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