.get().getLastIncludedIndex(),
965 snapshotHolder.get().getLastIncludedTerm(),
968 installSnapshotState.getTotalChunks(),
969 Optional.of(installSnapshotState.getLastChunkHashCode()),
971 ).toSerializable(followerLogInfo.getRaftVersion()),
976 private boolean resendSnapshotChunk(final ActorSelection followerActor,
977 final FollowerLogInformation followerLogInfo) {
978 if (!snapshotHolder.isPresent()) {
979 // Seems like we should never hit this case, but just in case we do, reset the snapshot progress so that it
980 // can restart from the next AppendEntries.
981 log.warn("{}: Attempting to resend snapshot with no snapshot holder present.", logName());
982 followerLogInfo.clearLeaderInstallSnapshotState();
986 LeaderInstallSnapshotState installSnapshotState = followerLogInfo.getInstallSnapshotState();
987 // we are resending, timer needs to be reset
988 installSnapshotState.resetChunkTimer();
989 installSnapshotState.markSendStatus(false);
991 sendSnapshotChunk(followerActor, followerLogInfo);
996 private void sendHeartBeat() {
997 if (!followerToLog.isEmpty()) {
998 log.trace("{}: Sending heartbeat", logName());
999 sendAppendEntries(context.getConfigParams().getHeartBeatInterval().toNanos(), true);
1001 appendEntriesMessageSlicer.checkExpiredSlicedMessageState();
1005 private void stopHeartBeat() {
1006 if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
1007 heartbeatSchedule.cancel();
1011 private void scheduleHeartBeat(final FiniteDuration interval) {
1012 if (followerToLog.isEmpty()) {
1013 // Optimization - do not bother scheduling a heartbeat as there are
1020 // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
1021 // message is sent to itself.
1022 // Scheduling the heartbeat only once here because heartbeats do not
1023 // need to be sent if there are other messages being sent to the remote
1025 heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
1026 interval, context.getActor(), SendHeartBeat.INSTANCE,
1027 context.getActorSystem().dispatcher(), context.getActor());
1031 public void close() {
1033 appendEntriesMessageSlicer.close();
1037 public final String getLeaderId() {
1038 return context.getId();
1042 public final short getLeaderPayloadVersion() {
1043 return context.getPayloadVersion();
1046 protected boolean isLeaderIsolated() {
1047 int minPresent = getMinIsolatedLeaderPeerCount();
1048 for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
1049 final PeerInfo peerInfo = context.getPeerInfo(followerLogInformation.getId());
1050 if (peerInfo != null && peerInfo.isVoting() && followerLogInformation.isFollowerActive()) {
1052 if (minPresent == 0) {
1057 return minPresent != 0;
1060 // called from example-actor for printing the follower-states
1061 public String printFollowerStates() {
1062 final StringBuilder sb = new StringBuilder();
1065 for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
1067 sb.append(followerLogInformation.getId());
1068 sb.append(" state:");
1069 sb.append(followerLogInformation.isFollowerActive());
1074 return sb.toString();
1078 public FollowerLogInformation getFollower(final String followerId) {
1079 return followerToLog.get(followerId);
1083 public int followerLogSize() {
1084 return followerToLog.size();
1087 static class SnapshotHolder {
1088 private final long lastIncludedTerm;
1089 private final long lastIncludedIndex;
1090 private final ByteSource snapshotBytes;
1092 SnapshotHolder(final Snapshot snapshot, final ByteSource snapshotBytes) {
1093 this.lastIncludedTerm = snapshot.getLastAppliedTerm();
1094 this.lastIncludedIndex = snapshot.getLastAppliedIndex();
1095 this.snapshotBytes = snapshotBytes;
1098 long getLastIncludedTerm() {
1099 return lastIncludedTerm;
1102 long getLastIncludedIndex() {
1103 return lastIncludedIndex;
1106 ByteSource getSnapshotBytes() {
1107 return snapshotBytes;