X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?p=controller.git;a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-akka-raft%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fraft%2Fbehaviors%2FCandidate.java;h=7c111d28de818e225eb4b1ab8e10d4f962198c72;hp=4a3e2c5d664406844edaddee6308abf112b0f79c;hb=1b0f84c4957e464bad6f7cb7350a8171c3d1621b;hpb=59c4f0b68407ad243976fa7e64ba19d39147abbe
diff --git a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Candidate.java b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Candidate.java
index 4a3e2c5d66..7c111d28de 100644
--- a/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Candidate.java
+++ b/opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Candidate.java
@@ -10,6 +10,9 @@ package org.opendaylight.controller.cluster.raft.behaviors;
import akka.actor.ActorRef;
import akka.actor.ActorSelection;
+import java.util.ArrayList;
+import java.util.Collection;
+import org.opendaylight.controller.cluster.raft.PeerInfo;
import org.opendaylight.controller.cluster.raft.RaftActorContext;
import org.opendaylight.controller.cluster.raft.RaftState;
import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
@@ -19,11 +22,10 @@ import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
import org.opendaylight.controller.cluster.raft.messages.RequestVote;
import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
-import java.util.Set;
-
/**
- * The behavior of a RaftActor when it is in the CandidateState
- *
+ * The behavior of a RaftActor when it is in the Candidate raft state.
+ *
+ *
* Candidates (§5.2):
*
* - On conversion to candidate, start election:
@@ -45,111 +47,126 @@ public class Candidate extends AbstractRaftActorBehavior {
private final int votesRequired;
- private final Set peers;
+ private final Collection votingPeers = new ArrayList<>();
public Candidate(RaftActorContext context) {
- super(context);
+ super(context, RaftState.Candidate);
- peers = context.getPeerAddresses().keySet();
-
- if(LOG.isDebugEnabled()) {
- LOG.debug("Election:Candidate has following peers: {}", peers);
+ for (PeerInfo peer: context.getPeers()) {
+ if (peer.isVoting()) {
+ votingPeers.add(peer.getId());
+ }
}
- if(peers.size() > 0) {
- // Votes are required from a majority of the peers including self.
- // The votesRequired field therefore stores a calculated value
- // of the number of votes required for this candidate to win an
- // election based on it's known peers.
- // If a peer was added during normal operation and raft replicas
- // came to know about them then the new peer would also need to be
- // taken into consideration when calculating this value.
- // Here are some examples for what the votesRequired would be for n
- // peers
- // 0 peers = 1 votesRequired (0 + 1) / 2 + 1 = 1
- // 2 peers = 2 votesRequired (2 + 1) / 2 + 1 = 2
- // 4 peers = 3 votesRequired (4 + 1) / 2 + 1 = 3
- int noOfPeers = peers.size();
- int self = 1;
- votesRequired = (noOfPeers + self) / 2 + 1;
+ log.debug("{}: Election: Candidate has following voting peers: {}", logName(), votingPeers);
+
+ votesRequired = getMajorityVoteCount(votingPeers.size());
+
+ startNewTerm();
+
+ if (votingPeers.isEmpty()) {
+ actor().tell(ElectionTimeout.INSTANCE, actor());
} else {
- votesRequired = 0;
+ scheduleElection(electionDuration());
}
+ }
- startNewTerm();
- scheduleElection(electionDuration());
+ @Override
+ public final String getLeaderId() {
+ return null;
+ }
+
+ @Override
+ public final short getLeaderPayloadVersion() {
+ return -1;
}
- @Override protected RaftActorBehavior handleAppendEntries(ActorRef sender,
- AppendEntries appendEntries) {
+ @Override
+ protected RaftActorBehavior handleAppendEntries(ActorRef sender, AppendEntries appendEntries) {
+
+ log.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
- if(LOG.isDebugEnabled()) {
- LOG.debug(appendEntries.toString());
+ // Some other candidate for the same term became a leader and sent us an append entry
+ if (currentTerm() == appendEntries.getTerm()) {
+ log.info("{}: New Leader {} sent an AppendEntries to Candidate for term {} - will switch to Follower",
+ logName(), appendEntries.getLeaderId(), currentTerm());
+
+ return switchBehavior(new Follower(context));
}
return this;
}
- @Override protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
- AppendEntriesReply appendEntriesReply) {
-
+ @Override
+ protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender, AppendEntriesReply appendEntriesReply) {
return this;
}
- @Override protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
- RequestVoteReply requestVoteReply) {
+ @Override
+ protected RaftActorBehavior handleRequestVoteReply(ActorRef sender, RequestVoteReply requestVoteReply) {
+ log.debug("{}: handleRequestVoteReply: {}, current voteCount: {}", logName(), requestVoteReply, voteCount);
if (requestVoteReply.isVoteGranted()) {
voteCount++;
}
if (voteCount >= votesRequired) {
- return switchBehavior(new Leader(context));
+ if (context.getLastApplied() < context.getReplicatedLog().lastIndex()) {
+ log.info("{}: LastApplied index {} is behind last index {} - switching to PreLeader",
+ logName(), context.getLastApplied(), context.getReplicatedLog().lastIndex());
+ return internalSwitchBehavior(RaftState.PreLeader);
+ } else {
+ return internalSwitchBehavior(RaftState.Leader);
+ }
}
return this;
}
- @Override public RaftState state() {
- return RaftState.Candidate;
- }
-
@Override
- public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
+ public RaftActorBehavior handleMessage(ActorRef sender, Object message) {
+ if (message instanceof ElectionTimeout) {
+ log.debug("{}: Received ElectionTimeout", logName());
- Object message = fromSerializableMessage(originalMessage);
+ if (votesRequired == 0) {
+ // If there are no peers then we should be a Leader
+ // We wait for the election timeout to occur before declare
+ // ourselves the leader. This gives enough time for a leader
+ // who we do not know about (as a peer)
+ // to send a message to the candidate
+
+ return internalSwitchBehavior(RaftState.Leader);
+ }
+
+ startNewTerm();
+ scheduleElection(electionDuration());
+ return this;
+ }
if (message instanceof RaftRPC) {
RaftRPC rpc = (RaftRPC) message;
- if(LOG.isDebugEnabled()) {
- LOG.debug("RaftRPC message received {} my term is {}", rpc, context.getTermInformation().getCurrentTerm());
- }
+ log.debug("{}: RaftRPC message received {}, my term is {}", logName(), rpc,
+ context.getTermInformation().getCurrentTerm());
// If RPC request or response contains term T > currentTerm:
// set currentTerm = T, convert to follower (§5.1)
// This applies to all RPC messages and responses
if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
+ log.info("{}: Term {} in \"{}\" message is greater than Candidate's term {} - switching to Follower",
+ logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
+
context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
- return switchBehavior(new Follower(context));
- }
- }
+ // The raft paper does not say whether or not a Candidate can/should process a RequestVote in
+ // this case but doing so gains quicker convergence when the sender's log is more up-to-date.
+ if (message instanceof RequestVote) {
+ super.handleMessage(sender, message);
+ }
- if (message instanceof ElectionTimeout) {
- if (votesRequired == 0) {
- // If there are no peers then we should be a Leader
- // We wait for the election timeout to occur before declare
- // ourselves the leader. This gives enough time for a leader
- // who we do not know about (as a peer)
- // to send a message to the candidate
-
- return switchBehavior(new Leader(context));
+ return internalSwitchBehavior(RaftState.Follower);
}
- startNewTerm();
- scheduleElection(electionDuration());
- return this;
}
return super.handleMessage(sender, message);
@@ -164,33 +181,32 @@ public class Candidate extends AbstractRaftActorBehavior {
// Increment the election term and vote for self
long currentTerm = context.getTermInformation().getCurrentTerm();
- context.getTermInformation().updateAndPersist(currentTerm + 1,
- context.getId());
+ long newTerm = currentTerm + 1;
+ context.getTermInformation().updateAndPersist(newTerm, context.getId());
- if(LOG.isDebugEnabled()) {
- LOG.debug("Starting new term {}", (currentTerm + 1));
- }
+ log.info("{}: Starting new election term {}", logName(), newTerm);
// Request for a vote
// TODO: Retry request for vote if replies do not arrive in a reasonable
// amount of time TBD
- for (String peerId : peers) {
+ for (String peerId : votingPeers) {
ActorSelection peerActor = context.getPeerActorSelection(peerId);
- if(peerActor != null) {
- peerActor.tell(new RequestVote(
+ if (peerActor != null) {
+ RequestVote requestVote = new RequestVote(
context.getTermInformation().getCurrentTerm(),
context.getId(),
context.getReplicatedLog().lastIndex(),
- context.getReplicatedLog().lastTerm()),
- context.getActor()
- );
- }
- }
+ context.getReplicatedLog().lastTerm());
+ log.debug("{}: Sending {} to peer {}", logName(), requestVote, peerId);
+ peerActor.tell(requestVote, context.getActor());
+ }
+ }
}
- @Override public void close() throws Exception {
+ @Override
+ public void close() {
stopElection();
}
}