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

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8
   9 package org.opendaylight.controller.cluster.raft.behaviors;
  10
  11 import akka.actor.ActorRef;
  12 import akka.actor.ActorSelection;
  13 import akka.actor.Cancellable;
  14 import com.google.common.annotations.VisibleForTesting;
  15 import com.google.common.base.Optional;
  16 import com.google.common.base.Preconditions;
  17 import com.google.protobuf.ByteString;
  18 import java.io.IOException;
  19 import java.util.Arrays;
  20 import java.util.Collection;
  21 import java.util.Collections;
  22 import java.util.HashMap;
  23 import java.util.Iterator;
  24 import java.util.LinkedList;
  25 import java.util.List;
  26 import java.util.Map;
  27 import java.util.Map.Entry;
  28 import java.util.Queue;
  29 import javax.annotation.Nullable;
  30 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  31 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  32 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  33 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  34 import org.opendaylight.controller.cluster.raft.PeerInfo;
  35 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  36 import org.opendaylight.controller.cluster.raft.RaftState;
  37 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  38 import org.opendaylight.controller.cluster.raft.Snapshot;
  39 import org.opendaylight.controller.cluster.raft.VotingState;
  40 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  41 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  42 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  43 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  44 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  45 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  46 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  47 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  48 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  49 import org.opendaylight.controller.cluster.raft.messages.UnInitializedFollowerSnapshotReply;
  50 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  51 import scala.concurrent.duration.FiniteDuration;
  52
  53 /**
  54  * The behavior of a RaftActor when it is in the Leader state
  55  * <p/>
  56  * Leaders:
  57  * <ul>
  58  * <li> Upon election: send initial empty AppendEntries RPCs
  59  * (heartbeat) to each server; repeat during idle periods to
  60  * prevent election timeouts (§5.2)
  61  * <li> If command received from client: append entry to local log,
  62  * respond after entry applied to state machine (§5.3)
  63  * <li> If last log index ≥ nextIndex for a follower: send
  64  * AppendEntries RPC with log entries starting at nextIndex
  65  * <ul>
  66  * <li> If successful: update nextIndex and matchIndex for
  67  * follower (§5.3)
  68  * <li> If AppendEntries fails because of log inconsistency:
  69  * decrement nextIndex and retry (§5.3)
  70  * </ul>
  71  * <li> If there exists an N such that N > commitIndex, a majority
  72  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  73  * set commitIndex = N (§5.3, §5.4).
  74  */
  75 public abstract class AbstractLeader extends AbstractRaftActorBehavior {
  76
  77     // The index of the first chunk that is sent when installing a snapshot
  78     public static final int FIRST_CHUNK_INDEX = 1;
  79
  80     // The index that the follower should respond with if it needs the install snapshot to be reset
  81     public static final int INVALID_CHUNK_INDEX = -1;
  82
  83     // This would be passed as the hash code of the last chunk when sending the first chunk
  84     public static final int INITIAL_LAST_CHUNK_HASH_CODE = -1;
  85
  86     private final Map<String, FollowerLogInformation> followerToLog = new HashMap<>();
  87     private final Map<String, FollowerToSnapshot> mapFollowerToSnapshot = new HashMap<>();
  88
  89     /**
  90      * Lookup table for request contexts based on journal index. We could use a {@link Map} here, but we really
  91      * expect the entries to be modified in sequence, hence we open-code the lookup.
  92      *
  93      * TODO: Evaluate the use of ArrayDeque(), as that has lower memory overhead. Non-head removals are more costly,
  94      *       but we already expect those to be far from frequent.
  95      */
  96     private final Queue<ClientRequestTracker> trackers = new LinkedList<>();
  97
  98     private Cancellable heartbeatSchedule = null;
  99     private Optional<SnapshotHolder> snapshot;
 100     private int minReplicationCount;
 101
 102     protected AbstractLeader(RaftActorContext context, RaftState state,
 103             @Nullable AbstractLeader initializeFromLeader) {
 104         super(context, state);
 105
 106         if(initializeFromLeader != null) {
 107             followerToLog.putAll(initializeFromLeader.followerToLog);
 108         } else {
 109             for(PeerInfo peerInfo: context.getPeers()) {
 110                 FollowerLogInformation followerLogInformation = new FollowerLogInformationImpl(peerInfo, -1, context);
 111                 followerToLog.put(peerInfo.getId(), followerLogInformation);
 112             }
 113         }
 114
 115         LOG.debug("{}: Election: Leader has following peers: {}", logName(), getFollowerIds());
 116
 117         updateMinReplicaCount();
 118
 119         snapshot = Optional.absent();
 120
 121         // Immediately schedule a heartbeat
 122         // Upon election: send initial empty AppendEntries RPCs
 123         // (heartbeat) to each server; repeat during idle periods to
 124         // prevent election timeouts (§5.2)
 125         sendAppendEntries(0, false);
 126
 127         // It is important to schedule this heartbeat here
 128         scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 129     }
 130
 131     protected AbstractLeader(RaftActorContext context, RaftState state) {
 132         this(context, state, null);
 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     public void addFollower(String followerId) {
 145         FollowerLogInformation followerLogInformation = new FollowerLogInformationImpl(
 146                 context.getPeerInfo(followerId), -1, context);
 147         followerToLog.put(followerId, followerLogInformation);
 148
 149         if(heartbeatSchedule == null) {
 150             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 151         }
 152     }
 153
 154     public void removeFollower(String followerId) {
 155         followerToLog.remove(followerId);
 156         mapFollowerToSnapshot.remove(followerId);
 157     }
 158
 159     public void updateMinReplicaCount() {
 160         int numVoting = 0;
 161         for(PeerInfo peer: context.getPeers()) {
 162             if(peer.isVoting()) {
 163                 numVoting++;
 164             }
 165         }
 166
 167         minReplicationCount = getMajorityVoteCount(numVoting);
 168     }
 169
 170     protected int getMinIsolatedLeaderPeerCount(){
 171       //the isolated Leader peer count will be 1 less than the majority vote count.
 172         //this is because the vote count has the self vote counted in it
 173         //for e.g
 174         //0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
 175         //2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
 176         //4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
 177
 178         return minReplicationCount > 0 ? (minReplicationCount - 1) : 0;
 179     }
 180
 181     @VisibleForTesting
 182     void setSnapshot(@Nullable Snapshot snapshot) {
 183         if(snapshot != null) {
 184             this.snapshot = Optional.of(new SnapshotHolder(snapshot));
 185         } else {
 186             this.snapshot = Optional.absent();
 187         }
 188     }
 189
 190     @Override
 191     protected RaftActorBehavior handleAppendEntries(ActorRef sender,
 192         AppendEntries appendEntries) {
 193
 194         LOG.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 195
 196         return this;
 197     }
 198
 199     @Override
 200     protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 201         AppendEntriesReply appendEntriesReply) {
 202
 203         if(LOG.isTraceEnabled()) {
 204             LOG.trace("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
 205         }
 206
 207         // Update the FollowerLogInformation
 208         String followerId = appendEntriesReply.getFollowerId();
 209         FollowerLogInformation followerLogInformation =
 210             followerToLog.get(followerId);
 211
 212         if(followerLogInformation == null){
 213             LOG.error("{}: handleAppendEntriesReply - unknown follower {}", logName(), followerId);
 214             return this;
 215         }
 216
 217         if(followerLogInformation.timeSinceLastActivity() >
 218                 context.getConfigParams().getElectionTimeOutInterval().toMillis()) {
 219             LOG.warn("{} : handleAppendEntriesReply delayed beyond election timeout, " +
 220                             "appendEntriesReply : {}, timeSinceLastActivity : {}, lastApplied : {}, commitIndex : {}",
 221                     logName(), appendEntriesReply, followerLogInformation.timeSinceLastActivity(),
 222                     context.getLastApplied(), context.getCommitIndex());
 223         }
 224
 225         followerLogInformation.markFollowerActive();
 226         followerLogInformation.setPayloadVersion(appendEntriesReply.getPayloadVersion());
 227         followerLogInformation.setRaftVersion(appendEntriesReply.getRaftVersion());
 228
 229         boolean updated = false;
 230         if(appendEntriesReply.getLogLastIndex() > context.getReplicatedLog().lastIndex()) {
 231             // The follower's log is actually ahead of the leader's log. Normally this doesn't happen
 232             // in raft as a node cannot become leader if it's log is behind another's. However, the
 233             // non-voting semantics deviate a bit from raft. Only voting members participate in
 234             // elections and can become leader so it's possible for a non-voting follower to be ahead
 235             // of the leader. This can happen if persistence is disabled and all voting members are
 236             // restarted. In this case, the voting leader will start out with an empty log however
 237             // the non-voting followers still retain the previous data in memory. On the first
 238             // AppendEntries, the non-voting follower returns a successful reply b/c the prevLogIndex
 239             // sent by the leader is -1 and thus the integrity checks pass. However the follower's returned
 240             // lastLogIndex may be higher in which case we want to reset the follower by installing a
 241             // snapshot. It's also possible that the follower's last log index is behind the leader's.
 242             // However in this case the log terms won't match and the logs will conflict - this is handled
 243             // elsewhere.
 244             LOG.debug("{}: handleAppendEntriesReply: follower {} lastIndex {} is ahead of our lastIndex {} - forcing install snaphot",
 245                     logName(), followerLogInformation.getId(), appendEntriesReply.getLogLastIndex(),
 246                     context.getReplicatedLog().lastIndex());
 247
 248             followerLogInformation.setMatchIndex(-1);
 249             followerLogInformation.setNextIndex(-1);
 250
 251             initiateCaptureSnapshot(followerId);
 252             updated = true;
 253         } else if (appendEntriesReply.isSuccess()) {
 254             updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
 255         } else {
 256             LOG.debug("{}: handleAppendEntriesReply: received unsuccessful reply: {}", logName(), appendEntriesReply);
 257
 258             long followerLastLogIndex = appendEntriesReply.getLogLastIndex();
 259             long followersLastLogTerm = getLogEntryTerm(followerLastLogIndex);
 260             if(appendEntriesReply.isForceInstallSnapshot()) {
 261                 // Reset the followers match and next index. This is to signal that this follower has nothing
 262                 // in common with this Leader and so would require a snapshot to be installed
 263                 followerLogInformation.setMatchIndex(-1);
 264                 followerLogInformation.setNextIndex(-1);
 265
 266                 // Force initiate a snapshot capture
 267                 initiateCaptureSnapshot(followerId);
 268             } else if(followerLastLogIndex < 0 || (followersLastLogTerm >= 0 &&
 269                     followersLastLogTerm == appendEntriesReply.getLogLastTerm())) {
 270                 // The follower's log is empty or the last entry is present in the leader's journal
 271                 // and the terms match so the follower is just behind the leader's journal from
 272                 // the last snapshot, if any. We'll catch up the follower quickly by starting at the
 273                 // follower's last log index.
 274
 275                 updated = updateFollowerLogInformation(followerLogInformation, appendEntriesReply);
 276             } else {
 277                 // The follower's log conflicts with leader's log so decrement follower's next index by 1
 278                 // in an attempt to find where the logs match.
 279
 280                 LOG.debug("{}: follower's last log term {} conflicts with the leader's {} - dec next index",
 281                         logName(), appendEntriesReply.getLogLastTerm(), followersLastLogTerm);
 282
 283                 followerLogInformation.decrNextIndex();
 284             }
 285         }
 286
 287         // Now figure out if this reply warrants a change in the commitIndex
 288         // If there exists an N such that N > commitIndex, a majority
 289         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 290         // set commitIndex = N (§5.3, §5.4).
 291         if(LOG.isTraceEnabled()) {
 292             LOG.trace("{}: handleAppendEntriesReply from {}: commitIndex: {}, lastAppliedIndex: {}, currentTerm: {}",
 293                     logName(), followerId, context.getCommitIndex(), context.getLastApplied(), currentTerm());
 294         }
 295
 296         for (long N = context.getCommitIndex() + 1; ; N++) {
 297             int replicatedCount = 1;
 298
 299             LOG.trace("{}: checking Nth index {}", logName(), N);
 300             for (FollowerLogInformation info : followerToLog.values()) {
 301                 final PeerInfo peerInfo = context.getPeerInfo(info.getId());
 302                 if(info.getMatchIndex() >= N && (peerInfo != null && peerInfo.isVoting())) {
 303                     replicatedCount++;
 304                 } else if(LOG.isTraceEnabled()) {
 305                     LOG.trace("{}: Not counting follower {} - matchIndex: {}, {}", logName(), info.getId(),
 306                             info.getMatchIndex(), peerInfo);
 307                 }
 308             }
 309
 310             if(LOG.isTraceEnabled()) {
 311                 LOG.trace("{}: replicatedCount {}, minReplicationCount: {}", logName(), replicatedCount, minReplicationCount);
 312             }
 313
 314             if (replicatedCount >= minReplicationCount) {
 315                 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(N);
 316                 if (replicatedLogEntry == null) {
 317                     LOG.debug("{}: ReplicatedLogEntry not found for index {} - snapshotIndex: {}, journal size: {}",
 318                             logName(), N, context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().size());
 319                     break;
 320                 }
 321
 322                 // Don't update the commit index if the log entry is from a previous term, as per §5.4.1:
 323                 // "Raft never commits log entries from previous terms by counting replicas".
 324                 // However we keep looping so we can make progress when new entries in the current term
 325                 // reach consensus, as per §5.4.1: "once an entry from the current term is committed by
 326                 // counting replicas, then all prior entries are committed indirectly".
 327                 if (replicatedLogEntry.getTerm() == currentTerm()) {
 328                     LOG.trace("{}: Setting commit index to {}", logName(), N);
 329                     context.setCommitIndex(N);
 330                 } else {
 331                     LOG.debug("{}: Not updating commit index to {} - retrieved log entry with index {}, term {} does not match the current term {}",
 332                             logName(), N, replicatedLogEntry.getIndex(), replicatedLogEntry.getTerm(), currentTerm());
 333                 }
 334             } else {
 335                 LOG.trace("{}: minReplicationCount not reached, actual {} - breaking", logName(), replicatedCount);
 336                 break;
 337             }
 338         }
 339
 340         // Apply the change to the state machine
 341         if (context.getCommitIndex() > context.getLastApplied()) {
 342             if(LOG.isDebugEnabled()) {
 343                 LOG.debug("{}: handleAppendEntriesReply from {}: applying to log - commitIndex: {}, lastAppliedIndex: {}",
 344                         logName(), followerId, context.getCommitIndex(), context.getLastApplied());
 345             }
 346
 347             applyLogToStateMachine(context.getCommitIndex());
 348         }
 349
 350         if (!context.getSnapshotManager().isCapturing()) {
 351             purgeInMemoryLog();
 352         }
 353
 354         //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
 355         sendUpdatesToFollower(followerId, followerLogInformation, false, !updated);
 356
 357         return this;
 358     }
 359
 360     private boolean updateFollowerLogInformation(FollowerLogInformation followerLogInformation,
 361             AppendEntriesReply appendEntriesReply) {
 362         boolean updated = followerLogInformation.setMatchIndex(appendEntriesReply.getLogLastIndex());
 363         updated = followerLogInformation.setNextIndex(appendEntriesReply.getLogLastIndex() + 1) || updated;
 364
 365         if(updated && LOG.isDebugEnabled()) {
 366             LOG.debug("{}: handleAppendEntriesReply - FollowerLogInformation for {} updated: matchIndex: {}, nextIndex: {}",
 367                     logName(), followerLogInformation.getId(), followerLogInformation.getMatchIndex(),
 368                     followerLogInformation.getNextIndex());
 369         }
 370         return updated;
 371     }
 372
 373     private void purgeInMemoryLog() {
 374         //find the lowest index across followers which has been replicated to all.
 375         // lastApplied if there are no followers, so that we keep clearing the log for single-node
 376         // we would delete the in-mem log from that index on, in-order to minimize mem usage
 377         // we would also share this info thru AE with the followers so that they can delete their log entries as well.
 378         long minReplicatedToAllIndex = followerToLog.isEmpty() ? context.getLastApplied() : Long.MAX_VALUE;
 379         for (FollowerLogInformation info : followerToLog.values()) {
 380             minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
 381         }
 382
 383         super.performSnapshotWithoutCapture(minReplicatedToAllIndex);
 384     }
 385
 386     @Override
 387     protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
 388         final Iterator<ClientRequestTracker> it = trackers.iterator();
 389         while (it.hasNext()) {
 390             final ClientRequestTracker t = it.next();
 391             if (t.getIndex() == logIndex) {
 392                 it.remove();
 393                 return t;
 394             }
 395         }
 396
 397         return null;
 398     }
 399
 400     @Override
 401     protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 402         RequestVoteReply requestVoteReply) {
 403         return this;
 404     }
 405
 406     protected void beforeSendHeartbeat(){}
 407
 408     @Override
 409     public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 410         Preconditions.checkNotNull(sender, "sender should not be null");
 411
 412         Object message = fromSerializableMessage(originalMessage);
 413
 414         if (message instanceof RaftRPC) {
 415             RaftRPC rpc = (RaftRPC) message;
 416             // If RPC request or response contains term T > currentTerm:
 417             // set currentTerm = T, convert to follower (§5.1)
 418             // This applies to all RPC messages and responses
 419             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 420                 LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {} - switching to Follower",
 421                         logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 422
 423                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 424
 425                 return internalSwitchBehavior(RaftState.Follower);
 426             }
 427         }
 428
 429         if (message instanceof SendHeartBeat) {
 430             beforeSendHeartbeat();
 431             sendHeartBeat();
 432             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 433         } else if(message instanceof SendInstallSnapshot) {
 434             // received from RaftActor
 435             setSnapshot(((SendInstallSnapshot) message).getSnapshot());
 436             sendInstallSnapshot();
 437         } else if (message instanceof Replicate) {
 438             replicate((Replicate) message);
 439         } else if (message instanceof InstallSnapshotReply) {
 440             handleInstallSnapshotReply((InstallSnapshotReply) message);
 441         } else {
 442             return super.handleMessage(sender, message);
 443         }
 444
 445         return this;
 446     }
 447
 448     private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
 449         LOG.debug("{}: handleInstallSnapshotReply: {}", logName(), reply);
 450
 451         String followerId = reply.getFollowerId();
 452         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 453
 454         if (followerToSnapshot == null) {
 455             LOG.error("{}: FollowerToSnapshot not found for follower {} in InstallSnapshotReply",
 456                     logName(), followerId);
 457             return;
 458         }
 459
 460         FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 461         if(followerLogInformation == null) {
 462             // This can happen during AddServer if it times out.
 463             LOG.error("{}: FollowerLogInformation not found for follower {} in InstallSnapshotReply",
 464                     logName(), followerId);
 465             mapFollowerToSnapshot.remove(followerId);
 466             return;
 467         }
 468
 469         followerLogInformation.markFollowerActive();
 470
 471         if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
 472             boolean wasLastChunk = false;
 473             if (reply.isSuccess()) {
 474                 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
 475                     //this was the last chunk reply
 476                     if(LOG.isDebugEnabled()) {
 477                         LOG.debug("{}: InstallSnapshotReply received, " +
 478                                 "last chunk received, Chunk: {}. Follower: {} Setting nextIndex: {}",
 479                                 logName(), reply.getChunkIndex(), followerId,
 480                             context.getReplicatedLog().getSnapshotIndex() + 1
 481                         );
 482                     }
 483
 484                     long followerMatchIndex = snapshot.get().getLastIncludedIndex();
 485                     followerLogInformation.setMatchIndex(followerMatchIndex);
 486                     followerLogInformation.setNextIndex(followerMatchIndex + 1);
 487                     mapFollowerToSnapshot.remove(followerId);
 488
 489                     LOG.debug("{}: follower: {}, matchIndex set to {}, nextIndex set to {}",
 490                         logName(), followerId, followerLogInformation.getMatchIndex(),
 491                         followerLogInformation.getNextIndex());
 492
 493                     if (mapFollowerToSnapshot.isEmpty()) {
 494                         // once there are no pending followers receiving snapshots
 495                         // we can remove snapshot from the memory
 496                         setSnapshot(null);
 497                     }
 498                     wasLastChunk = true;
 499                     if(context.getPeerInfo(followerId).getVotingState() == VotingState.VOTING_NOT_INITIALIZED){
 500                         UnInitializedFollowerSnapshotReply unInitFollowerSnapshotSuccess =
 501                                              new UnInitializedFollowerSnapshotReply(followerId);
 502                         context.getActor().tell(unInitFollowerSnapshotSuccess, context.getActor());
 503                         LOG.debug("Sent message UnInitializedFollowerSnapshotReply to self");
 504                     }
 505                 } else {
 506                     followerToSnapshot.markSendStatus(true);
 507                 }
 508             } else {
 509                 LOG.info("{}: InstallSnapshotReply received sending snapshot chunk failed, Will retry, Chunk: {}",
 510                         logName(), reply.getChunkIndex());
 511
 512                 followerToSnapshot.markSendStatus(false);
 513             }
 514
 515             if (wasLastChunk && !context.getSnapshotManager().isCapturing()) {
 516                 // Since the follower is now caught up try to purge the log.
 517                 purgeInMemoryLog();
 518             } else if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
 519                 ActorSelection followerActor = context.getPeerActorSelection(followerId);
 520                 if(followerActor != null) {
 521                     sendSnapshotChunk(followerActor, followerId);
 522                 }
 523             }
 524
 525         } else {
 526             LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
 527                     logName(), reply.getChunkIndex(), followerId,
 528                     followerToSnapshot.getChunkIndex());
 529
 530             if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
 531                 // Since the Follower did not find this index to be valid we should reset the follower snapshot
 532                 // so that Installing the snapshot can resume from the beginning
 533                 followerToSnapshot.reset();
 534             }
 535         }
 536     }
 537
 538     private void replicate(Replicate replicate) {
 539         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 540
 541         LOG.debug("{}: Replicate message: identifier: {}, logIndex: {}, payload: {}", logName(),
 542                 replicate.getIdentifier(), logIndex, replicate.getReplicatedLogEntry().getData().getClass());
 543
 544         // Create a tracker entry we will use this later to notify the
 545         // client actor
 546         if(replicate.getClientActor() != null) {
 547             trackers.add(new ClientRequestTrackerImpl(replicate.getClientActor(), replicate.getIdentifier(),
 548                     logIndex));
 549         }
 550
 551         boolean applyModificationToState = !context.anyVotingPeers()
 552                 || context.getRaftPolicy().applyModificationToStateBeforeConsensus();
 553
 554         if(applyModificationToState){
 555             context.setCommitIndex(logIndex);
 556             applyLogToStateMachine(logIndex);
 557         }
 558
 559         if (!followerToLog.isEmpty()) {
 560             sendAppendEntries(0, false);
 561         }
 562     }
 563
 564     protected void sendAppendEntries(long timeSinceLastActivityInterval, boolean isHeartbeat) {
 565         // Send an AppendEntries to all followers
 566         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 567             final String followerId = e.getKey();
 568             final FollowerLogInformation followerLogInformation = e.getValue();
 569             // This checks helps not to send a repeat message to the follower
 570             if(!followerLogInformation.isFollowerActive() ||
 571                     followerLogInformation.timeSinceLastActivity() >= timeSinceLastActivityInterval) {
 572                 sendUpdatesToFollower(followerId, followerLogInformation, true, isHeartbeat);
 573             }
 574         }
 575     }
 576
 577     /**
 578      *
 579      * This method checks if any update needs to be sent to the given follower. This includes append log entries,
 580      * sending next snapshot chunk, and initiating a snapshot.
 581      * @return true if any update is sent, false otherwise
 582      */
 583
 584     private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation,
 585                                        boolean sendHeartbeat, boolean isHeartbeat) {
 586
 587         ActorSelection followerActor = context.getPeerActorSelection(followerId);
 588         if (followerActor != null) {
 589             long followerNextIndex = followerLogInformation.getNextIndex();
 590             boolean isFollowerActive = followerLogInformation.isFollowerActive();
 591             boolean sendAppendEntries = false;
 592             List<ReplicatedLogEntry> entries = Collections.emptyList();
 593
 594             if (mapFollowerToSnapshot.get(followerId) != null) {
 595                 // if install snapshot is in process , then sent next chunk if possible
 596                 if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
 597                     sendSnapshotChunk(followerActor, followerId);
 598                 } else if(sendHeartbeat) {
 599                     // we send a heartbeat even if we have not received a reply for the last chunk
 600                     sendAppendEntries = true;
 601                 }
 602             } else {
 603                 long leaderLastIndex = context.getReplicatedLog().lastIndex();
 604                 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 605
 606                 if((!isHeartbeat && LOG.isDebugEnabled()) || LOG.isTraceEnabled()) {
 607                     LOG.debug("{}: Checking sendAppendEntries for follower {}: active: {}, followerNextIndex: {}, leaderLastIndex: {}, leaderSnapShotIndex: {}",
 608                             logName(), followerId, isFollowerActive, followerNextIndex, leaderLastIndex, leaderSnapShotIndex);
 609                 }
 610
 611                 if (isFollowerActive && context.getReplicatedLog().isPresent(followerNextIndex)) {
 612
 613                     LOG.debug("{}: sendAppendEntries: {} is present for follower {}", logName(),
 614                             followerNextIndex, followerId);
 615
 616                     if(followerLogInformation.okToReplicate()) {
 617                         // Try to send all the entries in the journal but not exceeding the max data size
 618                         // for a single AppendEntries message.
 619                         int maxEntries = (int) context.getReplicatedLog().size();
 620                         entries = context.getReplicatedLog().getFrom(followerNextIndex, maxEntries,
 621                                 context.getConfigParams().getSnapshotChunkSize());
 622                         sendAppendEntries = true;
 623                     }
 624                 } else if (isFollowerActive && followerNextIndex >= 0 &&
 625                     leaderLastIndex > followerNextIndex && !context.getSnapshotManager().isCapturing()) {
 626                     // if the followers next index is not present in the leaders log, and
 627                     // if the follower is just not starting and if leader's index is more than followers index
 628                     // then snapshot should be sent
 629
 630                     if (LOG.isDebugEnabled()) {
 631                         LOG.debug(String.format("%s: InitiateInstallSnapshot to follower: %s, " +
 632                                     "follower-nextIndex: %d, leader-snapshot-index: %d,  " +
 633                                     "leader-last-index: %d", logName(), followerId,
 634                                     followerNextIndex, leaderSnapShotIndex, leaderLastIndex));
 635                     }
 636
 637                     // Send heartbeat to follower whenever install snapshot is initiated.
 638                     sendAppendEntries = true;
 639                     if (canInstallSnapshot(followerNextIndex)) {
 640                         initiateCaptureSnapshot(followerId);
 641                     }
 642
 643                 } else if(sendHeartbeat) {
 644                     // we send an AppendEntries, even if the follower is inactive
 645                     // in-order to update the followers timestamp, in case it becomes active again
 646                     sendAppendEntries = true;
 647                 }
 648
 649             }
 650
 651             if(sendAppendEntries) {
 652                 sendAppendEntriesToFollower(followerActor, followerNextIndex,
 653                         entries, followerId);
 654             }
 655         }
 656     }
 657
 658     private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
 659         List<ReplicatedLogEntry> entries, String followerId) {
 660         AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
 661             getLogEntryIndex(followerNextIndex - 1),
 662             getLogEntryTerm(followerNextIndex - 1), entries,
 663             context.getCommitIndex(), super.getReplicatedToAllIndex(), context.getPayloadVersion());
 664
 665         if(!entries.isEmpty() || LOG.isTraceEnabled()) {
 666             LOG.debug("{}: Sending AppendEntries to follower {}: {}", logName(), followerId,
 667                     appendEntries);
 668         }
 669
 670         followerActor.tell(appendEntries, actor());
 671     }
 672
 673     /**
 674      * Install Snapshot works as follows
 675      * 1. Leader initiates the capture snapshot by sending a CaptureSnapshot message to actor
 676      * 2. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
 677      * and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
 678      * 3. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
 679      * 4. On complete, Follower sends back a InstallSnapshotReply.
 680      * 5. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
 681      * and replenishes the memory by deleting the snapshot in Replicated log.
 682      * 6. If another follower requires a snapshot and a snapshot has been collected (via CaptureSnapshotReply)
 683      * then send the existing snapshot in chunks to the follower.
 684      * @param followerId
 685      */
 686     public boolean initiateCaptureSnapshot(String followerId) {
 687         if (snapshot.isPresent()) {
 688             // if a snapshot is present in the memory, most likely another install is in progress
 689             // no need to capture snapshot.
 690             // This could happen if another follower needs an install when one is going on.
 691             final ActorSelection followerActor = context.getPeerActorSelection(followerId);
 692             sendSnapshotChunk(followerActor, followerId);
 693             return true;
 694         } else {
 695             return context.getSnapshotManager().captureToInstall(context.getReplicatedLog().last(),
 696                     this.getReplicatedToAllIndex(), followerId);
 697         }
 698     }
 699
 700     private boolean canInstallSnapshot(long nextIndex){
 701         // If the follower's nextIndex is -1 then we might as well send it a snapshot
 702         // Otherwise send it a snapshot only if the nextIndex is not present in the log but is present
 703         // in the snapshot
 704         return (nextIndex == -1 ||
 705                 (!context.getReplicatedLog().isPresent(nextIndex)
 706                         && context.getReplicatedLog().isInSnapshot(nextIndex)));
 707
 708     }
 709
 710
 711     private void sendInstallSnapshot() {
 712         LOG.debug("{}: sendInstallSnapshot", logName());
 713         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 714             String followerId = e.getKey();
 715             ActorSelection followerActor = context.getPeerActorSelection(followerId);
 716             FollowerLogInformation followerLogInfo = e.getValue();
 717
 718             if (followerActor != null) {
 719                 long nextIndex = followerLogInfo.getNextIndex();
 720                 if (context.getPeerInfo(followerId).getVotingState() == VotingState.VOTING_NOT_INITIALIZED ||
 721                         canInstallSnapshot(nextIndex)) {
 722                     sendSnapshotChunk(followerActor, followerId);
 723                 }
 724             }
 725         }
 726     }
 727
 728     /**
 729      *  Sends a snapshot chunk to a given follower
 730      *  InstallSnapshot should qualify as a heartbeat too.
 731      */
 732     private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
 733         try {
 734             if (snapshot.isPresent()) {
 735                 byte[] nextSnapshotChunk = getNextSnapshotChunk(followerId, snapshot.get().getSnapshotBytes());
 736
 737                 // Note: the previous call to getNextSnapshotChunk has the side-effect of adding
 738                 // followerId to the followerToSnapshot map.
 739                 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 740
 741                 int nextChunkIndex = followerToSnapshot.incrementChunkIndex();
 742                 Optional<ServerConfigurationPayload> serverConfig = Optional.absent();
 743                 if(followerToSnapshot.isLastChunk(nextChunkIndex)) {
 744                     serverConfig = Optional.fromNullable(context.getPeerServerInfo(true));
 745                 }
 746
 747                 followerActor.tell(
 748                     new InstallSnapshot(currentTerm(), context.getId(),
 749                         snapshot.get().getLastIncludedIndex(),
 750                         snapshot.get().getLastIncludedTerm(),
 751                         nextSnapshotChunk,
 752                         nextChunkIndex,
 753                         followerToSnapshot.getTotalChunks(),
 754                         Optional.of(followerToSnapshot.getLastChunkHashCode()),
 755                         serverConfig
 756                     ).toSerializable(followerToLog.get(followerId).getRaftVersion()),
 757                     actor()
 758                 );
 759
 760                 if(LOG.isDebugEnabled()) {
 761                     LOG.debug("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
 762                             logName(), followerActor.path(), followerToSnapshot.getChunkIndex(),
 763                             followerToSnapshot.getTotalChunks());
 764                 }
 765             }
 766         } catch (IOException e) {
 767             LOG.error("{}: InstallSnapshot failed for Leader.", logName(), e);
 768         }
 769     }
 770
 771     /**
 772      * Acccepts snaphot as ByteString, enters into map for future chunks
 773      * creates and return a ByteString chunk
 774      */
 775     private byte[] getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
 776         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 777         if (followerToSnapshot == null) {
 778             followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
 779             mapFollowerToSnapshot.put(followerId, followerToSnapshot);
 780         }
 781         byte[] nextChunk = followerToSnapshot.getNextChunk();
 782
 783         LOG.debug("{}: next snapshot chunk size for follower {}: {}", logName(), followerId, nextChunk.length);
 784
 785         return nextChunk;
 786     }
 787
 788     private void sendHeartBeat() {
 789         if (!followerToLog.isEmpty()) {
 790             LOG.trace("{}: Sending heartbeat", logName());
 791             sendAppendEntries(context.getConfigParams().getHeartBeatInterval().toMillis(), true);
 792         }
 793     }
 794
 795     private void stopHeartBeat() {
 796         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 797             heartbeatSchedule.cancel();
 798         }
 799     }
 800
 801     private void scheduleHeartBeat(FiniteDuration interval) {
 802         if (followerToLog.isEmpty()) {
 803             // Optimization - do not bother scheduling a heartbeat as there are
 804             // no followers
 805             return;
 806         }
 807
 808         stopHeartBeat();
 809
 810         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 811         // message is sent to itself.
 812         // Scheduling the heartbeat only once here because heartbeats do not
 813         // need to be sent if there are other messages being sent to the remote
 814         // actor.
 815         heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
 816             interval, context.getActor(), SendHeartBeat.INSTANCE,
 817             context.getActorSystem().dispatcher(), context.getActor());
 818     }
 819
 820     @Override
 821     public void close() {
 822         stopHeartBeat();
 823     }
 824
 825     @Override
 826     public final String getLeaderId() {
 827         return context.getId();
 828     }
 829
 830     @Override
 831     public final short getLeaderPayloadVersion() {
 832         return context.getPayloadVersion();
 833     }
 834
 835     protected boolean isLeaderIsolated() {
 836         int minPresent = getMinIsolatedLeaderPeerCount();
 837         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 838             final PeerInfo peerInfo = context.getPeerInfo(followerLogInformation.getId());
 839             if(peerInfo != null && peerInfo.isVoting() && followerLogInformation.isFollowerActive()) {
 840                 --minPresent;
 841                 if (minPresent == 0) {
 842                     return false;
 843                 }
 844             }
 845         }
 846         return (minPresent != 0);
 847     }
 848
 849     /**
 850      * Encapsulates the snapshot bytestring and handles the logic of sending
 851      * snapshot chunks
 852      */
 853     protected class FollowerToSnapshot {
 854         private final ByteString snapshotBytes;
 855         private int offset = 0;
 856         // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
 857         private int replyReceivedForOffset;
 858         // if replyStatus is false, the previous chunk is attempted
 859         private boolean replyStatus = false;
 860         private int chunkIndex;
 861         private final int totalChunks;
 862         private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 863         private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 864
 865         public FollowerToSnapshot(ByteString snapshotBytes) {
 866             this.snapshotBytes = snapshotBytes;
 867             int size = snapshotBytes.size();
 868             totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
 869                 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
 870             if(LOG.isDebugEnabled()) {
 871                 LOG.debug("{}: Snapshot {} bytes, total chunks to send:{}",
 872                         logName(), size, totalChunks);
 873             }
 874             replyReceivedForOffset = -1;
 875             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 876         }
 877
 878         public ByteString getSnapshotBytes() {
 879             return snapshotBytes;
 880         }
 881
 882         public int incrementOffset() {
 883             if(replyStatus) {
 884                 // if prev chunk failed, we would want to sent the same chunk again
 885                 offset = offset + context.getConfigParams().getSnapshotChunkSize();
 886             }
 887             return offset;
 888         }
 889
 890         public int incrementChunkIndex() {
 891             if (replyStatus) {
 892                 // if prev chunk failed, we would want to sent the same chunk again
 893                 chunkIndex =  chunkIndex + 1;
 894             }
 895             return chunkIndex;
 896         }
 897
 898         public int getChunkIndex() {
 899             return chunkIndex;
 900         }
 901
 902         public int getTotalChunks() {
 903             return totalChunks;
 904         }
 905
 906         public boolean canSendNextChunk() {
 907             // we only send a false if a chunk is sent but we have not received a reply yet
 908             return replyReceivedForOffset == offset;
 909         }
 910
 911         public boolean isLastChunk(int chunkIndex) {
 912             return totalChunks == chunkIndex;
 913         }
 914
 915         public void markSendStatus(boolean success) {
 916             if (success) {
 917                 // if the chunk sent was successful
 918                 replyReceivedForOffset = offset;
 919                 replyStatus = true;
 920                 lastChunkHashCode = nextChunkHashCode;
 921             } else {
 922                 // if the chunk sent was failure
 923                 replyReceivedForOffset = offset;
 924                 replyStatus = false;
 925             }
 926         }
 927
 928         public byte[] getNextChunk() {
 929             int snapshotLength = getSnapshotBytes().size();
 930             int start = incrementOffset();
 931             int size = context.getConfigParams().getSnapshotChunkSize();
 932             if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
 933                 size = snapshotLength;
 934             } else if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
 935                 size = snapshotLength - start;
 936             }
 937
 938             byte[] nextChunk = new byte[size];
 939             getSnapshotBytes().copyTo(nextChunk, start, 0, size);
 940             nextChunkHashCode = Arrays.hashCode(nextChunk);
 941
 942             LOG.debug("{}: Next chunk: total length={}, offset={}, size={}, hashCode={}", logName(),
 943                     snapshotLength, start, size, nextChunkHashCode);
 944             return nextChunk;
 945         }
 946
 947         /**
 948          * reset should be called when the Follower needs to be sent the snapshot from the beginning
 949          */
 950         public void reset(){
 951             offset = 0;
 952             replyStatus = false;
 953             replyReceivedForOffset = offset;
 954             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 955             lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 956         }
 957
 958         public int getLastChunkHashCode() {
 959             return lastChunkHashCode;
 960         }
 961     }
 962
 963     // called from example-actor for printing the follower-states
 964     public String printFollowerStates() {
 965         final StringBuilder sb = new StringBuilder();
 966
 967         sb.append('[');
 968         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 969             sb.append('{');
 970             sb.append(followerLogInformation.getId());
 971             sb.append(" state:");
 972             sb.append(followerLogInformation.isFollowerActive());
 973             sb.append("},");
 974         }
 975         sb.append(']');
 976
 977         return sb.toString();
 978     }
 979
 980     @VisibleForTesting
 981     public FollowerLogInformation getFollower(String followerId) {
 982         return followerToLog.get(followerId);
 983     }
 984
 985     @VisibleForTesting
 986     protected void setFollowerSnapshot(String followerId, FollowerToSnapshot snapshot) {
 987         mapFollowerToSnapshot.put(followerId, snapshot);
 988     }
 989
 990     @VisibleForTesting
 991     public int followerSnapshotSize() {
 992         return mapFollowerToSnapshot.size();
 993     }
 994
 995     @VisibleForTesting
 996     public int followerLogSize() {
 997         return followerToLog.size();
 998     }
 999
1000     private static class SnapshotHolder {
1001         private final long lastIncludedTerm;
1002         private final long lastIncludedIndex;
1003         private final ByteString snapshotBytes;
1004
1005         SnapshotHolder(Snapshot snapshot) {
1006             this.lastIncludedTerm = snapshot.getLastAppliedTerm();
1007             this.lastIncludedIndex = snapshot.getLastAppliedIndex();
1008             this.snapshotBytes = ByteString.copyFrom(snapshot.getState());
1009         }
1010
1011         long getLastIncludedTerm() {
1012             return lastIncludedTerm;
1013         }
1014
1015         long getLastIncludedIndex() {
1016             return lastIncludedIndex;
1017         }
1018
1019         ByteString getSnapshotBytes() {
1020             return snapshotBytes;
1021         }
1022     }
1023 }