package org.opendaylight.controller.cluster.raft.behaviors;
import akka.actor.ActorRef;
-import akka.actor.Cancellable;
+import akka.actor.ActorSelection;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Preconditions;
+import com.google.common.base.Stopwatch;
+import java.util.concurrent.TimeUnit;
+import javax.annotation.Nonnull;
+import javax.annotation.Nullable;
+import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
import org.opendaylight.controller.cluster.raft.RaftActorContext;
-import org.opendaylight.controller.cluster.raft.base.messages.InitiateInstallSnapshot;
+import org.opendaylight.controller.cluster.raft.RaftActorLeadershipTransferCohort;
+import org.opendaylight.controller.cluster.raft.RaftState;
+import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
import org.opendaylight.controller.cluster.raft.base.messages.IsolatedLeaderCheck;
-import scala.concurrent.duration.FiniteDuration;
+import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
/**
* The behavior of a RaftActor when it is in the Leader state
* set commitIndex = N (§5.3, §5.4).
*/
public class Leader extends AbstractLeader {
- private Cancellable installSnapshotSchedule = null;
- private Cancellable isolatedLeaderCheckSchedule = null;
+ private static final IsolatedLeaderCheck ISOLATED_LEADER_CHECK = new IsolatedLeaderCheck();
+ private final Stopwatch isolatedLeaderCheck;
+ private @Nullable LeadershipTransferContext leadershipTransferContext;
public Leader(RaftActorContext context) {
super(context);
-
- scheduleInstallSnapshotCheck(context.getConfigParams().getIsolatedCheckInterval());
-
- scheduleIsolatedLeaderCheck(
- new FiniteDuration(context.getConfigParams().getHeartBeatInterval().length() * 10,
- context.getConfigParams().getHeartBeatInterval().unit()));
+ isolatedLeaderCheck = Stopwatch.createStarted();
}
@Override public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
if (originalMessage instanceof IsolatedLeaderCheck) {
if (isLeaderIsolated()) {
- LOG.info("At least {} followers need to be active, Switching {} from Leader to IsolatedLeader",
- minIsolatedLeaderPeerCount, leaderId);
- return switchBehavior(new IsolatedLeader(context));
+ LOG.warn("{}: At least {} followers need to be active, Switching {} from Leader to IsolatedLeader",
+ context.getId(), getMinIsolatedLeaderPeerCount(), leaderId);
+
+ return internalSwitchBehavior(RaftState.IsolatedLeader);
}
}
return super.handleMessage(sender, originalMessage);
}
- protected void stopInstallSnapshotSchedule() {
- if (installSnapshotSchedule != null && !installSnapshotSchedule.isCancelled()) {
- installSnapshotSchedule.cancel();
+ @Override
+ protected void beforeSendHeartbeat(){
+ if(isolatedLeaderCheck.elapsed(TimeUnit.MILLISECONDS) > context.getConfigParams().getIsolatedCheckIntervalInMillis()){
+ context.getActor().tell(ISOLATED_LEADER_CHECK, context.getActor());
+ isolatedLeaderCheck.reset().start();
}
- }
- protected void scheduleInstallSnapshotCheck(FiniteDuration interval) {
- if (getFollowerIds().isEmpty()) {
- // Optimization - do not bother scheduling a heartbeat as there are
- // no followers
- return;
+ if(leadershipTransferContext != null && leadershipTransferContext.isExpired(
+ context.getConfigParams().getElectionTimeOutInterval().toMillis())) {
+ LOG.debug("{}: Leadership transfer expired", logName());
+ leadershipTransferContext = null;
}
+ }
- stopInstallSnapshotSchedule();
+ @Override
+ protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender, AppendEntriesReply appendEntriesReply) {
+ RaftActorBehavior returnBehavior = super.handleAppendEntriesReply(sender, appendEntriesReply);
+ tryToCompleteLeadershipTransfer(appendEntriesReply.getFollowerId());
+ return returnBehavior;
+ }
- // Schedule a message to send append entries to followers that can
- // accept an append entries with some data in it
- installSnapshotSchedule =
- context.getActorSystem().scheduler().scheduleOnce(
- interval,
- context.getActor(), new InitiateInstallSnapshot(),
- context.getActorSystem().dispatcher(), context.getActor());
+ /**
+ * Attempts to transfer leadership to a follower as per the raft paper (§3.10) as follows:
+ * <ul>
+ * <li>Start a timer (Stopwatch).</li>
+ * <li>Send an initial AppendEntries heartbeat to all followers.</li>
+ * <li>On AppendEntriesReply, check if the follower's new match Index matches the leader's last index</li>
+ * <li>If it matches, </li>
+ * <ul>
+ * <li>Send an additional AppendEntries to ensure the follower has applied all its log entries to its state.</li>
+ * <li>Send an ElectionTimeout to the follower to immediately start an election.</li>
+ * <li>Notify {@link RaftActorLeadershipTransferCohort#transferComplete}.</li>
+ * </ul>
+ * <li>Otherwise if the election time out period elapses, notify
+ * {@link RaftActorLeadershipTransferCohort#abortTtransfer}.</li>
+ * </ul>
+ *
+ * @param leadershipTransferCohort
+ */
+ public void transferLeadership(@Nonnull RaftActorLeadershipTransferCohort leadershipTransferCohort) {
+ LOG.debug("{}: Attempting to transfer leadership", logName());
+
+ leadershipTransferContext = new LeadershipTransferContext(leadershipTransferCohort);
+
+ // Send an immediate heart beat to the followers.
+ sendAppendEntries(0, false);
}
- protected void stopIsolatedLeaderCheckSchedule() {
- if (isolatedLeaderCheckSchedule != null && !isolatedLeaderCheckSchedule.isCancelled()) {
- isolatedLeaderCheckSchedule.cancel();
+ private void tryToCompleteLeadershipTransfer(String followerId) {
+ if(leadershipTransferContext == null) {
+ return;
}
- }
- protected void scheduleIsolatedLeaderCheck(FiniteDuration isolatedCheckInterval) {
- isolatedLeaderCheckSchedule = context.getActorSystem().scheduler().schedule(isolatedCheckInterval, isolatedCheckInterval,
- context.getActor(), new IsolatedLeaderCheck(),
- context.getActorSystem().dispatcher(), context.getActor());
+ FollowerLogInformation followerInfo = getFollower(followerId);
+ if(followerInfo == null) {
+ return;
+ }
+
+ long lastIndex = context.getReplicatedLog().lastIndex();
+
+ LOG.debug("{}: tryToCompleteLeadershipTransfer: followerId: {}, matchIndex: {}, lastIndex: {}",
+ logName(), followerId, followerInfo.getMatchIndex(), lastIndex);
+
+ if(followerInfo.getMatchIndex() == lastIndex) {
+ LOG.debug("{}: Follower's log matches - sending ElectionTimeout", logName());
+
+ // We can't be sure if the follower has applied all its log entries to its state so send an
+ // additional AppendEntries with the latest commit index.
+ sendAppendEntries(0, false);
+
+ // Now send an ElectionTimeout to the matching follower to immediately start an election.
+ ActorSelection followerActor = context.getPeerActorSelection(followerId);
+ followerActor.tell(new ElectionTimeout(), context.getActor());
+
+ LOG.debug("{}: Leader transfer complete", logName());
+
+ leadershipTransferContext.transferCohort.transferComplete();
+ leadershipTransferContext = null;
+ }
}
@Override
public void close() throws Exception {
- stopInstallSnapshotSchedule();
- stopIsolatedLeaderCheckSchedule();
+ if(leadershipTransferContext != null) {
+ leadershipTransferContext.transferCohort.abortTransfer();
+ }
+
super.close();
}
void markFollowerInActive(String followerId) {
getFollower(followerId).markFollowerInActive();
}
+
+ private static class LeadershipTransferContext {
+ RaftActorLeadershipTransferCohort transferCohort;
+ Stopwatch timer = Stopwatch.createStarted();
+
+ LeadershipTransferContext(RaftActorLeadershipTransferCohort transferCohort) {
+ this.transferCohort = transferCohort;
+ }
+
+ boolean isExpired(long timeout) {
+ if(timer.elapsed(TimeUnit.MILLISECONDS) >= timeout) {
+ transferCohort.abortTransfer();
+ return true;
+ }
+
+ return false;
+ }
+ }
}