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.japi.Procedure;
  13 import com.google.common.annotations.VisibleForTesting;
  14 import java.util.ArrayList;
  15 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  16 import org.opendaylight.controller.cluster.raft.RaftState;
  17 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  18 import org.opendaylight.controller.cluster.raft.ServerConfigurationPayload;
  19 import org.opendaylight.controller.cluster.raft.Snapshot;
  20 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  21 import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
  22 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  23 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  24 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  25 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  26 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  27 import org.opendaylight.controller.cluster.raft.messages.RequestVote;
  28 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  29
  30 /**
  31  * The behavior of a RaftActor in the Follower state
  32  * <p/>
  33  * <ul>
  34  * <li> Respond to RPCs from candidates and leaders
  35  * <li> If election timeout elapses without receiving AppendEntries
  36  * RPC from current leader or granting vote to candidate:
  37  * convert to candidate
  38  * </ul>
  39  */
  40 public class Follower extends AbstractRaftActorBehavior {
  41     private static final int SYNC_THRESHOLD = 10;
  42
  43     private final SyncStatusTracker initialSyncStatusTracker;
  44
  45     private final Procedure<ReplicatedLogEntry> appendAndPersistCallback = new Procedure<ReplicatedLogEntry>() {
  46         @Override
  47         public void apply(ReplicatedLogEntry logEntry) {
  48             context.getReplicatedLog().captureSnapshotIfReady(logEntry);
  49         }
  50     };
  51
  52     private SnapshotTracker snapshotTracker = null;
  53
  54     public Follower(RaftActorContext context) {
  55         this(context, null);
  56     }
  57
  58     public Follower(RaftActorContext context, String initialLeaderId) {
  59         super(context, RaftState.Follower);
  60         leaderId = initialLeaderId;
  61
  62         initialSyncStatusTracker = new SyncStatusTracker(context.getActor(), getId(), SYNC_THRESHOLD);
  63
  64         if(canStartElection()) {
  65             if (context.getPeerIds().isEmpty() && getLeaderId() == null) {
  66                 actor().tell(ELECTION_TIMEOUT, actor());
  67             } else {
  68                 scheduleElection(electionDuration());
  69             }
  70         }
  71
  72     }
  73
  74     private boolean isLogEntryPresent(long index){
  75         if(index == context.getReplicatedLog().getSnapshotIndex()){
  76             return true;
  77         }
  78
  79         ReplicatedLogEntry previousEntry = context.getReplicatedLog()
  80                 .get(index);
  81
  82         return previousEntry != null;
  83
  84     }
  85
  86     private long getLogEntryTerm(long index){
  87         if(index == context.getReplicatedLog().getSnapshotIndex()){
  88             return context.getReplicatedLog().getSnapshotTerm();
  89         }
  90
  91         ReplicatedLogEntry previousEntry = context.getReplicatedLog()
  92                 .get(index);
  93
  94         if(previousEntry != null){
  95             return previousEntry.getTerm();
  96         }
  97
  98         return -1;
  99     }
 100
 101     private void updateInitialSyncStatus(long currentLeaderCommit, String leaderId){
 102         initialSyncStatusTracker.update(leaderId, currentLeaderCommit, context.getCommitIndex());
 103     }
 104
 105     @Override protected RaftActorBehavior handleAppendEntries(ActorRef sender,
 106                                                               AppendEntries appendEntries) {
 107
 108         int numLogEntries = appendEntries.getEntries() != null ? appendEntries.getEntries().size() : 0;
 109         if(LOG.isTraceEnabled()) {
 110             LOG.trace("{}: handleAppendEntries: {}", logName(), appendEntries);
 111         } else if(LOG.isDebugEnabled() && numLogEntries > 0) {
 112             LOG.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 113         }
 114
 115         // TODO : Refactor this method into a bunch of smaller methods
 116         // to make it easier to read. Before refactoring ensure tests
 117         // cover the code properly
 118
 119         if (snapshotTracker != null || context.getSnapshotManager().isApplying()) {
 120             // if snapshot install is in progress, follower should just acknowledge append entries with a reply.
 121             AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 122                     lastIndex(), lastTerm(), context.getPayloadVersion());
 123
 124             if(LOG.isDebugEnabled()) {
 125                 LOG.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply);
 126             }
 127             sender.tell(reply, actor());
 128
 129             return this;
 130         }
 131
 132         // If we got here then we do appear to be talking to the leader
 133         leaderId = appendEntries.getLeaderId();
 134
 135         setLeaderPayloadVersion(appendEntries.getPayloadVersion());
 136
 137         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 138         // First check if the logs are in sync or not
 139         long lastIndex = lastIndex();
 140
 141         if (isOutOfSync(appendEntries)) {
 142             // We found that the log was out of sync so just send a negative
 143             // reply and return
 144
 145             LOG.debug("{}: Follower is out-of-sync, so sending negative reply, lastIndex: {}, lastTerm: {}",
 146                         logName(), lastIndex, lastTerm());
 147
 148             sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 149                     lastTerm(), context.getPayloadVersion()), actor());
 150             return this;
 151         }
 152
 153         if (appendEntries.getEntries() != null && appendEntries.getEntries().size() > 0) {
 154
 155             LOG.debug("{}: Number of entries to be appended = {}", logName(),
 156                         appendEntries.getEntries().size());
 157
 158             // 3. If an existing entry conflicts with a new one (same index
 159             // but different terms), delete the existing entry and all that
 160             // follow it (§5.3)
 161             int addEntriesFrom = 0;
 162             if (context.getReplicatedLog().size() > 0) {
 163
 164                 // Find the entry up until the one that is not in the follower's log
 165                 for (int i = 0;i < appendEntries.getEntries().size(); i++, addEntriesFrom++) {
 166                     ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i);
 167                     ReplicatedLogEntry newEntry = context.getReplicatedLog().get(matchEntry.getIndex());
 168
 169                     if (newEntry == null) {
 170                         //newEntry not found in the log
 171                         break;
 172                     }
 173
 174                     if (newEntry.getTerm() == matchEntry.getTerm()) {
 175                         continue;
 176                     }
 177
 178                     if(!context.getRaftPolicy().applyModificationToStateBeforeConsensus()) {
 179
 180                         LOG.debug("{}: Removing entries from log starting at {}", logName(),
 181                                 matchEntry.getIndex());
 182
 183                         // Entries do not match so remove all subsequent entries
 184                         context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex());
 185                         break;
 186                     } else {
 187                         sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 188                                 lastTerm(), context.getPayloadVersion(), true), actor());
 189                         return this;
 190                     }
 191                 }
 192             }
 193
 194             lastIndex = lastIndex();
 195             LOG.debug("{}: After cleanup entries to be added from = {}", logName(),
 196                         (addEntriesFrom + lastIndex));
 197
 198             // 4. Append any new entries not already in the log
 199             for (int i = addEntriesFrom; i < appendEntries.getEntries().size(); i++) {
 200                 ReplicatedLogEntry entry = appendEntries.getEntries().get(i);
 201
 202                 LOG.debug("{}: Append entry to log {}", logName(), entry.getData());
 203
 204                 context.getReplicatedLog().appendAndPersist(entry, appendAndPersistCallback);
 205
 206                 if(entry.getData() instanceof ServerConfigurationPayload) {
 207                     context.updatePeerIds((ServerConfigurationPayload)entry.getData());
 208                 }
 209             }
 210
 211             LOG.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size());
 212         }
 213
 214         // 5. If leaderCommit > commitIndex, set commitIndex =
 215         // min(leaderCommit, index of last new entry)
 216
 217         lastIndex = lastIndex();
 218         long prevCommitIndex = context.getCommitIndex();
 219
 220         context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex));
 221
 222         if (prevCommitIndex != context.getCommitIndex()) {
 223             LOG.debug("{}: Commit index set to {}", logName(), context.getCommitIndex());
 224         }
 225
 226         // If commitIndex > lastApplied: increment lastApplied, apply
 227         // log[lastApplied] to state machine (§5.3)
 228         // check if there are any entries to be applied. last-applied can be equal to last-index
 229         if (appendEntries.getLeaderCommit() > context.getLastApplied() &&
 230             context.getLastApplied() < lastIndex) {
 231             if(LOG.isDebugEnabled()) {
 232                 LOG.debug("{}: applyLogToStateMachine, " +
 233                         "appendEntries.getLeaderCommit(): {}," +
 234                         "context.getLastApplied(): {}, lastIndex(): {}", logName(),
 235                     appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex);
 236             }
 237
 238             applyLogToStateMachine(appendEntries.getLeaderCommit());
 239         }
 240
 241         AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 242             lastIndex, lastTerm(), context.getPayloadVersion());
 243
 244         if(LOG.isTraceEnabled()) {
 245             LOG.trace("{}: handleAppendEntries returning : {}", logName(), reply);
 246         } else if(LOG.isDebugEnabled() && numLogEntries > 0) {
 247             LOG.debug("{}: handleAppendEntries returning : {}", logName(), reply);
 248         }
 249
 250         sender.tell(reply, actor());
 251
 252         if (!context.getSnapshotManager().isCapturing()) {
 253             super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex());
 254         }
 255
 256         return this;
 257     }
 258
 259     private boolean isOutOfSync(AppendEntries appendEntries) {
 260
 261         long prevLogTerm = getLogEntryTerm(appendEntries.getPrevLogIndex());
 262         boolean prevEntryPresent = isLogEntryPresent(appendEntries.getPrevLogIndex());
 263         long lastIndex = lastIndex();
 264         int numLogEntries = appendEntries.getEntries() != null ? appendEntries.getEntries().size() : 0;
 265         boolean outOfSync = true;
 266
 267         if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) {
 268
 269             // The follower's log is out of sync because the leader does have
 270             // an entry at prevLogIndex and this follower has no entries in
 271             // it's log.
 272
 273             LOG.debug("{}: The followers log is empty and the senders prevLogIndex is {}",
 274                         logName(), appendEntries.getPrevLogIndex());
 275         } else if (lastIndex > -1 && appendEntries.getPrevLogIndex() != -1 && !prevEntryPresent) {
 276
 277             // The follower's log is out of sync because the Leader's
 278             // prevLogIndex entry was not found in it's log
 279
 280             LOG.debug("{}: The log is not empty but the prevLogIndex {} was not found in it - lastIndex: {}, snapshotIndex: {}",
 281                         logName(), appendEntries.getPrevLogIndex(), lastIndex, context.getReplicatedLog().getSnapshotIndex());
 282         } else if (lastIndex > -1 && prevEntryPresent && prevLogTerm != appendEntries.getPrevLogTerm()) {
 283
 284             // The follower's log is out of sync because the Leader's
 285             // prevLogIndex entry does exist in the follower's log but it has
 286             // a different term in it
 287
 288             LOG.debug(
 289                     "{}: Cannot append entries because previous entry term {}  is not equal to append entries prevLogTerm {}",
 290                     logName(), prevLogTerm, appendEntries.getPrevLogTerm());
 291         } else if(appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 292                 && appendEntries.getReplicatedToAllIndex() != -1
 293                 && !isLogEntryPresent(appendEntries.getReplicatedToAllIndex())
 294                 && !context.getReplicatedLog().isInSnapshot(appendEntries.getReplicatedToAllIndex())) {
 295             // This append entry comes from a leader who has it's log aggressively trimmed and so does not have
 296             // the previous entry in it's in-memory journal
 297
 298             LOG.debug(
 299                     "{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the in-memory journal",
 300                     logName(), appendEntries.getReplicatedToAllIndex());
 301         } else if(appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 302                 && appendEntries.getReplicatedToAllIndex() != -1 && numLogEntries > 0
 303                 && !isLogEntryPresent(appendEntries.getEntries().get(0).getIndex() - 1)
 304                 && !context.getReplicatedLog().isInSnapshot(appendEntries.getEntries().get(0).getIndex() - 1)) {
 305             LOG.debug(
 306                     "{}: Cannot append entries because the calculated previousIndex {} was not found in the in-memory journal",
 307                     logName(), appendEntries.getEntries().get(0).getIndex() - 1);
 308         } else {
 309             outOfSync = false;
 310         }
 311         return outOfSync;
 312     }
 313
 314     @Override protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 315         AppendEntriesReply appendEntriesReply) {
 316         return this;
 317     }
 318
 319     @Override protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 320         RequestVoteReply requestVoteReply) {
 321         return this;
 322     }
 323
 324     @Override public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 325
 326         Object message = fromSerializableMessage(originalMessage);
 327
 328         if (message instanceof RaftRPC) {
 329             RaftRPC rpc = (RaftRPC) message;
 330             // If RPC request or response contains term T > currentTerm:
 331             // set currentTerm = T, convert to follower (§5.1)
 332             // This applies to all RPC messages and responses
 333             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 334                 LOG.debug("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term",
 335                         logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 336
 337                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 338             }
 339         }
 340
 341         if (message instanceof ElectionTimeout) {
 342             if(canStartElection()) {
 343                 LOG.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
 344                 return internalSwitchBehavior(RaftState.Candidate);
 345             } else {
 346                 return this;
 347             }
 348
 349         } else if (message instanceof InstallSnapshot) {
 350             InstallSnapshot installSnapshot = (InstallSnapshot) message;
 351             handleInstallSnapshot(sender, installSnapshot);
 352         }
 353
 354         if(message instanceof RaftRPC && (!(message instanceof RequestVote) || (canGrantVote((RequestVote) message)))){
 355             scheduleElection(electionDuration());
 356         }
 357
 358         return super.handleMessage(sender, message);
 359     }
 360
 361     private void handleInstallSnapshot(final ActorRef sender, InstallSnapshot installSnapshot) {
 362
 363         LOG.debug("{}: InstallSnapshot received from leader {}, datasize: {} , Chunk: {}/{}",
 364                     logName(), installSnapshot.getLeaderId(), installSnapshot.getData().size(),
 365                     installSnapshot.getChunkIndex(), installSnapshot.getTotalChunks());
 366
 367         leaderId = installSnapshot.getLeaderId();
 368
 369         if(snapshotTracker == null){
 370             snapshotTracker = new SnapshotTracker(LOG, installSnapshot.getTotalChunks());
 371         }
 372
 373         updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId());
 374
 375         try {
 376             final InstallSnapshotReply reply = new InstallSnapshotReply(
 377                     currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true);
 378
 379             if(snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(),
 380                     installSnapshot.getLastChunkHashCode())){
 381                 Snapshot snapshot = Snapshot.create(snapshotTracker.getSnapshot(),
 382                         new ArrayList<ReplicatedLogEntry>(),
 383                         installSnapshot.getLastIncludedIndex(),
 384                         installSnapshot.getLastIncludedTerm(),
 385                         installSnapshot.getLastIncludedIndex(),
 386                         installSnapshot.getLastIncludedTerm(),
 387                         context.getTermInformation().getCurrentTerm(),
 388                         context.getTermInformation().getVotedFor(),
 389                         context.getPeerServerInfo(true));
 390
 391                 ApplySnapshot.Callback applySnapshotCallback = new ApplySnapshot.Callback() {
 392                     @Override
 393                     public void onSuccess() {
 394                         LOG.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 395
 396                         sender.tell(reply, actor());
 397                     }
 398
 399                     @Override
 400                     public void onFailure() {
 401                         sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor());
 402                     }
 403                 };
 404
 405                 actor().tell(new ApplySnapshot(snapshot, applySnapshotCallback), actor());
 406
 407                 snapshotTracker = null;
 408             } else {
 409                 LOG.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 410
 411                 sender.tell(reply, actor());
 412             }
 413         } catch (SnapshotTracker.InvalidChunkException e) {
 414             LOG.debug("{}: Exception in InstallSnapshot of follower", logName(), e);
 415
 416             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 417                     -1, false), actor());
 418             snapshotTracker = null;
 419
 420         } catch (Exception e){
 421             LOG.error("{}: Exception in InstallSnapshot of follower", logName(), e);
 422
 423             //send reply with success as false. The chunk will be sent again on failure
 424             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 425                     installSnapshot.getChunkIndex(), false), actor());
 426
 427         }
 428     }
 429
 430     @Override
 431     public void close() throws Exception {
 432         stopElection();
 433     }
 434
 435     @VisibleForTesting
 436     SnapshotTracker getSnapshotTracker(){
 437         return snapshotTracker;
 438     }
 439 }