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 com.google.common.annotations.VisibleForTesting;
  13 import java.util.ArrayList;
  14 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  15 import org.opendaylight.controller.cluster.raft.RaftState;
  16 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  17 import org.opendaylight.controller.cluster.raft.Snapshot;
  18 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  19 import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
  20 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  21 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  22 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  23 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  24 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  25 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  26 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  27
  28 /**
  29  * The behavior of a RaftActor in the Follower state
  30  * <p/>
  31  * <ul>
  32  * <li> Respond to RPCs from candidates and leaders
  33  * <li> If election timeout elapses without receiving AppendEntries
  34  * RPC from current leader or granting vote to candidate:
  35  * convert to candidate
  36  * </ul>
  37  */
  38 public class Follower extends AbstractRaftActorBehavior {
  39
  40
  41
  42     private SnapshotTracker snapshotTracker = null;
  43
  44     private final InitialSyncStatusTracker initialSyncStatusTracker;
  45
  46     public Follower(RaftActorContext context) {
  47         super(context, RaftState.Follower);
  48
  49         initialSyncStatusTracker = new InitialSyncStatusTracker(context.getActor());
  50
  51         if(context.getRaftPolicy().automaticElectionsEnabled()) {
  52             if (context.getPeerAddresses().isEmpty()) {
  53                 actor().tell(ELECTION_TIMEOUT, actor());
  54             } else {
  55                 scheduleElection(electionDuration());
  56             }
  57         }
  58
  59     }
  60
  61     private boolean isLogEntryPresent(long index){
  62         if(index == context.getReplicatedLog().getSnapshotIndex()){
  63             return true;
  64         }
  65
  66         ReplicatedLogEntry previousEntry = context.getReplicatedLog()
  67                 .get(index);
  68
  69         return previousEntry != null;
  70
  71     }
  72
  73     private long getLogEntryTerm(long index){
  74         if(index == context.getReplicatedLog().getSnapshotIndex()){
  75             return context.getReplicatedLog().getSnapshotTerm();
  76         }
  77
  78         ReplicatedLogEntry previousEntry = context.getReplicatedLog()
  79                 .get(index);
  80
  81         if(previousEntry != null){
  82             return previousEntry.getTerm();
  83         }
  84
  85         return -1;
  86     }
  87
  88     private void updateInitialSyncStatus(long currentLeaderCommit, String leaderId){
  89         initialSyncStatusTracker.update(leaderId, currentLeaderCommit, context.getCommitIndex());
  90     }
  91
  92     @Override protected RaftActorBehavior handleAppendEntries(ActorRef sender,
  93                                                               AppendEntries appendEntries) {
  94
  95         int numLogEntries = appendEntries.getEntries() != null ? appendEntries.getEntries().size() : 0;
  96         if(LOG.isTraceEnabled()) {
  97             LOG.trace("{}: handleAppendEntries: {}", logName(), appendEntries);
  98         } else if(LOG.isDebugEnabled() && numLogEntries > 0) {
  99             LOG.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 100         }
 101
 102         // TODO : Refactor this method into a bunch of smaller methods
 103         // to make it easier to read. Before refactoring ensure tests
 104         // cover the code properly
 105
 106         if (snapshotTracker != null) {
 107             // if snapshot install is in progress, follower should just acknowledge append entries with a reply.
 108             AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 109                     lastIndex(), lastTerm(), context.getPayloadVersion());
 110
 111             if(LOG.isDebugEnabled()) {
 112                 LOG.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply);
 113             }
 114             sender.tell(reply, actor());
 115
 116             return this;
 117         }
 118
 119         // If we got here then we do appear to be talking to the leader
 120         leaderId = appendEntries.getLeaderId();
 121
 122         setLeaderPayloadVersion(appendEntries.getPayloadVersion());
 123
 124         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 125         // First check if the logs are in sync or not
 126         long lastIndex = lastIndex();
 127
 128         if (isOutOfSync(appendEntries)) {
 129             // We found that the log was out of sync so just send a negative
 130             // reply and return
 131
 132             LOG.debug("{}: Follower is out-of-sync, so sending negative reply, lastIndex: {}, lastTerm: {}",
 133                         logName(), lastIndex, lastTerm());
 134
 135             sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 136                     lastTerm(), context.getPayloadVersion()), actor());
 137             return this;
 138         }
 139
 140         if (appendEntries.getEntries() != null && appendEntries.getEntries().size() > 0) {
 141
 142             LOG.debug("{}: Number of entries to be appended = {}", logName(),
 143                         appendEntries.getEntries().size());
 144
 145             // 3. If an existing entry conflicts with a new one (same index
 146             // but different terms), delete the existing entry and all that
 147             // follow it (§5.3)
 148             int addEntriesFrom = 0;
 149             if (context.getReplicatedLog().size() > 0) {
 150
 151                 // Find the entry up until which the one that is not in the follower's log
 152                 for (int i = 0;i < appendEntries.getEntries().size(); i++, addEntriesFrom++) {
 153                     ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i);
 154                     ReplicatedLogEntry newEntry = context.getReplicatedLog().get(matchEntry.getIndex());
 155
 156                     if (newEntry == null) {
 157                         //newEntry not found in the log
 158                         break;
 159                     }
 160
 161                     if (newEntry.getTerm() == matchEntry.getTerm()) {
 162                         continue;
 163                     }
 164
 165                     LOG.debug("{}: Removing entries from log starting at {}", logName(),
 166                                 matchEntry.getIndex());
 167
 168                     // Entries do not match so remove all subsequent entries
 169                     context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex());
 170                     break;
 171                 }
 172             }
 173
 174             lastIndex = lastIndex();
 175             LOG.debug("{}: After cleanup entries to be added from = {}", logName(),
 176                         (addEntriesFrom + lastIndex));
 177
 178             // 4. Append any new entries not already in the log
 179             for (int i = addEntriesFrom; i < appendEntries.getEntries().size(); i++) {
 180                 ReplicatedLogEntry entry = appendEntries.getEntries().get(i);
 181
 182                 LOG.debug("{}: Append entry to log {}", logName(), entry.getData());
 183
 184                 context.getReplicatedLog().appendAndPersist(entry);
 185             }
 186
 187             LOG.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size());
 188         }
 189
 190         // 5. If leaderCommit > commitIndex, set commitIndex =
 191         // min(leaderCommit, index of last new entry)
 192
 193         lastIndex = lastIndex();
 194         long prevCommitIndex = context.getCommitIndex();
 195
 196         context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex));
 197
 198         if (prevCommitIndex != context.getCommitIndex()) {
 199             LOG.debug("{}: Commit index set to {}", logName(), context.getCommitIndex());
 200         }
 201
 202         // If commitIndex > lastApplied: increment lastApplied, apply
 203         // log[lastApplied] to state machine (§5.3)
 204         // check if there are any entries to be applied. last-applied can be equal to last-index
 205         if (appendEntries.getLeaderCommit() > context.getLastApplied() &&
 206             context.getLastApplied() < lastIndex) {
 207             if(LOG.isDebugEnabled()) {
 208                 LOG.debug("{}: applyLogToStateMachine, " +
 209                         "appendEntries.getLeaderCommit(): {}," +
 210                         "context.getLastApplied(): {}, lastIndex(): {}", logName(),
 211                     appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex);
 212             }
 213
 214             applyLogToStateMachine(appendEntries.getLeaderCommit());
 215         }
 216
 217         AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 218             lastIndex, lastTerm(), context.getPayloadVersion());
 219
 220         if(LOG.isTraceEnabled()) {
 221             LOG.trace("{}: handleAppendEntries returning : {}", logName(), reply);
 222         } else if(LOG.isDebugEnabled() && numLogEntries > 0) {
 223             LOG.debug("{}: handleAppendEntries returning : {}", logName(), reply);
 224         }
 225
 226         sender.tell(reply, actor());
 227
 228         if (!context.getSnapshotManager().isCapturing()) {
 229             super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex());
 230         }
 231
 232         return this;
 233     }
 234
 235     private boolean isOutOfSync(AppendEntries appendEntries) {
 236
 237         long prevLogTerm = getLogEntryTerm(appendEntries.getPrevLogIndex());
 238         boolean prevEntryPresent = isLogEntryPresent(appendEntries.getPrevLogIndex());
 239         long lastIndex = lastIndex();
 240         int numLogEntries = appendEntries.getEntries() != null ? appendEntries.getEntries().size() : 0;
 241         boolean outOfSync = true;
 242
 243         if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) {
 244
 245             // The follower's log is out of sync because the leader does have
 246             // an entry at prevLogIndex and this follower has no entries in
 247             // it's log.
 248
 249             LOG.debug("{}: The followers log is empty and the senders prevLogIndex is {}",
 250                         logName(), appendEntries.getPrevLogIndex());
 251         } else if (lastIndex > -1 && appendEntries.getPrevLogIndex() != -1 && !prevEntryPresent) {
 252
 253             // The follower's log is out of sync because the Leader's
 254             // prevLogIndex entry was not found in it's log
 255
 256             LOG.debug("{}: The log is not empty but the prevLogIndex {} was not found in it",
 257                         logName(), appendEntries.getPrevLogIndex());
 258         } else if (lastIndex > -1 && prevEntryPresent && prevLogTerm != appendEntries.getPrevLogTerm()) {
 259
 260             // The follower's log is out of sync because the Leader's
 261             // prevLogIndex entry does exist in the follower's log but it has
 262             // a different term in it
 263
 264             LOG.debug(
 265                     "{}: Cannot append entries because previous entry term {}  is not equal to append entries prevLogTerm {}",
 266                     logName(), prevLogTerm, appendEntries.getPrevLogTerm());
 267         } else if(appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 268                 && appendEntries.getReplicatedToAllIndex() != -1
 269                 && !isLogEntryPresent(appendEntries.getReplicatedToAllIndex())) {
 270             // This append entry comes from a leader who has it's log aggressively trimmed and so does not have
 271             // the previous entry in it's in-memory journal
 272
 273             LOG.debug(
 274                     "{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the in-memory journal",
 275                     logName(), appendEntries.getReplicatedToAllIndex());
 276         } else if(appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 277                 && appendEntries.getReplicatedToAllIndex() != -1 && numLogEntries > 0 &&
 278                 !isLogEntryPresent(appendEntries.getEntries().get(0).getIndex() - 1)){
 279             LOG.debug(
 280                     "{}: Cannot append entries because the calculated previousIndex {} was not found in the in-memory journal",
 281                     logName(), appendEntries.getEntries().get(0).getIndex() - 1);
 282         } else {
 283             outOfSync = false;
 284         }
 285         return outOfSync;
 286     }
 287
 288     @Override protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 289         AppendEntriesReply appendEntriesReply) {
 290         return this;
 291     }
 292
 293     @Override protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 294         RequestVoteReply requestVoteReply) {
 295         return this;
 296     }
 297
 298     @Override public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 299
 300         Object message = fromSerializableMessage(originalMessage);
 301
 302         if (message instanceof RaftRPC) {
 303             RaftRPC rpc = (RaftRPC) message;
 304             // If RPC request or response contains term T > currentTerm:
 305             // set currentTerm = T, convert to follower (§5.1)
 306             // This applies to all RPC messages and responses
 307             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 308                 LOG.debug("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term",
 309                         logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 310
 311                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 312             }
 313         }
 314
 315         if (message instanceof ElectionTimeout) {
 316             LOG.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
 317             return switchBehavior(new Candidate(context));
 318
 319         } else if (message instanceof InstallSnapshot) {
 320             InstallSnapshot installSnapshot = (InstallSnapshot) message;
 321             handleInstallSnapshot(sender, installSnapshot);
 322         }
 323
 324         scheduleElection(electionDuration());
 325
 326         return super.handleMessage(sender, message);
 327     }
 328
 329     private void handleInstallSnapshot(ActorRef sender, InstallSnapshot installSnapshot) {
 330
 331         LOG.debug("{}: InstallSnapshot received from leader {}, datasize: {} , Chunk: {}/{}",
 332                     logName(), installSnapshot.getLeaderId(), installSnapshot.getData().size(),
 333                     installSnapshot.getChunkIndex(), installSnapshot.getTotalChunks());
 334
 335         if(snapshotTracker == null){
 336             snapshotTracker = new SnapshotTracker(LOG, installSnapshot.getTotalChunks());
 337         }
 338
 339         updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId());
 340
 341         try {
 342             if(snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(),
 343                     installSnapshot.getLastChunkHashCode())){
 344                 Snapshot snapshot = Snapshot.create(snapshotTracker.getSnapshot(),
 345                         new ArrayList<ReplicatedLogEntry>(),
 346                         installSnapshot.getLastIncludedIndex(),
 347                         installSnapshot.getLastIncludedTerm(),
 348                         installSnapshot.getLastIncludedIndex(),
 349                         installSnapshot.getLastIncludedTerm());
 350
 351                 actor().tell(new ApplySnapshot(snapshot), actor());
 352
 353                 snapshotTracker = null;
 354
 355             }
 356
 357             InstallSnapshotReply reply = new InstallSnapshotReply(
 358                     currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true);
 359
 360             LOG.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 361
 362             sender.tell(reply, actor());
 363
 364         } catch (SnapshotTracker.InvalidChunkException e) {
 365             LOG.debug("{}: Exception in InstallSnapshot of follower", logName(), e);
 366
 367             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 368                     -1, false), actor());
 369             snapshotTracker = null;
 370
 371         } catch (Exception e){
 372             LOG.error("{}: Exception in InstallSnapshot of follower", logName(), e);
 373
 374             //send reply with success as false. The chunk will be sent again on failure
 375             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 376                     installSnapshot.getChunkIndex(), false), actor());
 377
 378         }
 379     }
 380
 381     @Override
 382     public void close() throws Exception {
 383         stopElection();
 384     }
 385
 386     @VisibleForTesting
 387     SnapshotTracker getSnapshotTracker(){
 388         return snapshotTracker;
 389     }
 390
 391     private class InitialSyncStatusTracker {
 392
 393         private static final long INVALID_LOG_INDEX = -2L;
 394         private long initialLeaderCommit = INVALID_LOG_INDEX;
 395         private boolean initialSyncUpDone = false;
 396         private String syncedLeaderId = null;
 397         private final ActorRef actor;
 398
 399         public InitialSyncStatusTracker(ActorRef actor) {
 400             this.actor = actor;
 401         }
 402
 403         public void update(String leaderId, long leaderCommit, long commitIndex){
 404
 405             if(!leaderId.equals(syncedLeaderId)){
 406                 initialSyncUpDone = false;
 407                 initialLeaderCommit = INVALID_LOG_INDEX;
 408                 syncedLeaderId = leaderId;
 409             }
 410
 411             if(!initialSyncUpDone){
 412                 if(initialLeaderCommit == INVALID_LOG_INDEX){
 413                     actor.tell(new FollowerInitialSyncUpStatus(false, getId()), ActorRef.noSender());
 414                     initialLeaderCommit = leaderCommit;
 415                 } else if(commitIndex >= initialLeaderCommit){
 416                     actor.tell(new FollowerInitialSyncUpStatus(true, getId()), ActorRef.noSender());
 417                     initialSyncUpDone = true;
 418                 }
 419             }
 420         }
 421     }
 422 }