Merge "BUG 1623 - Clustering : Parsing Error thrown on startup"
[controller.git] / opendaylight / md-sal / sal-akka-raft / src / main / java / org / opendaylight / controller / cluster / raft / RaftActor.java
index 826faf7414395bdb1eefcc59906c3cd1f6ea0e16..8135d837d3ad86563ad0246022941bfc6134b59d 100644 (file)
@@ -14,69 +14,80 @@ import akka.event.Logging;
 import akka.event.LoggingAdapter;
 import akka.japi.Procedure;
 import akka.persistence.RecoveryCompleted;
+import akka.persistence.SaveSnapshotFailure;
+import akka.persistence.SaveSnapshotSuccess;
+import akka.persistence.SnapshotOffer;
+import akka.persistence.SnapshotSelectionCriteria;
 import akka.persistence.UntypedPersistentActor;
+import com.google.common.base.Optional;
+import com.google.protobuf.ByteString;
+import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
+import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
+import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
+import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
+import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
+import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
 import org.opendaylight.controller.cluster.raft.behaviors.Candidate;
 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
 import org.opendaylight.controller.cluster.raft.behaviors.Leader;
 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
+import org.opendaylight.controller.cluster.raft.client.messages.AddRaftPeer;
 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
-import org.opendaylight.controller.cluster.raft.internal.messages.ApplyState;
-import org.opendaylight.controller.cluster.raft.internal.messages.Replicate;
+import org.opendaylight.controller.cluster.raft.client.messages.RemoveRaftPeer;
+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.ArrayList;
-import java.util.List;
 import java.util.Map;
 
 /**
  * RaftActor encapsulates a state machine that needs to be kept synchronized
  * in a cluster. It implements the RAFT algorithm as described in the paper
  * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
- *     In Search of an Understandable Consensus Algorithm</a>
- * <p>
+ * In Search of an Understandable Consensus Algorithm</a>
+ * <p/>
  * RaftActor has 3 states and each state has a certain behavior associated
  * with it. A Raft actor can behave as,
  * <ul>
- *     <li> A Leader </li>
- *     <li> A Follower (or) </li>
- *     <li> A Candidate </li>
+ * <li> A Leader </li>
+ * <li> A Follower (or) </li>
+ * <li> A Candidate </li>
  * </ul>
- *
- * <p>
+ * <p/>
+ * <p/>
  * A RaftActor MUST be a Leader in order to accept requests from clients to
  * change the state of it's encapsulated state machine. Once a RaftActor becomes
  * a Leader it is also responsible for ensuring that all followers ultimately
  * have the same log and therefore the same state machine as itself.
- *
- * <p>
+ * <p/>
+ * <p/>
  * The current behavior of a RaftActor determines how election for leadership
  * is initiated and how peer RaftActors react to request for votes.
- *
- * <p>
+ * <p/>
+ * <p/>
  * Each RaftActor also needs to know the current election term. It uses this
  * information for a couple of things. One is to simply figure out who it
  * voted for in the last election. Another is to figure out if the message
  * it received to update it's state is stale.
- *
- * <p>
+ * <p/>
+ * <p/>
  * The RaftActor uses akka-persistence to store it's replicated log.
  * Furthermore through it's behaviors a Raft Actor determines
- *
+ * <p/>
  * <ul>
  * <li> when a log entry should be persisted </li>
  * <li> when a log entry should be applied to the state machine (and) </li>
  * <li> when a snapshot should be saved </li>
  * </ul>
- *
- * <a href="http://doc.akka.io/api/akka/2.3.3/index.html#akka.persistence.UntypedEventsourcedProcessor">UntypeEventSourceProcessor</a>
  */
 public abstract class RaftActor extends UntypedPersistentActor {
     protected final LoggingAdapter LOG =
         Logging.getLogger(getContext().system(), this);
 
     /**
-     *  The current state determines the current behavior of a RaftActor
+     * The current state determines the current behavior of a RaftActor
      * A Raft Actor always starts off in the Follower State
      */
     private RaftActorBehavior currentBehavior;
@@ -92,46 +103,324 @@ public abstract class RaftActor extends UntypedPersistentActor {
      */
     private ReplicatedLogImpl replicatedLog = new ReplicatedLogImpl();
 
+    private CaptureSnapshot captureSnapshot = null;
 
+    private volatile boolean hasSnapshotCaptureInitiated = false;
+
+    public RaftActor(String id, Map<String, String> peerAddresses) {
+        this(id, peerAddresses, Optional.<ConfigParams>absent());
+    }
+
+    public RaftActor(String id, Map<String, String> peerAddresses,
+         Optional<ConfigParams> configParams) {
 
-    public RaftActor(String id, Map<String, String> peerAddresses){
         context = new RaftActorContextImpl(this.getSelf(),
-            this.getContext(),
-            id, new ElectionTermImpl(getSelf().path().toString()),
-            -1, -1, replicatedLog, peerAddresses, LOG);
-        currentBehavior = switchBehavior(RaftState.Follower);
+            this.getContext(), id, new ElectionTermImpl(),
+            -1, -1, replicatedLog, peerAddresses,
+            (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
+            LOG);
     }
 
     @Override public void onReceiveRecover(Object message) {
-        if(message instanceof ReplicatedLogEntry) {
+        if (message instanceof SnapshotOffer) {
+            LOG.debug("SnapshotOffer called..");
+            SnapshotOffer offer = (SnapshotOffer) message;
+            Snapshot snapshot = (Snapshot) offer.snapshot();
+
+            // 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.setLastApplied(snapshot.getLastAppliedIndex());
+
+            LOG.debug("Applied snapshot to replicatedLog. " +
+                "snapshotIndex={}, snapshotTerm={}, journal-size={}",
+                replicatedLog.snapshotIndex, replicatedLog.snapshotTerm,
+                replicatedLog.size());
+
+            // Apply the snapshot to the actors state
+            applySnapshot(ByteString.copyFrom(snapshot.getState()));
+
+        } else if (message instanceof ReplicatedLogEntry) {
             replicatedLog.append((ReplicatedLogEntry) message);
-        } else if(message instanceof RecoveryCompleted){
-            LOG.debug("Log now has messages to index : " + replicatedLog.lastIndex());
+        } else if (message instanceof DeleteEntries) {
+            replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
+        } else if (message instanceof UpdateElectionTerm) {
+            context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(), ((UpdateElectionTerm) message).getVotedFor());
+        } else if (message instanceof RecoveryCompleted) {
+            LOG.debug(
+                "RecoveryCompleted - Switching actor to Follower - " +
+                    "Last index in log:{}, snapshotIndex={}, snapshotTerm={}, " +
+                    "journal-size={}",
+                replicatedLog.lastIndex(), replicatedLog.snapshotIndex,
+                replicatedLog.snapshotTerm, replicatedLog.size());
+            currentBehavior = switchBehavior(RaftState.Follower);
+            onStateChanged();
         }
     }
 
     @Override public void onReceiveCommand(Object message) {
-        if(message instanceof ApplyState){
+        if (message instanceof ApplyState){
+            ApplyState applyState = (ApplyState) message;
 
-            ApplyState applyState = (ApplyState)  message;
-
-            LOG.debug("Applying state for log index {}", applyState.getReplicatedLogEntry().getIndex());
+            LOG.debug("Applying state for log index {} data {}",
+                applyState.getReplicatedLogEntry().getIndex(),
+                applyState.getReplicatedLogEntry().getData());
 
             applyState(applyState.getClientActor(), applyState.getIdentifier(),
                 applyState.getReplicatedLogEntry().getData());
-        } else if(message instanceof FindLeader){
-            getSender().tell(new FindLeaderReply(
-                context.getPeerAddress(currentBehavior.getLeaderId())),
-                getSelf());
+
+        } else if(message instanceof ApplySnapshot ) {
+            Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
+
+            LOG.debug("ApplySnapshot called on Follower Actor " +
+                "snapshotIndex:{}, snapshotTerm:{}", snapshot.getLastAppliedIndex(),
+                snapshot.getLastAppliedTerm());
+            applySnapshot(ByteString.copyFrom(snapshot.getState()));
+
+            //clears the followers log, sets the snapshot index to ensure adjusted-index works
+            replicatedLog = new ReplicatedLogImpl(snapshot);
+            context.setReplicatedLog(replicatedLog);
+            context.setLastApplied(snapshot.getLastAppliedIndex());
+
+        } else if (message instanceof FindLeader) {
+            getSender().tell(
+                new FindLeaderReply(getLeaderAddress()),
+                getSelf()
+            );
+
+        } else if (message instanceof SaveSnapshotSuccess) {
+            SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
+            LOG.info("SaveSnapshotSuccess received for snapshot");
+
+            context.getReplicatedLog().snapshotCommit();
+
+            // TODO: Not sure if we want to be this aggressive with trimming stuff
+            trimPersistentData(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:");
+
+            context.getReplicatedLog().snapshotRollback();
+
+            LOG.info("Replicated Log rollbacked. Snapshot will be attempted in the next cycle." +
+                "snapshotIndex:{}, snapshotTerm:{}, log-size:{}",
+                context.getReplicatedLog().getSnapshotIndex(),
+                context.getReplicatedLog().getSnapshotTerm(),
+                context.getReplicatedLog().size());
+
+        } else if (message instanceof AddRaftPeer){
+
+            // FIXME : Do not add raft peers like this.
+            // When adding a new Peer we have to ensure that the a majority of
+            // the peers know about the new Peer. Doing it this way may cause
+            // a situation where multiple Leaders may emerge
+            AddRaftPeer arp = (AddRaftPeer)message;
+           context.addToPeers(arp.getName(), arp.getAddress());
+
+        } else if (message instanceof RemoveRaftPeer){
+
+            RemoveRaftPeer rrp = (RemoveRaftPeer)message;
+            context.removePeer(rrp.getName());
+
+        } else if (message instanceof CaptureSnapshot) {
+            LOG.debug("CaptureSnapshot received by actor");
+            CaptureSnapshot cs = (CaptureSnapshot)message;
+            captureSnapshot = cs;
+            createSnapshot();
+
+        } else if (message instanceof CaptureSnapshotReply){
+            LOG.debug("CaptureSnapshotReply received by actor");
+            CaptureSnapshotReply csr = (CaptureSnapshotReply) message;
+
+            ByteString stateInBytes = csr.getSnapshot();
+            LOG.debug("CaptureSnapshotReply stateInBytes size:{}", stateInBytes.size());
+            handleCaptureSnapshotReply(stateInBytes);
+
         } else {
+            if (!(message instanceof AppendEntriesMessages.AppendEntries)
+                && !(message instanceof AppendEntriesReply) && !(message instanceof SendHeartBeat)) {
+                LOG.debug("onReceiveCommand: message:" + message.getClass());
+            }
+
             RaftState state =
                 currentBehavior.handleMessage(getSender(), message);
+            RaftActorBehavior oldBehavior = currentBehavior;
             currentBehavior = switchBehavior(state);
+            if(oldBehavior != currentBehavior){
+                onStateChanged();
+            }
+        }
+    }
+
+    public java.util.Set<String> getPeers() {
+        return context.getPeerAddresses().keySet();
+    }
+
+    protected String getReplicatedLogState() {
+        return "snapshotIndex=" + context.getReplicatedLog().getSnapshotIndex()
+            + ", snapshotTerm=" + context.getReplicatedLog().getSnapshotTerm()
+            + ", im-mem journal size=" + context.getReplicatedLog().size();
+    }
+
+
+    /**
+     * When a derived RaftActor needs to persist something it must call
+     * persistData.
+     *
+     * @param clientActor
+     * @param identifier
+     * @param data
+     */
+    protected void persistData(ActorRef clientActor, String identifier,
+        Payload data) {
+
+        ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
+            context.getReplicatedLog().lastIndex() + 1,
+            context.getTermInformation().getCurrentTerm(), data);
+
+        LOG.debug("Persist data {}", replicatedLogEntry);
+
+        replicatedLog
+            .appendAndPersist(clientActor, identifier, replicatedLogEntry);
+    }
+
+    protected String getId() {
+        return context.getId();
+    }
+
+    /**
+     * Derived actors can call the isLeader method to check if the current
+     * RaftActor is the Leader or not
+     *
+     * @return true it this RaftActor is a Leader false otherwise
+     */
+    protected boolean isLeader() {
+        return context.getId().equals(currentBehavior.getLeaderId());
+    }
+
+    /**
+     * Derived actor can call getLeader if they need a reference to the Leader.
+     * This would be useful for example in forwarding a request to an actor
+     * which is the leader
+     *
+     * @return A reference to the leader if known, null otherwise
+     */
+    protected ActorSelection getLeader(){
+        String leaderAddress = getLeaderAddress();
+
+        if(leaderAddress == null){
+            return null;
         }
+
+        return context.actorSelection(leaderAddress);
+    }
+
+    /**
+     *
+     * @return the current leader's id
+     */
+    protected String getLeaderId(){
+        return currentBehavior.getLeaderId();
+    }
+
+    protected RaftState getRaftState() {
+        return currentBehavior.state();
+    }
+
+    protected ReplicatedLogEntry getLastLogEntry() {
+        return replicatedLog.last();
+    }
+
+    protected Long getCurrentTerm(){
+        return context.getTermInformation().getCurrentTerm();
     }
 
-    private RaftActorBehavior switchBehavior(RaftState state){
-        if(currentBehavior != null) {
+    protected Long getCommitIndex(){
+        return context.getCommitIndex();
+    }
+
+    protected Long getLastApplied(){
+        return context.getLastApplied();
+    }
+
+    /**
+     * setPeerAddress sets the address of a known peer at a later time.
+     * <p>
+     * This is to account for situations where a we know that a peer
+     * exists but we do not know an address up-front. This may also be used in
+     * situations where a known peer starts off in a different location and we
+     * need to change it's address
+     * <p>
+     * Note that if the peerId does not match the list of peers passed to
+     * this actor during construction an IllegalStateException will be thrown.
+     *
+     * @param peerId
+     * @param peerAddress
+     */
+    protected void setPeerAddress(String peerId, String peerAddress){
+        context.setPeerAddress(peerId, peerAddress);
+    }
+
+
+
+    /**
+     * The applyState method will be called by the RaftActor when some data
+     * needs to be applied to the actor's state
+     *
+     * @param clientActor A reference to the client who sent this message. This
+     *                    is the same reference that was passed to persistData
+     *                    by the derived actor. clientActor may be null when
+     *                    the RaftActor is behaving as a follower or during
+     *                    recovery.
+     * @param identifier  The identifier of the persisted data. This is also
+     *                    the same identifier that was passed to persistData by
+     *                    the derived actor. identifier may be null when
+     *                    the RaftActor is behaving as a follower or during
+     *                    recovery
+     * @param data        A piece of data that was persisted by the persistData call.
+     *                    This should NEVER be null.
+     */
+    protected abstract void applyState(ActorRef clientActor, String identifier,
+        Object data);
+
+    /**
+     * This method will be called by the RaftActor when a snapshot needs to be
+     * created. The derived actor should respond with its current state.
+     * <p/>
+     * During recovery the state that is returned by the derived actor will
+     * be passed back to it by calling the applySnapshot  method
+     *
+     * @return The current state of the actor
+     */
+    protected abstract void createSnapshot();
+
+    /**
+     * This method will be called by the RaftActor during recovery to
+     * reconstruct the state of the actor.
+     * <p/>
+     * This method may also be called at any other point during normal
+     * 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
+     */
+    protected abstract void applySnapshot(ByteString snapshot);
+
+    /**
+     * This method will be called by the RaftActor when the state of the
+     * RaftActor changes. The derived actor can then use methods like
+     * isLeader or getLeader to do something useful
+     */
+    protected abstract void onStateChanged();
+
+    private RaftActorBehavior switchBehavior(RaftState state) {
+        if (currentBehavior != null) {
             if (currentBehavior.state() == state) {
                 return currentBehavior;
             }
@@ -141,172 +430,239 @@ public abstract class RaftActor extends UntypedPersistentActor {
             try {
                 currentBehavior.close();
             } catch (Exception e) {
-                LOG.error(e, "Failed to close behavior : " + currentBehavior.state());
+                LOG.error(e,
+                    "Failed to close behavior : " + currentBehavior.state());
             }
 
         } else {
             LOG.info("Switching behavior to " + state);
         }
         RaftActorBehavior behavior = null;
-        if(state == RaftState.Candidate){
+        if (state == RaftState.Candidate) {
             behavior = new Candidate(context);
-        } else if(state == RaftState.Follower){
+        } else if (state == RaftState.Follower) {
             behavior = new Follower(context);
         } else {
             behavior = new Leader(context);
         }
-        return behavior;
-    }
-
-    /**
-     * When a derived RaftActor needs to persist something it must call
-     * persistData.
-     *
-     * @param clientActor
-     * @param identifier
-     * @param data
-     */
-    protected void persistData(ActorRef clientActor, String identifier, Object data){
-        LOG.debug("Persist data " + identifier);
-        ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
-            context.getReplicatedLog().lastIndex() + 1,
-            context.getTermInformation().getCurrentTerm(), data);
 
-        replicatedLog.appendAndPersist(clientActor, identifier, replicatedLogEntry);
-    }
 
-    protected abstract void applyState(ActorRef clientActor, String identifier, Object data);
 
-    protected String getId(){
-        return context.getId();
+        return behavior;
     }
 
-    protected boolean isLeader(){
-        return context.getId().equals(currentBehavior.getLeaderId());
+    private void trimPersistentData(long sequenceNumber) {
+        // Trim akka snapshots
+        // FIXME : Not sure how exactly the SnapshotSelectionCriteria is applied
+        // For now guessing that it is ANDed.
+        deleteSnapshots(new SnapshotSelectionCriteria(
+            sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), 43200000));
+
+        // Trim akka journal
+        deleteMessages(sequenceNumber);
     }
 
-    protected ActorSelection getLeader(){
+    private String getLeaderAddress(){
+        if(isLeader()){
+            return getSelf().path().toString();
+        }
         String leaderId = currentBehavior.getLeaderId();
+        if (leaderId == null) {
+            return null;
+        }
         String peerAddress = context.getPeerAddress(leaderId);
-        LOG.debug("getLeader leaderId = " + leaderId + " peerAddress = " + peerAddress);
-        return context.actorSelection(peerAddress);
+        LOG.debug("getLeaderAddress leaderId = " + leaderId + " peerAddress = "
+            + peerAddress);
+
+        return peerAddress;
     }
 
-    private class ReplicatedLogImpl implements ReplicatedLog {
-        private final List<ReplicatedLogEntry> journal = new ArrayList();
-        private long snapshotIndex = 0;
-        private Object snapShot = null;
+    private void handleCaptureSnapshotReply(ByteString stateInBytes) {
+        // 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(),
+            context.getReplicatedLog().getFrom(captureSnapshot.getLastAppliedIndex() + 1),
+            captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
+            captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
 
-        @Override public ReplicatedLogEntry get(long index) {
-            if(index < 0 || index >= journal.size()){
-                return null;
-            }
+        saveSnapshot(sn);
 
-            return journal.get((int) (index - snapshotIndex));
-        }
+        LOG.info("Persisting of snapshot done:{}", sn.getLogMessage());
 
-        @Override public ReplicatedLogEntry last() {
-            if(journal.size() == 0){
-                return null;
-            }
-            return get(journal.size() - 1);
-        }
+        //be greedy and remove entries from in-mem journal which are in the snapshot
+        // and update snapshotIndex and snapshotTerm without waiting for the success,
 
-        @Override public long lastIndex() {
-            if(journal.size() == 0){
-                return -1;
-            }
+        context.getReplicatedLog().snapshotPreCommit(stateInBytes,
+            captureSnapshot.getLastAppliedIndex(),
+            captureSnapshot.getLastAppliedTerm());
 
-            return last().getIndex();
-        }
+        LOG.info("Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
+            "and term:{}", captureSnapshot.getLastAppliedIndex(),
+            captureSnapshot.getLastAppliedTerm());
 
-        @Override public long lastTerm() {
-            if(journal.size() == 0){
-                return -1;
-            }
+        captureSnapshot = null;
+        hasSnapshotCaptureInitiated = false;
+    }
 
-            return last().getTerm();
-        }
 
+    private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
 
-        @Override public void removeFrom(long index) {
-            if(index < 0 || index >= journal.size()){
-                return;
-            }
-            for(int i= (int) (index - snapshotIndex) ; i < journal.size() ; i++){
-                deleteMessage(i);
-                journal.remove(i);
-            }
+        public ReplicatedLogImpl(Snapshot snapshot) {
+            super(ByteString.copyFrom(snapshot.getState()),
+                snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
+                snapshot.getUnAppliedEntries());
         }
 
-        @Override public void append(final ReplicatedLogEntry replicatedLogEntry) {
-            journal.add(replicatedLogEntry);
+        public ReplicatedLogImpl() {
+            super();
         }
 
-        @Override public List<ReplicatedLogEntry> getFrom(long index) {
-            List<ReplicatedLogEntry> entries = new ArrayList<>(100);
-            if(index < 0 || index >= journal.size()){
-                return entries;
-            }
-            for(int i= (int) (index - snapshotIndex); i < journal.size() ; i++){
-                entries.add(journal.get(i));
+        @Override public void removeFromAndPersist(long logEntryIndex) {
+            int adjustedIndex = adjustedIndex(logEntryIndex);
+
+            if (adjustedIndex < 0) {
+                return;
             }
-            return entries;
+
+            // FIXME: Maybe this should be done after the command is saved
+            journal.subList(adjustedIndex , journal.size()).clear();
+
+            persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>(){
+
+                @Override public void apply(DeleteEntries param)
+                    throws Exception {
+                    //FIXME : Doing nothing for now
+                }
+            });
         }
 
-        @Override public void appendAndPersist(final ReplicatedLogEntry replicatedLogEntry){
+        @Override public void appendAndPersist(
+            final ReplicatedLogEntry replicatedLogEntry) {
             appendAndPersist(null, null, replicatedLogEntry);
         }
 
-        public void appendAndPersist(final ActorRef clientActor, final String identifier, final ReplicatedLogEntry replicatedLogEntry){
-            context.getLogger().debug("Append log entry and persist" + replicatedLogEntry.getIndex());
+        public void appendAndPersist(final ActorRef clientActor,
+            final String identifier,
+            final ReplicatedLogEntry replicatedLogEntry) {
+            context.getLogger().debug(
+                "Append log entry and persist {} ", 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.
             persist(replicatedLogEntry,
                 new Procedure<ReplicatedLogEntry>() {
                     public void apply(ReplicatedLogEntry evt) throws Exception {
+                        // when a snaphsot is being taken, captureSnapshot != null
+                        if (hasSnapshotCaptureInitiated == false &&
+                            journal.size() % context.getConfigParams().getSnapshotBatchCount() == 0) {
+
+                            LOG.info("Initiating Snapshot Capture..");
+                            long lastAppliedIndex = -1;
+                            long lastAppliedTerm = -1;
+
+                            ReplicatedLogEntry lastAppliedEntry = get(context.getLastApplied());
+                            if (lastAppliedEntry != null) {
+                                lastAppliedIndex = lastAppliedEntry.getIndex();
+                                lastAppliedTerm = lastAppliedEntry.getTerm();
+                            }
+
+                            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);
+
+                            // send a CaptureSnapshot to self to make the expensive operation async.
+                            getSelf().tell(new CaptureSnapshot(
+                                lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm),
+                                null);
+                            hasSnapshotCaptureInitiated = true;
+                        }
                         // Send message for replication
-                        if(clientActor != null) {
+                        if (clientActor != null) {
                             currentBehavior.handleMessage(getSelf(),
                                 new Replicate(clientActor, identifier,
-                                    replicatedLogEntry));
+                                    replicatedLogEntry)
+                            );
                         }
                     }
-                });
+                }
+            );
         }
 
-        @Override public long size() {
-            return journal.size() + snapshotIndex;
-        }
     }
 
-    private static class ReplicatedLogImplEntry implements ReplicatedLogEntry,
-        Serializable {
+    private static class DeleteEntries implements Serializable {
+        private final int fromIndex;
 
-        private final long index;
-        private final long term;
-        private final Object payload;
 
-        public ReplicatedLogImplEntry(long index, long term, Object payload){
+        public DeleteEntries(int fromIndex) {
+            this.fromIndex = fromIndex;
+        }
 
-            this.index = index;
-            this.term = term;
-            this.payload = payload;
+        public int getFromIndex() {
+            return fromIndex;
         }
+    }
+
+
+    private class ElectionTermImpl implements ElectionTerm {
+        /**
+         * Identifier of the actor whose election term information this is
+         */
+        private long currentTerm = 0;
+        private String votedFor = null;
 
-        @Override public Object getData() {
-            return payload;
+        public long getCurrentTerm() {
+            return currentTerm;
         }
 
-        @Override public long getTerm() {
-            return term;
+        public String getVotedFor() {
+            return votedFor;
         }
 
-        @Override public long getIndex() {
-            return index;
+        @Override public void update(long currentTerm, String votedFor) {
+            LOG.debug("Set currentTerm={}, votedFor={}", currentTerm, votedFor);
+
+            this.currentTerm = currentTerm;
+            this.votedFor = votedFor;
+        }
+
+        @Override
+        public void updateAndPersist(long currentTerm, String votedFor){
+            update(currentTerm, votedFor);
+            // FIXME : Maybe first persist then update the state
+            persist(new UpdateElectionTerm(this.currentTerm, this.votedFor), new Procedure<UpdateElectionTerm>(){
+
+                @Override public void apply(UpdateElectionTerm param)
+                    throws Exception {
+
+                }
+            });
         }
     }
 
+    private static class UpdateElectionTerm implements Serializable {
+        private final long currentTerm;
+        private final String votedFor;
+
+        public UpdateElectionTerm(long currentTerm, String votedFor) {
+            this.currentTerm = currentTerm;
+            this.votedFor = votedFor;
+        }
+
+        public long getCurrentTerm() {
+            return currentTerm;
+        }
+
+        public String getVotedFor() {
+            return votedFor;
+        }
+    }
 
 }