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 package org.opendaylight.controller.cluster.raft.behaviors;
   9
  10 import akka.actor.ActorRef;
  11 import akka.actor.ActorSelection;
  12 import akka.actor.Address;
  13 import akka.cluster.Cluster;
  14 import akka.cluster.ClusterEvent.CurrentClusterState;
  15 import akka.cluster.Member;
  16 import akka.cluster.MemberStatus;
  17 import com.google.common.annotations.VisibleForTesting;
  18 import com.google.common.base.Stopwatch;
  19 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  20 import java.io.IOException;
  21 import java.util.HashSet;
  22 import java.util.List;
  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 java.util.function.Consumer;
  28 import org.eclipse.jdt.annotation.Nullable;
  29 import org.opendaylight.controller.cluster.messaging.MessageAssembler;
  30 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  31 import org.opendaylight.controller.cluster.raft.RaftState;
  32 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  33 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  34 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  35 import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
  36 import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow;
  37 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  38 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  39 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  40 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  41 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  42 import org.opendaylight.controller.cluster.raft.messages.RequestVote;
  43 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  44 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  45 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
  46
  47 /**
  48  * The behavior of a RaftActor in the Follower raft state.
  49  * <ul>
  50  * <li> Respond to RPCs from candidates and leaders
  51  * <li> If election timeout elapses without receiving AppendEntries
  52  * RPC from current leader or granting vote to candidate:
  53  * convert to candidate
  54  * </ul>
  55  */
  56 // Non-final for testing
  57 public class Follower extends AbstractRaftActorBehavior {
  58     private static final long MAX_ELECTION_TIMEOUT_FACTOR = 18;
  59
  60     private final SyncStatusTracker initialSyncStatusTracker;
  61
  62     private final MessageAssembler appendEntriesMessageAssembler;
  63
  64     private final Stopwatch lastLeaderMessageTimer = Stopwatch.createStarted();
  65     private SnapshotTracker snapshotTracker = null;
  66     private String leaderId;
  67     private short leaderPayloadVersion;
  68
  69     public Follower(final RaftActorContext context) {
  70         this(context, null, (short)-1);
  71     }
  72
  73     @SuppressFBWarnings(value = "MC_OVERRIDABLE_METHOD_CALL_IN_CONSTRUCTOR",
  74         justification = "electionDuration() is not final for Candidate override")
  75     public Follower(final RaftActorContext context, final String initialLeaderId,
  76             final short initialLeaderPayloadVersion) {
  77         super(context, RaftState.Follower);
  78         leaderId = initialLeaderId;
  79         leaderPayloadVersion = initialLeaderPayloadVersion;
  80
  81         initialSyncStatusTracker = new SyncStatusTracker(context.getActor(), getId(), context.getConfigParams()
  82             .getSyncIndexThreshold());
  83
  84         appendEntriesMessageAssembler = MessageAssembler.builder().logContext(logName())
  85                 .fileBackedStreamFactory(context.getFileBackedOutputStreamFactory())
  86                 .assembledMessageCallback((message, sender) -> handleMessage(sender, message)).build();
  87
  88         if (context.getPeerIds().isEmpty() && getLeaderId() == null) {
  89             actor().tell(TimeoutNow.INSTANCE, actor());
  90         } else {
  91             scheduleElection(electionDuration());
  92         }
  93     }
  94
  95     @Override
  96     public final String getLeaderId() {
  97         return leaderId;
  98     }
  99
 100     @VisibleForTesting
 101     protected final void setLeaderId(final @Nullable String leaderId) {
 102         this.leaderId = leaderId;
 103     }
 104
 105     @Override
 106     public short getLeaderPayloadVersion() {
 107         return leaderPayloadVersion;
 108     }
 109
 110     @VisibleForTesting
 111     protected final void setLeaderPayloadVersion(final short leaderPayloadVersion) {
 112         this.leaderPayloadVersion = leaderPayloadVersion;
 113     }
 114
 115     private void restartLastLeaderMessageTimer() {
 116         if (lastLeaderMessageTimer.isRunning()) {
 117             lastLeaderMessageTimer.reset();
 118         }
 119
 120         lastLeaderMessageTimer.start();
 121     }
 122
 123     private boolean isLogEntryPresent(final long index) {
 124         if (context.getReplicatedLog().isInSnapshot(index)) {
 125             return true;
 126         }
 127
 128         ReplicatedLogEntry entry = context.getReplicatedLog().get(index);
 129         return entry != null;
 130
 131     }
 132
 133     private void updateInitialSyncStatus(final long currentLeaderCommit, final String newLeaderId) {
 134         initialSyncStatusTracker.update(newLeaderId, currentLeaderCommit, context.getCommitIndex());
 135     }
 136
 137     @Override
 138     protected RaftActorBehavior handleAppendEntries(final ActorRef sender, final AppendEntries appendEntries) {
 139         int numLogEntries = appendEntries.getEntries().size();
 140         if (log.isTraceEnabled()) {
 141             log.trace("{}: handleAppendEntries: {}", logName(), appendEntries);
 142         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 143             log.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 144         }
 145
 146         if (snapshotTracker != null && !snapshotTracker.getLeaderId().equals(appendEntries.getLeaderId())) {
 147             log.debug("{}: snapshot install is in progress but the prior snapshot leaderId {} does not match the "
 148                 + "AppendEntries leaderId {}", logName(), snapshotTracker.getLeaderId(), appendEntries.getLeaderId());
 149             closeSnapshotTracker();
 150         }
 151
 152         if (snapshotTracker != null || context.getSnapshotManager().isApplying()) {
 153             // if snapshot install is in progress, follower should just acknowledge append entries with a reply.
 154             AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 155                     lastIndex(), lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(),
 156                     appendEntries.getLeaderRaftVersion());
 157
 158             log.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply);
 159             sender.tell(reply, actor());
 160
 161             return this;
 162         }
 163
 164         // If we got here then we do appear to be talking to the leader
 165         leaderId = appendEntries.getLeaderId();
 166         leaderPayloadVersion = appendEntries.getPayloadVersion();
 167
 168         if (appendEntries.getLeaderAddress().isPresent()) {
 169             final String address = appendEntries.getLeaderAddress().get();
 170             log.debug("New leader address: {}", address);
 171
 172             context.setPeerAddress(leaderId, address);
 173             context.getConfigParams().getPeerAddressResolver().setResolved(leaderId, address);
 174         }
 175
 176         // First check if the logs are in sync or not
 177         if (isOutOfSync(appendEntries, sender)) {
 178             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 179             return this;
 180         }
 181
 182         if (!processNewEntries(appendEntries, sender)) {
 183             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 184             return this;
 185         }
 186
 187         long lastIndex = lastIndex();
 188         long prevCommitIndex = context.getCommitIndex();
 189
 190         // If leaderCommit > commitIndex, set commitIndex = min(leaderCommit, index of last new entry)
 191         if (appendEntries.getLeaderCommit() > prevCommitIndex) {
 192             context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex));
 193         }
 194
 195         if (prevCommitIndex != context.getCommitIndex()) {
 196             log.debug("{}: Commit index set to {}", logName(), context.getCommitIndex());
 197         }
 198
 199         AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 200                 lastIndex, lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(),
 201                 appendEntries.getLeaderRaftVersion());
 202
 203         if (log.isTraceEnabled()) {
 204             log.trace("{}: handleAppendEntries returning : {}", logName(), reply);
 205         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 206             log.debug("{}: handleAppendEntries returning : {}", logName(), reply);
 207         }
 208
 209         // Reply to the leader before applying any previous state so as not to hold up leader consensus.
 210         sender.tell(reply, actor());
 211
 212         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 213
 214         // If leaderCommit > lastApplied, increment lastApplied and apply log[lastApplied] to state machine (§5.3).
 215         // lastApplied can be equal to lastIndex.
 216         if (appendEntries.getLeaderCommit() > context.getLastApplied() && context.getLastApplied() < lastIndex) {
 217             if (log.isDebugEnabled()) {
 218                 log.debug("{}: applyLogToStateMachine, appendEntries.getLeaderCommit(): {}, "
 219                         + "context.getLastApplied(): {}, lastIndex(): {}", logName(),
 220                     appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex);
 221             }
 222
 223             applyLogToStateMachine(appendEntries.getLeaderCommit());
 224         }
 225
 226         if (!context.getSnapshotManager().isCapturing()) {
 227             super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex());
 228         }
 229
 230         appendEntriesMessageAssembler.checkExpiredAssembledMessageState();
 231
 232         return this;
 233     }
 234
 235     private boolean processNewEntries(final AppendEntries appendEntries, final ActorRef sender) {
 236         int numLogEntries = appendEntries.getEntries().size();
 237         if (numLogEntries == 0) {
 238             return true;
 239         }
 240
 241         log.debug("{}: Number of entries to be appended = {}", logName(), numLogEntries);
 242
 243         long lastIndex = lastIndex();
 244         int addEntriesFrom = 0;
 245
 246         // First check for conflicting entries. If an existing entry conflicts with a new one (same index but different
 247         // term), delete the existing entry and all that follow it (§5.3)
 248         if (context.getReplicatedLog().size() > 0) {
 249             // Find the entry up until the one that is not in the follower's log
 250             for (int i = 0;i < numLogEntries; i++, addEntriesFrom++) {
 251                 ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i);
 252
 253                 if (!isLogEntryPresent(matchEntry.getIndex())) {
 254                     // newEntry not found in the log
 255                     break;
 256                 }
 257
 258                 long existingEntryTerm = getLogEntryTerm(matchEntry.getIndex());
 259
 260                 log.debug("{}: matchEntry {} is present: existingEntryTerm: {}", logName(), matchEntry,
 261                         existingEntryTerm);
 262
 263                 // existingEntryTerm == -1 means it's in the snapshot and not in the log. We don't know
 264                 // what the term was so we'll assume it matches.
 265                 if (existingEntryTerm == -1 || existingEntryTerm == matchEntry.getTerm()) {
 266                     continue;
 267                 }
 268
 269                 if (!context.getRaftPolicy().applyModificationToStateBeforeConsensus()) {
 270                     log.info("{}: Removing entries from log starting at {}, commitIndex: {}, lastApplied: {}",
 271                             logName(), matchEntry.getIndex(), context.getCommitIndex(), context.getLastApplied());
 272
 273                     // Entries do not match so remove all subsequent entries but only if the existing entries haven't
 274                     // been applied to the state yet.
 275                     if (matchEntry.getIndex() <= context.getLastApplied()
 276                             || !context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex())) {
 277                         // Could not remove the entries - this means the matchEntry index must be in the
 278                         // snapshot and not the log. In this case the prior entries are part of the state
 279                         // so we must send back a reply to force a snapshot to completely re-sync the
 280                         // follower's log and state.
 281
 282                         log.info("{}: Could not remove entries - sending reply to force snapshot", logName());
 283                         sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 284                                 lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(),
 285                                 appendEntries.getLeaderRaftVersion()), actor());
 286                         return false;
 287                     }
 288
 289                     break;
 290                 } else {
 291                     sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 292                             lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(),
 293                             appendEntries.getLeaderRaftVersion()), actor());
 294                     return false;
 295                 }
 296             }
 297         }
 298
 299         lastIndex = lastIndex();
 300         log.debug("{}: After cleanup, lastIndex: {}, entries to be added from: {}", logName(), lastIndex,
 301                 addEntriesFrom);
 302
 303         // When persistence successfully completes for each new log entry appended, we need to determine if we
 304         // should capture a snapshot to compact the persisted log. shouldCaptureSnapshot tracks whether or not
 305         // one of the log entries has exceeded the log size threshold whereby a snapshot should be taken. However
 306         // we don't initiate the snapshot at that log entry but rather after the last log entry has been persisted.
 307         // This is done because subsequent log entries after the one that tripped the threshold may have been
 308         // applied to the state already, as the persistence callback occurs async, and we want those entries
 309         // purged from the persisted log as well.
 310         final AtomicBoolean shouldCaptureSnapshot = new AtomicBoolean(false);
 311         final Consumer<ReplicatedLogEntry> appendAndPersistCallback = logEntry -> {
 312             final List<ReplicatedLogEntry> entries = appendEntries.getEntries();
 313             final ReplicatedLogEntry lastEntryToAppend = entries.get(entries.size() - 1);
 314             if (shouldCaptureSnapshot.get() && logEntry == lastEntryToAppend) {
 315                 context.getSnapshotManager().capture(context.getReplicatedLog().last(), getReplicatedToAllIndex());
 316             }
 317         };
 318
 319         // Append any new entries not already in the log
 320         for (int i = addEntriesFrom; i < numLogEntries; i++) {
 321             ReplicatedLogEntry entry = appendEntries.getEntries().get(i);
 322
 323             log.debug("{}: Append entry to log {}", logName(), entry.getData());
 324
 325             context.getReplicatedLog().appendAndPersist(entry, appendAndPersistCallback, false);
 326
 327             shouldCaptureSnapshot.compareAndSet(false,
 328                     context.getReplicatedLog().shouldCaptureSnapshot(entry.getIndex()));
 329
 330             if (entry.getData() instanceof ServerConfigurationPayload serverConfiguration) {
 331                 context.updatePeerIds(serverConfiguration);
 332             }
 333         }
 334
 335         log.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size());
 336
 337         return true;
 338     }
 339
 340     private boolean isOutOfSync(final AppendEntries appendEntries, final ActorRef sender) {
 341
 342         final long lastIndex = lastIndex();
 343         if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) {
 344
 345             // The follower's log is out of sync because the leader does have an entry at prevLogIndex and this
 346             // follower has no entries in it's log.
 347
 348             log.info("{}: The followers log is empty and the senders prevLogIndex is {}", logName(),
 349                 appendEntries.getPrevLogIndex());
 350
 351             sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 352             return true;
 353         }
 354
 355         if (lastIndex > -1) {
 356             if (isLogEntryPresent(appendEntries.getPrevLogIndex())) {
 357                 final long leadersPrevLogTermInFollowersLogOrSnapshot =
 358                         getLogEntryOrSnapshotTerm(appendEntries.getPrevLogIndex());
 359                 if (leadersPrevLogTermInFollowersLogOrSnapshot != appendEntries.getPrevLogTerm()) {
 360
 361                     // The follower's log is out of sync because the Leader's prevLogIndex entry does exist
 362                     // in the follower's log or snapshot but it has a different term.
 363
 364                     log.info("{}: The prevLogIndex {} was found in the log but the term {} is not equal to the append "
 365                         + "entries prevLogTerm {} - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 366                         appendEntries.getPrevLogIndex(), leadersPrevLogTermInFollowersLogOrSnapshot,
 367                         appendEntries.getPrevLogTerm(), lastIndex, context.getReplicatedLog().getSnapshotIndex(),
 368                         context.getReplicatedLog().getSnapshotTerm());
 369
 370                     sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 371                     return true;
 372                 }
 373             } else if (appendEntries.getPrevLogIndex() != -1) {
 374
 375                 // The follower's log is out of sync because the Leader's prevLogIndex entry was not found in it's log
 376
 377                 log.info("{}: The log is not empty but the prevLogIndex {} was not found in it - lastIndex: {}, "
 378                         + "snapshotIndex: {}, snapshotTerm: {}", logName(), appendEntries.getPrevLogIndex(), lastIndex,
 379                         context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm());
 380
 381                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 382                 return true;
 383             }
 384         }
 385
 386         if (appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 387                 && appendEntries.getReplicatedToAllIndex() != -1) {
 388             if (!isLogEntryPresent(appendEntries.getReplicatedToAllIndex())) {
 389                 // This append entry comes from a leader who has it's log aggressively trimmed and so does not have
 390                 // the previous entry in it's in-memory journal
 391
 392                 log.info("{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the "
 393                         + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 394                         appendEntries.getReplicatedToAllIndex(), lastIndex,
 395                         context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm());
 396
 397                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 398                 return true;
 399             }
 400
 401             final List<ReplicatedLogEntry> entries = appendEntries.getEntries();
 402             if (entries.size() > 0 && !isLogEntryPresent(entries.get(0).getIndex() - 1)) {
 403                 log.info("{}: Cannot append entries because the calculated previousIndex {} was not found in the "
 404                         + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 405                         entries.get(0).getIndex() - 1, lastIndex, context.getReplicatedLog().getSnapshotIndex(),
 406                         context.getReplicatedLog().getSnapshotTerm());
 407
 408                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 409                 return true;
 410             }
 411         }
 412
 413         return false;
 414     }
 415
 416     private void sendOutOfSyncAppendEntriesReply(final ActorRef sender, final boolean forceInstallSnapshot,
 417             final short leaderRaftVersion) {
 418         // We found that the log was out of sync so just send a negative reply.
 419         final AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex(),
 420                 lastTerm(), context.getPayloadVersion(), forceInstallSnapshot, needsLeaderAddress(),
 421                 leaderRaftVersion);
 422
 423         log.info("{}: Follower is out-of-sync so sending negative reply: {}", logName(), reply);
 424         sender.tell(reply, actor());
 425     }
 426
 427     private boolean needsLeaderAddress() {
 428         return context.getPeerAddress(leaderId) == null;
 429     }
 430
 431     @Override
 432     protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender,
 433         final AppendEntriesReply appendEntriesReply) {
 434         return this;
 435     }
 436
 437     @Override
 438     protected RaftActorBehavior handleRequestVoteReply(final ActorRef sender,
 439         final RequestVoteReply requestVoteReply) {
 440         return this;
 441     }
 442
 443     @Override
 444     final ApplyState getApplyStateFor(final ReplicatedLogEntry entry) {
 445         return new ApplyState(null, null, entry);
 446     }
 447
 448     @Override
 449     public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) {
 450         if (message instanceof ElectionTimeout || message instanceof TimeoutNow) {
 451             return handleElectionTimeout(message);
 452         }
 453
 454         if (appendEntriesMessageAssembler.handleMessage(message, actor())) {
 455             return this;
 456         }
 457
 458         if (!(message instanceof RaftRPC rpc)) {
 459             // The rest of the processing requires the message to be a RaftRPC
 460             return null;
 461         }
 462
 463         // If RPC request or response contains term T > currentTerm:
 464         // set currentTerm = T, convert to follower (§5.1)
 465         // This applies to all RPC messages and responses
 466         if (rpc.getTerm() > context.getTermInformation().getCurrentTerm() && shouldUpdateTerm(rpc)) {
 467             log.info("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term",
 468                 logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 469
 470             context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 471         }
 472
 473         if (rpc instanceof InstallSnapshot installSnapshot) {
 474             handleInstallSnapshot(sender, installSnapshot);
 475             restartLastLeaderMessageTimer();
 476             scheduleElection(electionDuration());
 477             return this;
 478         }
 479
 480         if (!(rpc instanceof RequestVote requestVote) || canGrantVote(requestVote)) {
 481             restartLastLeaderMessageTimer();
 482             scheduleElection(electionDuration());
 483         }
 484
 485         return super.handleMessage(sender, rpc);
 486     }
 487
 488     private RaftActorBehavior handleElectionTimeout(final Object message) {
 489         // If the message is ElectionTimeout, verify we haven't actually seen a message from the leader
 490         // during the election timeout interval. It may that the election timer expired b/c this actor
 491         // was busy and messages got delayed, in which case leader messages would be backed up in the
 492         // queue but would be processed before the ElectionTimeout message and thus would restart the
 493         // lastLeaderMessageTimer.
 494         long lastLeaderMessageInterval = lastLeaderMessageTimer.elapsed(TimeUnit.MILLISECONDS);
 495         long electionTimeoutInMillis = context.getConfigParams().getElectionTimeOutInterval().toMillis();
 496         boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning()
 497                 || lastLeaderMessageInterval >= electionTimeoutInMillis;
 498
 499         if (canStartElection()) {
 500             if (message instanceof TimeoutNow) {
 501                 log.debug("{}: Received TimeoutNow - switching to Candidate", logName());
 502                 return internalSwitchBehavior(RaftState.Candidate);
 503             } else if (noLeaderMessageReceived) {
 504                 // Check the cluster state to see if the leader is known to be up before we go to Candidate.
 505                 // However if we haven't heard from the leader in a long time even though the cluster state
 506                 // indicates it's up then something is wrong - leader might be stuck indefinitely - so switch
 507                 // to Candidate,
 508                 long maxElectionTimeout = electionTimeoutInMillis * MAX_ELECTION_TIMEOUT_FACTOR;
 509                 if (isLeaderAvailabilityKnown() && lastLeaderMessageInterval < maxElectionTimeout) {
 510                     log.debug("{}: Received ElectionTimeout but leader appears to be available", logName());
 511                     scheduleElection(electionDuration());
 512                 } else if (isThisFollowerIsolated()) {
 513                     log.debug("{}: this follower is isolated. Do not switch to Candidate for now.", logName());
 514                     setLeaderId(null);
 515                     scheduleElection(electionDuration());
 516                 } else {
 517                     log.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
 518                     return internalSwitchBehavior(RaftState.Candidate);
 519                 }
 520             } else {
 521                 log.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout {}",
 522                         logName(), lastLeaderMessageInterval, context.getConfigParams().getElectionTimeOutInterval());
 523                 scheduleElection(electionDuration());
 524             }
 525         } else if (message instanceof ElectionTimeout) {
 526             if (noLeaderMessageReceived) {
 527                 setLeaderId(null);
 528             }
 529
 530             scheduleElection(electionDuration());
 531         }
 532
 533         return this;
 534     }
 535
 536     private boolean isLeaderAvailabilityKnown() {
 537         if (leaderId == null) {
 538             return false;
 539         }
 540
 541         Optional<Cluster> cluster = context.getCluster();
 542         if (!cluster.isPresent()) {
 543             return false;
 544         }
 545
 546         ActorSelection leaderActor = context.getPeerActorSelection(leaderId);
 547         if (leaderActor == null) {
 548             return false;
 549         }
 550
 551         Address leaderAddress = leaderActor.anchorPath().address();
 552
 553         CurrentClusterState state = cluster.get().state();
 554         Set<Member> unreachable = state.getUnreachable();
 555
 556         log.debug("{}: Checking for leader {} in the cluster unreachable set {}", logName(), leaderAddress,
 557                 unreachable);
 558
 559         for (Member m: unreachable) {
 560             if (leaderAddress.equals(m.address())) {
 561                 log.info("{}: Leader {} is unreachable", logName(), leaderAddress);
 562                 return false;
 563             }
 564         }
 565
 566         for (Member m: state.getMembers()) {
 567             if (leaderAddress.equals(m.address())) {
 568                 if (m.status() == MemberStatus.up() || m.status() == MemberStatus.weaklyUp()) {
 569                     log.debug("{}: Leader {} cluster status is {} - leader is available", logName(),
 570                             leaderAddress, m.status());
 571                     return true;
 572                 } else {
 573                     log.debug("{}: Leader {} cluster status is {} - leader is unavailable", logName(),
 574                             leaderAddress, m.status());
 575                     return false;
 576                 }
 577             }
 578         }
 579
 580         log.debug("{}: Leader {} not found in the cluster member set", logName(), leaderAddress);
 581
 582         return false;
 583     }
 584
 585     private boolean isThisFollowerIsolated() {
 586         final Optional<Cluster> maybeCluster = context.getCluster();
 587         if (!maybeCluster.isPresent()) {
 588             return false;
 589         }
 590
 591         final Cluster cluster = maybeCluster.get();
 592         final Member selfMember = cluster.selfMember();
 593
 594         final CurrentClusterState state = cluster.state();
 595         final Set<Member> unreachable = state.getUnreachable();
 596         final Iterable<Member> members = state.getMembers();
 597
 598         log.debug("{}: Checking if this node is isolated in the cluster unreachable set {},"
 599                         + "all members {} self member: {}", logName(), unreachable, members, selfMember);
 600
 601         // no unreachable peers means we cannot be isolated
 602         if (unreachable.isEmpty()) {
 603             return false;
 604         }
 605
 606         final Set<Member> membersToCheck = new HashSet<>();
 607         members.forEach(membersToCheck::add);
 608
 609         membersToCheck.removeAll(unreachable);
 610
 611         // check if the only member not unreachable is us
 612         return membersToCheck.size() == 1 && membersToCheck.iterator().next().equals(selfMember);
 613     }
 614
 615     private void handleInstallSnapshot(final ActorRef sender, final InstallSnapshot installSnapshot) {
 616
 617         log.debug("{}: handleInstallSnapshot: {}", logName(), installSnapshot);
 618
 619         leaderId = installSnapshot.getLeaderId();
 620
 621         if (snapshotTracker == null) {
 622             snapshotTracker = new SnapshotTracker(log, installSnapshot.getTotalChunks(), installSnapshot.getLeaderId(),
 623                     context);
 624         }
 625
 626         updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId());
 627
 628         try {
 629             final InstallSnapshotReply reply = new InstallSnapshotReply(
 630                     currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true);
 631
 632             if (snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(),
 633                     installSnapshot.getLastChunkHashCode())) {
 634
 635                 log.info("{}: Snapshot installed from leader: {}", logName(), installSnapshot.getLeaderId());
 636
 637                 Snapshot snapshot = Snapshot.create(
 638                         context.getSnapshotManager().convertSnapshot(snapshotTracker.getSnapshotBytes()),
 639                         List.of(),
 640                         installSnapshot.getLastIncludedIndex(),
 641                         installSnapshot.getLastIncludedTerm(),
 642                         installSnapshot.getLastIncludedIndex(),
 643                         installSnapshot.getLastIncludedTerm(),
 644                         context.getTermInformation().getCurrentTerm(),
 645                         context.getTermInformation().getVotedFor(),
 646                         installSnapshot.getServerConfig().orElse(null));
 647
 648                 ApplySnapshot.Callback applySnapshotCallback = new ApplySnapshot.Callback() {
 649                     @Override
 650                     public void onSuccess() {
 651                         log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 652
 653                         sender.tell(reply, actor());
 654                     }
 655
 656                     @Override
 657                     public void onFailure() {
 658                         sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor());
 659                     }
 660                 };
 661
 662                 actor().tell(new ApplySnapshot(snapshot, applySnapshotCallback), actor());
 663
 664                 closeSnapshotTracker();
 665             } else {
 666                 log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 667
 668                 sender.tell(reply, actor());
 669             }
 670         } catch (IOException e) {
 671             log.debug("{}: Exception in InstallSnapshot of follower", logName(), e);
 672
 673             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor());
 674
 675             closeSnapshotTracker();
 676         }
 677     }
 678
 679     private void closeSnapshotTracker() {
 680         if (snapshotTracker != null) {
 681             snapshotTracker.close();
 682             snapshotTracker = null;
 683         }
 684     }
 685
 686     @Override
 687     public void close() {
 688         closeSnapshotTracker();
 689         stopElection();
 690         appendEntriesMessageAssembler.close();
 691     }
 692
 693     @VisibleForTesting
 694     SnapshotTracker getSnapshotTracker() {
 695         return snapshotTracker;
 696     }
 697 }