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.behaviors;
11 import akka.actor.ActorRef;
12 import akka.actor.ActorSelection;
13 import akka.actor.Cancellable;
14 import com.google.common.annotations.VisibleForTesting;
15 import com.google.common.base.Optional;
16 import com.google.common.base.Preconditions;
17 import com.google.common.collect.ImmutableMap;
18 import com.google.common.collect.ImmutableMap.Builder;
19 import com.google.protobuf.ByteString;
20 import java.io.IOException;
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 org.opendaylight.controller.cluster.raft.ClientRequestTracker;
30 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
31 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
32 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
33 import org.opendaylight.controller.cluster.raft.RaftActorContext;
34 import org.opendaylight.controller.cluster.raft.RaftState;
35 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
36 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
37 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
38 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
39 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
40 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
41 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
42 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
43 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
44 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
45 import scala.concurrent.duration.FiniteDuration;
48 * The behavior of a RaftActor when it is in the Leader state
52 * <li> Upon election: send initial empty AppendEntries RPCs
53 * (heartbeat) to each server; repeat during idle periods to
54 * prevent election timeouts (§5.2)
55 * <li> If command received from client: append entry to local log,
56 * respond after entry applied to state machine (§5.3)
57 * <li> If last log index ≥ nextIndex for a follower: send
58 * AppendEntries RPC with log entries starting at nextIndex
60 * <li> If successful: update nextIndex and matchIndex for
62 * <li> If AppendEntries fails because of log inconsistency:
63 * decrement nextIndex and retry (§5.3)
65 * <li> If there exists an N such that N > commitIndex, a majority
66 * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
67 * set commitIndex = N (§5.3, §5.4).
69 public abstract class AbstractLeader extends AbstractRaftActorBehavior {
71 // The index of the first chunk that is sent when installing a snapshot
72 public static final int FIRST_CHUNK_INDEX = 1;
74 // The index that the follower should respond with if it needs the install snapshot to be reset
75 public static final int INVALID_CHUNK_INDEX = -1;
77 // This would be passed as the hash code of the last chunk when sending the first chunk
78 public static final int INITIAL_LAST_CHUNK_HASH_CODE = -1;
80 private final Map<String, FollowerLogInformation> followerToLog;
81 private final Map<String, FollowerToSnapshot> mapFollowerToSnapshot = new HashMap<>();
83 private Cancellable heartbeatSchedule = null;
85 private final Collection<ClientRequestTracker> trackerList = new LinkedList<>();
87 protected final int minReplicationCount;
89 protected final int minIsolatedLeaderPeerCount;
91 private Optional<ByteString> snapshot;
93 public AbstractLeader(RaftActorContext context) {
94 super(context, RaftState.Leader);
96 setLeaderPayloadVersion(context.getPayloadVersion());
98 final Builder<String, FollowerLogInformation> ftlBuilder = ImmutableMap.builder();
99 for (String followerId : context.getPeerAddresses().keySet()) {
100 FollowerLogInformation followerLogInformation =
101 new FollowerLogInformationImpl(followerId, -1, context);
103 ftlBuilder.put(followerId, followerLogInformation);
105 followerToLog = ftlBuilder.build();
107 leaderId = context.getId();
109 LOG.debug("{}: Election: Leader has following peers: {}", logName(), getFollowerIds());
111 minReplicationCount = getMajorityVoteCount(getFollowerIds().size());
113 // the isolated Leader peer count will be 1 less than the majority vote count.
114 // this is because the vote count has the self vote counted in it
116 // 0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
117 // 2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
118 // 4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
119 minIsolatedLeaderPeerCount = minReplicationCount > 0 ? (minReplicationCount - 1) : 0;
121 snapshot = Optional.absent();
123 // Immediately schedule a heartbeat
124 // Upon election: send initial empty AppendEntries RPCs
125 // (heartbeat) to each server; repeat during idle periods to
126 // prevent election timeouts (§5.2)
127 sendAppendEntries(0, false);
129 // It is important to schedule this heartbeat here
130 scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
134 * Return an immutable collection of follower identifiers.
136 * @return Collection of follower IDs
138 public final Collection<String> getFollowerIds() {
139 return followerToLog.keySet();
143 void setSnapshot(Optional<ByteString> snapshot) {
144 this.snapshot = snapshot;
148 protected RaftActorBehavior handleAppendEntries(ActorRef sender,
149 AppendEntries appendEntries) {
151 LOG.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
157 protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
158 AppendEntriesReply appendEntriesReply) {
160 if(LOG.isTraceEnabled()) {
161 LOG.trace("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
164 // Update the FollowerLogInformation
165 String followerId = appendEntriesReply.getFollowerId();
166 FollowerLogInformation followerLogInformation =
167 followerToLog.get(followerId);
169 if(followerLogInformation == null){
170 LOG.error("{}: handleAppendEntriesReply - unknown follower {}", logName(), followerId);
174 if(followerLogInformation.timeSinceLastActivity() >
175 context.getConfigParams().getElectionTimeOutInterval().toMillis()) {
176 LOG.warn("{} : handleAppendEntriesReply delayed beyond election timeout, " +
177 "appendEntriesReply : {}, timeSinceLastActivity : {}, lastApplied : {}, commitIndex : {}",
178 logName(), appendEntriesReply, followerLogInformation.timeSinceLastActivity(),
179 context.getLastApplied(), context.getCommitIndex());
182 followerLogInformation.markFollowerActive();
183 followerLogInformation.setPayloadVersion(appendEntriesReply.getPayloadVersion());
185 boolean updated = false;
186 if (appendEntriesReply.isSuccess()) {
187 updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
189 LOG.debug("{}: handleAppendEntriesReply: received unsuccessful reply: {}", logName(), appendEntriesReply);
191 long followerLastLogIndex = appendEntriesReply.getLogLastIndex();
192 ReplicatedLogEntry followersLastLogEntry = context.getReplicatedLog().get(followerLastLogIndex);
193 if(followerLastLogIndex < 0 || (followersLastLogEntry != null &&
194 followersLastLogEntry.getTerm() == appendEntriesReply.getLogLastTerm())) {
195 // The follower's log is empty or the last entry is present in the leader's journal
196 // and the terms match so the follower is just behind the leader's journal from
197 // the last snapshot, if any. We'll catch up the follower quickly by starting at the
198 // follower's last log index.
200 updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
202 // TODO: When we find that the follower is out of sync with the
203 // Leader we simply decrement that followers next index by 1.
204 // Would it be possible to do better than this? The RAFT spec
205 // does not explicitly deal with it but may be something for us to
208 followerLogInformation.decrNextIndex();
212 // Now figure out if this reply warrants a change in the commitIndex
213 // If there exists an N such that N > commitIndex, a majority
214 // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
215 // set commitIndex = N (§5.3, §5.4).
216 for (long N = context.getCommitIndex() + 1; ; N++) {
217 int replicatedCount = 1;
219 for (FollowerLogInformation info : followerToLog.values()) {
220 if (info.getMatchIndex() >= N) {
225 if (replicatedCount >= minReplicationCount) {
226 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(N);
227 if (replicatedLogEntry != null &&
228 replicatedLogEntry.getTerm() == currentTerm()) {
229 context.setCommitIndex(N);
236 // Apply the change to the state machine
237 if (context.getCommitIndex() > context.getLastApplied()) {
238 if(LOG.isDebugEnabled()) {
239 LOG.debug("{}: handleAppendEntriesReply from {}: applying to log - commitIndex: {}, lastAppliedIndex: {}",
240 logName(), followerId, context.getCommitIndex(), context.getLastApplied());
243 applyLogToStateMachine(context.getCommitIndex());
246 if (!context.getSnapshotManager().isCapturing()) {
250 //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
251 sendUpdatesToFollower(followerId, followerLogInformation, false, !updated);
255 private boolean updateFollowerLogInformation(FollowerLogInformation followerLogInformation,
256 AppendEntriesReply appendEntriesReply) {
257 boolean updated = followerLogInformation.setMatchIndex(appendEntriesReply.getLogLastIndex());
258 updated = followerLogInformation.setNextIndex(appendEntriesReply.getLogLastIndex() + 1) || updated;
260 if(updated && LOG.isDebugEnabled()) {
261 LOG.debug("{}: handleAppendEntriesReply - FollowerLogInformation for {} updated: matchIndex: {}, nextIndex: {}",
262 logName(), followerLogInformation.getId(), followerLogInformation.getMatchIndex(),
263 followerLogInformation.getNextIndex());
268 private void purgeInMemoryLog() {
269 //find the lowest index across followers which has been replicated to all.
270 // lastApplied if there are no followers, so that we keep clearing the log for single-node
271 // we would delete the in-mem log from that index on, in-order to minimize mem usage
272 // we would also share this info thru AE with the followers so that they can delete their log entries as well.
273 long minReplicatedToAllIndex = followerToLog.isEmpty() ? context.getLastApplied() : Long.MAX_VALUE;
274 for (FollowerLogInformation info : followerToLog.values()) {
275 minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
278 super.performSnapshotWithoutCapture(minReplicatedToAllIndex);
282 protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
283 final Iterator<ClientRequestTracker> it = trackerList.iterator();
284 while (it.hasNext()) {
285 final ClientRequestTracker t = it.next();
286 if (t.getIndex() == logIndex) {
296 protected ClientRequestTracker findClientRequestTracker(long logIndex) {
297 for (ClientRequestTracker tracker : trackerList) {
298 if (tracker.getIndex() == logIndex) {
306 protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
307 RequestVoteReply requestVoteReply) {
311 protected void beforeSendHeartbeat(){}
314 public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
315 Preconditions.checkNotNull(sender, "sender should not be null");
317 Object message = fromSerializableMessage(originalMessage);
319 if (message instanceof RaftRPC) {
320 RaftRPC rpc = (RaftRPC) message;
321 // If RPC request or response contains term T > currentTerm:
322 // set currentTerm = T, convert to follower (§5.1)
323 // This applies to all RPC messages and responses
324 if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
325 LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {} - switching to Follower",
326 logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
328 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
330 return switchBehavior(new Follower(context));
334 if (message instanceof SendHeartBeat) {
335 beforeSendHeartbeat();
337 scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
340 } else if(message instanceof SendInstallSnapshot) {
341 // received from RaftActor
342 setSnapshot(Optional.of(((SendInstallSnapshot) message).getSnapshot()));
343 sendInstallSnapshot();
345 } else if (message instanceof Replicate) {
346 replicate((Replicate) message);
348 } else if (message instanceof InstallSnapshotReply){
349 handleInstallSnapshotReply((InstallSnapshotReply) message);
354 return super.handleMessage(sender, message);
357 private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
358 LOG.debug("{}: handleInstallSnapshotReply: {}", logName(), reply);
360 String followerId = reply.getFollowerId();
361 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
363 if (followerToSnapshot == null) {
364 LOG.error("{}: FollowerId {} in InstallSnapshotReply not known to Leader",
365 logName(), followerId);
369 FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
370 followerLogInformation.markFollowerActive();
372 if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
373 boolean wasLastChunk = false;
374 if (reply.isSuccess()) {
375 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
376 //this was the last chunk reply
377 if(LOG.isDebugEnabled()) {
378 LOG.debug("{}: InstallSnapshotReply received, " +
379 "last chunk received, Chunk: {}. Follower: {} Setting nextIndex: {}",
380 logName(), reply.getChunkIndex(), followerId,
381 context.getReplicatedLog().getSnapshotIndex() + 1
385 followerLogInformation.setMatchIndex(
386 context.getReplicatedLog().getSnapshotIndex());
387 followerLogInformation.setNextIndex(
388 context.getReplicatedLog().getSnapshotIndex() + 1);
389 mapFollowerToSnapshot.remove(followerId);
391 LOG.debug("{}: follower: {}, matchIndex set to {}, nextIndex set to {}",
392 logName(), followerId, followerLogInformation.getMatchIndex(),
393 followerLogInformation.getNextIndex());
395 if (mapFollowerToSnapshot.isEmpty()) {
396 // once there are no pending followers receiving snapshots
397 // we can remove snapshot from the memory
398 setSnapshot(Optional.<ByteString>absent());
403 followerToSnapshot.markSendStatus(true);
406 LOG.info("{}: InstallSnapshotReply received sending snapshot chunk failed, Will retry, Chunk: {}",
407 logName(), reply.getChunkIndex());
409 followerToSnapshot.markSendStatus(false);
412 if (wasLastChunk && !context.getSnapshotManager().isCapturing()) {
413 // Since the follower is now caught up try to purge the log.
415 } else if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
416 ActorSelection followerActor = context.getPeerActorSelection(followerId);
417 if(followerActor != null) {
418 sendSnapshotChunk(followerActor, followerId);
423 LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
424 logName(), reply.getChunkIndex(), followerId,
425 followerToSnapshot.getChunkIndex());
427 if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
428 // Since the Follower did not find this index to be valid we should reset the follower snapshot
429 // so that Installing the snapshot can resume from the beginning
430 followerToSnapshot.reset();
435 private void replicate(Replicate replicate) {
436 long logIndex = replicate.getReplicatedLogEntry().getIndex();
438 LOG.debug("{}: Replicate message: identifier: {}, logIndex: {}", logName(),
439 replicate.getIdentifier(), logIndex);
441 // Create a tracker entry we will use this later to notify the
444 new ClientRequestTrackerImpl(replicate.getClientActor(),
445 replicate.getIdentifier(),
449 if (followerToLog.isEmpty()) {
450 context.setCommitIndex(logIndex);
451 applyLogToStateMachine(logIndex);
453 sendAppendEntries(0, false);
457 private void sendAppendEntries(long timeSinceLastActivityInterval, boolean isHeartbeat) {
458 // Send an AppendEntries to all followers
459 for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
460 final String followerId = e.getKey();
461 final FollowerLogInformation followerLogInformation = e.getValue();
462 // This checks helps not to send a repeat message to the follower
463 if(!followerLogInformation.isFollowerActive() ||
464 followerLogInformation.timeSinceLastActivity() >= timeSinceLastActivityInterval) {
465 sendUpdatesToFollower(followerId, followerLogInformation, true, isHeartbeat);
472 * This method checks if any update needs to be sent to the given follower. This includes append log entries,
473 * sending next snapshot chunk, and initiating a snapshot.
474 * @return true if any update is sent, false otherwise
477 private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation,
478 boolean sendHeartbeat, boolean isHeartbeat) {
480 ActorSelection followerActor = context.getPeerActorSelection(followerId);
481 if (followerActor != null) {
482 long followerNextIndex = followerLogInformation.getNextIndex();
483 boolean isFollowerActive = followerLogInformation.isFollowerActive();
484 boolean sendAppendEntries = false;
485 List<ReplicatedLogEntry> entries = Collections.emptyList();
487 if (mapFollowerToSnapshot.get(followerId) != null) {
488 // if install snapshot is in process , then sent next chunk if possible
489 if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
490 sendSnapshotChunk(followerActor, followerId);
491 } else if(sendHeartbeat) {
492 // we send a heartbeat even if we have not received a reply for the last chunk
493 sendAppendEntries = true;
496 long leaderLastIndex = context.getReplicatedLog().lastIndex();
497 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
499 if((!isHeartbeat && LOG.isDebugEnabled()) || LOG.isTraceEnabled()) {
500 LOG.debug("{}: Checking sendAppendEntries for follower {}: active: {}, followerNextIndex: {}, leaderLastIndex: {}, leaderSnapShotIndex: {}",
501 logName(), followerId, isFollowerActive, followerNextIndex, leaderLastIndex, leaderSnapShotIndex);
504 if (isFollowerActive && context.getReplicatedLog().isPresent(followerNextIndex)) {
506 LOG.debug("{}: sendAppendEntries: {} is present for follower {}", logName(),
507 followerNextIndex, followerId);
509 if(followerLogInformation.okToReplicate()) {
510 // Try to send all the entries in the journal but not exceeding the max data size
511 // for a single AppendEntries message.
512 int maxEntries = (int) context.getReplicatedLog().size();
513 entries = context.getReplicatedLog().getFrom(followerNextIndex, maxEntries,
514 context.getConfigParams().getSnapshotChunkSize());
515 sendAppendEntries = true;
517 } else if (isFollowerActive && followerNextIndex >= 0 &&
518 leaderLastIndex > followerNextIndex && !context.getSnapshotManager().isCapturing()) {
519 // if the followers next index is not present in the leaders log, and
520 // if the follower is just not starting and if leader's index is more than followers index
521 // then snapshot should be sent
523 if (LOG.isDebugEnabled()) {
524 LOG.debug(String.format("%s: InitiateInstallSnapshot to follower: %s," +
525 "follower-nextIndex: %d, leader-snapshot-index: %d, " +
526 "leader-last-index: %d", logName(), followerId,
527 followerNextIndex, leaderSnapShotIndex, leaderLastIndex));
530 // Send heartbeat to follower whenever install snapshot is initiated.
531 sendAppendEntries = true;
532 initiateCaptureSnapshot(followerId, followerNextIndex);
534 } else if(sendHeartbeat) {
535 // we send an AppendEntries, even if the follower is inactive
536 // in-order to update the followers timestamp, in case it becomes active again
537 sendAppendEntries = true;
542 if(sendAppendEntries) {
543 sendAppendEntriesToFollower(followerActor, followerNextIndex,
544 entries, followerId);
549 private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
550 List<ReplicatedLogEntry> entries, String followerId) {
551 AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
552 prevLogIndex(followerNextIndex),
553 prevLogTerm(followerNextIndex), entries,
554 context.getCommitIndex(), super.getReplicatedToAllIndex(), context.getPayloadVersion());
556 if(!entries.isEmpty() || LOG.isTraceEnabled()) {
557 LOG.debug("{}: Sending AppendEntries to follower {}: {}", logName(), followerId,
561 followerActor.tell(appendEntries.toSerializable(), actor());
565 * Install Snapshot works as follows
566 * 1. Leader initiates the capture snapshot by sending a CaptureSnapshot message to actor
567 * 2. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
568 * and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
569 * 3. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
570 * 4. On complete, Follower sends back a InstallSnapshotReply.
571 * 5. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
572 * and replenishes the memory by deleting the snapshot in Replicated log.
573 * 6. If another follower requires a snapshot and a snapshot has been collected (via CaptureSnapshotReply)
574 * then send the existing snapshot in chunks to the follower.
576 * @param followerNextIndex
578 private void initiateCaptureSnapshot(String followerId, long followerNextIndex) {
579 if (!context.getReplicatedLog().isPresent(followerNextIndex) &&
580 context.getReplicatedLog().isInSnapshot(followerNextIndex)) {
582 if (snapshot.isPresent()) {
583 // if a snapshot is present in the memory, most likely another install is in progress
584 // no need to capture snapshot.
585 // This could happen if another follower needs an install when one is going on.
586 final ActorSelection followerActor = context.getPeerActorSelection(followerId);
587 sendSnapshotChunk(followerActor, followerId);
591 context.getSnapshotManager().captureToInstall(context.getReplicatedLog().last(),
592 this.getReplicatedToAllIndex(), followerId);
598 private void sendInstallSnapshot() {
599 LOG.debug("{}: sendInstallSnapshot", logName());
600 for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
601 ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
603 if (followerActor != null) {
604 long nextIndex = e.getValue().getNextIndex();
606 if (!context.getReplicatedLog().isPresent(nextIndex) &&
607 context.getReplicatedLog().isInSnapshot(nextIndex)) {
608 sendSnapshotChunk(followerActor, e.getKey());
615 * Sends a snapshot chunk to a given follower
616 * InstallSnapshot should qualify as a heartbeat too.
618 private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
620 if (snapshot.isPresent()) {
621 ByteString nextSnapshotChunk = getNextSnapshotChunk(followerId,snapshot.get());
623 // Note: the previous call to getNextSnapshotChunk has the side-effect of adding
624 // followerId to the followerToSnapshot map.
625 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
628 new InstallSnapshot(currentTerm(), context.getId(),
629 context.getReplicatedLog().getSnapshotIndex(),
630 context.getReplicatedLog().getSnapshotTerm(),
632 followerToSnapshot.incrementChunkIndex(),
633 followerToSnapshot.getTotalChunks(),
634 Optional.of(followerToSnapshot.getLastChunkHashCode())
639 if(LOG.isDebugEnabled()) {
640 LOG.debug("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
641 logName(), followerActor.path(), followerToSnapshot.getChunkIndex(),
642 followerToSnapshot.getTotalChunks());
645 } catch (IOException e) {
646 LOG.error("{}: InstallSnapshot failed for Leader.", logName(), e);
651 * Acccepts snaphot as ByteString, enters into map for future chunks
652 * creates and return a ByteString chunk
654 private ByteString getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
655 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
656 if (followerToSnapshot == null) {
657 followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
658 mapFollowerToSnapshot.put(followerId, followerToSnapshot);
660 ByteString nextChunk = followerToSnapshot.getNextChunk();
662 LOG.debug("{}: next snapshot chunk size for follower {}: {}", logName(), followerId, nextChunk.size());
667 private void sendHeartBeat() {
668 if (!followerToLog.isEmpty()) {
669 LOG.trace("{}: Sending heartbeat", logName());
670 sendAppendEntries(context.getConfigParams().getHeartBeatInterval().toMillis(), true);
674 private void stopHeartBeat() {
675 if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
676 heartbeatSchedule.cancel();
680 private void scheduleHeartBeat(FiniteDuration interval) {
681 if (followerToLog.isEmpty()) {
682 // Optimization - do not bother scheduling a heartbeat as there are
689 // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
690 // message is sent to itself.
691 // Scheduling the heartbeat only once here because heartbeats do not
692 // need to be sent if there are other messages being sent to the remote
694 heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
695 interval, context.getActor(), new SendHeartBeat(),
696 context.getActorSystem().dispatcher(), context.getActor());
700 public void close() throws Exception {
705 public String getLeaderId() {
706 return context.getId();
709 protected boolean isLeaderIsolated() {
710 int minPresent = minIsolatedLeaderPeerCount;
711 for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
712 if (followerLogInformation.isFollowerActive()) {
714 if (minPresent == 0) {
719 return (minPresent != 0);
723 * Encapsulates the snapshot bytestring and handles the logic of sending
726 protected class FollowerToSnapshot {
727 private final ByteString snapshotBytes;
728 private int offset = 0;
729 // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
730 private int replyReceivedForOffset;
731 // if replyStatus is false, the previous chunk is attempted
732 private boolean replyStatus = false;
733 private int chunkIndex;
734 private final int totalChunks;
735 private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
736 private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
738 public FollowerToSnapshot(ByteString snapshotBytes) {
739 this.snapshotBytes = snapshotBytes;
740 int size = snapshotBytes.size();
741 totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
742 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
743 if(LOG.isDebugEnabled()) {
744 LOG.debug("{}: Snapshot {} bytes, total chunks to send:{}",
745 logName(), size, totalChunks);
747 replyReceivedForOffset = -1;
748 chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
751 public ByteString getSnapshotBytes() {
752 return snapshotBytes;
755 public int incrementOffset() {
757 // if prev chunk failed, we would want to sent the same chunk again
758 offset = offset + context.getConfigParams().getSnapshotChunkSize();
763 public int incrementChunkIndex() {
765 // if prev chunk failed, we would want to sent the same chunk again
766 chunkIndex = chunkIndex + 1;
771 public int getChunkIndex() {
775 public int getTotalChunks() {
779 public boolean canSendNextChunk() {
780 // we only send a false if a chunk is sent but we have not received a reply yet
781 return replyReceivedForOffset == offset;
784 public boolean isLastChunk(int chunkIndex) {
785 return totalChunks == chunkIndex;
788 public void markSendStatus(boolean success) {
790 // if the chunk sent was successful
791 replyReceivedForOffset = offset;
793 lastChunkHashCode = nextChunkHashCode;
795 // if the chunk sent was failure
796 replyReceivedForOffset = offset;
801 public ByteString getNextChunk() {
802 int snapshotLength = getSnapshotBytes().size();
803 int start = incrementOffset();
804 int size = context.getConfigParams().getSnapshotChunkSize();
805 if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
806 size = snapshotLength;
808 if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
809 size = snapshotLength - start;
814 LOG.debug("{}: Next chunk: length={}, offset={},size={}", logName(),
815 snapshotLength, start, size);
817 ByteString substring = getSnapshotBytes().substring(start, start + size);
818 nextChunkHashCode = substring.hashCode();
823 * reset should be called when the Follower needs to be sent the snapshot from the beginning
828 replyReceivedForOffset = offset;
829 chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
830 lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
833 public int getLastChunkHashCode() {
834 return lastChunkHashCode;
838 // called from example-actor for printing the follower-states
839 public String printFollowerStates() {
840 final StringBuilder sb = new StringBuilder();
843 for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
845 sb.append(followerLogInformation.getId());
846 sb.append(" state:");
847 sb.append(followerLogInformation.isFollowerActive());
852 return sb.toString();
856 public FollowerLogInformation getFollower(String followerId) {
857 return followerToLog.get(followerId);
861 protected void setFollowerSnapshot(String followerId, FollowerToSnapshot snapshot) {
862 mapFollowerToSnapshot.put(followerId, snapshot);
866 public int followerSnapshotSize() {
867 return mapFollowerToSnapshot.size();
871 public int followerLogSize() {
872 return followerToLog.size();