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.common.collect.ImmutableMap;
  18 import com.google.common.collect.ImmutableMap.Builder;
  19 import com.google.protobuf.ByteString;
  20 import java.io.IOException;
  21 import java.util.Collection;
  22 import java.util.Collections;
  23 import java.util.HashMap;
  24 import java.util.Iterator;
  25 import java.util.LinkedList;
  26 import java.util.List;
  27 import java.util.Map;
  28 import java.util.Map.Entry;
  29 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  30 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  31 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  32 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  33 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  34 import org.opendaylight.controller.cluster.raft.RaftState;
  35 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  36 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  37 import org.opendaylight.controller.cluster.raft.base.messages.InitiateInstallSnapshot;
  38 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  39 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  40 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  41 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  42 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  43 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  44 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  45 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  46 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  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;
  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     protected final int minReplicationCount;
  90
  91     protected final int minIsolatedLeaderPeerCount;
  92
  93     private Optional<ByteString> snapshot;
  94
  95     private long replicatedToAllIndex = -1;
  96
  97     public AbstractLeader(RaftActorContext context) {
  98         super(context);
  99
 100         final Builder<String, FollowerLogInformation> ftlBuilder = ImmutableMap.builder();
 101         for (String followerId : context.getPeerAddresses().keySet()) {
 102             FollowerLogInformation followerLogInformation =
 103                 new FollowerLogInformationImpl(followerId,
 104                     context.getCommitIndex(), -1,
 105                     context.getConfigParams().getElectionTimeOutInterval());
 106
 107             ftlBuilder.put(followerId, followerLogInformation);
 108         }
 109         followerToLog = ftlBuilder.build();
 110
 111         leaderId = context.getId();
 112
 113         LOG.debug("{}: Election: Leader has following peers: {}", context.getId(), getFollowerIds());
 114
 115         minReplicationCount = getMajorityVoteCount(getFollowerIds().size());
 116
 117         // the isolated Leader peer count will be 1 less than the majority vote count.
 118         // this is because the vote count has the self vote counted in it
 119         // for e.g
 120         // 0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
 121         // 2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
 122         // 4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
 123         minIsolatedLeaderPeerCount = minReplicationCount > 0 ? (minReplicationCount - 1) : 0;
 124
 125         snapshot = Optional.absent();
 126
 127         // Immediately schedule a heartbeat
 128         // Upon election: send initial empty AppendEntries RPCs
 129         // (heartbeat) to each server; repeat during idle periods to
 130         // prevent election timeouts (§5.2)
 131         sendAppendEntries(0);
 132     }
 133
 134     /**
 135      * Return an immutable collection of follower identifiers.
 136      *
 137      * @return Collection of follower IDs
 138      */
 139     protected final Collection<String> getFollowerIds() {
 140         return followerToLog.keySet();
 141     }
 142
 143     private Optional<ByteString> getSnapshot() {
 144         return snapshot;
 145     }
 146
 147     @VisibleForTesting
 148     void setSnapshot(Optional<ByteString> snapshot) {
 149         this.snapshot = snapshot;
 150     }
 151
 152     @Override
 153     protected RaftActorBehavior handleAppendEntries(ActorRef sender,
 154         AppendEntries appendEntries) {
 155
 156         if(LOG.isDebugEnabled()) {
 157             LOG.debug("{}: handleAppendEntries: {}", context.getId(), appendEntries);
 158         }
 159
 160         return this;
 161     }
 162
 163     @Override
 164     protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 165         AppendEntriesReply appendEntriesReply) {
 166
 167         if(! appendEntriesReply.isSuccess()) {
 168             if(LOG.isDebugEnabled()) {
 169                 LOG.debug("{}: handleAppendEntriesReply: {}", context.getId(), appendEntriesReply);
 170             }
 171         }
 172
 173         // Update the FollowerLogInformation
 174         String followerId = appendEntriesReply.getFollowerId();
 175         FollowerLogInformation followerLogInformation =
 176             followerToLog.get(followerId);
 177
 178         if(followerLogInformation == null){
 179             LOG.error("{}: handleAppendEntriesReply - unknown follower {}", context.getId(), followerId);
 180             return this;
 181         }
 182
 183         followerLogInformation.markFollowerActive();
 184
 185         if (appendEntriesReply.isSuccess()) {
 186             followerLogInformation
 187                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 188             followerLogInformation
 189                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 190         } else {
 191
 192             // TODO: When we find that the follower is out of sync with the
 193             // Leader we simply decrement that followers next index by 1.
 194             // Would it be possible to do better than this? The RAFT spec
 195             // does not explicitly deal with it but may be something for us to
 196             // think about
 197
 198             followerLogInformation.decrNextIndex();
 199         }
 200
 201         // Now figure out if this reply warrants a change in the commitIndex
 202         // If there exists an N such that N > commitIndex, a majority
 203         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 204         // set commitIndex = N (§5.3, §5.4).
 205         for (long N = context.getCommitIndex() + 1; ; N++) {
 206             int replicatedCount = 1;
 207
 208             for (FollowerLogInformation info : followerToLog.values()) {
 209                 if (info.getMatchIndex() >= N) {
 210                     replicatedCount++;
 211                 }
 212             }
 213
 214             if (replicatedCount >= minReplicationCount) {
 215                 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(N);
 216                 if (replicatedLogEntry != null &&
 217                     replicatedLogEntry.getTerm() == currentTerm()) {
 218                     context.setCommitIndex(N);
 219                 }
 220             } else {
 221                 break;
 222             }
 223         }
 224
 225         // Apply the change to the state machine
 226         if (context.getCommitIndex() > context.getLastApplied()) {
 227             applyLogToStateMachine(context.getCommitIndex());
 228         }
 229
 230         if (!context.isSnapshotCaptureInitiated()) {
 231             purgeInMemoryLog();
 232         }
 233
 234         //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
 235         sendUpdatesToFollower(followerId, followerLogInformation, false);
 236         return this;
 237     }
 238
 239     private void purgeInMemoryLog() {
 240         //find the lowest index across followers which has been replicated to all. -1 if there are no followers.
 241         // we would delete the in-mem log from that index on, in-order to minimize mem usage
 242         // we would also share this info thru AE with the followers so that they can delete their log entries as well.
 243         long minReplicatedToAllIndex = followerToLog.isEmpty() ? -1 : Long.MAX_VALUE;
 244         for (FollowerLogInformation info : followerToLog.values()) {
 245             minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
 246         }
 247
 248         replicatedToAllIndex = fakeSnapshot(minReplicatedToAllIndex, replicatedToAllIndex);
 249     }
 250
 251     @Override
 252     protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
 253         final Iterator<ClientRequestTracker> it = trackerList.iterator();
 254         while (it.hasNext()) {
 255             final ClientRequestTracker t = it.next();
 256             if (t.getIndex() == logIndex) {
 257                 it.remove();
 258                 return t;
 259             }
 260         }
 261
 262         return null;
 263     }
 264
 265     @Override
 266     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 267         for (ClientRequestTracker tracker : trackerList) {
 268             if (tracker.getIndex() == logIndex) {
 269                 return tracker;
 270             }
 271         }
 272         return null;
 273     }
 274
 275     @Override
 276     protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 277         RequestVoteReply requestVoteReply) {
 278         return this;
 279     }
 280
 281     @Override
 282     public RaftState state() {
 283         return RaftState.Leader;
 284     }
 285
 286     @Override
 287     public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 288         Preconditions.checkNotNull(sender, "sender should not be null");
 289
 290         Object message = fromSerializableMessage(originalMessage);
 291
 292         if (message instanceof RaftRPC) {
 293             RaftRPC rpc = (RaftRPC) message;
 294             // If RPC request or response contains term T > currentTerm:
 295             // set currentTerm = T, convert to follower (§5.1)
 296             // This applies to all RPC messages and responses
 297             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 298                 LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {}", context.getId(),
 299                         rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 300
 301                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 302
 303                 return switchBehavior(new Follower(context));
 304             }
 305         }
 306
 307         try {
 308             if (message instanceof SendHeartBeat) {
 309                 sendHeartBeat();
 310                 return this;
 311
 312             } else if(message instanceof InitiateInstallSnapshot) {
 313                 installSnapshotIfNeeded();
 314
 315             } else if(message instanceof SendInstallSnapshot) {
 316                 // received from RaftActor
 317                 setSnapshot(Optional.of(((SendInstallSnapshot) message).getSnapshot()));
 318                 sendInstallSnapshot();
 319
 320             } else if (message instanceof Replicate) {
 321                 replicate((Replicate) message);
 322
 323             } else if (message instanceof InstallSnapshotReply){
 324                 handleInstallSnapshotReply((InstallSnapshotReply) message);
 325
 326             }
 327         } finally {
 328             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 329         }
 330
 331         return super.handleMessage(sender, message);
 332     }
 333
 334     private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
 335         String followerId = reply.getFollowerId();
 336         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 337
 338         if (followerToSnapshot == null) {
 339             LOG.error("{}: FollowerId {} in InstallSnapshotReply not known to Leader",
 340                     context.getId(), followerId);
 341             return;
 342         }
 343
 344         FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 345         followerLogInformation.markFollowerActive();
 346
 347         if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
 348             boolean wasLastChunk = false;
 349             if (reply.isSuccess()) {
 350                 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
 351                     //this was the last chunk reply
 352                     if(LOG.isDebugEnabled()) {
 353                         LOG.debug("{}: InstallSnapshotReply received, " +
 354                                 "last chunk received, Chunk:{}. Follower:{} Setting nextIndex:{}",
 355                                 context.getId(), reply.getChunkIndex(), followerId,
 356                             context.getReplicatedLog().getSnapshotIndex() + 1
 357                         );
 358                     }
 359
 360                     followerLogInformation.setMatchIndex(
 361                         context.getReplicatedLog().getSnapshotIndex());
 362                     followerLogInformation.setNextIndex(
 363                         context.getReplicatedLog().getSnapshotIndex() + 1);
 364                     mapFollowerToSnapshot.remove(followerId);
 365
 366                     if(LOG.isDebugEnabled()) {
 367                         LOG.debug("{}: followerToLog.get(followerId).getNextIndex()=" +
 368                                 context.getId(), followerToLog.get(followerId).getNextIndex());
 369                     }
 370
 371                     if (mapFollowerToSnapshot.isEmpty()) {
 372                         // once there are no pending followers receiving snapshots
 373                         // we can remove snapshot from the memory
 374                         setSnapshot(Optional.<ByteString>absent());
 375                     }
 376                     wasLastChunk = true;
 377
 378                 } else {
 379                     followerToSnapshot.markSendStatus(true);
 380                 }
 381             } else {
 382                 LOG.info("{}: InstallSnapshotReply received sending snapshot chunk failed, Will retry, Chunk: {}",
 383                         context.getId(), reply.getChunkIndex());
 384
 385                 followerToSnapshot.markSendStatus(false);
 386             }
 387
 388             if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
 389                 ActorSelection followerActor = context.getPeerActorSelection(followerId);
 390                 if(followerActor != null) {
 391                     sendSnapshotChunk(followerActor, followerId);
 392                 }
 393             }
 394
 395         } else {
 396             LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
 397                     context.getId(), reply.getChunkIndex(), followerId,
 398                     followerToSnapshot.getChunkIndex());
 399
 400             if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
 401                 // Since the Follower did not find this index to be valid we should reset the follower snapshot
 402                 // so that Installing the snapshot can resume from the beginning
 403                 followerToSnapshot.reset();
 404             }
 405         }
 406     }
 407
 408     private void replicate(Replicate replicate) {
 409         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 410
 411         if(LOG.isDebugEnabled()) {
 412             LOG.debug("{}: Replicate message {}", context.getId(), logIndex);
 413         }
 414
 415         // Create a tracker entry we will use this later to notify the
 416         // client actor
 417         trackerList.add(
 418             new ClientRequestTrackerImpl(replicate.getClientActor(),
 419                 replicate.getIdentifier(),
 420                 logIndex)
 421         );
 422
 423         if (followerToLog.isEmpty()) {
 424             context.setCommitIndex(logIndex);
 425             applyLogToStateMachine(logIndex);
 426         } else {
 427             sendAppendEntries(0);
 428         }
 429     }
 430
 431     private void sendAppendEntries(long timeSinceLastActivityInterval) {
 432         // Send an AppendEntries to all followers
 433         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 434             final String followerId = e.getKey();
 435             final FollowerLogInformation followerLogInformation = e.getValue();
 436             // This checks helps not to send a repeat message to the follower
 437             if(!followerLogInformation.isFollowerActive() ||
 438                     followerLogInformation.timeSinceLastActivity() >= timeSinceLastActivityInterval) {
 439                 sendUpdatesToFollower(followerId, followerLogInformation, true);
 440             }
 441         }
 442     }
 443
 444     /**
 445      *
 446      * This method checks if any update needs to be sent to the given follower. This includes append log entries,
 447      * sending next snapshot chunk, and initiating a snapshot.
 448      * @return true if any update is sent, false otherwise
 449      */
 450
 451     private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation,
 452                                           boolean sendHeartbeat) {
 453
 454         ActorSelection followerActor = context.getPeerActorSelection(followerId);
 455         if (followerActor != null) {
 456             long followerNextIndex = followerLogInformation.getNextIndex();
 457             boolean isFollowerActive = followerLogInformation.isFollowerActive();
 458
 459             if (mapFollowerToSnapshot.get(followerId) != null) {
 460                 // if install snapshot is in process , then sent next chunk if possible
 461                 if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
 462                     sendSnapshotChunk(followerActor, followerId);
 463                 } else if(sendHeartbeat) {
 464                     // we send a heartbeat even if we have not received a reply for the last chunk
 465                     sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
 466                         Collections.<ReplicatedLogEntry>emptyList(), followerId);
 467                 }
 468             } else {
 469                 long leaderLastIndex = context.getReplicatedLog().lastIndex();
 470                 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 471                 if (isFollowerActive &&
 472                     context.getReplicatedLog().isPresent(followerNextIndex)) {
 473                     // FIXME : Sending one entry at a time
 474                     final List<ReplicatedLogEntry> entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
 475
 476                     sendAppendEntriesToFollower(followerActor, followerNextIndex, entries, followerId);
 477
 478                 } else if (isFollowerActive && followerNextIndex >= 0 &&
 479                     leaderLastIndex >= followerNextIndex) {
 480                     // if the followers next index is not present in the leaders log, and
 481                     // if the follower is just not starting and if leader's index is more than followers index
 482                     // then snapshot should be sent
 483
 484                     if (LOG.isDebugEnabled()) {
 485                         LOG.debug("InitiateInstallSnapshot to follower:{}," +
 486                                 "follower-nextIndex:{}, leader-snapshot-index:{},  " +
 487                                 "leader-last-index:{}", followerId,
 488                             followerNextIndex, leaderSnapShotIndex, leaderLastIndex
 489                         );
 490                     }
 491                     actor().tell(new InitiateInstallSnapshot(), actor());
 492
 493                     // Send heartbeat to follower whenever install snapshot is initiated.
 494                     sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
 495                             Collections.<ReplicatedLogEntry>emptyList(), followerId);
 496
 497                 } else if(sendHeartbeat) {
 498                     //we send an AppendEntries, even if the follower is inactive
 499                     // in-order to update the followers timestamp, in case it becomes active again
 500                     sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
 501                         Collections.<ReplicatedLogEntry>emptyList(), followerId);
 502                 }
 503
 504             }
 505         }
 506     }
 507
 508     private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
 509         List<ReplicatedLogEntry> entries, String followerId) {
 510         AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
 511             prevLogIndex(followerNextIndex),
 512             prevLogTerm(followerNextIndex), entries,
 513             context.getCommitIndex(), replicatedToAllIndex);
 514
 515         if(!entries.isEmpty()) {
 516             LOG.debug("{}: Sending AppendEntries to follower {}: {}", context.getId(), followerId,
 517                     appendEntries);
 518         }
 519
 520         followerActor.tell(appendEntries.toSerializable(), actor());
 521     }
 522
 523     /**
 524      * An installSnapshot is scheduled at a interval that is a multiple of
 525      * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
 526      * snapshots at every heartbeat.
 527      *
 528      * Install Snapshot works as follows
 529      * 1. Leader sends a InitiateInstallSnapshot message to self
 530      * 2. Leader then initiates the capture snapshot by sending a CaptureSnapshot message to actor
 531      * 3. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
 532      * and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
 533      * 4. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
 534      * 5. On complete, Follower sends back a InstallSnapshotReply.
 535      * 6. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
 536      * and replenishes the memory by deleting the snapshot in Replicated log.
 537      *
 538      */
 539     private void installSnapshotIfNeeded() {
 540         if(LOG.isDebugEnabled()) {
 541             LOG.debug("{}: installSnapshotIfNeeded, followers {}", context.getId(), followerToLog.keySet());
 542         }
 543
 544         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 545             final ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
 546
 547             if (followerActor != null) {
 548                 long nextIndex = e.getValue().getNextIndex();
 549
 550                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 551                         context.getReplicatedLog().isInSnapshot(nextIndex)) {
 552                     LOG.info("{}: {} follower needs a snapshot install", context.getId(), e.getKey());
 553                     if (snapshot.isPresent()) {
 554                         // if a snapshot is present in the memory, most likely another install is in progress
 555                         // no need to capture snapshot
 556                         sendSnapshotChunk(followerActor, e.getKey());
 557
 558                     } else if (!context.isSnapshotCaptureInitiated()) {
 559                         initiateCaptureSnapshot();
 560                         //we just need 1 follower who would need snapshot to be installed.
 561                         // when we have the snapshot captured, we would again check (in SendInstallSnapshot)
 562                         // who needs an install and send to all who need
 563                         break;
 564                     }
 565
 566                 }
 567             }
 568         }
 569     }
 570
 571     // on every install snapshot, we try to capture the snapshot.
 572     // Once a capture is going on, another one issued will get ignored by RaftActor.
 573     private void initiateCaptureSnapshot() {
 574         LOG.info("{}: Initiating Snapshot Capture to Install Snapshot, Leader:{}", context.getId(), getLeaderId());
 575         ReplicatedLogEntry lastAppliedEntry = context.getReplicatedLog().get(context.getLastApplied());
 576         long lastAppliedIndex = -1;
 577         long lastAppliedTerm = -1;
 578
 579         if (lastAppliedEntry != null) {
 580             lastAppliedIndex = lastAppliedEntry.getIndex();
 581             lastAppliedTerm = lastAppliedEntry.getTerm();
 582         } else if (context.getReplicatedLog().getSnapshotIndex() > -1)  {
 583             lastAppliedIndex = context.getReplicatedLog().getSnapshotIndex();
 584             lastAppliedTerm = context.getReplicatedLog().getSnapshotTerm();
 585         }
 586
 587         boolean isInstallSnapshotInitiated = true;
 588         actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(),
 589                 lastAppliedIndex, lastAppliedTerm, isInstallSnapshotInitiated),
 590             actor());
 591         context.setSnapshotCaptureInitiated(true);
 592     }
 593
 594
 595     private void sendInstallSnapshot() {
 596         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 597             ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
 598
 599             if (followerActor != null) {
 600                 long nextIndex = e.getValue().getNextIndex();
 601
 602                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 603                     context.getReplicatedLog().isInSnapshot(nextIndex)) {
 604                     sendSnapshotChunk(followerActor, e.getKey());
 605                 }
 606             }
 607         }
 608     }
 609
 610     /**
 611      *  Sends a snapshot chunk to a given follower
 612      *  InstallSnapshot should qualify as a heartbeat too.
 613      */
 614     private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
 615         try {
 616             if (snapshot.isPresent()) {
 617                 ByteString nextSnapshotChunk = getNextSnapshotChunk(followerId,snapshot.get());
 618
 619                 // Note: the previous call to getNextSnapshotChunk has the side-effect of adding
 620                 // followerId to the followerToSnapshot map.
 621                 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 622
 623                 followerActor.tell(
 624                     new InstallSnapshot(currentTerm(), context.getId(),
 625                         context.getReplicatedLog().getSnapshotIndex(),
 626                         context.getReplicatedLog().getSnapshotTerm(),
 627                         nextSnapshotChunk,
 628                             followerToSnapshot.incrementChunkIndex(),
 629                             followerToSnapshot.getTotalChunks(),
 630                         Optional.of(followerToSnapshot.getLastChunkHashCode())
 631                     ).toSerializable(),
 632                     actor()
 633                 );
 634                 LOG.info("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
 635                         context.getId(), followerActor.path(),
 636                         followerToSnapshot.getChunkIndex(),
 637                         followerToSnapshot.getTotalChunks());
 638             }
 639         } catch (IOException e) {
 640             LOG.error("{}: InstallSnapshot failed for Leader.", context.getId(), e);
 641         }
 642     }
 643
 644     /**
 645      * Acccepts snaphot as ByteString, enters into map for future chunks
 646      * creates and return a ByteString chunk
 647      */
 648     private ByteString getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
 649         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 650         if (followerToSnapshot == null) {
 651             followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
 652             mapFollowerToSnapshot.put(followerId, followerToSnapshot);
 653         }
 654         ByteString nextChunk = followerToSnapshot.getNextChunk();
 655         if (LOG.isDebugEnabled()) {
 656             LOG.debug("{}: Leader's snapshot nextChunk size:{}", context.getId(), nextChunk.size());
 657         }
 658         return nextChunk;
 659     }
 660
 661     private void sendHeartBeat() {
 662         if (!followerToLog.isEmpty()) {
 663             sendAppendEntries(context.getConfigParams().getHeartBeatInterval().toMillis());
 664         }
 665     }
 666
 667     private void stopHeartBeat() {
 668         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 669             heartbeatSchedule.cancel();
 670         }
 671     }
 672
 673     private void scheduleHeartBeat(FiniteDuration interval) {
 674         if (followerToLog.isEmpty()) {
 675             // Optimization - do not bother scheduling a heartbeat as there are
 676             // no followers
 677             return;
 678         }
 679
 680         stopHeartBeat();
 681
 682         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 683         // message is sent to itself.
 684         // Scheduling the heartbeat only once here because heartbeats do not
 685         // need to be sent if there are other messages being sent to the remote
 686         // actor.
 687         heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
 688             interval, context.getActor(), new SendHeartBeat(),
 689             context.getActorSystem().dispatcher(), context.getActor());
 690     }
 691
 692     @Override
 693     public void close() throws Exception {
 694         stopHeartBeat();
 695     }
 696
 697     @Override
 698     public String getLeaderId() {
 699         return context.getId();
 700     }
 701
 702     protected boolean isLeaderIsolated() {
 703         int minPresent = minIsolatedLeaderPeerCount;
 704         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 705             if (followerLogInformation.isFollowerActive()) {
 706                 --minPresent;
 707                 if (minPresent == 0) {
 708                     break;
 709                 }
 710             }
 711         }
 712         return (minPresent != 0);
 713     }
 714
 715     /**
 716      * Encapsulates the snapshot bytestring and handles the logic of sending
 717      * snapshot chunks
 718      */
 719     protected class FollowerToSnapshot {
 720         private final ByteString snapshotBytes;
 721         private int offset = 0;
 722         // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
 723         private int replyReceivedForOffset;
 724         // if replyStatus is false, the previous chunk is attempted
 725         private boolean replyStatus = false;
 726         private int chunkIndex;
 727         private final int totalChunks;
 728         private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 729         private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 730
 731         public FollowerToSnapshot(ByteString snapshotBytes) {
 732             this.snapshotBytes = snapshotBytes;
 733             int size = snapshotBytes.size();
 734             totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
 735                 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
 736             if(LOG.isDebugEnabled()) {
 737                 LOG.debug("{}: Snapshot {} bytes, total chunks to send:{}",
 738                         context.getId(), size, totalChunks);
 739             }
 740             replyReceivedForOffset = -1;
 741             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 742         }
 743
 744         public ByteString getSnapshotBytes() {
 745             return snapshotBytes;
 746         }
 747
 748         public int incrementOffset() {
 749             if(replyStatus) {
 750                 // if prev chunk failed, we would want to sent the same chunk again
 751                 offset = offset + context.getConfigParams().getSnapshotChunkSize();
 752             }
 753             return offset;
 754         }
 755
 756         public int incrementChunkIndex() {
 757             if (replyStatus) {
 758                 // if prev chunk failed, we would want to sent the same chunk again
 759                 chunkIndex =  chunkIndex + 1;
 760             }
 761             return chunkIndex;
 762         }
 763
 764         public int getChunkIndex() {
 765             return chunkIndex;
 766         }
 767
 768         public int getTotalChunks() {
 769             return totalChunks;
 770         }
 771
 772         public boolean canSendNextChunk() {
 773             // we only send a false if a chunk is sent but we have not received a reply yet
 774             return replyReceivedForOffset == offset;
 775         }
 776
 777         public boolean isLastChunk(int chunkIndex) {
 778             return totalChunks == chunkIndex;
 779         }
 780
 781         public void markSendStatus(boolean success) {
 782             if (success) {
 783                 // if the chunk sent was successful
 784                 replyReceivedForOffset = offset;
 785                 replyStatus = true;
 786                 lastChunkHashCode = nextChunkHashCode;
 787             } else {
 788                 // if the chunk sent was failure
 789                 replyReceivedForOffset = offset;
 790                 replyStatus = false;
 791             }
 792         }
 793
 794         public ByteString getNextChunk() {
 795             int snapshotLength = getSnapshotBytes().size();
 796             int start = incrementOffset();
 797             int size = context.getConfigParams().getSnapshotChunkSize();
 798             if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
 799                 size = snapshotLength;
 800             } else {
 801                 if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
 802                     size = snapshotLength - start;
 803                 }
 804             }
 805
 806             if(LOG.isDebugEnabled()) {
 807                 LOG.debug("{}: Next chunk: length={}, offset={},size={}", context.getId(),
 808                     snapshotLength, start, size);
 809             }
 810             ByteString substring = getSnapshotBytes().substring(start, start + size);
 811             nextChunkHashCode = substring.hashCode();
 812             return substring;
 813         }
 814
 815         /**
 816          * reset should be called when the Follower needs to be sent the snapshot from the beginning
 817          */
 818         public void reset(){
 819             offset = 0;
 820             replyStatus = false;
 821             replyReceivedForOffset = offset;
 822             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 823             lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 824         }
 825
 826         public int getLastChunkHashCode() {
 827             return lastChunkHashCode;
 828         }
 829     }
 830
 831     // called from example-actor for printing the follower-states
 832     public String printFollowerStates() {
 833         final StringBuilder sb = new StringBuilder();
 834
 835         sb.append('[');
 836         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 837             sb.append('{');
 838             sb.append(followerLogInformation.getId());
 839             sb.append(" state:");
 840             sb.append(followerLogInformation.isFollowerActive());
 841             sb.append("},");
 842         }
 843         sb.append(']');
 844
 845         return sb.toString();
 846     }
 847
 848     @VisibleForTesting
 849     public FollowerLogInformation getFollower(String followerId) {
 850         return followerToLog.get(followerId);
 851     }
 852
 853     @VisibleForTesting
 854     protected void setFollowerSnapshot(String followerId, FollowerToSnapshot snapshot) {
 855         mapFollowerToSnapshot.put(followerId, snapshot);
 856     }
 857
 858     @VisibleForTesting
 859     public int followerSnapshotSize() {
 860         return mapFollowerToSnapshot.size();
 861     }
 862
 863     @VisibleForTesting
 864     public int followerLogSize() {
 865         return followerToLog.size();
 866     }
 867 }