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