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