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