opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/AbstractLeader.java

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   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
   7  */
   8
   9 package org.opendaylight.controller.cluster.raft.behaviors;
  10
  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.protobuf.ByteString;
  18 import java.io.IOException;
  19 import java.util.Arrays;
  20 import java.util.Collection;
  21 import java.util.Collections;
  22 import java.util.HashMap;
  23 import java.util.Iterator;
  24 import java.util.LinkedList;
  25 import java.util.List;
  26 import java.util.Map;
  27 import java.util.Map.Entry;
  28 import java.util.Queue;
  29 import javax.annotation.Nullable;
  30 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  31 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  32 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  33 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  34 import org.opendaylight.controller.cluster.raft.PeerInfo;
  35 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  36 import org.opendaylight.controller.cluster.raft.RaftState;
  37 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  38 import org.opendaylight.controller.cluster.raft.Snapshot;
  39 import org.opendaylight.controller.cluster.raft.VotingState;
  40 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  41 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  42 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  43 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  44 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  45 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  46 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  47 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  48 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  49 import org.opendaylight.controller.cluster.raft.messages.UnInitializedFollowerSnapshotReply;
  50 import scala.concurrent.duration.FiniteDuration;
  51
  52 /**
  53  * The behavior of a RaftActor when it is in the Leader state
  54  * <p/>
  55  * Leaders:
  56  * <ul>
  57  * <li> Upon election: send initial empty AppendEntries RPCs
  58  * (heartbeat) to each server; repeat during idle periods to
  59  * prevent election timeouts (§5.2)
  60  * <li> If command received from client: append entry to local log,
  61  * respond after entry applied to state machine (§5.3)
  62  * <li> If last log index ≥ nextIndex for a follower: send
  63  * AppendEntries RPC with log entries starting at nextIndex
  64  * <ul>
  65  * <li> If successful: update nextIndex and matchIndex for
  66  * follower (§5.3)
  67  * <li> If AppendEntries fails because of log inconsistency:
  68  * decrement nextIndex and retry (§5.3)
  69  * </ul>
  70  * <li> If there exists an N such that N > commitIndex, a majority
  71  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  72  * set commitIndex = N (§5.3, §5.4).
  73  */
  74 public abstract class AbstractLeader extends AbstractRaftActorBehavior {
  75
  76     // The index of the first chunk that is sent when installing a snapshot
  77     public static final int FIRST_CHUNK_INDEX = 1;
  78
  79     // The index that the follower should respond with if it needs the install snapshot to be reset
  80     public static final int INVALID_CHUNK_INDEX = -1;
  81
  82     // This would be passed as the hash code of the last chunk when sending the first chunk
  83     public static final int INITIAL_LAST_CHUNK_HASH_CODE = -1;
  84
  85     private final Map<String, FollowerLogInformation> followerToLog = new HashMap<>();
  86     private final Map<String, FollowerToSnapshot> mapFollowerToSnapshot = new HashMap<>();
  87
  88     /**
  89      * Lookup table for request contexts based on journal index. We could use a {@link Map} here, but we really
  90      * expect the entries to be modified in sequence, hence we open-code the lookup.
  91      *
  92      * TODO: Evaluate the use of ArrayDeque(), as that has lower memory overhead. Non-head removals are more costly,
  93      *       but we already expect those to be far from frequent.
  94      */
  95     private final Queue<ClientRequestTracker> trackers = new LinkedList<>();
  96
  97     private Cancellable heartbeatSchedule = null;
  98     private Optional<SnapshotHolder> snapshot;
  99     private int minReplicationCount;
 100
 101     protected AbstractLeader(RaftActorContext context, RaftState state) {
 102         super(context, state);
 103
 104         for(PeerInfo peerInfo: context.getPeers()) {
 105             FollowerLogInformation followerLogInformation = new FollowerLogInformationImpl(peerInfo, -1, context);
 106             followerToLog.put(peerInfo.getId(), followerLogInformation);
 107         }
 108
 109         LOG.debug("{}: Election: Leader has following peers: {}", logName(), getFollowerIds());
 110
 111         updateMinReplicaCount();
 112
 113         snapshot = Optional.absent();
 114
 115         // Immediately schedule a heartbeat
 116         // Upon election: send initial empty AppendEntries RPCs
 117         // (heartbeat) to each server; repeat during idle periods to
 118         // prevent election timeouts (§5.2)
 119         sendAppendEntries(0, false);
 120
 121         // It is important to schedule this heartbeat here
 122         scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 123     }
 124
 125     /**
 126      * Return an immutable collection of follower identifiers.
 127      *
 128      * @return Collection of follower IDs
 129      */
 130     public final Collection<String> getFollowerIds() {
 131         return followerToLog.keySet();
 132     }
 133
 134     public void addFollower(String followerId) {
 135         FollowerLogInformation followerLogInformation = new FollowerLogInformationImpl(
 136                 context.getPeerInfo(followerId), -1, context);
 137         followerToLog.put(followerId, followerLogInformation);
 138
 139         if(heartbeatSchedule == null) {
 140             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 141         }
 142     }
 143
 144     public void removeFollower(String followerId) {
 145         followerToLog.remove(followerId);
 146         mapFollowerToSnapshot.remove(followerId);
 147     }
 148
 149     public void updateMinReplicaCount() {
 150         int numVoting = 0;
 151         for(PeerInfo peer: context.getPeers()) {
 152             if(peer.isVoting()) {
 153                 numVoting++;
 154             }
 155         }
 156
 157         minReplicationCount = getMajorityVoteCount(numVoting);
 158     }
 159
 160     protected int getMinIsolatedLeaderPeerCount(){
 161       //the isolated Leader peer count will be 1 less than the majority vote count.
 162         //this is because the vote count has the self vote counted in it
 163         //for e.g
 164         //0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
 165         //2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
 166         //4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
 167
 168         return minReplicationCount > 0 ? (minReplicationCount - 1) : 0;
 169     }
 170
 171     @VisibleForTesting
 172     void setSnapshot(@Nullable Snapshot snapshot) {
 173         if(snapshot != null) {
 174             this.snapshot = Optional.of(new SnapshotHolder(snapshot));
 175         } else {
 176             this.snapshot = Optional.absent();
 177         }
 178     }
 179
 180     @Override
 181     protected RaftActorBehavior handleAppendEntries(ActorRef sender,
 182         AppendEntries appendEntries) {
 183
 184         LOG.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 185
 186         return this;
 187     }
 188
 189     @Override
 190     protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 191         AppendEntriesReply appendEntriesReply) {
 192
 193         if(LOG.isTraceEnabled()) {
 194             LOG.trace("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
 195         }
 196
 197         // Update the FollowerLogInformation
 198         String followerId = appendEntriesReply.getFollowerId();
 199         FollowerLogInformation followerLogInformation =
 200             followerToLog.get(followerId);
 201
 202         if(followerLogInformation == null){
 203             LOG.error("{}: handleAppendEntriesReply - unknown follower {}", logName(), followerId);
 204             return this;
 205         }
 206
 207         if(followerLogInformation.timeSinceLastActivity() >
 208                 context.getConfigParams().getElectionTimeOutInterval().toMillis()) {
 209             LOG.warn("{} : handleAppendEntriesReply delayed beyond election timeout, " +
 210                             "appendEntriesReply : {}, timeSinceLastActivity : {}, lastApplied : {}, commitIndex : {}",
 211                     logName(), appendEntriesReply, followerLogInformation.timeSinceLastActivity(),
 212                     context.getLastApplied(), context.getCommitIndex());
 213         }
 214
 215         followerLogInformation.markFollowerActive();
 216         followerLogInformation.setPayloadVersion(appendEntriesReply.getPayloadVersion());
 217         followerLogInformation.setRaftVersion(appendEntriesReply.getRaftVersion());
 218
 219         boolean updated = false;
 220         if (appendEntriesReply.isSuccess()) {
 221             updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
 222         } else {
 223             LOG.debug("{}: handleAppendEntriesReply: received unsuccessful reply: {}", logName(), appendEntriesReply);
 224
 225             long followerLastLogIndex = appendEntriesReply.getLogLastIndex();
 226             ReplicatedLogEntry followersLastLogEntry = context.getReplicatedLog().get(followerLastLogIndex);
 227             if(appendEntriesReply.isForceInstallSnapshot()) {
 228                 // Reset the followers match and next index. This is to signal that this follower has nothing
 229                 // in common with this Leader and so would require a snapshot to be installed
 230                 followerLogInformation.setMatchIndex(-1);
 231                 followerLogInformation.setNextIndex(-1);
 232
 233                 // Force initiate a snapshot capture
 234                 initiateCaptureSnapshot(followerId);
 235             } else if(followerLastLogIndex < 0 || (followersLastLogEntry != null &&
 236                     followersLastLogEntry.getTerm() == appendEntriesReply.getLogLastTerm())) {
 237                 // The follower's log is empty or the last entry is present in the leader's journal
 238                 // and the terms match so the follower is just behind the leader's journal from
 239                 // the last snapshot, if any. We'll catch up the follower quickly by starting at the
 240                 // follower's last log index.
 241
 242                 updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
 243             } else {
 244                 // TODO: When we find that the follower is out of sync with the
 245                 // Leader we simply decrement that followers next index by 1.
 246                 // Would it be possible to do better than this? The RAFT spec
 247                 // does not explicitly deal with it but may be something for us to
 248                 // think about.
 249
 250                 followerLogInformation.decrNextIndex();
 251             }
 252         }
 253
 254         // Now figure out if this reply warrants a change in the commitIndex
 255         // If there exists an N such that N > commitIndex, a majority
 256         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 257         // set commitIndex = N (§5.3, §5.4).
 258         if(LOG.isTraceEnabled()) {
 259             LOG.trace("{}: handleAppendEntriesReply from {}: commitIndex: {}, lastAppliedIndex: {}, currentTerm: {}",
 260                     logName(), followerId, context.getCommitIndex(), context.getLastApplied(), currentTerm());
 261         }
 262
 263         for (long N = context.getCommitIndex() + 1; ; N++) {
 264             int replicatedCount = 1;
 265
 266             LOG.trace("{}: checking Nth index {}", logName(), N);
 267             for (FollowerLogInformation info : followerToLog.values()) {
 268                 final PeerInfo peerInfo = context.getPeerInfo(info.getId());
 269                 if(info.getMatchIndex() >= N && (peerInfo != null && peerInfo.isVoting())) {
 270                     replicatedCount++;
 271                 } else if(LOG.isTraceEnabled()) {
 272                     LOG.trace("{}: Not counting follower {} - matchIndex: {}, {}", logName(), info.getId(),
 273                             info.getMatchIndex(), peerInfo);
 274                 }
 275             }
 276
 277             if(LOG.isTraceEnabled()) {
 278                 LOG.trace("{}: replicatedCount {}, minReplicationCount: {}", logName(), replicatedCount, minReplicationCount);
 279             }
 280
 281             if (replicatedCount >= minReplicationCount) {
 282                 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(N);
 283                 if (replicatedLogEntry == null) {
 284                     LOG.debug("{}: ReplicatedLogEntry not found for index {} - snapshotIndex: {}, journal size: {}",
 285                             logName(), N, context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().size());
 286                     break;
 287                 }
 288
 289                 // Don't update the commit index if the log entry is from a previous term, as per §5.4.1:
 290                 // "Raft never commits log entries from previous terms by counting replicas".
 291                 // However we keep looping so we can make progress when new entries in the current term
 292                 // reach consensus, as per §5.4.1: "once an entry from the current term is committed by
 293                 // counting replicas, then all prior entries are committed indirectly".
 294                 if (replicatedLogEntry.getTerm() == currentTerm()) {
 295                     LOG.trace("{}: Setting commit index to {}", logName(), N);
 296                     context.setCommitIndex(N);
 297                 } else {
 298                     LOG.debug("{}: Not updating commit index to {} - retrieved log entry with index {}, term {} does not match the current term {}",
 299                             logName(), N, replicatedLogEntry.getIndex(), replicatedLogEntry.getTerm(), currentTerm());
 300                 }
 301             } else {
 302                 LOG.trace("{}: minReplicationCount not reached, actual {} - breaking", logName(), replicatedCount);
 303                 break;
 304             }
 305         }
 306
 307         // Apply the change to the state machine
 308         if (context.getCommitIndex() > context.getLastApplied()) {
 309             if(LOG.isDebugEnabled()) {
 310                 LOG.debug("{}: handleAppendEntriesReply from {}: applying to log - commitIndex: {}, lastAppliedIndex: {}",
 311                         logName(), followerId, context.getCommitIndex(), context.getLastApplied());
 312             }
 313
 314             applyLogToStateMachine(context.getCommitIndex());
 315         }
 316
 317         if (!context.getSnapshotManager().isCapturing()) {
 318             purgeInMemoryLog();
 319         }
 320
 321         //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
 322         sendUpdatesToFollower(followerId, followerLogInformation, false, !updated);
 323         return this;
 324     }
 325
 326     private boolean updateFollowerLogInformation(FollowerLogInformation followerLogInformation,
 327             AppendEntriesReply appendEntriesReply) {
 328         boolean updated = followerLogInformation.setMatchIndex(appendEntriesReply.getLogLastIndex());
 329         updated = followerLogInformation.setNextIndex(appendEntriesReply.getLogLastIndex() + 1) || updated;
 330
 331         if(updated && LOG.isDebugEnabled()) {
 332             LOG.debug("{}: handleAppendEntriesReply - FollowerLogInformation for {} updated: matchIndex: {}, nextIndex: {}",
 333                     logName(), followerLogInformation.getId(), followerLogInformation.getMatchIndex(),
 334                     followerLogInformation.getNextIndex());
 335         }
 336         return updated;
 337     }
 338
 339     private void purgeInMemoryLog() {
 340         //find the lowest index across followers which has been replicated to all.
 341         // lastApplied if there are no followers, so that we keep clearing the log for single-node
 342         // we would delete the in-mem log from that index on, in-order to minimize mem usage
 343         // we would also share this info thru AE with the followers so that they can delete their log entries as well.
 344         long minReplicatedToAllIndex = followerToLog.isEmpty() ? context.getLastApplied() : Long.MAX_VALUE;
 345         for (FollowerLogInformation info : followerToLog.values()) {
 346             minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
 347         }
 348
 349         super.performSnapshotWithoutCapture(minReplicatedToAllIndex);
 350     }
 351
 352     @Override
 353     protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
 354         final Iterator<ClientRequestTracker> it = trackers.iterator();
 355         while (it.hasNext()) {
 356             final ClientRequestTracker t = it.next();
 357             if (t.getIndex() == logIndex) {
 358                 it.remove();
 359                 return t;
 360             }
 361         }
 362
 363         return null;
 364     }
 365
 366     @Override
 367     protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 368         RequestVoteReply requestVoteReply) {
 369         return this;
 370     }
 371
 372     protected void beforeSendHeartbeat(){}
 373
 374     @Override
 375     public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 376         Preconditions.checkNotNull(sender, "sender should not be null");
 377
 378         Object message = fromSerializableMessage(originalMessage);
 379
 380         if (message instanceof RaftRPC) {
 381             RaftRPC rpc = (RaftRPC) message;
 382             // If RPC request or response contains term T > currentTerm:
 383             // set currentTerm = T, convert to follower (§5.1)
 384             // This applies to all RPC messages and responses
 385             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 386                 LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {} - switching to Follower",
 387                         logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 388
 389                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 390
 391                 return internalSwitchBehavior(RaftState.Follower);
 392             }
 393         }
 394
 395         if (message instanceof SendHeartBeat) {
 396             beforeSendHeartbeat();
 397             sendHeartBeat();
 398             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 399         } else if(message instanceof SendInstallSnapshot) {
 400             // received from RaftActor
 401             setSnapshot(((SendInstallSnapshot) message).getSnapshot());
 402             sendInstallSnapshot();
 403         } else if (message instanceof Replicate) {
 404             replicate((Replicate) message);
 405         } else if (message instanceof InstallSnapshotReply) {
 406             handleInstallSnapshotReply((InstallSnapshotReply) message);
 407         } else {
 408             return super.handleMessage(sender, message);
 409         }
 410
 411         return this;
 412     }
 413
 414     private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
 415         LOG.debug("{}: handleInstallSnapshotReply: {}", logName(), reply);
 416
 417         String followerId = reply.getFollowerId();
 418         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 419
 420         if (followerToSnapshot == null) {
 421             LOG.error("{}: FollowerToSnapshot not found for follower {} in InstallSnapshotReply",
 422                     logName(), followerId);
 423             return;
 424         }
 425
 426         FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 427         if(followerLogInformation == null) {
 428             // This can happen during AddServer if it times out.
 429             LOG.error("{}: FollowerLogInformation not found for follower {} in InstallSnapshotReply",
 430                     logName(), followerId);
 431             mapFollowerToSnapshot.remove(followerId);
 432             return;
 433         }
 434
 435         followerLogInformation.markFollowerActive();
 436
 437         if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
 438             boolean wasLastChunk = false;
 439             if (reply.isSuccess()) {
 440                 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
 441                     //this was the last chunk reply
 442                     if(LOG.isDebugEnabled()) {
 443                         LOG.debug("{}: InstallSnapshotReply received, " +
 444                                 "last chunk received, Chunk: {}. Follower: {} Setting nextIndex: {}",
 445                                 logName(), reply.getChunkIndex(), followerId,
 446                             context.getReplicatedLog().getSnapshotIndex() + 1
 447                         );
 448                     }
 449
 450                     long followerMatchIndex = snapshot.get().getLastIncludedIndex();
 451                     followerLogInformation.setMatchIndex(followerMatchIndex);
 452                     followerLogInformation.setNextIndex(followerMatchIndex + 1);
 453                     mapFollowerToSnapshot.remove(followerId);
 454
 455                     LOG.debug("{}: follower: {}, matchIndex set to {}, nextIndex set to {}",
 456                         logName(), followerId, followerLogInformation.getMatchIndex(),
 457                         followerLogInformation.getNextIndex());
 458
 459                     if (mapFollowerToSnapshot.isEmpty()) {
 460                         // once there are no pending followers receiving snapshots
 461                         // we can remove snapshot from the memory
 462                         setSnapshot(null);
 463                     }
 464                     wasLastChunk = true;
 465                     if(context.getPeerInfo(followerId).getVotingState() == VotingState.VOTING_NOT_INITIALIZED){
 466                         UnInitializedFollowerSnapshotReply unInitFollowerSnapshotSuccess =
 467                                              new UnInitializedFollowerSnapshotReply(followerId);
 468                         context.getActor().tell(unInitFollowerSnapshotSuccess, context.getActor());
 469                         LOG.debug("Sent message UnInitializedFollowerSnapshotReply to self");
 470                     }
 471                 } else {
 472                     followerToSnapshot.markSendStatus(true);
 473                 }
 474             } else {
 475                 LOG.info("{}: InstallSnapshotReply received sending snapshot chunk failed, Will retry, Chunk: {}",
 476                         logName(), reply.getChunkIndex());
 477
 478                 followerToSnapshot.markSendStatus(false);
 479             }
 480
 481             if (wasLastChunk && !context.getSnapshotManager().isCapturing()) {
 482                 // Since the follower is now caught up try to purge the log.
 483                 purgeInMemoryLog();
 484             } else if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
 485                 ActorSelection followerActor = context.getPeerActorSelection(followerId);
 486                 if(followerActor != null) {
 487                     sendSnapshotChunk(followerActor, followerId);
 488                 }
 489             }
 490
 491         } else {
 492             LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
 493                     logName(), reply.getChunkIndex(), followerId,
 494                     followerToSnapshot.getChunkIndex());
 495
 496             if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
 497                 // Since the Follower did not find this index to be valid we should reset the follower snapshot
 498                 // so that Installing the snapshot can resume from the beginning
 499                 followerToSnapshot.reset();
 500             }
 501         }
 502     }
 503
 504     private void replicate(Replicate replicate) {
 505         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 506
 507         LOG.debug("{}: Replicate message: identifier: {}, logIndex: {}", logName(),
 508                 replicate.getIdentifier(), logIndex);
 509
 510         // Create a tracker entry we will use this later to notify the
 511         // client actor
 512         trackers.add(
 513             new ClientRequestTrackerImpl(replicate.getClientActor(),
 514                 replicate.getIdentifier(),
 515                 logIndex)
 516         );
 517
 518         boolean applyModificationToState = followerToLog.isEmpty()
 519                 || context.getRaftPolicy().applyModificationToStateBeforeConsensus();
 520
 521         if(applyModificationToState){
 522             context.setCommitIndex(logIndex);
 523             applyLogToStateMachine(logIndex);
 524         }
 525
 526         if (!followerToLog.isEmpty()) {
 527             sendAppendEntries(0, false);
 528         }
 529     }
 530
 531     protected void sendAppendEntries(long timeSinceLastActivityInterval, boolean isHeartbeat) {
 532         // Send an AppendEntries to all followers
 533         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 534             final String followerId = e.getKey();
 535             final FollowerLogInformation followerLogInformation = e.getValue();
 536             // This checks helps not to send a repeat message to the follower
 537             if(!followerLogInformation.isFollowerActive() ||
 538                     followerLogInformation.timeSinceLastActivity() >= timeSinceLastActivityInterval) {
 539                 sendUpdatesToFollower(followerId, followerLogInformation, true, isHeartbeat);
 540             }
 541         }
 542     }
 543
 544     /**
 545      *
 546      * This method checks if any update needs to be sent to the given follower. This includes append log entries,
 547      * sending next snapshot chunk, and initiating a snapshot.
 548      * @return true if any update is sent, false otherwise
 549      */
 550
 551     private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation,
 552                                        boolean sendHeartbeat, boolean isHeartbeat) {
 553
 554         ActorSelection followerActor = context.getPeerActorSelection(followerId);
 555         if (followerActor != null) {
 556             long followerNextIndex = followerLogInformation.getNextIndex();
 557             boolean isFollowerActive = followerLogInformation.isFollowerActive();
 558             boolean sendAppendEntries = false;
 559             List<ReplicatedLogEntry> entries = Collections.emptyList();
 560
 561             if (mapFollowerToSnapshot.get(followerId) != null) {
 562                 // if install snapshot is in process , then sent next chunk if possible
 563                 if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
 564                     sendSnapshotChunk(followerActor, followerId);
 565                 } else if(sendHeartbeat) {
 566                     // we send a heartbeat even if we have not received a reply for the last chunk
 567                     sendAppendEntries = true;
 568                 }
 569             } else {
 570                 long leaderLastIndex = context.getReplicatedLog().lastIndex();
 571                 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 572
 573                 if((!isHeartbeat && LOG.isDebugEnabled()) || LOG.isTraceEnabled()) {
 574                     LOG.debug("{}: Checking sendAppendEntries for follower {}: active: {}, followerNextIndex: {}, leaderLastIndex: {}, leaderSnapShotIndex: {}",
 575                             logName(), followerId, isFollowerActive, followerNextIndex, leaderLastIndex, leaderSnapShotIndex);
 576                 }
 577
 578                 if (isFollowerActive && context.getReplicatedLog().isPresent(followerNextIndex)) {
 579
 580                     LOG.debug("{}: sendAppendEntries: {} is present for follower {}", logName(),
 581                             followerNextIndex, followerId);
 582
 583                     if(followerLogInformation.okToReplicate()) {
 584                         // Try to send all the entries in the journal but not exceeding the max data size
 585                         // for a single AppendEntries message.
 586                         int maxEntries = (int) context.getReplicatedLog().size();
 587                         entries = context.getReplicatedLog().getFrom(followerNextIndex, maxEntries,
 588                                 context.getConfigParams().getSnapshotChunkSize());
 589                         sendAppendEntries = true;
 590                     }
 591                 } else if (isFollowerActive && followerNextIndex >= 0 &&
 592                     leaderLastIndex > followerNextIndex && !context.getSnapshotManager().isCapturing()) {
 593                     // if the followers next index is not present in the leaders log, and
 594                     // if the follower is just not starting and if leader's index is more than followers index
 595                     // then snapshot should be sent
 596
 597                     if (LOG.isDebugEnabled()) {
 598                         LOG.debug(String.format("%s: InitiateInstallSnapshot to follower: %s," +
 599                                     "follower-nextIndex: %d, leader-snapshot-index: %d,  " +
 600                                     "leader-last-index: %d", logName(), followerId,
 601                                     followerNextIndex, leaderSnapShotIndex, leaderLastIndex));
 602                     }
 603
 604                     // Send heartbeat to follower whenever install snapshot is initiated.
 605                     sendAppendEntries = true;
 606                     if (canInstallSnapshot(followerNextIndex)) {
 607                         initiateCaptureSnapshot(followerId);
 608                     }
 609
 610                 } else if(sendHeartbeat) {
 611                     // we send an AppendEntries, even if the follower is inactive
 612                     // in-order to update the followers timestamp, in case it becomes active again
 613                     sendAppendEntries = true;
 614                 }
 615
 616             }
 617
 618             if(sendAppendEntries) {
 619                 sendAppendEntriesToFollower(followerActor, followerNextIndex,
 620                         entries, followerId);
 621             }
 622         }
 623     }
 624
 625     private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
 626         List<ReplicatedLogEntry> entries, String followerId) {
 627         AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
 628             prevLogIndex(followerNextIndex),
 629             prevLogTerm(followerNextIndex), entries,
 630             context.getCommitIndex(), super.getReplicatedToAllIndex(), context.getPayloadVersion());
 631
 632         if(!entries.isEmpty() || LOG.isTraceEnabled()) {
 633             LOG.debug("{}: Sending AppendEntries to follower {}: {}", logName(), followerId,
 634                     appendEntries);
 635         }
 636
 637         followerActor.tell(appendEntries, actor());
 638     }
 639
 640     /**
 641      * Install Snapshot works as follows
 642      * 1. Leader initiates the capture snapshot by sending a CaptureSnapshot message to actor
 643      * 2. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
 644      * and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
 645      * 3. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
 646      * 4. On complete, Follower sends back a InstallSnapshotReply.
 647      * 5. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
 648      * and replenishes the memory by deleting the snapshot in Replicated log.
 649      * 6. If another follower requires a snapshot and a snapshot has been collected (via CaptureSnapshotReply)
 650      * then send the existing snapshot in chunks to the follower.
 651      * @param followerId
 652      */
 653     public boolean initiateCaptureSnapshot(String followerId) {
 654         if (snapshot.isPresent()) {
 655             // if a snapshot is present in the memory, most likely another install is in progress
 656             // no need to capture snapshot.
 657             // This could happen if another follower needs an install when one is going on.
 658             final ActorSelection followerActor = context.getPeerActorSelection(followerId);
 659             sendSnapshotChunk(followerActor, followerId);
 660             return true;
 661         } else {
 662             return context.getSnapshotManager().captureToInstall(context.getReplicatedLog().last(),
 663                     this.getReplicatedToAllIndex(), followerId);
 664         }
 665     }
 666
 667     private boolean canInstallSnapshot(long nextIndex){
 668         // If the follower's nextIndex is -1 then we might as well send it a snapshot
 669         // Otherwise send it a snapshot only if the nextIndex is not present in the log but is present
 670         // in the snapshot
 671         return (nextIndex == -1 ||
 672                 (!context.getReplicatedLog().isPresent(nextIndex)
 673                         && context.getReplicatedLog().isInSnapshot(nextIndex)));
 674
 675     }
 676
 677
 678     private void sendInstallSnapshot() {
 679         LOG.debug("{}: sendInstallSnapshot", logName());
 680         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 681             String followerId = e.getKey();
 682             ActorSelection followerActor = context.getPeerActorSelection(followerId);
 683             FollowerLogInformation followerLogInfo = e.getValue();
 684
 685             if (followerActor != null) {
 686                 long nextIndex = followerLogInfo.getNextIndex();
 687                 if (context.getPeerInfo(followerId).getVotingState() == VotingState.VOTING_NOT_INITIALIZED ||
 688                         canInstallSnapshot(nextIndex)) {
 689                     sendSnapshotChunk(followerActor, followerId);
 690                 }
 691             }
 692         }
 693     }
 694
 695     /**
 696      *  Sends a snapshot chunk to a given follower
 697      *  InstallSnapshot should qualify as a heartbeat too.
 698      */
 699     private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
 700         try {
 701             if (snapshot.isPresent()) {
 702                 byte[] nextSnapshotChunk = getNextSnapshotChunk(followerId, snapshot.get().getSnapshotBytes());
 703
 704                 // Note: the previous call to getNextSnapshotChunk has the side-effect of adding
 705                 // followerId to the followerToSnapshot map.
 706                 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 707
 708                 followerActor.tell(
 709                     new InstallSnapshot(currentTerm(), context.getId(),
 710                         snapshot.get().getLastIncludedIndex(),
 711                         snapshot.get().getLastIncludedTerm(),
 712                         nextSnapshotChunk,
 713                         followerToSnapshot.incrementChunkIndex(),
 714                         followerToSnapshot.getTotalChunks(),
 715                         Optional.of(followerToSnapshot.getLastChunkHashCode())
 716                     ).toSerializable(followerToLog.get(followerId).getRaftVersion()),
 717                     actor()
 718                 );
 719
 720                 if(LOG.isDebugEnabled()) {
 721                     LOG.debug("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
 722                             logName(), followerActor.path(), followerToSnapshot.getChunkIndex(),
 723                             followerToSnapshot.getTotalChunks());
 724                 }
 725             }
 726         } catch (IOException e) {
 727             LOG.error("{}: InstallSnapshot failed for Leader.", logName(), e);
 728         }
 729     }
 730
 731     /**
 732      * Acccepts snaphot as ByteString, enters into map for future chunks
 733      * creates and return a ByteString chunk
 734      */
 735     private byte[] getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
 736         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 737         if (followerToSnapshot == null) {
 738             followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
 739             mapFollowerToSnapshot.put(followerId, followerToSnapshot);
 740         }
 741         byte[] nextChunk = followerToSnapshot.getNextChunk();
 742
 743         LOG.debug("{}: next snapshot chunk size for follower {}: {}", logName(), followerId, nextChunk.length);
 744
 745         return nextChunk;
 746     }
 747
 748     private void sendHeartBeat() {
 749         if (!followerToLog.isEmpty()) {
 750             LOG.trace("{}: Sending heartbeat", logName());
 751             sendAppendEntries(context.getConfigParams().getHeartBeatInterval().toMillis(), true);
 752         }
 753     }
 754
 755     private void stopHeartBeat() {
 756         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 757             heartbeatSchedule.cancel();
 758         }
 759     }
 760
 761     private void scheduleHeartBeat(FiniteDuration interval) {
 762         if (followerToLog.isEmpty()) {
 763             // Optimization - do not bother scheduling a heartbeat as there are
 764             // no followers
 765             return;
 766         }
 767
 768         stopHeartBeat();
 769
 770         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 771         // message is sent to itself.
 772         // Scheduling the heartbeat only once here because heartbeats do not
 773         // need to be sent if there are other messages being sent to the remote
 774         // actor.
 775         heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
 776             interval, context.getActor(), SendHeartBeat.INSTANCE,
 777             context.getActorSystem().dispatcher(), context.getActor());
 778     }
 779
 780     @Override
 781     public void close() {
 782         stopHeartBeat();
 783     }
 784
 785     @Override
 786     public final String getLeaderId() {
 787         return context.getId();
 788     }
 789
 790     @Override
 791     public final short getLeaderPayloadVersion() {
 792         return context.getPayloadVersion();
 793     }
 794
 795     protected boolean isLeaderIsolated() {
 796         int minPresent = getMinIsolatedLeaderPeerCount();
 797         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 798             final PeerInfo peerInfo = context.getPeerInfo(followerLogInformation.getId());
 799             if(peerInfo != null && peerInfo.isVoting() && followerLogInformation.isFollowerActive()) {
 800                 --minPresent;
 801                 if (minPresent == 0) {
 802                     return false;
 803                 }
 804             }
 805         }
 806         return (minPresent != 0);
 807     }
 808
 809     /**
 810      * Encapsulates the snapshot bytestring and handles the logic of sending
 811      * snapshot chunks
 812      */
 813     protected class FollowerToSnapshot {
 814         private final ByteString snapshotBytes;
 815         private int offset = 0;
 816         // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
 817         private int replyReceivedForOffset;
 818         // if replyStatus is false, the previous chunk is attempted
 819         private boolean replyStatus = false;
 820         private int chunkIndex;
 821         private final int totalChunks;
 822         private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 823         private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 824
 825         public FollowerToSnapshot(ByteString snapshotBytes) {
 826             this.snapshotBytes = snapshotBytes;
 827             int size = snapshotBytes.size();
 828             totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
 829                 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
 830             if(LOG.isDebugEnabled()) {
 831                 LOG.debug("{}: Snapshot {} bytes, total chunks to send:{}",
 832                         logName(), size, totalChunks);
 833             }
 834             replyReceivedForOffset = -1;
 835             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 836         }
 837
 838         public ByteString getSnapshotBytes() {
 839             return snapshotBytes;
 840         }
 841
 842         public int incrementOffset() {
 843             if(replyStatus) {
 844                 // if prev chunk failed, we would want to sent the same chunk again
 845                 offset = offset + context.getConfigParams().getSnapshotChunkSize();
 846             }
 847             return offset;
 848         }
 849
 850         public int incrementChunkIndex() {
 851             if (replyStatus) {
 852                 // if prev chunk failed, we would want to sent the same chunk again
 853                 chunkIndex =  chunkIndex + 1;
 854             }
 855             return chunkIndex;
 856         }
 857
 858         public int getChunkIndex() {
 859             return chunkIndex;
 860         }
 861
 862         public int getTotalChunks() {
 863             return totalChunks;
 864         }
 865
 866         public boolean canSendNextChunk() {
 867             // we only send a false if a chunk is sent but we have not received a reply yet
 868             return replyReceivedForOffset == offset;
 869         }
 870
 871         public boolean isLastChunk(int chunkIndex) {
 872             return totalChunks == chunkIndex;
 873         }
 874
 875         public void markSendStatus(boolean success) {
 876             if (success) {
 877                 // if the chunk sent was successful
 878                 replyReceivedForOffset = offset;
 879                 replyStatus = true;
 880                 lastChunkHashCode = nextChunkHashCode;
 881             } else {
 882                 // if the chunk sent was failure
 883                 replyReceivedForOffset = offset;
 884                 replyStatus = false;
 885             }
 886         }
 887
 888         public byte[] getNextChunk() {
 889             int snapshotLength = getSnapshotBytes().size();
 890             int start = incrementOffset();
 891             int size = context.getConfigParams().getSnapshotChunkSize();
 892             if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
 893                 size = snapshotLength;
 894             } else if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
 895                 size = snapshotLength - start;
 896             }
 897
 898             byte[] nextChunk = new byte[size];
 899             getSnapshotBytes().copyTo(nextChunk, start, 0, size);
 900             nextChunkHashCode = Arrays.hashCode(nextChunk);
 901
 902             LOG.debug("{}: Next chunk: total length={}, offset={}, size={}, hashCode={}", logName(),
 903                     snapshotLength, start, size, nextChunkHashCode);
 904             return nextChunk;
 905         }
 906
 907         /**
 908          * reset should be called when the Follower needs to be sent the snapshot from the beginning
 909          */
 910         public void reset(){
 911             offset = 0;
 912             replyStatus = false;
 913             replyReceivedForOffset = offset;
 914             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 915             lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 916         }
 917
 918         public int getLastChunkHashCode() {
 919             return lastChunkHashCode;
 920         }
 921     }
 922
 923     // called from example-actor for printing the follower-states
 924     public String printFollowerStates() {
 925         final StringBuilder sb = new StringBuilder();
 926
 927         sb.append('[');
 928         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 929             sb.append('{');
 930             sb.append(followerLogInformation.getId());
 931             sb.append(" state:");
 932             sb.append(followerLogInformation.isFollowerActive());
 933             sb.append("},");
 934         }
 935         sb.append(']');
 936
 937         return sb.toString();
 938     }
 939
 940     @VisibleForTesting
 941     public FollowerLogInformation getFollower(String followerId) {
 942         return followerToLog.get(followerId);
 943     }
 944
 945     @VisibleForTesting
 946     protected void setFollowerSnapshot(String followerId, FollowerToSnapshot snapshot) {
 947         mapFollowerToSnapshot.put(followerId, snapshot);
 948     }
 949
 950     @VisibleForTesting
 951     public int followerSnapshotSize() {
 952         return mapFollowerToSnapshot.size();
 953     }
 954
 955     @VisibleForTesting
 956     public int followerLogSize() {
 957         return followerToLog.size();
 958     }
 959
 960     private static class SnapshotHolder {
 961         private final long lastIncludedTerm;
 962         private final long lastIncludedIndex;
 963         private final ByteString snapshotBytes;
 964
 965         SnapshotHolder(Snapshot snapshot) {
 966             this.lastIncludedTerm = snapshot.getLastAppliedTerm();
 967             this.lastIncludedIndex = snapshot.getLastAppliedIndex();
 968             this.snapshotBytes = ByteString.copyFrom(snapshot.getState());
 969         }
 970
 971         long getLastIncludedTerm() {
 972             return lastIncludedTerm;
 973         }
 974
 975         long getLastIncludedIndex() {
 976             return lastIncludedIndex;
 977         }
 978
 979         ByteString getSnapshotBytes() {
 980             return snapshotBytes;
 981         }
 982     }
 983 }