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.internal.messages.ApplyState;
  23 import org.opendaylight.controller.cluster.raft.internal.messages.Replicate;
  24 import org.opendaylight.controller.cluster.raft.internal.messages.SendHeartBeat;
  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.RequestVoteReply;
  28 import scala.concurrent.duration.FiniteDuration;
  29
  30 import java.util.ArrayList;
  31 import java.util.HashMap;
  32 import java.util.List;
  33 import java.util.Map;
  34 import java.util.concurrent.TimeUnit;
  35 import java.util.concurrent.atomic.AtomicLong;
  36
  37 /**
  38  * The behavior of a RaftActor when it is in the Leader state
  39  * <p/>
  40  * Leaders:
  41  * <ul>
  42  * <li> Upon election: send initial empty AppendEntries RPCs
  43  * (heartbeat) to each server; repeat during idle periods to
  44  * prevent election timeouts (§5.2)
  45  * <li> If command received from client: append entry to local log,
  46  * respond after entry applied to state machine (§5.3)
  47  * <li> If last log index ≥ nextIndex for a follower: send
  48  * AppendEntries RPC with log entries starting at nextIndex
  49  * <ul>
  50  * <li> If successful: update nextIndex and matchIndex for
  51  * follower (§5.3)
  52  * <li> If AppendEntries fails because of log inconsistency:
  53  * decrement nextIndex and retry (§5.3)
  54  * </ul>
  55  * <li> If there exists an N such that N > commitIndex, a majority
  56  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  57  * set commitIndex = N (§5.3, §5.4).
  58  */
  59 public class Leader extends AbstractRaftActorBehavior {
  60
  61
  62     private final Map<String, FollowerLogInformation> followerToLog =
  63         new HashMap();
  64
  65     private final Map<String, ActorSelection> followerToActor = new HashMap<>();
  66
  67     private Cancellable heartbeatCancel = null;
  68
  69     private List<ClientRequestTracker> trackerList = new ArrayList<>();
  70
  71     private final int minReplicationCount;
  72
  73     public Leader(RaftActorContext context) {
  74         super(context);
  75
  76         if (lastIndex() >= 0) {
  77             context.setCommitIndex(lastIndex());
  78         }
  79
  80         for (String followerId : context.getPeerAddresses().keySet()) {
  81             FollowerLogInformation followerLogInformation =
  82                 new FollowerLogInformationImpl(followerId,
  83                     new AtomicLong(lastIndex()),
  84                     new AtomicLong(-1));
  85
  86             followerToActor.put(followerId,
  87                 context.actorSelection(context.getPeerAddress(followerId)));
  88
  89             followerToLog.put(followerId, followerLogInformation);
  90
  91         }
  92
  93         if (followerToActor.size() > 0) {
  94             minReplicationCount = (followerToActor.size() + 1) / 2 + 1;
  95         } else {
  96             minReplicationCount = 0;
  97         }
  98
  99
 100         // Immediately schedule a heartbeat
 101         // Upon election: send initial empty AppendEntries RPCs
 102         // (heartbeat) to each server; repeat during idle periods to
 103         // prevent election timeouts (§5.2)
 104         scheduleHeartBeat(new FiniteDuration(0, TimeUnit.SECONDS));
 105
 106
 107     }
 108
 109     @Override protected RaftState handleAppendEntries(ActorRef sender,
 110         AppendEntries appendEntries, RaftState suggestedState) {
 111
 112         context.getLogger()
 113             .error("An unexpected AppendEntries received in state " + state());
 114
 115         return suggestedState;
 116     }
 117
 118     @Override protected RaftState handleAppendEntriesReply(ActorRef sender,
 119         AppendEntriesReply appendEntriesReply, RaftState suggestedState) {
 120
 121         // Do not take any other action since a behavior change is coming
 122         if (suggestedState != state())
 123             return suggestedState;
 124
 125         // Update the FollowerLogInformation
 126         String followerId = appendEntriesReply.getFollowerId();
 127         FollowerLogInformation followerLogInformation =
 128             followerToLog.get(followerId);
 129         if (appendEntriesReply.isSuccess()) {
 130             followerLogInformation
 131                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 132             followerLogInformation
 133                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 134         } else {
 135             followerLogInformation.decrNextIndex();
 136         }
 137
 138         // Now figure out if this reply warrants a change in the commitIndex
 139         // If there exists an N such that N > commitIndex, a majority
 140         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 141         // set commitIndex = N (§5.3, §5.4).
 142         for (long N = context.getCommitIndex() + 1; ; N++) {
 143             int replicatedCount = 1;
 144
 145             for (FollowerLogInformation info : followerToLog.values()) {
 146                 if (info.getMatchIndex().get() >= N) {
 147                     replicatedCount++;
 148                 }
 149             }
 150
 151             if (replicatedCount >= minReplicationCount) {
 152                 ReplicatedLogEntry replicatedLogEntry =
 153                     context.getReplicatedLog().get(N);
 154                 if (replicatedLogEntry != null
 155                     && replicatedLogEntry.getTerm()
 156                     == currentTerm()) {
 157                     context.setCommitIndex(N);
 158                 }
 159             } else {
 160                 break;
 161             }
 162         }
 163
 164         // Apply the change to the state machine
 165         if (context.getCommitIndex() > context.getLastApplied()) {
 166             applyLogToStateMachine(context.getCommitIndex());
 167         }
 168
 169         return suggestedState;
 170     }
 171
 172     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 173         for (ClientRequestTracker tracker : trackerList) {
 174             if (tracker.getIndex() == logIndex) {
 175                 return tracker;
 176             }
 177         }
 178
 179         return null;
 180     }
 181
 182     @Override protected RaftState handleRequestVoteReply(ActorRef sender,
 183         RequestVoteReply requestVoteReply, RaftState suggestedState) {
 184         return suggestedState;
 185     }
 186
 187     @Override public RaftState state() {
 188         return RaftState.Leader;
 189     }
 190
 191     @Override public RaftState handleMessage(ActorRef sender, Object message) {
 192         Preconditions.checkNotNull(sender, "sender should not be null");
 193
 194         try {
 195             if (message instanceof SendHeartBeat) {
 196                 return sendHeartBeat();
 197             } else if (message instanceof Replicate) {
 198                 replicate((Replicate) message);
 199             }
 200         } finally {
 201             scheduleHeartBeat(HEART_BEAT_INTERVAL);
 202         }
 203
 204         return super.handleMessage(sender, message);
 205     }
 206
 207     private void replicate(Replicate replicate) {
 208         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 209
 210         context.getLogger().debug("Replicate message " + logIndex);
 211
 212         if (followerToActor.size() == 0) {
 213             context.setCommitIndex(
 214                 replicate.getReplicatedLogEntry().getIndex());
 215
 216             context.getActor()
 217                 .tell(new ApplyState(replicate.getClientActor(),
 218                         replicate.getIdentifier(),
 219                         replicate.getReplicatedLogEntry()),
 220                     context.getActor()
 221                 );
 222         } else {
 223
 224             // Create a tracker entry we will use this later to notify the
 225             // client actor
 226             trackerList.add(
 227                 new ClientRequestTrackerImpl(replicate.getClientActor(),
 228                     replicate.getIdentifier(),
 229                     logIndex)
 230             );
 231
 232             sendAppendEntries();
 233         }
 234     }
 235
 236     private void sendAppendEntries() {
 237         // Send an AppendEntries to all followers
 238         for (String followerId : followerToActor.keySet()) {
 239             ActorSelection followerActor =
 240                 followerToActor.get(followerId);
 241
 242             FollowerLogInformation followerLogInformation =
 243                 followerToLog.get(followerId);
 244
 245             long nextIndex = followerLogInformation.getNextIndex().get();
 246
 247             List<ReplicatedLogEntry> entries =
 248                 context.getReplicatedLog().getFrom(nextIndex);
 249
 250             followerActor.tell(
 251                 new AppendEntries(currentTerm(), context.getId(),
 252                     prevLogIndex(nextIndex), prevLogTerm(nextIndex),
 253                     entries, context.getCommitIndex()
 254                 ),
 255                 actor()
 256             );
 257         }
 258     }
 259
 260     private RaftState sendHeartBeat() {
 261         if (followerToActor.size() > 0) {
 262             sendAppendEntries();
 263         }
 264         return state();
 265     }
 266
 267     private void stopHeartBeat() {
 268         if (heartbeatCancel != null && !heartbeatCancel.isCancelled()) {
 269             heartbeatCancel.cancel();
 270         }
 271     }
 272
 273     private void scheduleHeartBeat(FiniteDuration interval) {
 274         if(followerToActor.keySet().size() == 0){
 275             // Optimization - do not bother scheduling a heartbeat as there are
 276             // no followers
 277             return;
 278         }
 279
 280         stopHeartBeat();
 281
 282         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 283         // message is sent to itself.
 284         // Scheduling the heartbeat only once here because heartbeats do not
 285         // need to be sent if there are other messages being sent to the remote
 286         // actor.
 287         heartbeatCancel =
 288             context.getActorSystem().scheduler().scheduleOnce(
 289                 interval,
 290                 context.getActor(), new SendHeartBeat(),
 291                 context.getActorSystem().dispatcher(), context.getActor());
 292     }
 293
 294     @Override public void close() throws Exception {
 295         stopHeartBeat();
 296     }
 297
 298     @Override public String getLeaderId() {
 299         return context.getId();
 300     }
 301
 302 }