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 org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  16 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  17 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  18 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  19 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  20 import org.opendaylight.controller.cluster.raft.RaftState;
  21 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  22 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  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.util.ArrayList;
  35 import java.util.Collections;
  36 import java.util.HashMap;
  37 import java.util.List;
  38 import java.util.Map;
  39 import java.util.Set;
  40 import java.util.concurrent.TimeUnit;
  41 import java.util.concurrent.atomic.AtomicLong;
  42
  43 /**
  44  * The behavior of a RaftActor when it is in the Leader state
  45  * <p/>
  46  * Leaders:
  47  * <ul>
  48  * <li> Upon election: send initial empty AppendEntries RPCs
  49  * (heartbeat) to each server; repeat during idle periods to
  50  * prevent election timeouts (§5.2)
  51  * <li> If command received from client: append entry to local log,
  52  * respond after entry applied to state machine (§5.3)
  53  * <li> If last log index ≥ nextIndex for a follower: send
  54  * AppendEntries RPC with log entries starting at nextIndex
  55  * <ul>
  56  * <li> If successful: update nextIndex and matchIndex for
  57  * follower (§5.3)
  58  * <li> If AppendEntries fails because of log inconsistency:
  59  * decrement nextIndex and retry (§5.3)
  60  * </ul>
  61  * <li> If there exists an N such that N > commitIndex, a majority
  62  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  63  * set commitIndex = N (§5.3, §5.4).
  64  */
  65 public class Leader extends AbstractRaftActorBehavior {
  66
  67
  68     private final Map<String, FollowerLogInformation> followerToLog =
  69         new HashMap();
  70
  71     private final Set<String> followers;
  72
  73     private Cancellable heartbeatSchedule = null;
  74     private Cancellable appendEntriesSchedule = null;
  75     private Cancellable installSnapshotSchedule = null;
  76
  77     private List<ClientRequestTracker> trackerList = new ArrayList<>();
  78
  79     private final int minReplicationCount;
  80
  81     public Leader(RaftActorContext context) {
  82         super(context);
  83
  84         if (lastIndex() >= 0) {
  85             context.setCommitIndex(lastIndex());
  86         }
  87
  88         followers = context.getPeerAddresses().keySet();
  89
  90         for (String followerId : followers) {
  91             FollowerLogInformation followerLogInformation =
  92                 new FollowerLogInformationImpl(followerId,
  93                     new AtomicLong(lastIndex()),
  94                     new AtomicLong(-1));
  95
  96             followerToLog.put(followerId, followerLogInformation);
  97         }
  98
  99         context.getLogger().debug("Election:Leader has following peers:"+ followers);
 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 RaftState handleAppendEntries(ActorRef sender,
 122         AppendEntries appendEntries) {
 123
 124         context.getLogger().info("Leader: Received {}", appendEntries.toString());
 125
 126         return state();
 127     }
 128
 129     @Override protected RaftState handleAppendEntriesReply(ActorRef sender,
 130         AppendEntriesReply appendEntriesReply) {
 131
 132         if(! appendEntriesReply.isSuccess()) {
 133             context.getLogger()
 134                 .info("Leader: Received {}", appendEntriesReply.toString());
 135         }
 136
 137         // Update the FollowerLogInformation
 138         String followerId = appendEntriesReply.getFollowerId();
 139         FollowerLogInformation followerLogInformation =
 140             followerToLog.get(followerId);
 141
 142         if(followerLogInformation == null){
 143             context.getLogger().error("Unknown follower {}", followerId);
 144             return state();
 145         }
 146
 147         if (appendEntriesReply.isSuccess()) {
 148             followerLogInformation
 149                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 150             followerLogInformation
 151                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 152         } else {
 153
 154             // TODO: When we find that the follower is out of sync with the
 155             // Leader we simply decrement that followers next index by 1.
 156             // Would it be possible to do better than this? The RAFT spec
 157             // does not explicitly deal with it but may be something for us to
 158             // think about
 159
 160             followerLogInformation.decrNextIndex();
 161         }
 162
 163         // Now figure out if this reply warrants a change in the commitIndex
 164         // If there exists an N such that N > commitIndex, a majority
 165         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 166         // set commitIndex = N (§5.3, §5.4).
 167         for (long N = context.getCommitIndex() + 1; ; N++) {
 168             int replicatedCount = 1;
 169
 170             for (FollowerLogInformation info : followerToLog.values()) {
 171                 if (info.getMatchIndex().get() >= N) {
 172                     replicatedCount++;
 173                 }
 174             }
 175
 176             if (replicatedCount >= minReplicationCount) {
 177                 ReplicatedLogEntry replicatedLogEntry =
 178                     context.getReplicatedLog().get(N);
 179                 if (replicatedLogEntry != null
 180                     && replicatedLogEntry.getTerm()
 181                     == currentTerm()) {
 182                     context.setCommitIndex(N);
 183                 }
 184             } else {
 185                 break;
 186             }
 187         }
 188
 189         // Apply the change to the state machine
 190         if (context.getCommitIndex() > context.getLastApplied()) {
 191             applyLogToStateMachine(context.getCommitIndex());
 192         }
 193
 194         return state();
 195     }
 196
 197     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 198         for (ClientRequestTracker tracker : trackerList) {
 199             if (tracker.getIndex() == logIndex) {
 200                 return tracker;
 201             }
 202         }
 203
 204         return null;
 205     }
 206
 207     @Override protected RaftState handleRequestVoteReply(ActorRef sender,
 208         RequestVoteReply requestVoteReply) {
 209         return state();
 210     }
 211
 212     @Override public RaftState state() {
 213         return RaftState.Leader;
 214     }
 215
 216     @Override public RaftState handleMessage(ActorRef sender, Object originalMessage) {
 217         Preconditions.checkNotNull(sender, "sender should not be null");
 218
 219         Object message = fromSerializableMessage(originalMessage);
 220
 221         if (message instanceof RaftRPC) {
 222             RaftRPC rpc = (RaftRPC) message;
 223             // If RPC request or response contains term T > currentTerm:
 224             // set currentTerm = T, convert to follower (§5.1)
 225             // This applies to all RPC messages and responses
 226             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 227                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 228                 return RaftState.Follower;
 229             }
 230         }
 231
 232         try {
 233             if (message instanceof SendHeartBeat) {
 234                 return sendHeartBeat();
 235             } else if(message instanceof SendInstallSnapshot) {
 236                 installSnapshotIfNeeded();
 237             } else if (message instanceof Replicate) {
 238                 replicate((Replicate) message);
 239             } else if (message instanceof InstallSnapshotReply){
 240                 handleInstallSnapshotReply(
 241                     (InstallSnapshotReply) message);
 242             }
 243         } finally {
 244             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 245         }
 246
 247         return super.handleMessage(sender, message);
 248     }
 249
 250     private void handleInstallSnapshotReply(InstallSnapshotReply message) {
 251         InstallSnapshotReply reply = message;
 252         String followerId = reply.getFollowerId();
 253         FollowerLogInformation followerLogInformation =
 254             followerToLog.get(followerId);
 255
 256         followerLogInformation
 257             .setMatchIndex(context.getReplicatedLog().getSnapshotIndex());
 258         followerLogInformation
 259             .setNextIndex(context.getReplicatedLog().getSnapshotIndex() + 1);
 260     }
 261
 262     private void replicate(Replicate replicate) {
 263         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 264
 265         context.getLogger().debug("Replicate message " + logIndex);
 266
 267         if (followers.size() == 0) {
 268             context.setCommitIndex(
 269                 replicate.getReplicatedLogEntry().getIndex());
 270
 271             context.getActor()
 272                 .tell(new ApplyState(replicate.getClientActor(),
 273                         replicate.getIdentifier(),
 274                         replicate.getReplicatedLogEntry()),
 275                     context.getActor()
 276                 );
 277         } else {
 278
 279             // Create a tracker entry we will use this later to notify the
 280             // client actor
 281             trackerList.add(
 282                 new ClientRequestTrackerImpl(replicate.getClientActor(),
 283                     replicate.getIdentifier(),
 284                     logIndex)
 285             );
 286
 287             sendAppendEntries();
 288         }
 289     }
 290
 291     private void sendAppendEntries() {
 292         // Send an AppendEntries to all followers
 293         for (String followerId : followers) {
 294             ActorSelection followerActor =
 295                 context.getPeerActorSelection(followerId);
 296
 297             if (followerActor != null) {
 298                 FollowerLogInformation followerLogInformation =
 299                     followerToLog.get(followerId);
 300
 301                 long nextIndex = followerLogInformation.getNextIndex().get();
 302
 303                 List<ReplicatedLogEntry> entries = Collections.emptyList();
 304
 305                 if (context.getReplicatedLog().isPresent(nextIndex)) {
 306                     // TODO: Instead of sending all entries from nextIndex
 307                     // only send a fixed number of entries to each follower
 308                     // This is to avoid the situation where there are a lot of
 309                     // entries to install for a fresh follower or to a follower
 310                     // that has fallen too far behind with the log but yet is not
 311                     // eligible to receive a snapshot
 312                     entries =
 313                         context.getReplicatedLog().getFrom(nextIndex);
 314                 }
 315
 316                 followerActor.tell(
 317                     new AppendEntries(currentTerm(), context.getId(),
 318                         prevLogIndex(nextIndex),
 319                         prevLogTerm(nextIndex), entries,
 320                         context.getCommitIndex()).toSerializable(),
 321                     actor()
 322                 );
 323             }
 324         }
 325     }
 326
 327     /**
 328      * An installSnapshot is scheduled at a interval that is a multiple of
 329      * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
 330      * snapshots at every heartbeat.
 331      */
 332     private void installSnapshotIfNeeded(){
 333         for (String followerId : followers) {
 334             ActorSelection followerActor =
 335                 context.getPeerActorSelection(followerId);
 336
 337             if(followerActor != null) {
 338                 FollowerLogInformation followerLogInformation =
 339                     followerToLog.get(followerId);
 340
 341                 long nextIndex = followerLogInformation.getNextIndex().get();
 342
 343                 if (!context.getReplicatedLog().isPresent(nextIndex) && context
 344                     .getReplicatedLog().isInSnapshot(nextIndex)) {
 345                     followerActor.tell(
 346                         new InstallSnapshot(currentTerm(), context.getId(),
 347                             context.getReplicatedLog().getSnapshotIndex(),
 348                             context.getReplicatedLog().getSnapshotTerm(),
 349                             context.getReplicatedLog().getSnapshot()
 350                         ),
 351                         actor()
 352                     );
 353                 }
 354             }
 355         }
 356     }
 357
 358     private RaftState sendHeartBeat() {
 359         if (followers.size() > 0) {
 360             sendAppendEntries();
 361         }
 362         return state();
 363     }
 364
 365     private void stopHeartBeat() {
 366         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 367             heartbeatSchedule.cancel();
 368         }
 369     }
 370
 371     private void stopInstallSnapshotSchedule() {
 372         if (installSnapshotSchedule != null && !installSnapshotSchedule.isCancelled()) {
 373             installSnapshotSchedule.cancel();
 374         }
 375     }
 376
 377     private void scheduleHeartBeat(FiniteDuration interval) {
 378         if(followers.size() == 0){
 379             // Optimization - do not bother scheduling a heartbeat as there are
 380             // no followers
 381             return;
 382         }
 383
 384         stopHeartBeat();
 385
 386         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 387         // message is sent to itself.
 388         // Scheduling the heartbeat only once here because heartbeats do not
 389         // need to be sent if there are other messages being sent to the remote
 390         // actor.
 391         heartbeatSchedule =
 392             context.getActorSystem().scheduler().scheduleOnce(
 393                 interval,
 394                 context.getActor(), new SendHeartBeat(),
 395                 context.getActorSystem().dispatcher(), context.getActor());
 396     }
 397
 398
 399     private void scheduleInstallSnapshotCheck(FiniteDuration interval) {
 400         if(followers.size() == 0){
 401             // Optimization - do not bother scheduling a heartbeat as there are
 402             // no followers
 403             return;
 404         }
 405
 406         stopInstallSnapshotSchedule();
 407
 408         // Schedule a message to send append entries to followers that can
 409         // accept an append entries with some data in it
 410         installSnapshotSchedule =
 411             context.getActorSystem().scheduler().scheduleOnce(
 412                 interval,
 413                 context.getActor(), new SendInstallSnapshot(),
 414                 context.getActorSystem().dispatcher(), context.getActor());
 415     }
 416
 417
 418
 419     @Override public void close() throws Exception {
 420         stopHeartBeat();
 421     }
 422
 423     @Override public String getLeaderId() {
 424         return context.getId();
 425     }
 426
 427 }