import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
import org.opendaylight.controller.cluster.raft.behaviors.AbstractLeader;
-import org.opendaylight.controller.cluster.raft.behaviors.AbstractRaftActorBehavior;
+import org.opendaylight.controller.cluster.raft.behaviors.DelegatingRaftActorBehavior;
import org.opendaylight.controller.cluster.raft.behaviors.Follower;
import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
* The current state determines the current behavior of a RaftActor
* A Raft Actor always starts off in the Follower State
*/
- private RaftActorBehavior currentBehavior;
+ private final DelegatingRaftActorBehavior currentBehavior = new DelegatingRaftActorBehavior();
/**
* This context should NOT be passed directly to any other actor it is
private final Procedure<Void> createSnapshotProcedure = new CreateSnapshotProcedure();
- /**
- * The in-memory journal
- */
- private ReplicatedLogImpl replicatedLog = new ReplicatedLogImpl();
-
private Stopwatch recoveryTimer;
private int currentRecoveryBatchCount;
context = new RaftActorContextImpl(this.getSelf(),
this.getContext(), id, new ElectionTermImpl(delegatingPersistenceProvider, id, LOG),
- -1, -1, replicatedLog, peerAddresses,
- (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
- LOG);
+ -1, -1, peerAddresses,
+ (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()), LOG);
+
+ context.setReplicatedLog(ReplicatedLogImpl.newInstance(context, delegatingPersistenceProvider, currentBehavior));
}
private void initRecoveryTimer() {
} else if (message instanceof ApplyJournalEntries) {
onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
} else if (message instanceof DeleteEntries) {
- replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
+ replicatedLog().removeFrom(((DeleteEntries) message).getFromIndex());
} else if (message instanceof UpdateElectionTerm) {
context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
((UpdateElectionTerm) message).getVotedFor());
// Create a replicated log with the snapshot information
// The replicated log can be used later on to retrieve this snapshot
// when we need to install it on a peer
- replicatedLog = new ReplicatedLogImpl(snapshot);
- context.setReplicatedLog(replicatedLog);
+ context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, delegatingPersistenceProvider,
+ currentBehavior));
context.setLastApplied(snapshot.getLastAppliedIndex());
context.setCommitIndex(snapshot.getLastAppliedIndex());
timer.stop();
LOG.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size=" +
- replicatedLog.size(), persistenceId(), timer.toString(),
- replicatedLog.getSnapshotIndex(), replicatedLog.getSnapshotTerm());
+ replicatedLog().size(), persistenceId(), timer.toString(),
+ replicatedLog().getSnapshotIndex(), replicatedLog().getSnapshotTerm());
}
private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
LOG.debug("{}: Received ReplicatedLogEntry for recovery: {}", persistenceId(), logEntry.getIndex());
}
- replicatedLog.append(logEntry);
+ replicatedLog().append(logEntry);
}
private void onRecoveredApplyLogEntries(long toIndex) {
}
for (long i = context.getLastApplied() + 1; i <= toIndex; i++) {
- batchRecoveredLogEntry(replicatedLog.get(i));
+ batchRecoveredLogEntry(replicatedLog().get(i));
}
context.setLastApplied(toIndex);
"Persistence Id = " + persistenceId() +
" Last index in log={}, snapshotIndex={}, snapshotTerm={}, " +
"journal-size={}",
- replicatedLog.lastIndex(), replicatedLog.getSnapshotIndex(),
- replicatedLog.getSnapshotTerm(), replicatedLog.size());
+ replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
+ replicatedLog().getSnapshotTerm(), replicatedLog().size());
initializeBehavior();
}
}
protected void changeCurrentBehavior(RaftActorBehavior newBehavior){
- reusableBehaviorStateHolder.init(currentBehavior);
- currentBehavior = newBehavior;
- handleBehaviorChange(reusableBehaviorStateHolder, currentBehavior);
+ reusableBehaviorStateHolder.init(getCurrentBehavior());
+ setCurrentBehavior(newBehavior);
+ handleBehaviorChange(reusableBehaviorStateHolder, getCurrentBehavior());
}
@Override public void handleCommand(Object message) {
applySnapshot(snapshot.getState());
//clears the followers log, sets the snapshot index to ensure adjusted-index works
- replicatedLog = new ReplicatedLogImpl(snapshot);
- context.setReplicatedLog(replicatedLog);
+ context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, delegatingPersistenceProvider,
+ currentBehavior));
context.setLastApplied(snapshot.getLastAppliedIndex());
} else if (message instanceof FindLeader) {
} else if (message.equals(COMMIT_SNAPSHOT)) {
commitSnapshot(-1);
} else {
- reusableBehaviorStateHolder.init(currentBehavior);
+ reusableBehaviorStateHolder.init(getCurrentBehavior());
- currentBehavior = currentBehavior.handleMessage(getSender(), message);
+ setCurrentBehavior(currentBehavior.handleMessage(getSender(), message));
- handleBehaviorChange(reusableBehaviorStateHolder, currentBehavior);
+ handleBehaviorChange(reusableBehaviorStateHolder, getCurrentBehavior());
}
}
OnDemandRaftState.Builder builder = OnDemandRaftState.builder()
.commitIndex(context.getCommitIndex())
.currentTerm(context.getTermInformation().getCurrentTerm())
- .inMemoryJournalDataSize(replicatedLog.dataSize())
- .inMemoryJournalLogSize(replicatedLog.size())
+ .inMemoryJournalDataSize(replicatedLog().dataSize())
+ .inMemoryJournalLogSize(replicatedLog().size())
.isSnapshotCaptureInitiated(context.getSnapshotManager().isCapturing())
.lastApplied(context.getLastApplied())
- .lastIndex(replicatedLog.lastIndex())
- .lastTerm(replicatedLog.lastTerm())
+ .lastIndex(replicatedLog().lastIndex())
+ .lastTerm(replicatedLog().lastTerm())
.leader(getLeaderId())
.raftState(currentBehavior.state().toString())
.replicatedToAllIndex(currentBehavior.getReplicatedToAllIndex())
- .snapshotIndex(replicatedLog.getSnapshotIndex())
- .snapshotTerm(replicatedLog.getSnapshotTerm())
+ .snapshotIndex(replicatedLog().getSnapshotIndex())
+ .snapshotTerm(replicatedLog().getSnapshotTerm())
.votedFor(context.getTermInformation().getVotedFor())
.peerAddresses(ImmutableMap.copyOf(context.getPeerAddresses()));
builder.lastLogTerm(lastLogEntry.getTerm());
}
- if(currentBehavior instanceof AbstractLeader) {
- AbstractLeader leader = (AbstractLeader)currentBehavior;
+ if(getCurrentBehavior() instanceof AbstractLeader) {
+ AbstractLeader leader = (AbstractLeader)getCurrentBehavior();
Collection<String> followerIds = leader.getFollowerIds();
List<FollowerInfo> followerInfoList = Lists.newArrayListWithCapacity(followerIds.size());
for(String id: followerIds) {
final RaftActorContext raftContext = getRaftActorContext();
- replicatedLog
- .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());
+ replicatedLog().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);
+ // Apply the state immediately
+ applyState(clientActor, identifier, data);
- // Send a ApplyJournalEntries message so that we write the fact that we applied
- // the state to durable storage
- self().tell(new ApplyJournalEntries(replicatedLogEntry.getIndex()), self());
+ // Send a ApplyJournalEntries message so that we write the fact that we applied
+ // the state to durable storage
+ self().tell(new ApplyJournalEntries(replicatedLogEntry.getIndex()), self());
- context.getSnapshotManager().trimLog(context.getLastApplied(), currentBehavior);
+ context.getSnapshotManager().trimLog(context.getLastApplied(), currentBehavior);
- } else if (clientActor != null) {
- // Send message for replication
- currentBehavior.handleMessage(getSelf(),
- new Replicate(clientActor, identifier,
- replicatedLogEntry)
- );
- }
+ } else if (clientActor != null) {
+ // Send message for replication
+ currentBehavior.handleMessage(getSelf(),
+ new Replicate(clientActor, identifier, replicatedLogEntry));
+ }
+ }
+ });
+ }
- }
- }); }
+ private ReplicatedLog replicatedLog() {
+ return context.getReplicatedLog();
+ }
protected String getId() {
return context.getId();
}
+ @VisibleForTesting
+ void setCurrentBehavior(RaftActorBehavior behavior) {
+ currentBehavior.setDelegate(behavior);
+ }
+
+ protected RaftActorBehavior getCurrentBehavior() {
+ return currentBehavior.getDelegate();
+ }
+
/**
* Derived actors can call the isLeader method to check if the current
* RaftActor is the Leader or not
}
protected ReplicatedLogEntry getLastLogEntry() {
- return replicatedLog.last();
+ return replicatedLog().last();
}
protected Long getCurrentTerm(){
private void handleCaptureSnapshotReply(byte[] snapshotBytes) {
LOG.debug("{}: CaptureSnapshotReply received by actor: snapshot size {}", persistenceId(), snapshotBytes.length);
- context.getSnapshotManager().persist(persistence(), snapshotBytes, currentBehavior, getTotalMemory());
- }
-
- protected long getTotalMemory() {
- return Runtime.getRuntime().totalMemory();
+ context.getSnapshotManager().persist(persistence(), snapshotBytes, currentBehavior, context.getTotalMemory());
}
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 = 0L;
-
-
- public ReplicatedLogImpl(Snapshot snapshot) {
- super(snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
- snapshot.getUnAppliedEntries());
- }
-
- public ReplicatedLogImpl() {
- super();
- }
-
- @Override public void removeFromAndPersist(long logEntryIndex) {
- int adjustedIndex = adjustedIndex(logEntryIndex);
-
- if (adjustedIndex < 0) {
- return;
- }
-
- // FIXME: Maybe this should be done after the command is saved
- journal.subList(adjustedIndex , journal.size()).clear();
-
- persistence().persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>() {
-
- @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(replicatedLogEntry, null);
- }
-
- public void appendAndPersist(
- final ReplicatedLogEntry replicatedLogEntry,
- final Procedure<ReplicatedLogEntry> callback) {
-
- if(LOG.isDebugEnabled()) {
- 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
- journal.add(replicatedLogEntry);
-
- // When persisting events with persist it is guaranteed that the
- // persistent actor will not receive further commands between the
- // persist call and the execution(s) of the associated event
- // handler. This also holds for multiple persist calls in context
- // of a single command.
- persistence().persist(replicatedLogEntry,
- new Procedure<ReplicatedLogEntry>() {
- @Override
- public void apply(ReplicatedLogEntry evt) throws Exception {
- int logEntrySize = replicatedLogEntry.size();
-
- dataSize += logEntrySize;
- long dataSizeForCheck = dataSize;
-
- dataSizeSinceLastSnapshot += logEntrySize;
-
- 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 journalSize = replicatedLogEntry.getIndex() + 1;
- long dataThreshold = getTotalMemory() *
- context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
-
- if ((journalSize % context.getConfigParams().getSnapshotBatchCount() == 0
- || dataSizeForCheck > dataThreshold)) {
-
- boolean started = context.getSnapshotManager().capture(replicatedLogEntry,
- currentBehavior.getReplicatedToAllIndex());
-
- if(started){
- dataSizeSinceLastSnapshot = 0;
- }
-
- }
-
- if (callback != null){
- callback.apply(replicatedLogEntry);
- }
- }
- }
- );
- }
-
+ return getRaftActorContext().hasFollowers();
}
static class DeleteEntries implements Serializable {
}
}
- @VisibleForTesting
- void setCurrentBehavior(AbstractRaftActorBehavior behavior) {
- currentBehavior = behavior;
- }
-
- protected RaftActorBehavior getCurrentBehavior() {
- return currentBehavior;
- }
-
private static class BehaviorStateHolder {
private RaftActorBehavior behavior;
private String leaderId;
--- /dev/null
+/*
+ * Copyright (c) 2015 Cisco Systems, Inc. and others. All rights reserved.
+ *
+ * This program and the accompanying materials are made available under the
+ * 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;
+
+import akka.japi.Procedure;
+import java.util.Collections;
+import java.util.List;
+import org.opendaylight.controller.cluster.DataPersistenceProvider;
+import org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries;
+import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
+
+/**
+ * Implementation of ReplicatedLog used by the RaftActor.
+ */
+class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
+ private static final int DATA_SIZE_DIVIDER = 5;
+
+ private long dataSizeSinceLastSnapshot = 0L;
+ private final RaftActorContext context;
+ private final DataPersistenceProvider persistence;
+ private final RaftActorBehavior currentBehavior;
+
+ private final Procedure<DeleteEntries> deleteProcedure = new Procedure<DeleteEntries>() {
+ @Override
+ public void apply(DeleteEntries param) {
+ dataSize = 0;
+ for (ReplicatedLogEntry entry : journal) {
+ dataSize += entry.size();
+ }
+ }
+ };
+
+ static ReplicatedLog newInstance(Snapshot snapshot, RaftActorContext context,
+ DataPersistenceProvider persistence, RaftActorBehavior currentBehavior) {
+ return new ReplicatedLogImpl(snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
+ snapshot.getUnAppliedEntries(), context, persistence, currentBehavior);
+ }
+
+ static ReplicatedLog newInstance(RaftActorContext context,
+ DataPersistenceProvider persistence, RaftActorBehavior currentBehavior) {
+ return new ReplicatedLogImpl(-1L, -1L, Collections.<ReplicatedLogEntry>emptyList(), context,
+ persistence, currentBehavior);
+ }
+
+ private ReplicatedLogImpl(long snapshotIndex, long snapshotTerm, List<ReplicatedLogEntry> unAppliedEntries,
+ RaftActorContext context, DataPersistenceProvider persistence, RaftActorBehavior currentBehavior) {
+ super(snapshotIndex, snapshotTerm, unAppliedEntries);
+ this.context = context;
+ this.persistence = persistence;
+ this.currentBehavior = currentBehavior;
+ }
+
+ @Override
+ public void removeFromAndPersist(long logEntryIndex) {
+ int adjustedIndex = adjustedIndex(logEntryIndex);
+
+ if (adjustedIndex < 0) {
+ return;
+ }
+
+ // FIXME: Maybe this should be done after the command is saved
+ journal.subList(adjustedIndex , journal.size()).clear();
+
+ persistence.persist(new DeleteEntries(adjustedIndex), deleteProcedure);
+ }
+
+ @Override
+ public void appendAndPersist(final ReplicatedLogEntry replicatedLogEntry) {
+ appendAndPersist(replicatedLogEntry, null);
+ }
+
+ @Override
+ public void appendAndPersist(final ReplicatedLogEntry replicatedLogEntry,
+ final Procedure<ReplicatedLogEntry> callback) {
+
+ if(context.getLogger().isDebugEnabled()) {
+ context.getLogger().debug("{}: Append log entry and persist {} ", context.getId(), replicatedLogEntry);
+ }
+
+ // FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
+ journal.add(replicatedLogEntry);
+
+ // When persisting events with persist it is guaranteed that the
+ // persistent actor will not receive further commands between the
+ // persist call and the execution(s) of the associated event
+ // handler. This also holds for multiple persist calls in context
+ // of a single command.
+ persistence.persist(replicatedLogEntry,
+ new Procedure<ReplicatedLogEntry>() {
+ @Override
+ public void apply(ReplicatedLogEntry evt) throws Exception {
+ int logEntrySize = replicatedLogEntry.size();
+
+ dataSize += logEntrySize;
+ long dataSizeForCheck = dataSize;
+
+ dataSizeSinceLastSnapshot += logEntrySize;
+
+ if (!context.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 journalSize = replicatedLogEntry.getIndex() + 1;
+ long dataThreshold = context.getTotalMemory() *
+ context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
+
+ if ((journalSize % context.getConfigParams().getSnapshotBatchCount() == 0
+ || dataSizeForCheck > dataThreshold)) {
+
+ boolean started = context.getSnapshotManager().capture(replicatedLogEntry,
+ currentBehavior.getReplicatedToAllIndex());
+
+ if(started){
+ dataSizeSinceLastSnapshot = 0;
+ }
+ }
+
+ if (callback != null){
+ callback.apply(replicatedLogEntry);
+ }
+ }
+ }
+ );
+ }
+}
\ No newline at end of file
Code Review / controller.git / commitdiff