import akka.actor.ActorRef;
import akka.actor.ActorSelection;
-import akka.event.Logging;
-import akka.event.LoggingAdapter;
import akka.japi.Procedure;
import akka.persistence.RecoveryCompleted;
import akka.persistence.SaveSnapshotFailure;
import com.google.common.base.Optional;
import com.google.common.base.Stopwatch;
import com.google.protobuf.ByteString;
+import java.io.Serializable;
+import java.util.Map;
import org.opendaylight.controller.cluster.DataPersistenceProvider;
import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActor;
+import org.opendaylight.controller.cluster.notifications.RoleChanged;
import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
import org.opendaylight.controller.protobuff.messages.cluster.raft.AppendEntriesMessages;
-
-import java.io.Serializable;
-import java.util.Map;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
/**
* RaftActor encapsulates a state machine that needs to be kept synchronized
* </ul>
*/
public abstract class RaftActor extends AbstractUntypedPersistentActor {
- protected final LoggingAdapter LOG =
- Logging.getLogger(getContext().system(), this);
+ protected final Logger LOG = LoggerFactory.getLogger(getClass());
/**
* The current state determines the current behavior of a RaftActor
private CaptureSnapshot captureSnapshot = null;
- private volatile boolean hasSnapshotCaptureInitiated = false;
-
private Stopwatch recoveryTimer;
private int currentRecoveryBatchCount;
deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), scala.Long.MaxValue()));
onRecoveryComplete();
- currentBehavior = new Follower(context);
- onStateChanged();
+
+ initializeBehavior();
}
}
}
private void onRecoveredSnapshot(SnapshotOffer offer) {
if(LOG.isDebugEnabled()) {
- LOG.debug("SnapshotOffer called..");
+ LOG.debug("{}: SnapshotOffer called..", persistenceId());
}
initRecoveryTimer();
timer.start();
// Apply the snapshot to the actors state
- applyRecoverySnapshot(ByteString.copyFrom(snapshot.getState()));
+ applyRecoverySnapshot(snapshot.getState());
timer.stop();
LOG.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size=" +
replicatedLog.size(), persistenceId(), timer.toString(),
- replicatedLog.snapshotIndex, replicatedLog.snapshotTerm);
+ replicatedLog.getSnapshotIndex(), replicatedLog.getSnapshotTerm());
}
private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
if(LOG.isDebugEnabled()) {
- LOG.debug("Received ReplicatedLogEntry for recovery: {}", logEntry.getIndex());
+ LOG.debug("{}: Received ReplicatedLogEntry for recovery: {}", persistenceId(), logEntry.getIndex());
}
replicatedLog.append(logEntry);
private void onRecoveredApplyLogEntries(ApplyLogEntries ale) {
if(LOG.isDebugEnabled()) {
- LOG.debug("Received ApplyLogEntries for recovery, applying to state: {} to {}",
- context.getLastApplied() + 1, ale.getToIndex());
+ LOG.debug("{}: Received ApplyLogEntries for recovery, applying to state: {} to {}",
+ persistenceId(), context.getLastApplied() + 1, ale.getToIndex());
}
for (long i = context.getLastApplied() + 1; i <= ale.getToIndex(); i++) {
"Persistence Id = " + persistenceId() +
" Last index in log={}, snapshotIndex={}, snapshotTerm={}, " +
"journal-size={}",
- replicatedLog.lastIndex(), replicatedLog.snapshotIndex,
- replicatedLog.snapshotTerm, replicatedLog.size());
+ replicatedLog.lastIndex(), replicatedLog.getSnapshotIndex(),
+ replicatedLog.getSnapshotTerm(), replicatedLog.size());
- currentBehavior = new Follower(context);
- onStateChanged();
+ initializeBehavior();
+ }
+
+ protected void initializeBehavior(){
+ changeCurrentBehavior(new Follower(context));
+ }
+
+ protected void changeCurrentBehavior(RaftActorBehavior newBehavior){
+ RaftActorBehavior oldBehavior = currentBehavior;
+ currentBehavior = newBehavior;
+ handleBehaviorChange(oldBehavior, currentBehavior);
}
@Override public void handleCommand(Object message) {
ApplyState applyState = (ApplyState) message;
if(LOG.isDebugEnabled()) {
- LOG.debug("Applying state for log index {} data {}",
- applyState.getReplicatedLogEntry().getIndex(),
+ LOG.debug("{}: Applying state for log index {} data {}",
+ persistenceId(), applyState.getReplicatedLogEntry().getIndex(),
applyState.getReplicatedLogEntry().getData());
}
} else if (message instanceof ApplyLogEntries){
ApplyLogEntries ale = (ApplyLogEntries) message;
if(LOG.isDebugEnabled()) {
- LOG.debug("Persisting ApplyLogEntries with index={}", ale.getToIndex());
+ LOG.debug("{}: Persisting ApplyLogEntries with index={}", persistenceId(), ale.getToIndex());
}
persistence().persist(new ApplyLogEntries(ale.getToIndex()), new Procedure<ApplyLogEntries>() {
@Override
Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
if(LOG.isDebugEnabled()) {
- LOG.debug("ApplySnapshot called on Follower Actor " +
- "snapshotIndex:{}, snapshotTerm:{}", snapshot.getLastAppliedIndex(),
+ LOG.debug("{}: ApplySnapshot called on Follower Actor " +
+ "snapshotIndex:{}, snapshotTerm:{}", persistenceId(), snapshot.getLastAppliedIndex(),
snapshot.getLastAppliedTerm()
);
}
- applySnapshot(ByteString.copyFrom(snapshot.getState()));
+
+ applySnapshot(snapshot.getState());
//clears the followers log, sets the snapshot index to ensure adjusted-index works
replicatedLog = new ReplicatedLogImpl(snapshot);
} else if (message instanceof SaveSnapshotSuccess) {
SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
- LOG.info("SaveSnapshotSuccess received for snapshot");
+ LOG.info("{}: SaveSnapshotSuccess received for snapshot", persistenceId());
long sequenceNumber = success.metadata().sequenceNr();
} else if (message instanceof SaveSnapshotFailure) {
SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message;
- LOG.info("saveSnapshotFailure.metadata():{}", saveSnapshotFailure.metadata().toString());
- LOG.error(saveSnapshotFailure.cause(), "SaveSnapshotFailure received for snapshot Cause:");
+ LOG.error("{}: SaveSnapshotFailure received for snapshot Cause:",
+ persistenceId(), saveSnapshotFailure.cause());
context.getReplicatedLog().snapshotRollback();
- LOG.info("Replicated Log rollbacked. Snapshot will be attempted in the next cycle." +
- "snapshotIndex:{}, snapshotTerm:{}, log-size:{}",
+ LOG.info("{}: Replicated Log rollbacked. Snapshot will be attempted in the next cycle." +
+ "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
context.getReplicatedLog().getSnapshotIndex(),
context.getReplicatedLog().getSnapshotTerm(),
context.getReplicatedLog().size());
} else if (message instanceof CaptureSnapshot) {
- LOG.info("CaptureSnapshot received by actor");
- CaptureSnapshot cs = (CaptureSnapshot)message;
- captureSnapshot = cs;
- createSnapshot();
+ LOG.info("{}: CaptureSnapshot received by actor", persistenceId());
- } else if (message instanceof CaptureSnapshotReply){
- LOG.info("CaptureSnapshotReply received by actor");
- CaptureSnapshotReply csr = (CaptureSnapshotReply) message;
+ if(captureSnapshot == null) {
+ captureSnapshot = (CaptureSnapshot)message;
+ createSnapshot();
+ }
- ByteString stateInBytes = csr.getSnapshot();
- LOG.info("CaptureSnapshotReply stateInBytes size:{}", stateInBytes.size());
- handleCaptureSnapshotReply(stateInBytes);
+ } else if (message instanceof CaptureSnapshotReply){
+ handleCaptureSnapshotReply(((CaptureSnapshotReply) message).getSnapshot());
} else {
if (!(message instanceof AppendEntriesMessages.AppendEntries)
&& !(message instanceof AppendEntriesReply) && !(message instanceof SendHeartBeat)) {
if(LOG.isDebugEnabled()) {
- LOG.debug("onReceiveCommand: message: {}", message.getClass());
+ LOG.debug("{}: onReceiveCommand: message: {}", persistenceId(), message.getClass());
}
}
RaftActorBehavior oldBehavior = currentBehavior;
currentBehavior = currentBehavior.handleMessage(getSender(), message);
- if(oldBehavior != currentBehavior){
- onStateChanged();
- }
-
- onLeaderChanged(oldBehavior.getLeaderId(), currentBehavior.getLeaderId());
+ handleBehaviorChange(oldBehavior, currentBehavior);
}
}
- public java.util.Set<String> getPeers() {
+ private void handleBehaviorChange(RaftActorBehavior oldBehavior, RaftActorBehavior currentBehavior) {
+ if (oldBehavior != currentBehavior){
+ onStateChanged();
+ }
- return context.getPeerAddresses().keySet();
- }
+ String oldBehaviorLeaderId = oldBehavior == null? null : oldBehavior.getLeaderId();
+ String oldBehaviorState = oldBehavior == null? null : oldBehavior.state().name();
- protected String getReplicatedLogState() {
- return "snapshotIndex=" + context.getReplicatedLog().getSnapshotIndex()
- + ", snapshotTerm=" + context.getReplicatedLog().getSnapshotTerm()
- + ", im-mem journal size=" + context.getReplicatedLog().size();
- }
+ // it can happen that the state has not changed but the leader has changed.
+ onLeaderChanged(oldBehaviorLeaderId, currentBehavior.getLeaderId());
+ if (getRoleChangeNotifier().isPresent() &&
+ (oldBehavior == null || (oldBehavior.state() != currentBehavior.state()))) {
+ getRoleChangeNotifier().get().tell(
+ new RoleChanged(getId(), oldBehaviorState , currentBehavior.state().name()),
+ getSelf());
+ }
+ }
/**
* When a derived RaftActor needs to persist something it must call
* @param identifier
* @param data
*/
- protected void persistData(ActorRef clientActor, String identifier,
- Payload data) {
+ protected void persistData(final ActorRef clientActor, final String identifier,
+ final Payload data) {
ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
context.getReplicatedLog().lastIndex() + 1,
context.getTermInformation().getCurrentTerm(), data);
if(LOG.isDebugEnabled()) {
- LOG.debug("Persist data {}", replicatedLogEntry);
+ LOG.debug("{}: Persist data {}", persistenceId(), replicatedLogEntry);
}
+ final RaftActorContext raftContext = getRaftActorContext();
+
replicatedLog
- .appendAndPersist(clientActor, identifier, replicatedLogEntry);
- }
+ .appendAndPersist(replicatedLogEntry, new Procedure<ReplicatedLogEntry>() {
+ @Override
+ public void apply(ReplicatedLogEntry replicatedLogEntry) throws Exception {
+ if(!hasFollowers()){
+ // Increment the Commit Index and the Last Applied values
+ raftContext.setCommitIndex(replicatedLogEntry.getIndex());
+ raftContext.setLastApplied(replicatedLogEntry.getIndex());
+
+ // Apply the state immediately
+ applyState(clientActor, identifier, data);
+
+ // Send a ApplyLogEntries message so that we write the fact that we applied
+ // the state to durable storage
+ self().tell(new ApplyLogEntries((int) replicatedLogEntry.getIndex()), self());
+
+ // Check if the "real" snapshot capture has been initiated. If no then do the fake snapshot
+ if(!context.isSnapshotCaptureInitiated()){
+ raftContext.getReplicatedLog().snapshotPreCommit(raftContext.getLastApplied(),
+ raftContext.getTermInformation().getCurrentTerm());
+ raftContext.getReplicatedLog().snapshotCommit();
+ } else {
+ LOG.debug("{}: Skipping fake snapshotting for {} because real snapshotting is in progress",
+ persistenceId(), getId());
+ }
+ } else if (clientActor != null) {
+ // Send message for replication
+ currentBehavior.handleMessage(getSelf(),
+ new Replicate(clientActor, identifier,
+ replicatedLogEntry)
+ );
+ }
+
+ }
+ }); }
protected String getId() {
return context.getId();
*
* @param snapshot A snapshot of the state of the actor
*/
- protected abstract void applyRecoverySnapshot(ByteString snapshot);
+ protected abstract void applyRecoverySnapshot(byte[] snapshotBytes);
/**
* This method is called during recovery at the end of a batch to apply the current batched
* operations when the derived actor is out of sync with it's peers
* and the only way to bring it in sync is by applying a snapshot
*
- * @param snapshot A snapshot of the state of the actor
+ * @param snapshotBytes A snapshot of the state of the actor
*/
- protected abstract void applySnapshot(ByteString snapshot);
+ protected abstract void applySnapshot(byte[] snapshotBytes);
/**
* This method will be called by the RaftActor when the state of the
protected abstract DataPersistenceProvider persistence();
+ /**
+ * Notifier Actor for this RaftActor to notify when a role change happens
+ * @return ActorRef - ActorRef of the notifier or Optional.absent if none.
+ */
+ protected abstract Optional<ActorRef> getRoleChangeNotifier();
+
protected void onLeaderChanged(String oldLeader, String newLeader){};
private void trimPersistentData(long sequenceNumber) {
}
String peerAddress = context.getPeerAddress(leaderId);
if(LOG.isDebugEnabled()) {
- LOG.debug("getLeaderAddress leaderId = {} peerAddress = {}",
- leaderId, peerAddress);
+ LOG.debug("{}: getLeaderAddress leaderId = {} peerAddress = {}",
+ persistenceId(), leaderId, peerAddress);
}
return peerAddress;
}
- private void handleCaptureSnapshotReply(ByteString stateInBytes) {
+ private void handleCaptureSnapshotReply(byte[] snapshotBytes) {
+ LOG.info("{}: CaptureSnapshotReply received by actor: snapshot size {}", persistenceId(), snapshotBytes.length);
+
// create a snapshot object from the state provided and save it
// when snapshot is saved async, SaveSnapshotSuccess is raised.
- Snapshot sn = Snapshot.create(stateInBytes.toByteArray(),
+ Snapshot sn = Snapshot.create(snapshotBytes,
context.getReplicatedLog().getFrom(captureSnapshot.getLastAppliedIndex() + 1),
captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
persistence().saveSnapshot(sn);
- LOG.info("Persisting of snapshot done:{}", sn.getLogMessage());
+ LOG.info("{}: Persisting of snapshot done:{}", persistenceId(), sn.getLogMessage());
//be greedy and remove entries from in-mem journal which are in the snapshot
// and update snapshotIndex and snapshotTerm without waiting for the success,
captureSnapshot.getLastAppliedIndex(),
captureSnapshot.getLastAppliedTerm());
- LOG.info("Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
- "and term:{}", captureSnapshot.getLastAppliedIndex(),
+ LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
+ "and term:{}", persistenceId(), captureSnapshot.getLastAppliedIndex(),
captureSnapshot.getLastAppliedTerm());
if (isLeader() && captureSnapshot.isInstallSnapshotInitiated()) {
// this would be call straight to the leader and won't initiate in serialization
- currentBehavior.handleMessage(getSelf(), new SendInstallSnapshot(stateInBytes));
+ currentBehavior.handleMessage(getSelf(), new SendInstallSnapshot(
+ ByteString.copyFrom(snapshotBytes)));
}
captureSnapshot = null;
- hasSnapshotCaptureInitiated = false;
+ context.setSnapshotCaptureInitiated(false);
+ }
+
+ protected boolean hasFollowers(){
+ return getRaftActorContext().getPeerAddresses().keySet().size() > 0;
}
private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
+ private static final int DATA_SIZE_DIVIDER = 5;
+ private long dataSizeSinceLastSnapshot = 0;
+
public ReplicatedLogImpl(Snapshot snapshot) {
super(snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
snapshot.getUnAppliedEntries());
@Override public void apply(DeleteEntries param)
throws Exception {
//FIXME : Doing nothing for now
+ dataSize = 0;
+ for(ReplicatedLogEntry entry : journal){
+ dataSize += entry.size();
+ }
}
});
}
@Override public void appendAndPersist(
final ReplicatedLogEntry replicatedLogEntry) {
- appendAndPersist(null, null, replicatedLogEntry);
+ appendAndPersist(replicatedLogEntry, null);
}
- public void appendAndPersist(final ActorRef clientActor,
- final String identifier,
- final ReplicatedLogEntry replicatedLogEntry) {
+ @Override
+ public int dataSize() {
+ return dataSize;
+ }
+
+ public void appendAndPersist(
+ final ReplicatedLogEntry replicatedLogEntry,
+ final Procedure<ReplicatedLogEntry> callback) {
if(LOG.isDebugEnabled()) {
- LOG.debug("Append log entry and persist {} ", replicatedLogEntry);
+ LOG.debug("{}: Append log entry and persist {} ", persistenceId(), replicatedLogEntry);
}
// FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
new Procedure<ReplicatedLogEntry>() {
@Override
public void apply(ReplicatedLogEntry evt) throws Exception {
+ int logEntrySize = replicatedLogEntry.size();
+
+ dataSize += logEntrySize;
+ long dataSizeForCheck = dataSize;
+
+ dataSizeSinceLastSnapshot += logEntrySize;
+ long journalSize = lastIndex()+1;
+
+ if(!hasFollowers()) {
+ // When we do not have followers we do not maintain an in-memory log
+ // due to this the journalSize will never become anything close to the
+ // snapshot batch count. In fact will mostly be 1.
+ // Similarly since the journal's dataSize depends on the entries in the
+ // journal the journal's dataSize will never reach a value close to the
+ // memory threshold.
+ // By maintaining the dataSize outside the journal we are tracking essentially
+ // what we have written to the disk however since we no longer are in
+ // need of doing a snapshot just for the sake of freeing up memory we adjust
+ // the real size of data by the DATA_SIZE_DIVIDER so that we do not snapshot as often
+ // as if we were maintaining a real snapshot
+ dataSizeForCheck = dataSizeSinceLastSnapshot / DATA_SIZE_DIVIDER;
+ }
+
+ long dataThreshold = Runtime.getRuntime().totalMemory() *
+ getRaftActorContext().getConfigParams().getSnapshotDataThresholdPercentage() / 100;
+
// when a snaphsot is being taken, captureSnapshot != null
- if (hasSnapshotCaptureInitiated == false &&
- journal.size() % context.getConfigParams().getSnapshotBatchCount() == 0) {
+ if (!context.isSnapshotCaptureInitiated() &&
+ ( journalSize % context.getConfigParams().getSnapshotBatchCount() == 0 ||
+ dataSizeForCheck > dataThreshold)) {
+
+ dataSizeSinceLastSnapshot = 0;
- LOG.info("Initiating Snapshot Capture..");
+ LOG.info("{}: Initiating Snapshot Capture..", persistenceId());
long lastAppliedIndex = -1;
long lastAppliedTerm = -1;
ReplicatedLogEntry lastAppliedEntry = get(context.getLastApplied());
- if (lastAppliedEntry != null) {
+ if (!hasFollowers()) {
+ lastAppliedIndex = replicatedLogEntry.getIndex();
+ lastAppliedTerm = replicatedLogEntry.getTerm();
+ } else if (lastAppliedEntry != null) {
lastAppliedIndex = lastAppliedEntry.getIndex();
lastAppliedTerm = lastAppliedEntry.getTerm();
}
if(LOG.isDebugEnabled()) {
- LOG.debug("Snapshot Capture logSize: {}", journal.size());
- LOG.debug("Snapshot Capture lastApplied:{} ",
- context.getLastApplied());
- LOG.debug("Snapshot Capture lastAppliedIndex:{}", lastAppliedIndex);
- LOG.debug("Snapshot Capture lastAppliedTerm:{}", lastAppliedTerm);
+ LOG.debug("{}: Snapshot Capture logSize: {}", persistenceId(), journal.size());
+ LOG.debug("{}: Snapshot Capture lastApplied:{} ",
+ persistenceId(), context.getLastApplied());
+ LOG.debug("{}: Snapshot Capture lastAppliedIndex:{}", persistenceId(),
+ lastAppliedIndex);
+ LOG.debug("{}: Snapshot Capture lastAppliedTerm:{}", persistenceId(),
+ lastAppliedTerm);
}
// send a CaptureSnapshot to self to make the expensive operation async.
getSelf().tell(new CaptureSnapshot(
lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm),
null);
- hasSnapshotCaptureInitiated = true;
+ context.setSnapshotCaptureInitiated(true);
}
- // Send message for replication
- if (clientActor != null) {
- currentBehavior.handleMessage(getSelf(),
- new Replicate(clientActor, identifier,
- replicatedLogEntry)
- );
+ if(callback != null){
+ callback.apply(replicatedLogEntry);
}
}
}
@Override public void update(long currentTerm, String votedFor) {
if(LOG.isDebugEnabled()) {
- LOG.debug("Set currentTerm={}, votedFor={}", currentTerm, votedFor);
+ LOG.debug("{}: Set currentTerm={}, votedFor={}", persistenceId(), currentTerm, votedFor);
}
this.currentTerm = currentTerm;
this.votedFor = votedFor;
protected RaftActorBehavior getCurrentBehavior() {
return currentBehavior;
}
+
}