* terms of the Eclipse Public License v1.0 which accompanies this distribution,
* and is available at http://www.eclipse.org/legal/epl-v10.html
*/
-
package org.opendaylight.controller.cluster.raft.behaviors;
+import static java.util.Objects.requireNonNull;
+
import akka.actor.ActorRef;
import akka.actor.Cancellable;
-import java.util.Random;
+import akka.cluster.Cluster;
+import akka.cluster.Member;
+import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
+import java.util.Optional;
+import java.util.Set;
+import java.util.concurrent.ThreadLocalRandom;
import java.util.concurrent.TimeUnit;
-import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
import org.opendaylight.controller.cluster.raft.RaftActorContext;
import org.opendaylight.controller.cluster.raft.RaftState;
import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
-import org.opendaylight.controller.cluster.raft.SerializationUtils;
-import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
+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 org.opendaylight.controller.cluster.raft.persisted.ApplyJournalEntries;
import org.slf4j.Logger;
import scala.concurrent.duration.FiniteDuration;
/**
- * Abstract class that represents the behavior of a RaftActor
- * <p/>
- * All Servers:
- * <ul>
- * <li> If commitIndex > lastApplied: increment lastApplied, apply
- * log[lastApplied] to state machine (§5.3)
- * <li> If RPC request or response contains term T > currentTerm:
- * set currentTerm = T, convert to follower (§5.1)
+ * Abstract class that provides common code for a RaftActor behavior.
*/
public abstract class AbstractRaftActorBehavior implements RaftActorBehavior {
-
/**
- * Information about the RaftActor whose behavior this class represents
+ * Information about the RaftActor whose behavior this class represents.
*/
protected final RaftActorContext context;
/**
- *
+ * Used for message logging.
*/
- protected final Logger LOG;
+ @SuppressFBWarnings("SLF4J_LOGGER_SHOULD_BE_PRIVATE")
+ protected final Logger log;
/**
- *
+ * Prepended to log messages to provide appropriate context.
*/
- private Cancellable electionCancel = null;
+ private final String logName;
/**
- *
+ * The RaftState corresponding to his behavior.
*/
- protected String leaderId = null;
-
- private long replicatedToAllIndex = -1;
+ private final RaftState state;
- private final String logName;
+ /**
+ * Used to cancel a scheduled election.
+ */
+ private Cancellable electionCancel = null;
- private final RaftState state;
+ /**
+ * The index of the last log entry that has been replicated to all raft peers.
+ */
+ private long replicatedToAllIndex = -1;
- protected AbstractRaftActorBehavior(RaftActorContext context, RaftState state) {
- this.context = context;
- this.state = state;
- this.LOG = context.getLogger();
+ AbstractRaftActorBehavior(final RaftActorContext context, final RaftState state) {
+ this.context = requireNonNull(context);
+ this.state = requireNonNull(state);
+ log = context.getLogger();
logName = String.format("%s (%s)", context.getId(), state);
}
+ public static RaftActorBehavior createBehavior(final RaftActorContext context, final RaftState state) {
+ switch (state) {
+ case Candidate:
+ return new Candidate(context);
+ case Follower:
+ return new Follower(context);
+ case IsolatedLeader:
+ return new IsolatedLeader(context);
+ case Leader:
+ return new Leader(context);
+ case PreLeader:
+ return new PreLeader(context);
+ default:
+ throw new IllegalArgumentException("Unhandled state " + state);
+ }
+ }
+
@Override
- public RaftState state() {
+ public final RaftState state() {
return state;
}
- public String logName() {
+ protected final String logName() {
return logName;
}
@Override
- public void setReplicatedToAllIndex(long replicatedToAllIndex) {
+ public void setReplicatedToAllIndex(final long replicatedToAllIndex) {
this.replicatedToAllIndex = replicatedToAllIndex;
}
*
* @param sender The actor that sent this message
* @param appendEntries The AppendEntries message
- * @return
+ * @return a new behavior if it was changed or the current behavior
*/
protected abstract RaftActorBehavior handleAppendEntries(ActorRef sender,
AppendEntries appendEntries);
-
/**
- * appendEntries first processes the AppendEntries message and then
- * delegates handling to a specific behavior
+ * Handles the common logic for the AppendEntries message and delegates handling to the derived class.
*
- * @param sender
- * @param appendEntries
- * @return
+ * @param sender the ActorRef that sent the message
+ * @param appendEntries the message
+ * @return a new behavior if it was changed or the current behavior
*/
- protected RaftActorBehavior appendEntries(ActorRef sender,
- AppendEntries appendEntries) {
+ protected RaftActorBehavior appendEntries(final ActorRef sender, final AppendEntries appendEntries) {
// 1. Reply false if term < currentTerm (§5.1)
if (appendEntries.getTerm() < currentTerm()) {
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: Cannot append entries because sender term {} is less than {}",
- logName(), appendEntries.getTerm(), currentTerm());
- }
+ log.info("{}: Cannot append entries because sender's term {} is less than {}", logName(),
+ appendEntries.getTerm(), currentTerm());
- sender.tell(
- new AppendEntriesReply(context.getId(), currentTerm(), false,
- lastIndex(), lastTerm()), actor()
- );
+ sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex(), lastTerm(),
+ context.getPayloadVersion(), false, false, appendEntries.getLeaderRaftVersion()), actor());
return this;
}
*
* @param sender The actor that sent this message
* @param appendEntriesReply The AppendEntriesReply message
- * @return
+ * @return a new behavior if it was changed or the current behavior
*/
protected abstract RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
AppendEntriesReply appendEntriesReply);
/**
- * requestVote handles the RequestVote message. This logic is common
- * for all behaviors
+ * Handles the logic for the RequestVote message that is common for all behaviors.
*
- * @param sender
- * @param requestVote
- * @return
+ * @param sender the ActorRef that sent the message
+ * @param requestVote the message
+ * @return a new behavior if it was changed or the current behavior
*/
- protected RaftActorBehavior requestVote(ActorRef sender, RequestVote requestVote) {
+ protected RaftActorBehavior requestVote(final ActorRef sender, final RequestVote requestVote) {
- LOG.debug("{}: In requestVote: {}", logName(), requestVote);
+ log.debug("{}: In requestVote: {} - currentTerm: {}, votedFor: {}, lastIndex: {}, lastTerm: {}", logName(),
+ requestVote, currentTerm(), votedFor(), lastIndex(), lastTerm());
+ boolean grantVote = canGrantVote(requestVote);
+
+ if (grantVote) {
+ context.getTermInformation().updateAndPersist(requestVote.getTerm(), requestVote.getCandidateId());
+ }
+
+ RequestVoteReply reply = new RequestVoteReply(currentTerm(), grantVote);
+
+ log.debug("{}: requestVote returning: {}", logName(), reply);
+
+ sender.tell(reply, actor());
+
+ return this;
+ }
+
+ protected boolean canGrantVote(final RequestVote requestVote) {
boolean grantVote = false;
// Reply false if term < currentTerm (§5.1)
// If votedFor is null or candidateId, and candidate’s log is at
// least as up-to-date as receiver’s log, grant vote (§5.2, §5.4)
} else if (votedFor() == null || votedFor()
- .equals(requestVote.getCandidateId())) {
+ .equals(requestVote.getCandidateId())) {
boolean candidateLatest = false;
// the log with the later term is more up-to-date. If the logs
// end with the same term, then whichever log is longer is
// more up-to-date.
- if (requestVote.getLastLogTerm() > lastTerm()) {
- candidateLatest = true;
- } else if ((requestVote.getLastLogTerm() == lastTerm())
- && requestVote.getLastLogIndex() >= lastIndex()) {
+ if (requestVote.getLastLogTerm() > lastTerm()
+ || requestVote.getLastLogTerm() == lastTerm() && requestVote.getLastLogIndex() >= lastIndex()) {
candidateLatest = true;
}
if (candidateLatest) {
grantVote = true;
- context.getTermInformation().updateAndPersist(requestVote.getTerm(),
- requestVote.getCandidateId());
}
}
-
- RequestVoteReply reply = new RequestVoteReply(currentTerm(), grantVote);
-
- LOG.debug("{}: requestVote returning: {}", logName(), reply);
-
- sender.tell(reply, actor());
-
- return this;
+ return grantVote;
}
/**
*
* @param sender The actor that sent this message
* @param requestVoteReply The RequestVoteReply message
- * @return
+ * @return a new behavior if it was changed or the current behavior
*/
protected abstract RaftActorBehavior handleRequestVoteReply(ActorRef sender,
RequestVoteReply requestVoteReply);
/**
- * Creates a random election duration
+ * Returns a duration for election with an additional variance for randomness.
*
- * @return
+ * @return a random election duration
*/
protected FiniteDuration electionDuration() {
- long variance = new Random().nextInt(context.getConfigParams().getElectionTimeVariance());
+ long variance = ThreadLocalRandom.current().nextInt(context.getConfigParams().getElectionTimeVariance());
return context.getConfigParams().getElectionTimeOutInterval().$plus(
new FiniteDuration(variance, TimeUnit.MILLISECONDS));
}
/**
- * stop the scheduled election
+ * Stops the currently scheduled election.
*/
protected void stopElection() {
if (electionCancel != null && !electionCancel.isCancelled()) {
}
}
+ protected boolean canStartElection() {
+ return context.getRaftPolicy().automaticElectionsEnabled() && context.isVotingMember();
+ }
+
/**
- * schedule a new election
+ * Schedule a new election.
*
- * @param interval
+ * @param interval the duration after which we should trigger a new election
*/
- protected void scheduleElection(FiniteDuration interval) {
+ // Non-final for testing
+ protected void scheduleElection(final FiniteDuration interval) {
stopElection();
- // Schedule an election. When the scheduler triggers an ElectionTimeout
- // message is sent to itself
- electionCancel =
- context.getActorSystem().scheduler().scheduleOnce(interval,
- context.getActor(), new ElectionTimeout(),
- context.getActorSystem().dispatcher(), context.getActor());
+ // Schedule an election. When the scheduler triggers an ElectionTimeout message is sent to itself
+ electionCancel = context.getActorSystem().scheduler().scheduleOnce(interval, context.getActor(),
+ ElectionTimeout.INSTANCE, context.getActorSystem().dispatcher(), context.getActor());
}
/**
- * Get the current term
- * @return
+ * Returns the current election term.
+ *
+ * @return the current term
*/
protected long currentTerm() {
return context.getTermInformation().getCurrentTerm();
}
/**
- * Get the candidate for whom we voted in the current term
- * @return
+ * Returns the id of the candidate that this server voted for in current term.
+ *
+ * @return the candidate for whom we voted in the current term
*/
protected String votedFor() {
return context.getTermInformation().getVotedFor();
}
/**
- * Get the actor associated with this behavior
- * @return
+ * Returns the actor associated with this behavior.
+ *
+ * @return the actor
*/
- protected ActorRef actor() {
+ protected final ActorRef actor() {
return context.getActor();
}
/**
- * Get the term from the last entry in the log
+ * Returns the term of the last entry in the log.
*
- * @return
+ * @return the term
*/
protected long lastTerm() {
return context.getReplicatedLog().lastTerm();
}
/**
- * Get the index from the last entry in the log
+ * Returns the index of the last entry in the log.
*
- * @return
+ * @return the index
*/
protected long lastIndex() {
return context.getReplicatedLog().lastIndex();
}
/**
- * Find the client request tracker for a specific logIndex
+ * Returns the actual index of the entry in replicated log for the given index or -1 if not found.
*
- * @param logIndex
- * @return
+ * @return the log entry index or -1 if not found
*/
- protected ClientRequestTracker findClientRequestTracker(long logIndex) {
- return null;
- }
+ protected long getLogEntryIndex(final long index) {
+ if (index == context.getReplicatedLog().getSnapshotIndex()) {
+ return context.getReplicatedLog().getSnapshotIndex();
+ }
- /**
- * Find the client request tracker for a specific logIndex
- *
- * @param logIndex
- * @return
- */
- protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
- return null;
- }
+ ReplicatedLogEntry entry = context.getReplicatedLog().get(index);
+ if (entry != null) {
+ return entry.getIndex();
+ }
+ return -1;
+ }
/**
- * Find the log index from the previous to last entry in the log
+ * Returns the actual term of the entry in the replicated log for the given index or -1 if not found.
*
- * @return
+ * @return the log entry term or -1 if not found
*/
- protected long prevLogIndex(long index){
- ReplicatedLogEntry prevEntry =
- context.getReplicatedLog().get(index - 1);
- if (prevEntry != null) {
- return prevEntry.getIndex();
+ protected long getLogEntryTerm(final long index) {
+ if (index == context.getReplicatedLog().getSnapshotIndex()) {
+ return context.getReplicatedLog().getSnapshotTerm();
+ }
+
+ ReplicatedLogEntry entry = context.getReplicatedLog().get(index);
+ if (entry != null) {
+ return entry.getTerm();
}
+
return -1;
}
/**
- * Find the log term from the previous to last entry in the log
- * @return
+ * Returns the actual term of the entry in the replicated log for the given index or, if not present, returns the
+ * snapshot term if the given index is in the snapshot or -1 otherwise.
+ *
+ * @return the term or -1 otherwise
*/
- protected long prevLogTerm(long index){
- ReplicatedLogEntry prevEntry =
- context.getReplicatedLog().get(index - 1);
- if (prevEntry != null) {
- return prevEntry.getTerm();
+ protected long getLogEntryOrSnapshotTerm(final long index) {
+ if (context.getReplicatedLog().isInSnapshot(index)) {
+ return context.getReplicatedLog().getSnapshotTerm();
}
- return -1;
+
+ return getLogEntryTerm(index);
}
/**
- * Apply the provided index to the state machine
+ * Applies the log entries up to the specified index that is known to be committed to the state machine.
*
- * @param index a log index that is known to be committed
+ * @param index the log index
*/
protected void applyLogToStateMachine(final long index) {
- long newLastApplied = context.getLastApplied();
// Now maybe we apply to the state machine
- for (long i = context.getLastApplied() + 1;
- i < index + 1; i++) {
- ActorRef clientActor = null;
- String identifier = null;
- ClientRequestTracker tracker = removeClientRequestTracker(i);
-
- if (tracker != null) {
- clientActor = tracker.getClientActor();
- identifier = tracker.getIdentifier();
- }
- ReplicatedLogEntry replicatedLogEntry =
- context.getReplicatedLog().get(i);
+ for (long i = context.getLastApplied() + 1; i < index + 1; i++) {
+ ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(i);
if (replicatedLogEntry != null) {
// Send a local message to the local RaftActor (it's derived class to be
// specific to apply the log to it's index)
- actor().tell(new ApplyState(clientActor, identifier,
- replicatedLogEntry), actor());
- newLastApplied = i;
+
+ final ApplyState applyState = getApplyStateFor(replicatedLogEntry);
+
+ log.debug("{}: Setting last applied to {}", logName(), i);
+
+ context.setLastApplied(i);
+ context.getApplyStateConsumer().accept(applyState);
} else {
//if one index is not present in the log, no point in looping
// around as the rest wont be present either
- LOG.warn(
- "{}: Missing index {} from log. Cannot apply state. Ignoring {} to {}",
+ log.warn("{}: Missing index {} from log. Cannot apply state. Ignoring {} to {}",
logName(), i, i, index);
break;
}
}
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: Setting last applied to {}", logName(), newLastApplied);
- }
- context.setLastApplied(newLastApplied);
// send a message to persist a ApplyLogEntries marker message into akka's persistent journal
// will be used during recovery
//in case if the above code throws an error and this message is not sent, it would be fine
// as the append entries received later would initiate add this message to the journal
- actor().tell(new ApplyLogEntries((int) context.getLastApplied()), actor());
+ actor().tell(new ApplyJournalEntries(context.getLastApplied()), actor());
}
- protected Object fromSerializableMessage(Object serializable){
- return SerializationUtils.fromSerializable(serializable);
- }
+ /**
+ * Create an ApplyState message for a particular log entry so we can determine how to apply this entry.
+ *
+ * @param entry the log entry
+ * @return ApplyState for this entry
+ */
+ abstract ApplyState getApplyStateFor(ReplicatedLogEntry entry);
@Override
- public RaftActorBehavior handleMessage(ActorRef sender, Object message) {
+ public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) {
if (message instanceof AppendEntries) {
return appendEntries(sender, (AppendEntries) message);
} else if (message instanceof AppendEntriesReply) {
return requestVote(sender, (RequestVote) message);
} else if (message instanceof RequestVoteReply) {
return handleRequestVoteReply(sender, (RequestVoteReply) message);
+ } else {
+ return null;
}
- return this;
}
- @Override public String getLeaderId() {
- return leaderId;
+ @Override
+ public RaftActorBehavior switchBehavior(final RaftActorBehavior behavior) {
+ return internalSwitchBehavior(behavior);
+ }
+
+ protected RaftActorBehavior internalSwitchBehavior(final RaftState newState) {
+ return internalSwitchBehavior(createBehavior(context, newState));
}
- protected RaftActorBehavior switchBehavior(RaftActorBehavior behavior) {
- LOG.info("{} :- Switching from behavior {} to {}", logName(), this.state(), behavior.state());
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ protected RaftActorBehavior internalSwitchBehavior(final RaftActorBehavior newBehavior) {
+ if (!context.getRaftPolicy().automaticElectionsEnabled()) {
+ return this;
+ }
+
+ log.info("{} :- Switching from behavior {} to {}, election term: {}", logName(), this.state(),
+ newBehavior.state(), context.getTermInformation().getCurrentTerm());
try {
close();
- } catch (Exception e) {
- LOG.error("{}: Failed to close behavior : {}", logName(), this.state(), e);
+ } catch (RuntimeException e) {
+ log.error("{}: Failed to close behavior : {}", logName(), this.state(), e);
}
-
- return behavior;
+ return newBehavior;
}
- protected int getMajorityVoteCount(int numPeers) {
+
+ protected int getMajorityVoteCount(final int numPeers) {
// Votes are required from a majority of the peers including self.
// The numMajority field therefore stores a calculated value
// of the number of votes required for this candidate to win an
/**
- * Performs a snapshot with no capture on the replicated log.
- * It clears the log from the supplied index or last-applied-1 which ever is minimum.
+ * Performs a snapshot with no capture on the replicated log. It clears the log from the supplied index or
+ * lastApplied-1 which ever is minimum.
*
- * @param snapshotCapturedIndex
+ * @param snapshotCapturedIndex the index from which to clear
*/
protected void performSnapshotWithoutCapture(final long snapshotCapturedIndex) {
- // we would want to keep the lastApplied as its used while capturing snapshots
- long lastApplied = context.getLastApplied();
- long tempMin = Math.min(snapshotCapturedIndex, (lastApplied > -1 ? lastApplied - 1 : -1));
-
- if (tempMin > -1 && context.getReplicatedLog().isPresent(tempMin)) {
- LOG.debug("{}: fakeSnapshot purging log to {} for term {}", logName(), tempMin,
- context.getTermInformation().getCurrentTerm());
-
- //use the term of the temp-min, since we check for isPresent, entry will not be null
- ReplicatedLogEntry entry = context.getReplicatedLog().get(tempMin);
- context.getReplicatedLog().snapshotPreCommit(tempMin, entry.getTerm());
- context.getReplicatedLog().snapshotCommit();
- setReplicatedToAllIndex(tempMin);
+ long actualIndex = context.getSnapshotManager().trimLog(snapshotCapturedIndex);
+
+ if (actualIndex != -1) {
+ setReplicatedToAllIndex(actualIndex);
}
}
+ protected final String getId() {
+ return context.getId();
+ }
+
+ // Check whether we should update the term. In case of half-connected nodes, we want to ignore RequestVote
+ // messages, as the candidate is not able to receive our response.
+ protected boolean shouldUpdateTerm(final RaftRPC rpc) {
+ if (!(rpc instanceof RequestVote)) {
+ return true;
+ }
+
+ final RequestVote requestVote = (RequestVote) rpc;
+ log.debug("{}: Found higher term in RequestVote rpc, verifying whether it's safe to update term.", logName());
+ final Optional<Cluster> maybeCluster = context.getCluster();
+ if (!maybeCluster.isPresent()) {
+ return true;
+ }
+
+ final Cluster cluster = maybeCluster.get();
+
+ final Set<Member> unreachable = cluster.state().getUnreachable();
+ log.debug("{}: Cluster state: {}", logName(), unreachable);
+
+ for (Member member : unreachable) {
+ for (String role : member.getRoles()) {
+ if (requestVote.getCandidateId().startsWith(role)) {
+ log.debug("{}: Unreachable member: {}, matches candidateId in: {}, not updating term", logName(),
+ member, requestVote);
+ return false;
+ }
+ }
+ }
+
+ log.debug("{}: Candidate in requestVote:{} with higher term appears reachable, updating term.", logName(),
+ requestVote);
+ return true;
+ }
}
Code Review / controller.git / blobdiff