import akka.actor.ActorSelection;
import akka.actor.ActorSystem;
import akka.actor.Props;
+import akka.cluster.Cluster;
import com.google.common.annotations.VisibleForTesting;
import java.util.Collection;
+import java.util.Optional;
import java.util.function.LongSupplier;
import javax.annotation.Nullable;
import org.opendaylight.controller.cluster.DataPersistenceProvider;
*/
ActorRef getActor();
+ /**
+ * The akka Cluster singleton for the actor system if one is configured.
+ *
+ * @return an Optional containing the CLuster instance is present.
+ */
+ Optional<Cluster> getCluster();
+
/**
* @return the ElectionTerm information
*/
import akka.actor.ActorSelection;
import akka.actor.ActorSystem;
import akka.actor.Props;
+import akka.cluster.Cluster;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Preconditions;
import java.util.ArrayList;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
+import java.util.Optional;
import java.util.Set;
import java.util.function.LongSupplier;
import org.opendaylight.controller.cluster.DataPersistenceProvider;
private int numVotingPeers = -1;
+ private Optional<Cluster> cluster;
+
public RaftActorContextImpl(ActorRef actor, ActorContext context, String id,
ElectionTerm termInformation, long commitIndex, long lastApplied, Map<String, String> peerAddresses,
ConfigParams configParams, DataPersistenceProvider persistenceProvider, Logger logger) {
return actor;
}
+ @Override
+ public Optional<Cluster> getCluster() {
+ if(cluster == null) {
+ try {
+ cluster = Optional.of(Cluster.get(getActorSystem()));
+ } catch(Exception e) {
+ // An exception means there's no cluster configured. This will only happen in unit tests.
+ LOG.debug("{}: Could not obtain Cluster: {}", getId(), e);
+ cluster = Optional.empty();
+ }
+ }
+
+ return cluster;
+ }
+
@Override
public ElectionTerm getTermInformation() {
return termInformation;
package org.opendaylight.controller.cluster.raft.behaviors;
import akka.actor.ActorRef;
+import akka.actor.ActorSelection;
+import akka.actor.Address;
+import akka.cluster.Cluster;
+import akka.cluster.ClusterEvent.CurrentClusterState;
+import akka.cluster.Member;
+import akka.cluster.MemberStatus;
import akka.japi.Procedure;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Stopwatch;
import java.util.ArrayList;
+import java.util.Optional;
+import java.util.Set;
import java.util.concurrent.TimeUnit;
import javax.annotation.Nullable;
import org.opendaylight.controller.cluster.raft.RaftActorContext;
public class Follower extends AbstractRaftActorBehavior {
private static final int SYNC_THRESHOLD = 10;
+ private static final long MAX_ELECTION_TIMEOUT_FACTOR = 18;
+
private final SyncStatusTracker initialSyncStatusTracker;
private final Procedure<ReplicatedLogEntry> appendAndPersistCallback = new Procedure<ReplicatedLogEntry>() {
}
};
- private final Stopwatch lastLeaderMessageTimer = Stopwatch.createUnstarted();
+ private final Stopwatch lastLeaderMessageTimer = Stopwatch.createStarted();
private SnapshotTracker snapshotTracker = null;
private String leaderId;
private short leaderPayloadVersion;
// queue but would be processed before the ElectionTimeout message and thus would restart the
// lastLeaderMessageTimer.
long lastLeaderMessageInterval = lastLeaderMessageTimer.elapsed(TimeUnit.MILLISECONDS);
- boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning() || lastLeaderMessageInterval >=
- context.getConfigParams().getElectionTimeOutInterval().toMillis();
+ long electionTimeoutInMillis = context.getConfigParams().getElectionTimeOutInterval().toMillis();
+ boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning() ||
+ lastLeaderMessageInterval >= electionTimeoutInMillis;
if(canStartElection()) {
- if(message instanceof TimeoutNow || noLeaderMessageReceived) {
- LOG.debug("{}: Received {} - switching to Candidate", logName(), message.getClass().getSimpleName());
+ if(message instanceof TimeoutNow) {
+ LOG.debug("{}: Received TimeoutNow - switching to Candidate", logName());
return internalSwitchBehavior(RaftState.Candidate);
+ } else if(noLeaderMessageReceived) {
+ // Check the cluster state to see if the leader is known to be up before we go to Candidate.
+ // However if we haven't heard from the leader in a long time even though the cluster state
+ // indicates it's up then something is wrong - leader might be stuck indefinitely - so switch
+ // to Candidate,
+ long maxElectionTimeout = electionTimeoutInMillis * MAX_ELECTION_TIMEOUT_FACTOR;
+ if(isLeaderAvailabilityKnown() && lastLeaderMessageInterval < maxElectionTimeout) {
+ LOG.debug("{}: Received ElectionTimeout but leader appears to be available", logName());
+ scheduleElection(electionDuration());
+ } else {
+ LOG.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
+ return internalSwitchBehavior(RaftState.Candidate);
+ }
} else {
- LOG.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout",
- logName(), lastLeaderMessageInterval);
+ LOG.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout {}",
+ logName(), lastLeaderMessageInterval, context.getConfigParams().getElectionTimeOutInterval());
scheduleElection(electionDuration());
}
} else if(message instanceof ElectionTimeout) {
return this;
}
+ private boolean isLeaderAvailabilityKnown() {
+ if(leaderId == null) {
+ return false;
+ }
+
+ Optional<Cluster> cluster = context.getCluster();
+ if(!cluster.isPresent()) {
+ return false;
+ }
+
+ ActorSelection leaderActor = context.getPeerActorSelection(leaderId);
+ if(leaderActor == null) {
+ return false;
+ }
+
+ Address leaderAddress = leaderActor.anchorPath().address();
+
+ CurrentClusterState state = cluster.get().state();
+ Set<Member> unreachable = state.getUnreachable();
+
+ LOG.debug("{}: Checking for leader {} in the cluster unreachable set {}", logName(), leaderAddress,
+ unreachable);
+
+ for(Member m: unreachable) {
+ if(leaderAddress.equals(m.address())) {
+ LOG.info("{}: Leader {} is unreachable", logName(), leaderAddress);
+ return false;
+ }
+ }
+
+ for(Member m: state.getMembers()) {
+ if(leaderAddress.equals(m.address())) {
+ if(m.status() == MemberStatus.up() || m.status() == MemberStatus.weaklyUp()) {
+ LOG.debug("{}: Leader {} cluster status is {} - leader is available", logName(),
+ leaderAddress, m.status());
+ return true;
+ } else {
+ LOG.debug("{}: Leader {} cluster status is {} - leader is unavailable", logName(),
+ leaderAddress, m.status());
+ return false;
+ }
+ }
+ }
+
+ LOG.debug("{}: Leader {} not found in the cluster member set", logName(), leaderAddress);
+
+ return false;
+ }
+
private void handleInstallSnapshot(final ActorRef sender, InstallSnapshot installSnapshot) {
LOG.debug("{}: handleInstallSnapshot: {}", logName(), installSnapshot);
import static org.mockito.Mockito.doReturn;
import static org.mockito.Mockito.mock;
import static org.opendaylight.controller.cluster.datastore.DataStoreVersions.CURRENT_VERSION;
-import static org.opendaylight.controller.cluster.datastore.ShardDataTreeMocking.immediateCanCommit;
-import static org.opendaylight.controller.cluster.datastore.ShardDataTreeMocking.immediateCommit;
-import static org.opendaylight.controller.cluster.datastore.ShardDataTreeMocking.immediatePreCommit;
import static org.opendaylight.controller.cluster.datastore.ShardDataTreeMocking.successfulCanCommit;
import static org.opendaylight.controller.cluster.datastore.ShardDataTreeMocking.successfulCommit;
import static org.opendaylight.controller.cluster.datastore.ShardDataTreeMocking.successfulPreCommit;
import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
+import org.opendaylight.controller.cluster.datastore.modification.MergeModification;
import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
import org.opendaylight.controller.cluster.datastore.modification.WriteModification;
import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
}
public static void writeToStore(final ShardDataTree store, final YangInstanceIdentifier id,
- final NormalizedNode<?,?> node) throws InterruptedException, ExecutionException {
- final ReadWriteShardDataTreeTransaction transaction = store.newReadWriteTransaction(nextTransactionId());
-
- transaction.getSnapshot().write(id, node);
- final ShardDataTreeCohort cohort = transaction.ready();
- immediateCanCommit(cohort);
- immediatePreCommit(cohort);
- immediateCommit(cohort);
+ final NormalizedNode<?,?> node) throws Exception {
+ BatchedModifications batched = newBatchedModifications(nextTransactionId(), id, node, true, true, 1);
+ DataTreeModification modification = store.getDataTree().takeSnapshot().newModification();
+ batched.apply(modification);
+ modification.ready();
+ store.applyForeignCandidate(batched.getTransactionID(), store.getDataTree().prepare(modification));
}
public void mergeToStore(final ShardDataTree store, final YangInstanceIdentifier id,
- final NormalizedNode<?,?> node) throws InterruptedException, ExecutionException {
- final ReadWriteShardDataTreeTransaction transaction = store.newReadWriteTransaction(nextTransactionId());
-
- transaction.getSnapshot().merge(id, node);
- final ShardDataTreeCohort cohort = transaction.ready();
- immediateCanCommit(cohort);
- immediatePreCommit(cohort);
- immediateCommit(cohort);
+ final NormalizedNode<?,?> node) throws Exception {
+ final BatchedModifications batched = new BatchedModifications(nextTransactionId(), CURRENT_VERSION);
+ batched.addModification(new MergeModification(id, node));
+ batched.setReady(true);
+ batched.setDoCommitOnReady(true);
+ batched.setTotalMessagesSent(1);
+
+ DataTreeModification modification = store.getDataTree().takeSnapshot().newModification();
+ batched.apply(modification);
+ modification.ready();
+ store.applyForeignCandidate(batched.getTransactionID(), store.getDataTree().prepare(modification));
}
public static void writeToStore(final DataTree store, final YangInstanceIdentifier id,
private Shard createShard() {
TestActorRef<Shard> actor = actorFactory.createTestActor(newShardProps());
- ShardTestKit.waitUntilLeader(actor);
-
return actor.underlyingActor();
}
}
shardElectionTimeoutFactor(1).customRaftPolicyImplementation(null));
JavaTestKit.shutdownActorSystem(leaderSystem, null, true);
+ Cluster.get(followerSystem).leave(MEMBER_1_ADDRESS);
followerTestKit.waitUntilNoLeader(followerDistributedDataStore.getActorContext(), CARS);
customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).
shardElectionTimeoutFactor(10));
+ Cluster.get(followerSystem).leave(MEMBER_1_ADDRESS);
leaderTestKit.waitUntilNoLeader(leaderDistributedDataStore.getActorContext(), "cars");
// Submit all tx's - the messages should get queued for retry.
JavaTestKit.shutdownActorSystem(leaderSystem, null, true);
+ Cluster.get(followerSystem).leave(MEMBER_1_ADDRESS);
+
Uninterruptibles.sleepUninterruptibly(100, TimeUnit.MILLISECONDS);
sendDatastoreContextUpdate(followerDistributedDataStore, followerDatastoreContextBuilder.
JavaTestKit.shutdownActorSystem(leaderSystem, null, true);
+ Cluster.get(followerSystem).leave(MEMBER_1_ADDRESS);
+
sendDatastoreContextUpdate(followerDistributedDataStore, followerDatastoreContextBuilder.
operationTimeoutInMillis(500).shardElectionTimeoutFactor(1).customRaftPolicyImplementation(null));