opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Follower.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.Address;
  14 import akka.cluster.Cluster;
  15 import akka.cluster.ClusterEvent.CurrentClusterState;
  16 import akka.cluster.Member;
  17 import akka.cluster.MemberStatus;
  18 import akka.japi.Procedure;
  19 import com.google.common.annotations.VisibleForTesting;
  20 import com.google.common.base.Stopwatch;
  21 import java.io.IOException;
  22 import java.util.ArrayList;
  23 import java.util.Optional;
  24 import java.util.Set;
  25 import java.util.concurrent.TimeUnit;
  26 import java.util.concurrent.atomic.AtomicBoolean;
  27 import javax.annotation.Nullable;
  28 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  29 import org.opendaylight.controller.cluster.raft.RaftState;
  30 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  31 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  32 import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
  33 import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow;
  34 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  35 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  36 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  37 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  38 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  39 import org.opendaylight.controller.cluster.raft.messages.RequestVote;
  40 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  41 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  42 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
  43
  44 /**
  45  * The behavior of a RaftActor in the Follower raft state.
  46  * <ul>
  47  * <li> Respond to RPCs from candidates and leaders
  48  * <li> If election timeout elapses without receiving AppendEntries
  49  * RPC from current leader or granting vote to candidate:
  50  * convert to candidate
  51  * </ul>
  52  */
  53 public class Follower extends AbstractRaftActorBehavior {
  54     private static final long MAX_ELECTION_TIMEOUT_FACTOR = 18;
  55
  56     private final SyncStatusTracker initialSyncStatusTracker;
  57
  58     private final Stopwatch lastLeaderMessageTimer = Stopwatch.createStarted();
  59     private SnapshotTracker snapshotTracker = null;
  60     private String leaderId;
  61     private short leaderPayloadVersion;
  62
  63     public Follower(final RaftActorContext context) {
  64         this(context, null, (short)-1);
  65     }
  66
  67     public Follower(final RaftActorContext context, final String initialLeaderId,
  68             final short initialLeaderPayloadVersion) {
  69         super(context, RaftState.Follower);
  70         this.leaderId = initialLeaderId;
  71         this.leaderPayloadVersion = initialLeaderPayloadVersion;
  72
  73         initialSyncStatusTracker = new SyncStatusTracker(context.getActor(), getId(), context.getConfigParams()
  74             .getSyncIndexThreshold());
  75
  76         if (context.getPeerIds().isEmpty() && getLeaderId() == null) {
  77             actor().tell(TimeoutNow.INSTANCE, actor());
  78         } else {
  79             scheduleElection(electionDuration());
  80         }
  81     }
  82
  83     @Override
  84     public final String getLeaderId() {
  85         return leaderId;
  86     }
  87
  88     @VisibleForTesting
  89     protected final void setLeaderId(@Nullable final String leaderId) {
  90         this.leaderId = leaderId;
  91     }
  92
  93     @Override
  94     public short getLeaderPayloadVersion() {
  95         return leaderPayloadVersion;
  96     }
  97
  98     @VisibleForTesting
  99     protected final void setLeaderPayloadVersion(final short leaderPayloadVersion) {
 100         this.leaderPayloadVersion = leaderPayloadVersion;
 101     }
 102
 103     private void restartLastLeaderMessageTimer() {
 104         if (lastLeaderMessageTimer.isRunning()) {
 105             lastLeaderMessageTimer.reset();
 106         }
 107
 108         lastLeaderMessageTimer.start();
 109     }
 110
 111     private boolean isLogEntryPresent(final long index) {
 112         if (context.getReplicatedLog().isInSnapshot(index)) {
 113             return true;
 114         }
 115
 116         ReplicatedLogEntry entry = context.getReplicatedLog().get(index);
 117         return entry != null;
 118
 119     }
 120
 121     private void updateInitialSyncStatus(final long currentLeaderCommit, final String newLeaderId) {
 122         initialSyncStatusTracker.update(newLeaderId, currentLeaderCommit, context.getCommitIndex());
 123     }
 124
 125     @Override
 126     protected RaftActorBehavior handleAppendEntries(final ActorRef sender, final AppendEntries appendEntries) {
 127
 128         int numLogEntries = appendEntries.getEntries() != null ? appendEntries.getEntries().size() : 0;
 129         if (log.isTraceEnabled()) {
 130             log.trace("{}: handleAppendEntries: {}", logName(), appendEntries);
 131         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 132             log.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 133         }
 134
 135         // TODO : Refactor this method into a bunch of smaller methods
 136         // to make it easier to read. Before refactoring ensure tests
 137         // cover the code properly
 138
 139         if (snapshotTracker != null && !snapshotTracker.getLeaderId().equals(appendEntries.getLeaderId())) {
 140             log.debug("{}: snapshot install is in progress but the prior snapshot leaderId {} does not match the "
 141                 + "AppendEntries leaderId {}", logName(), snapshotTracker.getLeaderId(), appendEntries.getLeaderId());
 142             closeSnapshotTracker();
 143         }
 144
 145         if (snapshotTracker != null || context.getSnapshotManager().isApplying()) {
 146             // if snapshot install is in progress, follower should just acknowledge append entries with a reply.
 147             AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 148                     lastIndex(), lastTerm(), context.getPayloadVersion());
 149
 150             log.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply);
 151             sender.tell(reply, actor());
 152
 153             return this;
 154         }
 155
 156         // If we got here then we do appear to be talking to the leader
 157         leaderId = appendEntries.getLeaderId();
 158         leaderPayloadVersion = appendEntries.getPayloadVersion();
 159
 160         // First check if the logs are in sync or not
 161         long lastIndex = lastIndex();
 162
 163         if (isOutOfSync(appendEntries)) {
 164             // We found that the log was out of sync so just send a negative
 165             // reply and return
 166
 167             final AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 168                     lastTerm(), context.getPayloadVersion());
 169
 170             log.info("{}: Follower is out-of-sync so sending negative reply: {}", logName(), reply);
 171             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 172             sender.tell(reply, actor());
 173             return this;
 174         }
 175
 176         if (appendEntries.getEntries() != null && appendEntries.getEntries().size() > 0) {
 177
 178             log.debug("{}: Number of entries to be appended = {}", logName(),
 179                         appendEntries.getEntries().size());
 180
 181             // 3. If an existing entry conflicts with a new one (same index
 182             // but different terms), delete the existing entry and all that
 183             // follow it (§5.3)
 184             int addEntriesFrom = 0;
 185             if (context.getReplicatedLog().size() > 0) {
 186
 187                 // Find the entry up until the one that is not in the follower's log
 188                 for (int i = 0;i < appendEntries.getEntries().size(); i++, addEntriesFrom++) {
 189                     ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i);
 190
 191                     if (!isLogEntryPresent(matchEntry.getIndex())) {
 192                         // newEntry not found in the log
 193                         break;
 194                     }
 195
 196                     long existingEntryTerm = getLogEntryTerm(matchEntry.getIndex());
 197
 198                     log.debug("{}: matchEntry {} is present: existingEntryTerm: {}", logName(), matchEntry,
 199                             existingEntryTerm);
 200
 201                     // existingEntryTerm == -1 means it's in the snapshot and not in the log. We don't know
 202                     // what the term was so we'll assume it matches.
 203                     if (existingEntryTerm == -1 || existingEntryTerm == matchEntry.getTerm()) {
 204                         continue;
 205                     }
 206
 207                     if (!context.getRaftPolicy().applyModificationToStateBeforeConsensus()) {
 208
 209                         log.info("{}: Removing entries from log starting at {}", logName(), matchEntry.getIndex());
 210
 211                         // Entries do not match so remove all subsequent entries
 212                         if (!context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex())) {
 213                             // Could not remove the entries - this means the matchEntry index must be in the
 214                             // snapshot and not the log. In this case the prior entries are part of the state
 215                             // so we must send back a reply to force a snapshot to completely re-sync the
 216                             // follower's log and state.
 217
 218                             log.info("{}: Could not remove entries - sending reply to force snapshot", logName());
 219                             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 220                             sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 221                                     lastTerm(), context.getPayloadVersion(), true), actor());
 222                             return this;
 223                         }
 224
 225                         break;
 226                     } else {
 227                         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 228                         sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 229                                 lastTerm(), context.getPayloadVersion(), true), actor());
 230                         return this;
 231                     }
 232                 }
 233             }
 234
 235             lastIndex = lastIndex();
 236             log.debug("{}: After cleanup, lastIndex: {}, entries to be added from: {}", logName(),
 237                     lastIndex, addEntriesFrom);
 238
 239             // When persistence successfully completes for each new log entry appended, we need to determine if we
 240             // should capture a snapshot to compact the persisted log. shouldCaptureSnapshot tracks whether or not
 241             // one of the log entries has exceeded the log size threshold whereby a snapshot should be taken. However
 242             // we don't initiate the snapshot at that log entry but rather after the last log entry has been persisted.
 243             // This is done because subsequent log entries after the one that tripped the threshold may have been
 244             // applied to the state already, as the persistence callback occurs async, and we want those entries
 245             // purged from the persisted log as well.
 246             final AtomicBoolean shouldCaptureSnapshot = new AtomicBoolean(false);
 247             final Procedure<ReplicatedLogEntry> appendAndPersistCallback = logEntry -> {
 248                 final ReplicatedLogEntry lastEntryToAppend = appendEntries.getEntries().get(
 249                         appendEntries.getEntries().size() - 1);
 250                 if (shouldCaptureSnapshot.get() && logEntry == lastEntryToAppend) {
 251                     context.getSnapshotManager().capture(context.getReplicatedLog().last(), getReplicatedToAllIndex());
 252                 }
 253             };
 254
 255             // 4. Append any new entries not already in the log
 256             for (int i = addEntriesFrom; i < appendEntries.getEntries().size(); i++) {
 257                 ReplicatedLogEntry entry = appendEntries.getEntries().get(i);
 258
 259                 log.debug("{}: Append entry to log {}", logName(), entry.getData());
 260
 261                 context.getReplicatedLog().appendAndPersist(entry, appendAndPersistCallback, false);
 262
 263                 shouldCaptureSnapshot.compareAndSet(false,
 264                         context.getReplicatedLog().shouldCaptureSnapshot(entry.getIndex()));
 265
 266                 if (entry.getData() instanceof ServerConfigurationPayload) {
 267                     context.updatePeerIds((ServerConfigurationPayload)entry.getData());
 268                 }
 269             }
 270
 271             log.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size());
 272         }
 273
 274         // 5. If leaderCommit > commitIndex, set commitIndex =
 275         // min(leaderCommit, index of last new entry)
 276
 277         lastIndex = lastIndex();
 278         long prevCommitIndex = context.getCommitIndex();
 279
 280         if (appendEntries.getLeaderCommit() > prevCommitIndex) {
 281             context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex));
 282         }
 283
 284         if (prevCommitIndex != context.getCommitIndex()) {
 285             log.debug("{}: Commit index set to {}", logName(), context.getCommitIndex());
 286         }
 287
 288         AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 289                 lastIndex, lastTerm(), context.getPayloadVersion());
 290
 291         if (log.isTraceEnabled()) {
 292             log.trace("{}: handleAppendEntries returning : {}", logName(), reply);
 293         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 294             log.debug("{}: handleAppendEntries returning : {}", logName(), reply);
 295         }
 296
 297         // Reply to the leader before applying any previous state so as not to hold up leader consensus.
 298         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 299         sender.tell(reply, actor());
 300
 301         // If commitIndex > lastApplied: increment lastApplied, apply
 302         // log[lastApplied] to state machine (§5.3)
 303         // check if there are any entries to be applied. last-applied can be equal to last-index
 304         if (appendEntries.getLeaderCommit() > context.getLastApplied()
 305                 && context.getLastApplied() < lastIndex) {
 306             if (log.isDebugEnabled()) {
 307                 log.debug("{}: applyLogToStateMachine, appendEntries.getLeaderCommit(): {},"
 308                         + "context.getLastApplied(): {}, lastIndex(): {}", logName(),
 309                     appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex);
 310             }
 311
 312             applyLogToStateMachine(appendEntries.getLeaderCommit());
 313         }
 314
 315         if (!context.getSnapshotManager().isCapturing()) {
 316             super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex());
 317         }
 318
 319         return this;
 320     }
 321
 322     private boolean isOutOfSync(final AppendEntries appendEntries) {
 323
 324         long prevLogTerm = getLogEntryTerm(appendEntries.getPrevLogIndex());
 325         boolean prevEntryPresent = isLogEntryPresent(appendEntries.getPrevLogIndex());
 326         long lastIndex = lastIndex();
 327         int numLogEntries = appendEntries.getEntries() != null ? appendEntries.getEntries().size() : 0;
 328         boolean outOfSync = true;
 329
 330         if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) {
 331
 332             // The follower's log is out of sync because the leader does have
 333             // an entry at prevLogIndex and this follower has no entries in
 334             // it's log.
 335
 336             log.info("{}: The followers log is empty and the senders prevLogIndex is {}",
 337                         logName(), appendEntries.getPrevLogIndex());
 338         } else if (lastIndex > -1 && appendEntries.getPrevLogIndex() != -1 && !prevEntryPresent) {
 339
 340             // The follower's log is out of sync because the Leader's
 341             // prevLogIndex entry was not found in it's log
 342
 343             log.info("{}: The log is not empty but the prevLogIndex {} was not found in it - "
 344                     + "lastIndex: {}, snapshotIndex: {}", logName(), appendEntries.getPrevLogIndex(), lastIndex,
 345                     context.getReplicatedLog().getSnapshotIndex());
 346         } else if (lastIndex > -1 && prevEntryPresent && prevLogTerm != appendEntries.getPrevLogTerm()) {
 347
 348             // The follower's log is out of sync because the Leader's
 349             // prevLogIndex entry does exist in the follower's log but it has
 350             // a different term in it
 351
 352             log.info("{}: The prevLogIndex {} was found in the log but the term {} is not equal to the append entries"
 353                       + "prevLogTerm {} - lastIndex: {}, snapshotIndex: {}", logName(), appendEntries.getPrevLogIndex(),
 354                       prevLogTerm, appendEntries.getPrevLogTerm(), lastIndex,
 355                       context.getReplicatedLog().getSnapshotIndex());
 356         } else if (appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 357                 && appendEntries.getReplicatedToAllIndex() != -1
 358                 && !isLogEntryPresent(appendEntries.getReplicatedToAllIndex())) {
 359             // This append entry comes from a leader who has it's log aggressively trimmed and so does not have
 360             // the previous entry in it's in-memory journal
 361
 362             log.info("{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the"
 363                     + " in-memory journal", logName(), appendEntries.getReplicatedToAllIndex());
 364         } else if (appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 365                 && appendEntries.getReplicatedToAllIndex() != -1 && numLogEntries > 0
 366                 && !isLogEntryPresent(appendEntries.getEntries().get(0).getIndex() - 1)) {
 367             log.info("{}: Cannot append entries because the calculated previousIndex {} was not found in the "
 368                     + " in-memory journal", logName(), appendEntries.getEntries().get(0).getIndex() - 1);
 369         } else {
 370             outOfSync = false;
 371         }
 372         return outOfSync;
 373     }
 374
 375     @Override
 376     protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender,
 377         final AppendEntriesReply appendEntriesReply) {
 378         return this;
 379     }
 380
 381     @Override
 382     protected RaftActorBehavior handleRequestVoteReply(final ActorRef sender,
 383         final RequestVoteReply requestVoteReply) {
 384         return this;
 385     }
 386
 387     @Override
 388     public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) {
 389         if (message instanceof ElectionTimeout || message instanceof TimeoutNow) {
 390             return handleElectionTimeout(message);
 391         }
 392
 393         if (!(message instanceof RaftRPC)) {
 394             // The rest of the processing requires the message to be a RaftRPC
 395             return null;
 396         }
 397
 398         final RaftRPC rpc = (RaftRPC) message;
 399         // If RPC request or response contains term T > currentTerm:
 400         // set currentTerm = T, convert to follower (§5.1)
 401         // This applies to all RPC messages and responses
 402         if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 403             log.info("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term",
 404                 logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 405
 406             context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 407         }
 408
 409         if (rpc instanceof InstallSnapshot) {
 410             handleInstallSnapshot(sender, (InstallSnapshot) rpc);
 411             restartLastLeaderMessageTimer();
 412             scheduleElection(electionDuration());
 413             return this;
 414         }
 415
 416         if (!(rpc instanceof RequestVote) || canGrantVote((RequestVote) rpc)) {
 417             restartLastLeaderMessageTimer();
 418             scheduleElection(electionDuration());
 419         }
 420
 421         return super.handleMessage(sender, rpc);
 422     }
 423
 424     private RaftActorBehavior handleElectionTimeout(final Object message) {
 425         // If the message is ElectionTimeout, verify we haven't actually seen a message from the leader
 426         // during the election timeout interval. It may that the election timer expired b/c this actor
 427         // was busy and messages got delayed, in which case leader messages would be backed up in the
 428         // queue but would be processed before the ElectionTimeout message and thus would restart the
 429         // lastLeaderMessageTimer.
 430         long lastLeaderMessageInterval = lastLeaderMessageTimer.elapsed(TimeUnit.MILLISECONDS);
 431         long electionTimeoutInMillis = context.getConfigParams().getElectionTimeOutInterval().toMillis();
 432         boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning()
 433                 || lastLeaderMessageInterval >= electionTimeoutInMillis;
 434
 435         if (canStartElection()) {
 436             if (message instanceof TimeoutNow) {
 437                 log.debug("{}: Received TimeoutNow - switching to Candidate", logName());
 438                 return internalSwitchBehavior(RaftState.Candidate);
 439             } else if (noLeaderMessageReceived) {
 440                 // Check the cluster state to see if the leader is known to be up before we go to Candidate.
 441                 // However if we haven't heard from the leader in a long time even though the cluster state
 442                 // indicates it's up then something is wrong - leader might be stuck indefinitely - so switch
 443                 // to Candidate,
 444                 long maxElectionTimeout = electionTimeoutInMillis * MAX_ELECTION_TIMEOUT_FACTOR;
 445                 if (isLeaderAvailabilityKnown() && lastLeaderMessageInterval < maxElectionTimeout) {
 446                     log.debug("{}: Received ElectionTimeout but leader appears to be available", logName());
 447                     scheduleElection(electionDuration());
 448                 } else {
 449                     log.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
 450                     return internalSwitchBehavior(RaftState.Candidate);
 451                 }
 452             } else {
 453                 log.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout {}",
 454                         logName(), lastLeaderMessageInterval, context.getConfigParams().getElectionTimeOutInterval());
 455                 scheduleElection(electionDuration());
 456             }
 457         } else if (message instanceof ElectionTimeout) {
 458             if (noLeaderMessageReceived) {
 459                 setLeaderId(null);
 460             }
 461
 462             scheduleElection(electionDuration());
 463         }
 464
 465         return this;
 466     }
 467
 468     private boolean isLeaderAvailabilityKnown() {
 469         if (leaderId == null) {
 470             return false;
 471         }
 472
 473         Optional<Cluster> cluster = context.getCluster();
 474         if (!cluster.isPresent()) {
 475             return false;
 476         }
 477
 478         ActorSelection leaderActor = context.getPeerActorSelection(leaderId);
 479         if (leaderActor == null) {
 480             return false;
 481         }
 482
 483         Address leaderAddress = leaderActor.anchorPath().address();
 484
 485         CurrentClusterState state = cluster.get().state();
 486         Set<Member> unreachable = state.getUnreachable();
 487
 488         log.debug("{}: Checking for leader {} in the cluster unreachable set {}", logName(), leaderAddress,
 489                 unreachable);
 490
 491         for (Member m: unreachable) {
 492             if (leaderAddress.equals(m.address())) {
 493                 log.info("{}: Leader {} is unreachable", logName(), leaderAddress);
 494                 return false;
 495             }
 496         }
 497
 498         for (Member m: state.getMembers()) {
 499             if (leaderAddress.equals(m.address())) {
 500                 if (m.status() == MemberStatus.up() || m.status() == MemberStatus.weaklyUp()) {
 501                     log.debug("{}: Leader {} cluster status is {} - leader is available", logName(),
 502                             leaderAddress, m.status());
 503                     return true;
 504                 } else {
 505                     log.debug("{}: Leader {} cluster status is {} - leader is unavailable", logName(),
 506                             leaderAddress, m.status());
 507                     return false;
 508                 }
 509             }
 510         }
 511
 512         log.debug("{}: Leader {} not found in the cluster member set", logName(), leaderAddress);
 513
 514         return false;
 515     }
 516
 517     private void handleInstallSnapshot(final ActorRef sender, final InstallSnapshot installSnapshot) {
 518
 519         log.debug("{}: handleInstallSnapshot: {}", logName(), installSnapshot);
 520
 521         leaderId = installSnapshot.getLeaderId();
 522
 523         if (snapshotTracker == null) {
 524             snapshotTracker = new SnapshotTracker(log, installSnapshot.getTotalChunks(), installSnapshot.getLeaderId(),
 525                     context);
 526         }
 527
 528         updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId());
 529
 530         try {
 531             final InstallSnapshotReply reply = new InstallSnapshotReply(
 532                     currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true);
 533
 534             if (snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(),
 535                     installSnapshot.getLastChunkHashCode())) {
 536
 537                 log.info("{}: Snapshot installed from leader: {}", logName(), installSnapshot.getLeaderId());
 538
 539                 Snapshot snapshot = Snapshot.create(
 540                         context.getSnapshotManager().convertSnapshot(snapshotTracker.getSnapshotBytes()),
 541                         new ArrayList<>(),
 542                         installSnapshot.getLastIncludedIndex(),
 543                         installSnapshot.getLastIncludedTerm(),
 544                         installSnapshot.getLastIncludedIndex(),
 545                         installSnapshot.getLastIncludedTerm(),
 546                         context.getTermInformation().getCurrentTerm(),
 547                         context.getTermInformation().getVotedFor(),
 548                         installSnapshot.getServerConfig().orNull());
 549
 550                 ApplySnapshot.Callback applySnapshotCallback = new ApplySnapshot.Callback() {
 551                     @Override
 552                     public void onSuccess() {
 553                         log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 554
 555                         sender.tell(reply, actor());
 556                     }
 557
 558                     @Override
 559                     public void onFailure() {
 560                         sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor());
 561                     }
 562                 };
 563
 564                 actor().tell(new ApplySnapshot(snapshot, applySnapshotCallback), actor());
 565
 566                 closeSnapshotTracker();
 567             } else {
 568                 log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 569
 570                 sender.tell(reply, actor());
 571             }
 572         } catch (IOException e) {
 573             log.debug("{}: Exception in InstallSnapshot of follower", logName(), e);
 574
 575             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 576                     -1, false), actor());
 577
 578             closeSnapshotTracker();
 579         }
 580     }
 581
 582     private void closeSnapshotTracker() {
 583         if (snapshotTracker != null) {
 584             snapshotTracker.close();
 585             snapshotTracker = null;
 586         }
 587     }
 588
 589     @Override
 590     public void close() {
 591         closeSnapshotTracker();
 592         stopElection();
 593     }
 594
 595     @VisibleForTesting
 596     SnapshotTracker getSnapshotTracker() {
 597         return snapshotTracker;
 598     }
 599 }