X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-akka-raft%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fraft%2Fbehaviors%2FLeader.java;h=3534ac5cf142eda058ceb1e39b90d311b096b20b;hb=9b319f491af1c65705b69e8a182aab5006a2f959;hp=ebcdcd40fb078ebcc16439ec2feaa87b6f62eca4;hpb=d55de2cfa7bc8fe6c85d091a46f08219972b88d2;p=controller.git
diff --git a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Leader.java b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Leader.java
index ebcdcd40fb..3534ac5cf1 100644
--- a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Leader.java
+++ b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Leader.java
@@ -7,17 +7,27 @@
*/
package org.opendaylight.controller.cluster.raft.behaviors;
+import static java.util.Objects.requireNonNull;
+
import akka.actor.ActorRef;
+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.Optional;
import java.util.concurrent.TimeUnit;
+import org.eclipse.jdt.annotation.NonNull;
+import org.eclipse.jdt.annotation.Nullable;
+import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
import org.opendaylight.controller.cluster.raft.RaftActorContext;
-import org.opendaylight.controller.cluster.raft.base.messages.IsolatedLeaderCheck;
+import org.opendaylight.controller.cluster.raft.RaftActorLeadershipTransferCohort;
+import org.opendaylight.controller.cluster.raft.RaftState;
+import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow;
+import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
/**
- * The behavior of a RaftActor when it is in the Leader state
- *
+ * The behavior of a RaftActor when it is in the Leader state.
+ *
+ *
* Leaders:
*
* - Upon election: send initial empty AppendEntries RPCs
@@ -27,61 +37,178 @@ import org.opendaylight.controller.cluster.raft.base.messages.IsolatedLeaderChec
* respond after entry applied to state machine (§5.3)
*
- If last log index ⥠nextIndex for a follower: send
* AppendEntries RPC with log entries starting at nextIndex
- *
* - If successful: update nextIndex and matchIndex for
* follower (§5.3)
*
- If AppendEntries fails because of log inconsistency:
* decrement nextIndex and retry (§5.3)
- *
- * - If there exists an N such that N > commitIndex, a majority
+ *
- If there exists an N such that N > commitIndex, a majority
* of matchIndex[i] ⥠N, and log[N].term == currentTerm:
* set commitIndex = N (§5.3, §5.4).
+ *
*/
public class Leader extends AbstractLeader {
- private static final IsolatedLeaderCheck ISOLATED_LEADER_CHECK = new IsolatedLeaderCheck();
- private final Stopwatch isolatedLeaderCheck;
+ /**
+ * Internal message sent to periodically check if this leader has become isolated and should transition
+ * to {@link IsolatedLeader}.
+ */
+ @VisibleForTesting
+ static final Object ISOLATED_LEADER_CHECK = new Object();
- public Leader(RaftActorContext context) {
- super(context);
- isolatedLeaderCheck = Stopwatch.createStarted();
+ private final Stopwatch isolatedLeaderCheck = Stopwatch.createStarted();
+ private @Nullable LeadershipTransferContext leadershipTransferContext;
+
+ Leader(final RaftActorContext context, @Nullable final AbstractLeader initializeFromLeader) {
+ super(context, RaftState.Leader, initializeFromLeader);
}
- @Override public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
- Preconditions.checkNotNull(sender, "sender should not be null");
+ public Leader(final RaftActorContext context) {
+ this(context, null);
+ }
- if (originalMessage instanceof IsolatedLeaderCheck) {
- if (isLeaderIsolated()) {
- LOG.warn("{}: At least {} followers need to be active, Switching {} from Leader to IsolatedLeader",
- context.getId(), minIsolatedLeaderPeerCount, leaderId);
+ @Override
+ public RaftActorBehavior handleMessage(final ActorRef sender, final Object originalMessage) {
+ requireNonNull(sender, "sender should not be null");
- return switchBehavior(new IsolatedLeader(context));
+ if (ISOLATED_LEADER_CHECK.equals(originalMessage)) {
+ if (isLeaderIsolated()) {
+ log.warn("{}: At least {} followers need to be active, Switching {} from Leader to IsolatedLeader",
+ context.getId(), getMinIsolatedLeaderPeerCount(), getLeaderId());
+ return internalSwitchBehavior(new IsolatedLeader(context, this));
+ } else {
+ return this;
}
+ } else {
+ return super.handleMessage(sender, originalMessage);
}
-
- return super.handleMessage(sender, originalMessage);
}
@Override
- protected void beforeSendHeartbeat(){
- if(isolatedLeaderCheck.elapsed(TimeUnit.MILLISECONDS) > context.getConfigParams().getIsolatedCheckIntervalInMillis()){
+ protected void beforeSendHeartbeat() {
+ if (isolatedLeaderCheck.elapsed(TimeUnit.MILLISECONDS)
+ > context.getConfigParams().getIsolatedCheckIntervalInMillis()) {
context.getActor().tell(ISOLATED_LEADER_CHECK, context.getActor());
isolatedLeaderCheck.reset().start();
}
+ if (leadershipTransferContext != null && leadershipTransferContext.isExpired(
+ context.getConfigParams().getElectionTimeOutInterval().toMillis())) {
+ log.debug("{}: Leadership transfer expired", logName());
+ leadershipTransferContext = null;
+ }
}
@Override
- public void close() throws Exception {
+ protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender,
+ final AppendEntriesReply appendEntriesReply) {
+ RaftActorBehavior returnBehavior = super.handleAppendEntriesReply(sender, appendEntriesReply);
+ tryToCompleteLeadershipTransfer(appendEntriesReply.getFollowerId());
+ return returnBehavior;
+ }
+
+ /**
+ * Attempts to transfer leadership to a follower as per the raft paper (§3.10) as follows:
+ *
+ * - Start a timer (Stopwatch).
+ * - Send an initial AppendEntries heartbeat to all followers.
+ * - On AppendEntriesReply, check if the follower's new match Index matches the leader's last index
+ * - If it matches,
+ *
+ * - Send an additional AppendEntries to ensure the follower has applied all its log entries to its state.
+ * - Send an ElectionTimeout to the follower to immediately start an election.
+ * - Notify {@link RaftActorLeadershipTransferCohort#transferComplete}.
+ *
+ * - Otherwise if the election time out period elapses, notify
+ * {@link RaftActorLeadershipTransferCohort#abortTransfer}.
+ *
+ *
+ * @param leadershipTransferCohort the cohort participating in the leadership transfer
+ */
+ public void transferLeadership(@NonNull final RaftActorLeadershipTransferCohort leadershipTransferCohort) {
+ log.debug("{}: Attempting to transfer leadership", logName());
+
+ leadershipTransferContext = new LeadershipTransferContext(leadershipTransferCohort);
+
+ // Send an immediate heart beat to the followers.
+ sendAppendEntries(0, false);
+ }
+
+ private void tryToCompleteLeadershipTransfer(final String followerId) {
+ if (leadershipTransferContext == null) {
+ return;
+ }
+
+ final Optional requestedFollowerIdOptional
+ = leadershipTransferContext.transferCohort.getRequestedFollowerId();
+ if (requestedFollowerIdOptional.isPresent() && !requestedFollowerIdOptional.get().equals(followerId)) {
+ // we want to transfer leadership to specific follower
+ return;
+ }
+
+ FollowerLogInformation followerInfo = getFollower(followerId);
+ if (followerInfo == null) {
+ return;
+ }
+
+ long lastIndex = context.getReplicatedLog().lastIndex();
+ boolean isVoting = context.getPeerInfo(followerId).isVoting();
+
+ log.debug("{}: tryToCompleteLeadershipTransfer: followerId: {}, matchIndex: {}, lastIndex: {}, isVoting: {}",
+ logName(), followerId, followerInfo.getMatchIndex(), lastIndex, isVoting);
+
+ if (isVoting && 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 a TimeoutNow message to the matching follower to immediately start an election.
+ ActorSelection followerActor = context.getPeerActorSelection(followerId);
+ followerActor.tell(TimeoutNow.INSTANCE, context.getActor());
+
+ log.debug("{}: Leader transfer complete", logName());
+
+ leadershipTransferContext.transferCohort.transferComplete();
+ leadershipTransferContext = null;
+ }
+ }
+
+ @Override
+ public void close() {
+ if (leadershipTransferContext != null) {
+ LeadershipTransferContext localLeadershipTransferContext = leadershipTransferContext;
+ leadershipTransferContext = null;
+ localLeadershipTransferContext.transferCohort.abortTransfer();
+ }
+
super.close();
}
@VisibleForTesting
- void markFollowerActive(String followerId) {
+ void markFollowerActive(final String followerId) {
getFollower(followerId).markFollowerActive();
}
@VisibleForTesting
- void markFollowerInActive(String followerId) {
+ void markFollowerInActive(final String followerId) {
getFollower(followerId).markFollowerInActive();
}
+
+ private static class LeadershipTransferContext {
+ RaftActorLeadershipTransferCohort transferCohort;
+ Stopwatch timer = Stopwatch.createStarted();
+
+ LeadershipTransferContext(final RaftActorLeadershipTransferCohort transferCohort) {
+ this.transferCohort = transferCohort;
+ }
+
+ boolean isExpired(final long timeout) {
+ if (timer.elapsed(TimeUnit.MILLISECONDS) >= timeout) {
+ transferCohort.abortTransfer();
+ return true;
+ }
+
+ return false;
+ }
+ }
}