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.Replicate;
  23 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  24 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  25 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  26 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  27 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  28 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  29 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  30 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  31 import scala.concurrent.duration.FiniteDuration;
  32
  33 import java.util.ArrayList;
  34 import java.util.Collections;
  35 import java.util.HashMap;
  36 import java.util.List;
  37 import java.util.Map;
  38 import java.util.Set;
  39 import java.util.concurrent.TimeUnit;
  40 import java.util.concurrent.atomic.AtomicLong;
  41
  42 /**
  43  * The behavior of a RaftActor when it is in the Leader state
  44  * <p/>
  45  * Leaders:
  46  * <ul>
  47  * <li> Upon election: send initial empty AppendEntries RPCs
  48  * (heartbeat) to each server; repeat during idle periods to
  49  * prevent election timeouts (§5.2)
  50  * <li> If command received from client: append entry to local log,
  51  * respond after entry applied to state machine (§5.3)
  52  * <li> If last log index ≥ nextIndex for a follower: send
  53  * AppendEntries RPC with log entries starting at nextIndex
  54  * <ul>
  55  * <li> If successful: update nextIndex and matchIndex for
  56  * follower (§5.3)
  57  * <li> If AppendEntries fails because of log inconsistency:
  58  * decrement nextIndex and retry (§5.3)
  59  * </ul>
  60  * <li> If there exists an N such that N > commitIndex, a majority
  61  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  62  * set commitIndex = N (§5.3, §5.4).
  63  */
  64 public class Leader extends AbstractRaftActorBehavior {
  65
  66
  67     private final Map<String, FollowerLogInformation> followerToLog =
  68         new HashMap();
  69
  70     private final Set<String> followers;
  71
  72     private Cancellable heartbeatSchedule = null;
  73     private Cancellable appendEntriesSchedule = null;
  74     private Cancellable installSnapshotSchedule = null;
  75
  76     private List<ClientRequestTracker> trackerList = new ArrayList<>();
  77
  78     private final int minReplicationCount;
  79
  80     public Leader(RaftActorContext context) {
  81         super(context);
  82
  83         if (lastIndex() >= 0) {
  84             context.setCommitIndex(lastIndex());
  85         }
  86
  87         followers = context.getPeerAddresses().keySet();
  88
  89         for (String followerId : followers) {
  90             FollowerLogInformation followerLogInformation =
  91                 new FollowerLogInformationImpl(followerId,
  92                     new AtomicLong(lastIndex()),
  93                     new AtomicLong(-1));
  94
  95             followerToLog.put(followerId, followerLogInformation);
  96         }
  97
  98         context.getLogger().debug("Election:Leader has following peers:"+ followers);
  99
 100         if (followers.size() > 0) {
 101             minReplicationCount = (followers.size() + 1) / 2 + 1;
 102         } else {
 103             minReplicationCount = 0;
 104         }
 105
 106
 107         // Immediately schedule a heartbeat
 108         // Upon election: send initial empty AppendEntries RPCs
 109         // (heartbeat) to each server; repeat during idle periods to
 110         // prevent election timeouts (§5.2)
 111         scheduleHeartBeat(new FiniteDuration(0, TimeUnit.SECONDS));
 112
 113         scheduleInstallSnapshotCheck(
 114             new FiniteDuration(context.getConfigParams().getHeartBeatInterval().length() * 1000,
 115                 context.getConfigParams().getHeartBeatInterval().unit())
 116         );
 117
 118     }
 119
 120     @Override protected RaftState handleAppendEntries(ActorRef sender,
 121         AppendEntries appendEntries) {
 122
 123         context.getLogger().debug(appendEntries.toString());
 124
 125         return state();
 126     }
 127
 128     @Override protected RaftState handleAppendEntriesReply(ActorRef sender,
 129         AppendEntriesReply appendEntriesReply) {
 130
 131         if(! appendEntriesReply.isSuccess()) {
 132             context.getLogger()
 133                 .debug(appendEntriesReply.toString());
 134         }
 135
 136         // Update the FollowerLogInformation
 137         String followerId = appendEntriesReply.getFollowerId();
 138         FollowerLogInformation followerLogInformation =
 139             followerToLog.get(followerId);
 140
 141         if(followerLogInformation == null){
 142             context.getLogger().error("Unknown follower {}", followerId);
 143             return state();
 144         }
 145
 146         if (appendEntriesReply.isSuccess()) {
 147             followerLogInformation
 148                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 149             followerLogInformation
 150                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 151         } else {
 152
 153             // TODO: When we find that the follower is out of sync with the
 154             // Leader we simply decrement that followers next index by 1.
 155             // Would it be possible to do better than this? The RAFT spec
 156             // does not explicitly deal with it but may be something for us to
 157             // think about
 158
 159             followerLogInformation.decrNextIndex();
 160         }
 161
 162         // Now figure out if this reply warrants a change in the commitIndex
 163         // If there exists an N such that N > commitIndex, a majority
 164         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 165         // set commitIndex = N (§5.3, §5.4).
 166         for (long N = context.getCommitIndex() + 1; ; N++) {
 167             int replicatedCount = 1;
 168
 169             for (FollowerLogInformation info : followerToLog.values()) {
 170                 if (info.getMatchIndex().get() >= N) {
 171                     replicatedCount++;
 172                 }
 173             }
 174
 175             if (replicatedCount >= minReplicationCount) {
 176                 ReplicatedLogEntry replicatedLogEntry =
 177                     context.getReplicatedLog().get(N);
 178                 if (replicatedLogEntry != null
 179                     && replicatedLogEntry.getTerm()
 180                     == currentTerm()) {
 181                     context.setCommitIndex(N);
 182                 }
 183             } else {
 184                 break;
 185             }
 186         }
 187
 188         // Apply the change to the state machine
 189         if (context.getCommitIndex() > context.getLastApplied()) {
 190             applyLogToStateMachine(context.getCommitIndex());
 191         }
 192
 193         return state();
 194     }
 195
 196     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 197         for (ClientRequestTracker tracker : trackerList) {
 198             if (tracker.getIndex() == logIndex) {
 199                 return tracker;
 200             }
 201         }
 202
 203         return null;
 204     }
 205
 206     @Override protected RaftState handleRequestVoteReply(ActorRef sender,
 207         RequestVoteReply requestVoteReply) {
 208         return state();
 209     }
 210
 211     @Override public RaftState state() {
 212         return RaftState.Leader;
 213     }
 214
 215     @Override public RaftState handleMessage(ActorRef sender, Object originalMessage) {
 216         Preconditions.checkNotNull(sender, "sender should not be null");
 217
 218         Object message = fromSerializableMessage(originalMessage);
 219
 220         if (message instanceof RaftRPC) {
 221             RaftRPC rpc = (RaftRPC) message;
 222             // If RPC request or response contains term T > currentTerm:
 223             // set currentTerm = T, convert to follower (§5.1)
 224             // This applies to all RPC messages and responses
 225             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 226                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 227                 return RaftState.Follower;
 228             }
 229         }
 230
 231         try {
 232             if (message instanceof SendHeartBeat) {
 233                 return sendHeartBeat();
 234             } else if(message instanceof SendInstallSnapshot) {
 235                 installSnapshotIfNeeded();
 236             } else if (message instanceof Replicate) {
 237                 replicate((Replicate) message);
 238             } else if (message instanceof InstallSnapshotReply){
 239                 handleInstallSnapshotReply(
 240                     (InstallSnapshotReply) message);
 241             }
 242         } finally {
 243             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 244         }
 245
 246         return super.handleMessage(sender, message);
 247     }
 248
 249     private void handleInstallSnapshotReply(InstallSnapshotReply message) {
 250         InstallSnapshotReply reply = message;
 251         String followerId = reply.getFollowerId();
 252         FollowerLogInformation followerLogInformation =
 253             followerToLog.get(followerId);
 254
 255         followerLogInformation
 256             .setMatchIndex(context.getReplicatedLog().getSnapshotIndex());
 257         followerLogInformation
 258             .setNextIndex(context.getReplicatedLog().getSnapshotIndex() + 1);
 259     }
 260
 261     private void replicate(Replicate replicate) {
 262         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 263
 264         context.getLogger().debug("Replicate message " + logIndex);
 265
 266         // Create a tracker entry we will use this later to notify the
 267         // client actor
 268         trackerList.add(
 269             new ClientRequestTrackerImpl(replicate.getClientActor(),
 270                 replicate.getIdentifier(),
 271                 logIndex)
 272         );
 273
 274         if (followers.size() == 0) {
 275             context.setCommitIndex(logIndex);
 276             applyLogToStateMachine(logIndex);
 277         } else {
 278             sendAppendEntries();
 279         }
 280     }
 281
 282     private void sendAppendEntries() {
 283         // Send an AppendEntries to all followers
 284         for (String followerId : followers) {
 285             ActorSelection followerActor =
 286                 context.getPeerActorSelection(followerId);
 287
 288             if (followerActor != null) {
 289                 FollowerLogInformation followerLogInformation =
 290                     followerToLog.get(followerId);
 291
 292                 long nextIndex = followerLogInformation.getNextIndex().get();
 293
 294                 List<ReplicatedLogEntry> entries = Collections.emptyList();
 295
 296                 if (context.getReplicatedLog().isPresent(nextIndex)) {
 297                     // FIXME : Sending one entry at a time
 298                     entries =
 299                         context.getReplicatedLog().getFrom(nextIndex, 1);
 300                 }
 301
 302                 followerActor.tell(
 303                     new AppendEntries(currentTerm(), context.getId(),
 304                         prevLogIndex(nextIndex),
 305                         prevLogTerm(nextIndex), entries,
 306                         context.getCommitIndex()).toSerializable(),
 307                     actor()
 308                 );
 309             }
 310         }
 311     }
 312
 313     /**
 314      * An installSnapshot is scheduled at a interval that is a multiple of
 315      * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
 316      * snapshots at every heartbeat.
 317      */
 318     private void installSnapshotIfNeeded(){
 319         for (String followerId : followers) {
 320             ActorSelection followerActor =
 321                 context.getPeerActorSelection(followerId);
 322
 323             if(followerActor != null) {
 324                 FollowerLogInformation followerLogInformation =
 325                     followerToLog.get(followerId);
 326
 327                 long nextIndex = followerLogInformation.getNextIndex().get();
 328
 329                 if (!context.getReplicatedLog().isPresent(nextIndex) && context
 330                     .getReplicatedLog().isInSnapshot(nextIndex)) {
 331                     followerActor.tell(
 332                         new InstallSnapshot(currentTerm(), context.getId(),
 333                             context.getReplicatedLog().getSnapshotIndex(),
 334                             context.getReplicatedLog().getSnapshotTerm(),
 335                             context.getReplicatedLog().getSnapshot()
 336                         ),
 337                         actor()
 338                     );
 339                 }
 340             }
 341         }
 342     }
 343
 344     private RaftState sendHeartBeat() {
 345         if (followers.size() > 0) {
 346             sendAppendEntries();
 347         }
 348         return state();
 349     }
 350
 351     private void stopHeartBeat() {
 352         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 353             heartbeatSchedule.cancel();
 354         }
 355     }
 356
 357     private void stopInstallSnapshotSchedule() {
 358         if (installSnapshotSchedule != null && !installSnapshotSchedule.isCancelled()) {
 359             installSnapshotSchedule.cancel();
 360         }
 361     }
 362
 363     private void scheduleHeartBeat(FiniteDuration interval) {
 364         if(followers.size() == 0){
 365             // Optimization - do not bother scheduling a heartbeat as there are
 366             // no followers
 367             return;
 368         }
 369
 370         stopHeartBeat();
 371
 372         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 373         // message is sent to itself.
 374         // Scheduling the heartbeat only once here because heartbeats do not
 375         // need to be sent if there are other messages being sent to the remote
 376         // actor.
 377         heartbeatSchedule =
 378             context.getActorSystem().scheduler().scheduleOnce(
 379                 interval,
 380                 context.getActor(), new SendHeartBeat(),
 381                 context.getActorSystem().dispatcher(), context.getActor());
 382     }
 383
 384
 385     private void scheduleInstallSnapshotCheck(FiniteDuration interval) {
 386         if(followers.size() == 0){
 387             // Optimization - do not bother scheduling a heartbeat as there are
 388             // no followers
 389             return;
 390         }
 391
 392         stopInstallSnapshotSchedule();
 393
 394         // Schedule a message to send append entries to followers that can
 395         // accept an append entries with some data in it
 396         installSnapshotSchedule =
 397             context.getActorSystem().scheduler().scheduleOnce(
 398                 interval,
 399                 context.getActor(), new SendInstallSnapshot(),
 400                 context.getActorSystem().dispatcher(), context.getActor());
 401     }
 402
 403
 404
 405     @Override public void close() throws Exception {
 406         stopHeartBeat();
 407     }
 408
 409     @Override public String getLeaderId() {
 410         return context.getId();
 411     }
 412
 413 }