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 package org.opendaylight.controller.cluster.raft.behaviors;
   9
  10 import akka.actor.ActorRef;
  11 import akka.actor.ActorSelection;
  12 import com.google.common.annotations.VisibleForTesting;
  13 import com.google.common.base.Optional;
  14 import com.google.common.base.Preconditions;
  15 import com.google.common.base.Stopwatch;
  16 import java.util.concurrent.TimeUnit;
  17 import javax.annotation.Nonnull;
  18 import javax.annotation.Nullable;
  19 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  20 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  21 import org.opendaylight.controller.cluster.raft.RaftActorLeadershipTransferCohort;
  22 import org.opendaylight.controller.cluster.raft.RaftState;
  23 import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow;
  24 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  25
  26 /**
  27  * The behavior of a RaftActor when it is in the Leader state.
  28  *
  29  * <p>
  30  * Leaders:
  31  * <ul>
  32  * <li> Upon election: send initial empty AppendEntries RPCs
  33  * (heartbeat) to each server; repeat during idle periods to
  34  * prevent election timeouts (§5.2)
  35  * <li> If command received from client: append entry to local log,
  36  * respond after entry applied to state machine (§5.3)
  37  * <li> If last log index ≥ nextIndex for a follower: send
  38  * AppendEntries RPC with log entries starting at nextIndex
  39  * <li> If successful: update nextIndex and matchIndex for
  40  * follower (§5.3)
  41  * <li> If AppendEntries fails because of log inconsistency:
  42  * decrement nextIndex and retry (§5.3)
  43  * <li> If there exists an N such that N &gt; commitIndex, a majority
  44  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  45  * set commitIndex = N (§5.3, §5.4).
  46  * </ul>
  47  */
  48 public class Leader extends AbstractLeader {
  49     /**
  50      * Internal message sent to periodically check if this leader has become isolated and should transition
  51      * to {@link IsolatedLeader}.
  52      */
  53     @VisibleForTesting
  54     static final Object ISOLATED_LEADER_CHECK = new Object();
  55
  56     private final Stopwatch isolatedLeaderCheck = Stopwatch.createStarted();
  57     @Nullable private LeadershipTransferContext leadershipTransferContext;
  58
  59     Leader(RaftActorContext context, @Nullable AbstractLeader initializeFromLeader) {
  60         super(context, RaftState.Leader, initializeFromLeader);
  61     }
  62
  63     public Leader(RaftActorContext context) {
  64         this(context, null);
  65     }
  66
  67     @Override
  68     public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
  69         Preconditions.checkNotNull(sender, "sender should not be null");
  70
  71         if (ISOLATED_LEADER_CHECK.equals(originalMessage)) {
  72             if (isLeaderIsolated()) {
  73                 log.warn("{}: At least {} followers need to be active, Switching {} from Leader to IsolatedLeader",
  74                     context.getId(), getMinIsolatedLeaderPeerCount(), getLeaderId());
  75                 return internalSwitchBehavior(new IsolatedLeader(context, this));
  76             } else {
  77                 return this;
  78             }
  79         } else {
  80             return super.handleMessage(sender, originalMessage);
  81         }
  82     }
  83
  84     @Override
  85     protected void beforeSendHeartbeat() {
  86         if (isolatedLeaderCheck.elapsed(TimeUnit.MILLISECONDS)
  87                 > context.getConfigParams().getIsolatedCheckIntervalInMillis()) {
  88             context.getActor().tell(ISOLATED_LEADER_CHECK, context.getActor());
  89             isolatedLeaderCheck.reset().start();
  90         }
  91
  92         if (leadershipTransferContext != null && leadershipTransferContext.isExpired(
  93                 context.getConfigParams().getElectionTimeOutInterval().toMillis())) {
  94             log.debug("{}: Leadership transfer expired", logName());
  95             leadershipTransferContext = null;
  96         }
  97     }
  98
  99     @Override
 100     protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender, AppendEntriesReply appendEntriesReply) {
 101         RaftActorBehavior returnBehavior = super.handleAppendEntriesReply(sender, appendEntriesReply);
 102         tryToCompleteLeadershipTransfer(appendEntriesReply.getFollowerId());
 103         return returnBehavior;
 104     }
 105
 106     /**
 107      * Attempts to transfer leadership to a follower as per the raft paper (§3.10) as follows:
 108      * <ul>
 109      * <li>Start a timer (Stopwatch).</li>
 110      * <li>Send an initial AppendEntries heartbeat to all followers.</li>
 111      * <li>On AppendEntriesReply, check if the follower's new match Index matches the leader's last index</li>
 112      * <li>If it matches,
 113      *   <ul>
 114      *   <li>Send an additional AppendEntries to ensure the follower has applied all its log entries to its state.</li>
 115      *   <li>Send an ElectionTimeout to the follower to immediately start an election.</li>
 116      *   <li>Notify {@link RaftActorLeadershipTransferCohort#transferComplete}.</li>
 117      *   </ul></li>
 118      * <li>Otherwise if the election time out period elapses, notify
 119      *     {@link RaftActorLeadershipTransferCohort#abortTransfer}.</li>
 120      * </ul>
 121      *
 122      * @param leadershipTransferCohort the cohort participating in the leadership transfer
 123      */
 124     public void transferLeadership(@Nonnull RaftActorLeadershipTransferCohort leadershipTransferCohort) {
 125         log.debug("{}: Attempting to transfer leadership", logName());
 126
 127         leadershipTransferContext = new LeadershipTransferContext(leadershipTransferCohort);
 128
 129         // Send an immediate heart beat to the followers.
 130         sendAppendEntries(0, false);
 131     }
 132
 133     private void tryToCompleteLeadershipTransfer(String followerId) {
 134         if (leadershipTransferContext == null) {
 135             return;
 136         }
 137
 138         final Optional<String> requestedFollowerIdOptional
 139                 = leadershipTransferContext.transferCohort.getRequestedFollowerId();
 140         if (requestedFollowerIdOptional.isPresent() && !requestedFollowerIdOptional.get().equals(followerId)) {
 141             // we want to transfer leadership to specific follower
 142             return;
 143         }
 144
 145         FollowerLogInformation followerInfo = getFollower(followerId);
 146         if (followerInfo == null) {
 147             return;
 148         }
 149
 150         long lastIndex = context.getReplicatedLog().lastIndex();
 151         boolean isVoting = context.getPeerInfo(followerId).isVoting();
 152
 153         log.debug("{}: tryToCompleteLeadershipTransfer: followerId: {}, matchIndex: {}, lastIndex: {}, isVoting: {}",
 154                 logName(), followerId, followerInfo.getMatchIndex(), lastIndex, isVoting);
 155
 156         if (isVoting && followerInfo.getMatchIndex() == lastIndex) {
 157             log.debug("{}: Follower's log matches - sending ElectionTimeout", logName());
 158
 159             // We can't be sure if the follower has applied all its log entries to its state so send an
 160             // additional AppendEntries with the latest commit index.
 161             sendAppendEntries(0, false);
 162
 163             // Now send a TimeoutNow message to the matching follower to immediately start an election.
 164             ActorSelection followerActor = context.getPeerActorSelection(followerId);
 165             followerActor.tell(TimeoutNow.INSTANCE, context.getActor());
 166
 167             log.debug("{}: Leader transfer complete", logName());
 168
 169             leadershipTransferContext.transferCohort.transferComplete();
 170             leadershipTransferContext = null;
 171         }
 172     }
 173
 174     @Override
 175     public void close() {
 176         if (leadershipTransferContext != null) {
 177             LeadershipTransferContext localLeadershipTransferContext = leadershipTransferContext;
 178             leadershipTransferContext = null;
 179             localLeadershipTransferContext.transferCohort.abortTransfer();
 180         }
 181
 182         super.close();
 183     }
 184
 185     @VisibleForTesting
 186     void markFollowerActive(String followerId) {
 187         getFollower(followerId).markFollowerActive();
 188     }
 189
 190     @VisibleForTesting
 191     void markFollowerInActive(String followerId) {
 192         getFollower(followerId).markFollowerInActive();
 193     }
 194
 195     private static class LeadershipTransferContext {
 196         RaftActorLeadershipTransferCohort transferCohort;
 197         Stopwatch timer = Stopwatch.createStarted();
 198
 199         LeadershipTransferContext(RaftActorLeadershipTransferCohort transferCohort) {
 200             this.transferCohort = transferCohort;
 201         }
 202
 203         boolean isExpired(long timeout) {
 204             if (timer.elapsed(TimeUnit.MILLISECONDS) >= timeout) {
 205                 transferCohort.abortTransfer();
 206                 return true;
 207             }
 208
 209             return false;
 210         }
 211     }
 212 }