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.ArrayList;
  22 import java.util.HashSet;
  23 import java.util.List;
  24 import java.util.Optional;
  25 import java.util.Set;
  26 import java.util.concurrent.TimeUnit;
  27 import java.util.concurrent.atomic.AtomicBoolean;
  28 import java.util.function.Consumer;
  29 import org.eclipse.jdt.annotation.Nullable;
  30 import org.opendaylight.controller.cluster.messaging.MessageAssembler;
  31 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  32 import org.opendaylight.controller.cluster.raft.RaftState;
  33 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  34 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  35 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  36 import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
  37 import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow;
  38 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  39 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  40 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  41 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  42 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  43 import org.opendaylight.controller.cluster.raft.messages.RequestVote;
  44 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  45 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  46 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
  47
  48 /**
  49  * The behavior of a RaftActor in the Follower raft state.
  50  * <ul>
  51  * <li> Respond to RPCs from candidates and leaders
  52  * <li> If election timeout elapses without receiving AppendEntries
  53  * RPC from current leader or granting vote to candidate:
  54  * convert to candidate
  55  * </ul>
  56  */
  57 // Non-final for testing
  58 public class Follower extends AbstractRaftActorBehavior {
  59     private static final long MAX_ELECTION_TIMEOUT_FACTOR = 18;
  60
  61     private final SyncStatusTracker initialSyncStatusTracker;
  62
  63     private final MessageAssembler appendEntriesMessageAssembler;
  64
  65     private final Stopwatch lastLeaderMessageTimer = Stopwatch.createStarted();
  66     private SnapshotTracker snapshotTracker = null;
  67     private String leaderId;
  68     private short leaderPayloadVersion;
  69
  70     public Follower(final RaftActorContext context) {
  71         this(context, null, (short)-1);
  72     }
  73
  74     @SuppressFBWarnings(value = "MC_OVERRIDABLE_METHOD_CALL_IN_CONSTRUCTOR",
  75         justification = "electionDuration() is not final for Candidate override")
  76     public Follower(final RaftActorContext context, final String initialLeaderId,
  77             final short initialLeaderPayloadVersion) {
  78         super(context, RaftState.Follower);
  79         leaderId = initialLeaderId;
  80         leaderPayloadVersion = initialLeaderPayloadVersion;
  81
  82         initialSyncStatusTracker = new SyncStatusTracker(context.getActor(), getId(), context.getConfigParams()
  83             .getSyncIndexThreshold());
  84
  85         appendEntriesMessageAssembler = MessageAssembler.builder().logContext(logName())
  86                 .fileBackedStreamFactory(context.getFileBackedOutputStreamFactory())
  87                 .assembledMessageCallback((message, sender) -> handleMessage(sender, message)).build();
  88
  89         if (context.getPeerIds().isEmpty() && getLeaderId() == null) {
  90             actor().tell(TimeoutNow.INSTANCE, actor());
  91         } else {
  92             scheduleElection(electionDuration());
  93         }
  94     }
  95
  96     @Override
  97     public final String getLeaderId() {
  98         return leaderId;
  99     }
 100
 101     @VisibleForTesting
 102     protected final void setLeaderId(final @Nullable String leaderId) {
 103         this.leaderId = leaderId;
 104     }
 105
 106     @Override
 107     public short getLeaderPayloadVersion() {
 108         return leaderPayloadVersion;
 109     }
 110
 111     @VisibleForTesting
 112     protected final void setLeaderPayloadVersion(final short leaderPayloadVersion) {
 113         this.leaderPayloadVersion = leaderPayloadVersion;
 114     }
 115
 116     private void restartLastLeaderMessageTimer() {
 117         if (lastLeaderMessageTimer.isRunning()) {
 118             lastLeaderMessageTimer.reset();
 119         }
 120
 121         lastLeaderMessageTimer.start();
 122     }
 123
 124     private boolean isLogEntryPresent(final long index) {
 125         if (context.getReplicatedLog().isInSnapshot(index)) {
 126             return true;
 127         }
 128
 129         ReplicatedLogEntry entry = context.getReplicatedLog().get(index);
 130         return entry != null;
 131
 132     }
 133
 134     private void updateInitialSyncStatus(final long currentLeaderCommit, final String newLeaderId) {
 135         initialSyncStatusTracker.update(newLeaderId, currentLeaderCommit, context.getCommitIndex());
 136     }
 137
 138     @Override
 139     protected RaftActorBehavior handleAppendEntries(final ActorRef sender, final AppendEntries appendEntries) {
 140         int numLogEntries = appendEntries.getEntries().size();
 141         if (log.isTraceEnabled()) {
 142             log.trace("{}: handleAppendEntries: {}", logName(), appendEntries);
 143         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 144             log.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 145         }
 146
 147         if (snapshotTracker != null && !snapshotTracker.getLeaderId().equals(appendEntries.getLeaderId())) {
 148             log.debug("{}: snapshot install is in progress but the prior snapshot leaderId {} does not match the "
 149                 + "AppendEntries leaderId {}", logName(), snapshotTracker.getLeaderId(), appendEntries.getLeaderId());
 150             closeSnapshotTracker();
 151         }
 152
 153         if (snapshotTracker != null || context.getSnapshotManager().isApplying()) {
 154             // if snapshot install is in progress, follower should just acknowledge append entries with a reply.
 155             AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 156                     lastIndex(), lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(),
 157                     appendEntries.getLeaderRaftVersion());
 158
 159             log.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply);
 160             sender.tell(reply, actor());
 161
 162             return this;
 163         }
 164
 165         // If we got here then we do appear to be talking to the leader
 166         leaderId = appendEntries.getLeaderId();
 167         leaderPayloadVersion = appendEntries.getPayloadVersion();
 168
 169         if (appendEntries.getLeaderAddress().isPresent()) {
 170             final String address = appendEntries.getLeaderAddress().get();
 171             log.debug("New leader address: {}", address);
 172
 173             context.setPeerAddress(leaderId, address);
 174             context.getConfigParams().getPeerAddressResolver().setResolved(leaderId, address);
 175         }
 176
 177         // First check if the logs are in sync or not
 178         if (isOutOfSync(appendEntries, sender)) {
 179             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 180             return this;
 181         }
 182
 183         if (!processNewEntries(appendEntries, sender)) {
 184             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 185             return this;
 186         }
 187
 188         long lastIndex = lastIndex();
 189         long prevCommitIndex = context.getCommitIndex();
 190
 191         // If leaderCommit > commitIndex, set commitIndex = min(leaderCommit, index of last new entry)
 192         if (appendEntries.getLeaderCommit() > prevCommitIndex) {
 193             context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex));
 194         }
 195
 196         if (prevCommitIndex != context.getCommitIndex()) {
 197             log.debug("{}: Commit index set to {}", logName(), context.getCommitIndex());
 198         }
 199
 200         AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 201                 lastIndex, lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(),
 202                 appendEntries.getLeaderRaftVersion());
 203
 204         if (log.isTraceEnabled()) {
 205             log.trace("{}: handleAppendEntries returning : {}", logName(), reply);
 206         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 207             log.debug("{}: handleAppendEntries returning : {}", logName(), reply);
 208         }
 209
 210         // Reply to the leader before applying any previous state so as not to hold up leader consensus.
 211         sender.tell(reply, actor());
 212
 213         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 214
 215         // If leaderCommit > lastApplied, increment lastApplied and apply log[lastApplied] to state machine (§5.3).
 216         // lastApplied can be equal to lastIndex.
 217         if (appendEntries.getLeaderCommit() > context.getLastApplied() && context.getLastApplied() < lastIndex) {
 218             if (log.isDebugEnabled()) {
 219                 log.debug("{}: applyLogToStateMachine, appendEntries.getLeaderCommit(): {}, "
 220                         + "context.getLastApplied(): {}, lastIndex(): {}", logName(),
 221                     appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex);
 222             }
 223
 224             applyLogToStateMachine(appendEntries.getLeaderCommit());
 225         }
 226
 227         if (!context.getSnapshotManager().isCapturing()) {
 228             super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex());
 229         }
 230
 231         appendEntriesMessageAssembler.checkExpiredAssembledMessageState();
 232
 233         return this;
 234     }
 235
 236     private boolean processNewEntries(final AppendEntries appendEntries, final ActorRef sender) {
 237         int numLogEntries = appendEntries.getEntries().size();
 238         if (numLogEntries == 0) {
 239             return true;
 240         }
 241
 242         log.debug("{}: Number of entries to be appended = {}", logName(), numLogEntries);
 243
 244         long lastIndex = lastIndex();
 245         int addEntriesFrom = 0;
 246
 247         // First check for conflicting entries. If an existing entry conflicts with a new one (same index but different
 248         // term), delete the existing entry and all that follow it (§5.3)
 249         if (context.getReplicatedLog().size() > 0) {
 250             // Find the entry up until the one that is not in the follower's log
 251             for (int i = 0;i < numLogEntries; i++, addEntriesFrom++) {
 252                 ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i);
 253
 254                 if (!isLogEntryPresent(matchEntry.getIndex())) {
 255                     // newEntry not found in the log
 256                     break;
 257                 }
 258
 259                 long existingEntryTerm = getLogEntryTerm(matchEntry.getIndex());
 260
 261                 log.debug("{}: matchEntry {} is present: existingEntryTerm: {}", logName(), matchEntry,
 262                         existingEntryTerm);
 263
 264                 // existingEntryTerm == -1 means it's in the snapshot and not in the log. We don't know
 265                 // what the term was so we'll assume it matches.
 266                 if (existingEntryTerm == -1 || existingEntryTerm == matchEntry.getTerm()) {
 267                     continue;
 268                 }
 269
 270                 if (!context.getRaftPolicy().applyModificationToStateBeforeConsensus()) {
 271                     log.info("{}: Removing entries from log starting at {}, commitIndex: {}, lastApplied: {}",
 272                             logName(), matchEntry.getIndex(), context.getCommitIndex(), context.getLastApplied());
 273
 274                     // Entries do not match so remove all subsequent entries but only if the existing entries haven't
 275                     // been applied to the state yet.
 276                     if (matchEntry.getIndex() <= context.getLastApplied()
 277                             || !context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex())) {
 278                         // Could not remove the entries - this means the matchEntry index must be in the
 279                         // snapshot and not the log. In this case the prior entries are part of the state
 280                         // so we must send back a reply to force a snapshot to completely re-sync the
 281                         // follower's log and state.
 282
 283                         log.info("{}: Could not remove entries - sending reply to force snapshot", logName());
 284                         sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 285                                 lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(),
 286                                 appendEntries.getLeaderRaftVersion()), actor());
 287                         return false;
 288                     }
 289
 290                     break;
 291                 } else {
 292                     sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 293                             lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(),
 294                             appendEntries.getLeaderRaftVersion()), actor());
 295                     return false;
 296                 }
 297             }
 298         }
 299
 300         lastIndex = lastIndex();
 301         log.debug("{}: After cleanup, lastIndex: {}, entries to be added from: {}", logName(), lastIndex,
 302                 addEntriesFrom);
 303
 304         // When persistence successfully completes for each new log entry appended, we need to determine if we
 305         // should capture a snapshot to compact the persisted log. shouldCaptureSnapshot tracks whether or not
 306         // one of the log entries has exceeded the log size threshold whereby a snapshot should be taken. However
 307         // we don't initiate the snapshot at that log entry but rather after the last log entry has been persisted.
 308         // This is done because subsequent log entries after the one that tripped the threshold may have been
 309         // applied to the state already, as the persistence callback occurs async, and we want those entries
 310         // purged from the persisted log as well.
 311         final AtomicBoolean shouldCaptureSnapshot = new AtomicBoolean(false);
 312         final Consumer<ReplicatedLogEntry> appendAndPersistCallback = logEntry -> {
 313             final List<ReplicatedLogEntry> entries = appendEntries.getEntries();
 314             final ReplicatedLogEntry lastEntryToAppend = entries.get(entries.size() - 1);
 315             if (shouldCaptureSnapshot.get() && logEntry == lastEntryToAppend) {
 316                 context.getSnapshotManager().capture(context.getReplicatedLog().last(), getReplicatedToAllIndex());
 317             }
 318         };
 319
 320         // Append any new entries not already in the log
 321         for (int i = addEntriesFrom; i < numLogEntries; i++) {
 322             ReplicatedLogEntry entry = appendEntries.getEntries().get(i);
 323
 324             log.debug("{}: Append entry to log {}", logName(), entry.getData());
 325
 326             context.getReplicatedLog().appendAndPersist(entry, appendAndPersistCallback, false);
 327
 328             shouldCaptureSnapshot.compareAndSet(false,
 329                     context.getReplicatedLog().shouldCaptureSnapshot(entry.getIndex()));
 330
 331             if (entry.getData() instanceof ServerConfigurationPayload) {
 332                 context.updatePeerIds((ServerConfigurationPayload)entry.getData());
 333             }
 334         }
 335
 336         log.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size());
 337
 338         return true;
 339     }
 340
 341     private boolean isOutOfSync(final AppendEntries appendEntries, final ActorRef sender) {
 342
 343         final long lastIndex = lastIndex();
 344         if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) {
 345
 346             // The follower's log is out of sync because the leader does have an entry at prevLogIndex and this
 347             // follower has no entries in it's log.
 348
 349             log.info("{}: The followers log is empty and the senders prevLogIndex is {}", logName(),
 350                 appendEntries.getPrevLogIndex());
 351
 352             sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 353             return true;
 354         }
 355
 356         if (lastIndex > -1) {
 357             if (isLogEntryPresent(appendEntries.getPrevLogIndex())) {
 358                 final long leadersPrevLogTermInFollowersLogOrSnapshot =
 359                         getLogEntryOrSnapshotTerm(appendEntries.getPrevLogIndex());
 360                 if (leadersPrevLogTermInFollowersLogOrSnapshot != appendEntries.getPrevLogTerm()) {
 361
 362                     // The follower's log is out of sync because the Leader's prevLogIndex entry does exist
 363                     // in the follower's log or snapshot but it has a different term.
 364
 365                     log.info("{}: The prevLogIndex {} was found in the log but the term {} is not equal to the append "
 366                         + "entries prevLogTerm {} - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 367                         appendEntries.getPrevLogIndex(), leadersPrevLogTermInFollowersLogOrSnapshot,
 368                         appendEntries.getPrevLogTerm(), lastIndex, context.getReplicatedLog().getSnapshotIndex(),
 369                         context.getReplicatedLog().getSnapshotTerm());
 370
 371                     sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 372                     return true;
 373                 }
 374             } else if (appendEntries.getPrevLogIndex() != -1) {
 375
 376                 // The follower's log is out of sync because the Leader's prevLogIndex entry was not found in it's log
 377
 378                 log.info("{}: The log is not empty but the prevLogIndex {} was not found in it - lastIndex: {}, "
 379                         + "snapshotIndex: {}, snapshotTerm: {}", logName(), appendEntries.getPrevLogIndex(), lastIndex,
 380                         context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm());
 381
 382                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 383                 return true;
 384             }
 385         }
 386
 387         if (appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 388                 && appendEntries.getReplicatedToAllIndex() != -1) {
 389             if (!isLogEntryPresent(appendEntries.getReplicatedToAllIndex())) {
 390                 // This append entry comes from a leader who has it's log aggressively trimmed and so does not have
 391                 // the previous entry in it's in-memory journal
 392
 393                 log.info("{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the "
 394                         + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 395                         appendEntries.getReplicatedToAllIndex(), lastIndex,
 396                         context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm());
 397
 398                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 399                 return true;
 400             }
 401
 402             final List<ReplicatedLogEntry> entries = appendEntries.getEntries();
 403             if (entries.size() > 0 && !isLogEntryPresent(entries.get(0).getIndex() - 1)) {
 404                 log.info("{}: Cannot append entries because the calculated previousIndex {} was not found in the "
 405                         + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 406                         entries.get(0).getIndex() - 1, lastIndex, context.getReplicatedLog().getSnapshotIndex(),
 407                         context.getReplicatedLog().getSnapshotTerm());
 408
 409                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 410                 return true;
 411             }
 412         }
 413
 414         return false;
 415     }
 416
 417     private void sendOutOfSyncAppendEntriesReply(final ActorRef sender, final boolean forceInstallSnapshot,
 418             final short leaderRaftVersion) {
 419         // We found that the log was out of sync so just send a negative reply.
 420         final AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex(),
 421                 lastTerm(), context.getPayloadVersion(), forceInstallSnapshot, needsLeaderAddress(),
 422                 leaderRaftVersion);
 423
 424         log.info("{}: Follower is out-of-sync so sending negative reply: {}", logName(), reply);
 425         sender.tell(reply, actor());
 426     }
 427
 428     private boolean needsLeaderAddress() {
 429         return context.getPeerAddress(leaderId) == null;
 430     }
 431
 432     @Override
 433     protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender,
 434         final AppendEntriesReply appendEntriesReply) {
 435         return this;
 436     }
 437
 438     @Override
 439     protected RaftActorBehavior handleRequestVoteReply(final ActorRef sender,
 440         final RequestVoteReply requestVoteReply) {
 441         return this;
 442     }
 443
 444     @Override
 445     final ApplyState getApplyStateFor(final ReplicatedLogEntry entry) {
 446         return new ApplyState(null, null, entry);
 447     }
 448
 449     @Override
 450     public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) {
 451         if (message instanceof ElectionTimeout || message instanceof TimeoutNow) {
 452             return handleElectionTimeout(message);
 453         }
 454
 455         if (appendEntriesMessageAssembler.handleMessage(message, actor())) {
 456             return this;
 457         }
 458
 459         if (!(message instanceof RaftRPC)) {
 460             // The rest of the processing requires the message to be a RaftRPC
 461             return null;
 462         }
 463
 464         final RaftRPC rpc = (RaftRPC) message;
 465         // If RPC request or response contains term T > currentTerm:
 466         // set currentTerm = T, convert to follower (§5.1)
 467         // This applies to all RPC messages and responses
 468         if (rpc.getTerm() > context.getTermInformation().getCurrentTerm() && shouldUpdateTerm(rpc)) {
 469             log.info("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term",
 470                 logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 471
 472             context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 473         }
 474
 475         if (rpc instanceof InstallSnapshot) {
 476             handleInstallSnapshot(sender, (InstallSnapshot) rpc);
 477             restartLastLeaderMessageTimer();
 478             scheduleElection(electionDuration());
 479             return this;
 480         }
 481
 482         if (!(rpc instanceof RequestVote) || canGrantVote((RequestVote) rpc)) {
 483             restartLastLeaderMessageTimer();
 484             scheduleElection(electionDuration());
 485         }
 486
 487         return super.handleMessage(sender, rpc);
 488     }
 489
 490     private RaftActorBehavior handleElectionTimeout(final Object message) {
 491         // If the message is ElectionTimeout, verify we haven't actually seen a message from the leader
 492         // during the election timeout interval. It may that the election timer expired b/c this actor
 493         // was busy and messages got delayed, in which case leader messages would be backed up in the
 494         // queue but would be processed before the ElectionTimeout message and thus would restart the
 495         // lastLeaderMessageTimer.
 496         long lastLeaderMessageInterval = lastLeaderMessageTimer.elapsed(TimeUnit.MILLISECONDS);
 497         long electionTimeoutInMillis = context.getConfigParams().getElectionTimeOutInterval().toMillis();
 498         boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning()
 499                 || lastLeaderMessageInterval >= electionTimeoutInMillis;
 500
 501         if (canStartElection()) {
 502             if (message instanceof TimeoutNow) {
 503                 log.debug("{}: Received TimeoutNow - switching to Candidate", logName());
 504                 return internalSwitchBehavior(RaftState.Candidate);
 505             } else if (noLeaderMessageReceived) {
 506                 // Check the cluster state to see if the leader is known to be up before we go to Candidate.
 507                 // However if we haven't heard from the leader in a long time even though the cluster state
 508                 // indicates it's up then something is wrong - leader might be stuck indefinitely - so switch
 509                 // to Candidate,
 510                 long maxElectionTimeout = electionTimeoutInMillis * MAX_ELECTION_TIMEOUT_FACTOR;
 511                 if (isLeaderAvailabilityKnown() && lastLeaderMessageInterval < maxElectionTimeout) {
 512                     log.debug("{}: Received ElectionTimeout but leader appears to be available", logName());
 513                     scheduleElection(electionDuration());
 514                 } else if (isThisFollowerIsolated()) {
 515                     log.debug("{}: this follower is isolated. Do not switch to Candidate for now.", logName());
 516                     setLeaderId(null);
 517                     scheduleElection(electionDuration());
 518                 } else {
 519                     log.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
 520                     return internalSwitchBehavior(RaftState.Candidate);
 521                 }
 522             } else {
 523                 log.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout {}",
 524                         logName(), lastLeaderMessageInterval, context.getConfigParams().getElectionTimeOutInterval());
 525                 scheduleElection(electionDuration());
 526             }
 527         } else if (message instanceof ElectionTimeout) {
 528             if (noLeaderMessageReceived) {
 529                 setLeaderId(null);
 530             }
 531
 532             scheduleElection(electionDuration());
 533         }
 534
 535         return this;
 536     }
 537
 538     private boolean isLeaderAvailabilityKnown() {
 539         if (leaderId == null) {
 540             return false;
 541         }
 542
 543         Optional<Cluster> cluster = context.getCluster();
 544         if (!cluster.isPresent()) {
 545             return false;
 546         }
 547
 548         ActorSelection leaderActor = context.getPeerActorSelection(leaderId);
 549         if (leaderActor == null) {
 550             return false;
 551         }
 552
 553         Address leaderAddress = leaderActor.anchorPath().address();
 554
 555         CurrentClusterState state = cluster.get().state();
 556         Set<Member> unreachable = state.getUnreachable();
 557
 558         log.debug("{}: Checking for leader {} in the cluster unreachable set {}", logName(), leaderAddress,
 559                 unreachable);
 560
 561         for (Member m: unreachable) {
 562             if (leaderAddress.equals(m.address())) {
 563                 log.info("{}: Leader {} is unreachable", logName(), leaderAddress);
 564                 return false;
 565             }
 566         }
 567
 568         for (Member m: state.getMembers()) {
 569             if (leaderAddress.equals(m.address())) {
 570                 if (m.status() == MemberStatus.up() || m.status() == MemberStatus.weaklyUp()) {
 571                     log.debug("{}: Leader {} cluster status is {} - leader is available", logName(),
 572                             leaderAddress, m.status());
 573                     return true;
 574                 } else {
 575                     log.debug("{}: Leader {} cluster status is {} - leader is unavailable", logName(),
 576                             leaderAddress, m.status());
 577                     return false;
 578                 }
 579             }
 580         }
 581
 582         log.debug("{}: Leader {} not found in the cluster member set", logName(), leaderAddress);
 583
 584         return false;
 585     }
 586
 587     private boolean isThisFollowerIsolated() {
 588         final Optional<Cluster> maybeCluster = context.getCluster();
 589         if (!maybeCluster.isPresent()) {
 590             return false;
 591         }
 592
 593         final Cluster cluster = maybeCluster.get();
 594         final Member selfMember = cluster.selfMember();
 595
 596         final CurrentClusterState state = cluster.state();
 597         final Set<Member> unreachable = state.getUnreachable();
 598         final Iterable<Member> members = state.getMembers();
 599
 600         log.debug("{}: Checking if this node is isolated in the cluster unreachable set {},"
 601                         + "all members {} self member: {}", logName(), unreachable, members, selfMember);
 602
 603         // no unreachable peers means we cannot be isolated
 604         if (unreachable.size() == 0) {
 605             return false;
 606         }
 607
 608         final Set<Member> membersToCheck = new HashSet<>();
 609         members.forEach(membersToCheck::add);
 610
 611         membersToCheck.removeAll(unreachable);
 612
 613         // check if the only member not unreachable is us
 614         if (membersToCheck.size() == 1 && membersToCheck.iterator().next().equals(selfMember)) {
 615             return true;
 616         }
 617
 618         return false;
 619     }
 620
 621     private void handleInstallSnapshot(final ActorRef sender, final InstallSnapshot installSnapshot) {
 622
 623         log.debug("{}: handleInstallSnapshot: {}", logName(), installSnapshot);
 624
 625         leaderId = installSnapshot.getLeaderId();
 626
 627         if (snapshotTracker == null) {
 628             snapshotTracker = new SnapshotTracker(log, installSnapshot.getTotalChunks(), installSnapshot.getLeaderId(),
 629                     context);
 630         }
 631
 632         updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId());
 633
 634         try {
 635             final InstallSnapshotReply reply = new InstallSnapshotReply(
 636                     currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true);
 637
 638             if (snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(),
 639                     installSnapshot.getLastChunkHashCode())) {
 640
 641                 log.info("{}: Snapshot installed from leader: {}", logName(), installSnapshot.getLeaderId());
 642
 643                 Snapshot snapshot = Snapshot.create(
 644                         context.getSnapshotManager().convertSnapshot(snapshotTracker.getSnapshotBytes()),
 645                         new ArrayList<>(),
 646                         installSnapshot.getLastIncludedIndex(),
 647                         installSnapshot.getLastIncludedTerm(),
 648                         installSnapshot.getLastIncludedIndex(),
 649                         installSnapshot.getLastIncludedTerm(),
 650                         context.getTermInformation().getCurrentTerm(),
 651                         context.getTermInformation().getVotedFor(),
 652                         installSnapshot.getServerConfig().orElse(null));
 653
 654                 ApplySnapshot.Callback applySnapshotCallback = new ApplySnapshot.Callback() {
 655                     @Override
 656                     public void onSuccess() {
 657                         log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 658
 659                         sender.tell(reply, actor());
 660                     }
 661
 662                     @Override
 663                     public void onFailure() {
 664                         sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor());
 665                     }
 666                 };
 667
 668                 actor().tell(new ApplySnapshot(snapshot, applySnapshotCallback), actor());
 669
 670                 closeSnapshotTracker();
 671             } else {
 672                 log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 673
 674                 sender.tell(reply, actor());
 675             }
 676         } catch (IOException e) {
 677             log.debug("{}: Exception in InstallSnapshot of follower", logName(), e);
 678
 679             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 680                     -1, false), actor());
 681
 682             closeSnapshotTracker();
 683         }
 684     }
 685
 686     private void closeSnapshotTracker() {
 687         if (snapshotTracker != null) {
 688             snapshotTracker.close();
 689             snapshotTracker = null;
 690         }
 691     }
 692
 693     @Override
 694     public void close() {
 695         closeSnapshotTracker();
 696         stopElection();
 697         appendEntriesMessageAssembler.close();
 698     }
 699
 700     @VisibleForTesting
 701     SnapshotTracker getSnapshotTracker() {
 702         return snapshotTracker;
 703     }
 704 }