opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Leader.java

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8
   9 package org.opendaylight.controller.cluster.raft.behaviors;
  10
  11 import akka.actor.ActorRef;
  12 import akka.actor.ActorSelection;
  13 import akka.actor.Cancellable;
  14 import akka.event.LoggingAdapter;
  15 import com.google.common.base.Preconditions;
  16 import com.google.protobuf.ByteString;
  17 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  18 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  19 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  20 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  21 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  22 import org.opendaylight.controller.cluster.raft.RaftState;
  23 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  24 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  25 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  26 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  27 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  28 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  29 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  30 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  31 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  32 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  33 import scala.concurrent.duration.FiniteDuration;
  34
  35 import java.io.IOException;
  36 import java.util.ArrayList;
  37 import java.util.Collections;
  38 import java.util.HashMap;
  39 import java.util.List;
  40 import java.util.Map;
  41 import java.util.Set;
  42 import java.util.concurrent.TimeUnit;
  43 import java.util.concurrent.atomic.AtomicLong;
  44
  45 /**
  46  * The behavior of a RaftActor when it is in the Leader state
  47  * <p/>
  48  * Leaders:
  49  * <ul>
  50  * <li> Upon election: send initial empty AppendEntries RPCs
  51  * (heartbeat) to each server; repeat during idle periods to
  52  * prevent election timeouts (§5.2)
  53  * <li> If command received from client: append entry to local log,
  54  * respond after entry applied to state machine (§5.3)
  55  * <li> If last log index ≥ nextIndex for a follower: send
  56  * AppendEntries RPC with log entries starting at nextIndex
  57  * <ul>
  58  * <li> If successful: update nextIndex and matchIndex for
  59  * follower (§5.3)
  60  * <li> If AppendEntries fails because of log inconsistency:
  61  * decrement nextIndex and retry (§5.3)
  62  * </ul>
  63  * <li> If there exists an N such that N > commitIndex, a majority
  64  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  65  * set commitIndex = N (§5.3, §5.4).
  66  */
  67 public class Leader extends AbstractRaftActorBehavior {
  68
  69
  70     protected final Map<String, FollowerLogInformation> followerToLog =
  71         new HashMap();
  72     protected final Map<String, FollowerToSnapshot> mapFollowerToSnapshot = new HashMap<>();
  73
  74     private final Set<String> followers;
  75
  76     private Cancellable heartbeatSchedule = null;
  77     private Cancellable appendEntriesSchedule = null;
  78     private Cancellable installSnapshotSchedule = null;
  79
  80     private List<ClientRequestTracker> trackerList = new ArrayList<>();
  81
  82     private final int minReplicationCount;
  83
  84     private final LoggingAdapter LOG;
  85
  86     public Leader(RaftActorContext context) {
  87         super(context);
  88
  89         LOG = context.getLogger();
  90
  91         followers = context.getPeerAddresses().keySet();
  92
  93         for (String followerId : followers) {
  94             FollowerLogInformation followerLogInformation =
  95                 new FollowerLogInformationImpl(followerId,
  96                     new AtomicLong(context.getCommitIndex()),
  97                     new AtomicLong(-1));
  98
  99             followerToLog.put(followerId, followerLogInformation);
 100         }
 101
 102         if(LOG.isDebugEnabled()) {
 103             LOG.debug("Election:Leader has following peers:" + followers);
 104         }
 105
 106         if (followers.size() > 0) {
 107             minReplicationCount = (followers.size() + 1) / 2 + 1;
 108         } else {
 109             minReplicationCount = 0;
 110         }
 111
 112
 113         // Immediately schedule a heartbeat
 114         // Upon election: send initial empty AppendEntries RPCs
 115         // (heartbeat) to each server; repeat during idle periods to
 116         // prevent election timeouts (§5.2)
 117         scheduleHeartBeat(new FiniteDuration(0, TimeUnit.SECONDS));
 118
 119         scheduleInstallSnapshotCheck(
 120             new FiniteDuration(context.getConfigParams().getHeartBeatInterval().length() * 1000,
 121                 context.getConfigParams().getHeartBeatInterval().unit())
 122         );
 123
 124     }
 125
 126     @Override protected RaftState handleAppendEntries(ActorRef sender,
 127         AppendEntries appendEntries) {
 128
 129         if(LOG.isDebugEnabled()) {
 130             LOG.debug(appendEntries.toString());
 131         }
 132
 133         return state();
 134     }
 135
 136     @Override protected RaftState handleAppendEntriesReply(ActorRef sender,
 137         AppendEntriesReply appendEntriesReply) {
 138
 139         if(! appendEntriesReply.isSuccess()) {
 140             if(LOG.isDebugEnabled()) {
 141                 LOG.debug(appendEntriesReply.toString());
 142             }
 143         }
 144
 145         // Update the FollowerLogInformation
 146         String followerId = appendEntriesReply.getFollowerId();
 147         FollowerLogInformation followerLogInformation =
 148             followerToLog.get(followerId);
 149
 150         if(followerLogInformation == null){
 151             LOG.error("Unknown follower {}", followerId);
 152             return state();
 153         }
 154
 155         if (appendEntriesReply.isSuccess()) {
 156             followerLogInformation
 157                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 158             followerLogInformation
 159                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 160         } else {
 161
 162             // TODO: When we find that the follower is out of sync with the
 163             // Leader we simply decrement that followers next index by 1.
 164             // Would it be possible to do better than this? The RAFT spec
 165             // does not explicitly deal with it but may be something for us to
 166             // think about
 167
 168             followerLogInformation.decrNextIndex();
 169         }
 170
 171         // Now figure out if this reply warrants a change in the commitIndex
 172         // If there exists an N such that N > commitIndex, a majority
 173         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 174         // set commitIndex = N (§5.3, §5.4).
 175         for (long N = context.getCommitIndex() + 1; ; N++) {
 176             int replicatedCount = 1;
 177
 178             for (FollowerLogInformation info : followerToLog.values()) {
 179                 if (info.getMatchIndex().get() >= N) {
 180                     replicatedCount++;
 181                 }
 182             }
 183
 184             if (replicatedCount >= minReplicationCount) {
 185                 ReplicatedLogEntry replicatedLogEntry =
 186                     context.getReplicatedLog().get(N);
 187                 if (replicatedLogEntry != null
 188                     && replicatedLogEntry.getTerm()
 189                     == currentTerm()) {
 190                     context.setCommitIndex(N);
 191                 }
 192             } else {
 193                 break;
 194             }
 195         }
 196
 197         // Apply the change to the state machine
 198         if (context.getCommitIndex() > context.getLastApplied()) {
 199             applyLogToStateMachine(context.getCommitIndex());
 200         }
 201
 202         return state();
 203     }
 204
 205     protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
 206
 207         ClientRequestTracker toRemove = findClientRequestTracker(logIndex);
 208         if(toRemove != null) {
 209             trackerList.remove(toRemove);
 210         }
 211
 212         return toRemove;
 213     }
 214
 215     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 216         for (ClientRequestTracker tracker : trackerList) {
 217             if (tracker.getIndex() == logIndex) {
 218                 return tracker;
 219             }
 220         }
 221
 222         return null;
 223     }
 224
 225     @Override protected RaftState handleRequestVoteReply(ActorRef sender,
 226         RequestVoteReply requestVoteReply) {
 227         return state();
 228     }
 229
 230     @Override public RaftState state() {
 231         return RaftState.Leader;
 232     }
 233
 234     @Override public RaftState handleMessage(ActorRef sender, Object originalMessage) {
 235         Preconditions.checkNotNull(sender, "sender should not be null");
 236
 237         Object message = fromSerializableMessage(originalMessage);
 238
 239         if (message instanceof RaftRPC) {
 240             RaftRPC rpc = (RaftRPC) message;
 241             // If RPC request or response contains term T > currentTerm:
 242             // set currentTerm = T, convert to follower (§5.1)
 243             // This applies to all RPC messages and responses
 244             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 245                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 246                 return RaftState.Follower;
 247             }
 248         }
 249
 250         try {
 251             if (message instanceof SendHeartBeat) {
 252                 return sendHeartBeat();
 253             } else if(message instanceof SendInstallSnapshot) {
 254                 installSnapshotIfNeeded();
 255             } else if (message instanceof Replicate) {
 256                 replicate((Replicate) message);
 257             } else if (message instanceof InstallSnapshotReply){
 258                 handleInstallSnapshotReply(
 259                     (InstallSnapshotReply) message);
 260             }
 261         } finally {
 262             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 263         }
 264
 265         return super.handleMessage(sender, message);
 266     }
 267
 268     private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
 269         String followerId = reply.getFollowerId();
 270         FollowerToSnapshot followerToSnapshot =
 271             mapFollowerToSnapshot.get(followerId);
 272
 273         if (followerToSnapshot != null &&
 274             followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
 275
 276             if (reply.isSuccess()) {
 277                 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
 278                     //this was the last chunk reply
 279                     if(LOG.isDebugEnabled()) {
 280                         LOG.debug("InstallSnapshotReply received, " +
 281                                 "last chunk received, Chunk:{}. Follower:{} Setting nextIndex:{}",
 282                             reply.getChunkIndex(), followerId,
 283                             context.getReplicatedLog().getSnapshotIndex() + 1
 284                         );
 285                     }
 286
 287                     FollowerLogInformation followerLogInformation =
 288                         followerToLog.get(followerId);
 289                     followerLogInformation.setMatchIndex(
 290                         context.getReplicatedLog().getSnapshotIndex());
 291                     followerLogInformation.setNextIndex(
 292                         context.getReplicatedLog().getSnapshotIndex() + 1);
 293                     mapFollowerToSnapshot.remove(followerId);
 294
 295                     if(LOG.isDebugEnabled()) {
 296                         LOG.debug("followerToLog.get(followerId).getNextIndex().get()=" +
 297                             followerToLog.get(followerId).getNextIndex().get());
 298                     }
 299
 300                 } else {
 301                     followerToSnapshot.markSendStatus(true);
 302                 }
 303             } else {
 304                 LOG.info("InstallSnapshotReply received, " +
 305                         "sending snapshot chunk failed, Will retry, Chunk:{}",
 306                     reply.getChunkIndex()
 307                 );
 308                 followerToSnapshot.markSendStatus(false);
 309             }
 310
 311         } else {
 312             LOG.error("ERROR!!" +
 313                     "FollowerId in InstallSnapshotReply not known to Leader" +
 314                     " or Chunk Index in InstallSnapshotReply not matching {} != {}",
 315                 followerToSnapshot.getChunkIndex(), reply.getChunkIndex()
 316             );
 317         }
 318     }
 319
 320     private void replicate(Replicate replicate) {
 321         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 322
 323         if(LOG.isDebugEnabled()) {
 324             LOG.debug("Replicate message " + logIndex);
 325         }
 326
 327         // Create a tracker entry we will use this later to notify the
 328         // client actor
 329         trackerList.add(
 330             new ClientRequestTrackerImpl(replicate.getClientActor(),
 331                 replicate.getIdentifier(),
 332                 logIndex)
 333         );
 334
 335         if (followers.size() == 0) {
 336             context.setCommitIndex(logIndex);
 337             applyLogToStateMachine(logIndex);
 338         } else {
 339             sendAppendEntries();
 340         }
 341     }
 342
 343     private void sendAppendEntries() {
 344         // Send an AppendEntries to all followers
 345         for (String followerId : followers) {
 346             ActorSelection followerActor = context.getPeerActorSelection(followerId);
 347
 348             if (followerActor != null) {
 349                 FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 350                 long followerNextIndex = followerLogInformation.getNextIndex().get();
 351                 List<ReplicatedLogEntry> entries = Collections.emptyList();
 352
 353                 if (mapFollowerToSnapshot.get(followerId) != null) {
 354                     if (mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
 355                         sendSnapshotChunk(followerActor, followerId);
 356                     }
 357
 358                 } else {
 359
 360                     if (context.getReplicatedLog().isPresent(followerNextIndex)) {
 361                         // FIXME : Sending one entry at a time
 362                         entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
 363
 364                         followerActor.tell(
 365                             new AppendEntries(currentTerm(), context.getId(),
 366                                 prevLogIndex(followerNextIndex),
 367                                 prevLogTerm(followerNextIndex), entries,
 368                                 context.getCommitIndex()).toSerializable(),
 369                             actor()
 370                         );
 371
 372                     } else {
 373                         // if the followers next index is not present in the leaders log, then snapshot should be sent
 374                         long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 375                         long leaderLastIndex = context.getReplicatedLog().lastIndex();
 376                         if (followerNextIndex >= 0 && leaderLastIndex >= followerNextIndex ) {
 377                             // if the follower is just not starting and leader's index
 378                             // is more than followers index
 379                             if(LOG.isDebugEnabled()) {
 380                                 LOG.debug("SendInstallSnapshot to follower:{}," +
 381                                         "follower-nextIndex:{}, leader-snapshot-index:{},  " +
 382                                         "leader-last-index:{}", followerId,
 383                                     followerNextIndex, leaderSnapShotIndex, leaderLastIndex
 384                                 );
 385                             }
 386
 387                             actor().tell(new SendInstallSnapshot(), actor());
 388                         } else {
 389                             followerActor.tell(
 390                                 new AppendEntries(currentTerm(), context.getId(),
 391                                     prevLogIndex(followerNextIndex),
 392                                     prevLogTerm(followerNextIndex), entries,
 393                                     context.getCommitIndex()).toSerializable(),
 394                                 actor()
 395                             );
 396                         }
 397                     }
 398                 }
 399             }
 400         }
 401     }
 402
 403     /**
 404      * An installSnapshot is scheduled at a interval that is a multiple of
 405      * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
 406      * snapshots at every heartbeat.
 407      */
 408     private void installSnapshotIfNeeded(){
 409         for (String followerId : followers) {
 410             ActorSelection followerActor =
 411                 context.getPeerActorSelection(followerId);
 412
 413             if(followerActor != null) {
 414                 FollowerLogInformation followerLogInformation =
 415                     followerToLog.get(followerId);
 416
 417                 long nextIndex = followerLogInformation.getNextIndex().get();
 418
 419                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 420                     context.getReplicatedLog().isInSnapshot(nextIndex)) {
 421                     sendSnapshotChunk(followerActor, followerId);
 422                 }
 423             }
 424         }
 425     }
 426
 427     /**
 428      *  Sends a snapshot chunk to a given follower
 429      *  InstallSnapshot should qualify as a heartbeat too.
 430      */
 431     private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
 432         try {
 433             followerActor.tell(
 434                 new InstallSnapshot(currentTerm(), context.getId(),
 435                     context.getReplicatedLog().getSnapshotIndex(),
 436                     context.getReplicatedLog().getSnapshotTerm(),
 437                     getNextSnapshotChunk(followerId,
 438                         context.getReplicatedLog().getSnapshot()),
 439                     mapFollowerToSnapshot.get(followerId).incrementChunkIndex(),
 440                     mapFollowerToSnapshot.get(followerId).getTotalChunks()
 441                 ).toSerializable(),
 442                 actor()
 443             );
 444             LOG.info("InstallSnapshot sent to follower {}, Chunk: {}/{}",
 445                 followerActor.path(), mapFollowerToSnapshot.get(followerId).getChunkIndex(),
 446                 mapFollowerToSnapshot.get(followerId).getTotalChunks());
 447         } catch (IOException e) {
 448             LOG.error("InstallSnapshot failed for Leader.", e);
 449         }
 450     }
 451
 452     /**
 453      * Acccepts snaphot as ByteString, enters into map for future chunks
 454      * creates and return a ByteString chunk
 455      */
 456     private ByteString getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
 457         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 458         if (followerToSnapshot == null) {
 459             followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
 460             mapFollowerToSnapshot.put(followerId, followerToSnapshot);
 461         }
 462         ByteString nextChunk = followerToSnapshot.getNextChunk();
 463         if(LOG.isDebugEnabled()) {
 464             LOG.debug("Leader's snapshot nextChunk size:{}", nextChunk.size());
 465         }
 466
 467         return nextChunk;
 468     }
 469
 470     private RaftState sendHeartBeat() {
 471         if (followers.size() > 0) {
 472             sendAppendEntries();
 473         }
 474         return state();
 475     }
 476
 477     private void stopHeartBeat() {
 478         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 479             heartbeatSchedule.cancel();
 480         }
 481     }
 482
 483     private void stopInstallSnapshotSchedule() {
 484         if (installSnapshotSchedule != null && !installSnapshotSchedule.isCancelled()) {
 485             installSnapshotSchedule.cancel();
 486         }
 487     }
 488
 489     private void scheduleHeartBeat(FiniteDuration interval) {
 490         if(followers.size() == 0){
 491             // Optimization - do not bother scheduling a heartbeat as there are
 492             // no followers
 493             return;
 494         }
 495
 496         stopHeartBeat();
 497
 498         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 499         // message is sent to itself.
 500         // Scheduling the heartbeat only once here because heartbeats do not
 501         // need to be sent if there are other messages being sent to the remote
 502         // actor.
 503         heartbeatSchedule =
 504             context.getActorSystem().scheduler().scheduleOnce(
 505                 interval,
 506                 context.getActor(), new SendHeartBeat(),
 507                 context.getActorSystem().dispatcher(), context.getActor());
 508     }
 509
 510
 511     private void scheduleInstallSnapshotCheck(FiniteDuration interval) {
 512         if(followers.size() == 0){
 513             // Optimization - do not bother scheduling a heartbeat as there are
 514             // no followers
 515             return;
 516         }
 517
 518         stopInstallSnapshotSchedule();
 519
 520         // Schedule a message to send append entries to followers that can
 521         // accept an append entries with some data in it
 522         installSnapshotSchedule =
 523             context.getActorSystem().scheduler().scheduleOnce(
 524                 interval,
 525                 context.getActor(), new SendInstallSnapshot(),
 526                 context.getActorSystem().dispatcher(), context.getActor());
 527     }
 528
 529
 530
 531     @Override public void close() throws Exception {
 532         stopHeartBeat();
 533     }
 534
 535     @Override public String getLeaderId() {
 536         return context.getId();
 537     }
 538
 539     /**
 540      * Encapsulates the snapshot bytestring and handles the logic of sending
 541      * snapshot chunks
 542      */
 543     protected class FollowerToSnapshot {
 544         private ByteString snapshotBytes;
 545         private int offset = 0;
 546         // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
 547         private int replyReceivedForOffset;
 548         // if replyStatus is false, the previous chunk is attempted
 549         private boolean replyStatus = false;
 550         private int chunkIndex;
 551         private int totalChunks;
 552
 553         public FollowerToSnapshot(ByteString snapshotBytes) {
 554             this.snapshotBytes = snapshotBytes;
 555             replyReceivedForOffset = -1;
 556             chunkIndex = 1;
 557             int size = snapshotBytes.size();
 558             totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
 559                 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
 560             if(LOG.isDebugEnabled()) {
 561                 LOG.debug("Snapshot {} bytes, total chunks to send:{}",
 562                     size, totalChunks);
 563             }
 564         }
 565
 566         public ByteString getSnapshotBytes() {
 567             return snapshotBytes;
 568         }
 569
 570         public int incrementOffset() {
 571             if(replyStatus) {
 572                 // if prev chunk failed, we would want to sent the same chunk again
 573                 offset = offset + context.getConfigParams().getSnapshotChunkSize();
 574             }
 575             return offset;
 576         }
 577
 578         public int incrementChunkIndex() {
 579             if (replyStatus) {
 580                 // if prev chunk failed, we would want to sent the same chunk again
 581                 chunkIndex =  chunkIndex + 1;
 582             }
 583             return chunkIndex;
 584         }
 585
 586         public int getChunkIndex() {
 587             return chunkIndex;
 588         }
 589
 590         public int getTotalChunks() {
 591             return totalChunks;
 592         }
 593
 594         public boolean canSendNextChunk() {
 595             // we only send a false if a chunk is sent but we have not received a reply yet
 596             return replyReceivedForOffset == offset;
 597         }
 598
 599         public boolean isLastChunk(int chunkIndex) {
 600             return totalChunks == chunkIndex;
 601         }
 602
 603         public void markSendStatus(boolean success) {
 604             if (success) {
 605                 // if the chunk sent was successful
 606                 replyReceivedForOffset = offset;
 607                 replyStatus = true;
 608             } else {
 609                 // if the chunk sent was failure
 610                 replyReceivedForOffset = offset;
 611                 replyStatus = false;
 612             }
 613         }
 614
 615         public ByteString getNextChunk() {
 616             int snapshotLength = getSnapshotBytes().size();
 617             int start = incrementOffset();
 618             int size = context.getConfigParams().getSnapshotChunkSize();
 619             if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
 620                 size = snapshotLength;
 621             } else {
 622                 if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
 623                     size = snapshotLength - start;
 624                 }
 625             }
 626
 627             if(LOG.isDebugEnabled()) {
 628                 LOG.debug("length={}, offset={},size={}",
 629                     snapshotLength, start, size);
 630             }
 631             return getSnapshotBytes().substring(start, start + size);
 632
 633         }
 634     }
 635
 636 }