import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
-import java.util.concurrent.TimeUnit;
import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
import org.opendaylight.controller.cluster.raft.RaftState;
import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
-import org.opendaylight.controller.cluster.raft.base.messages.InitiateInstallSnapshot;
import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
private Optional<ByteString> snapshot;
public AbstractLeader(RaftActorContext context) {
- super(context);
+ super(context, RaftState.Leader);
final Builder<String, FollowerLogInformation> ftlBuilder = ImmutableMap.builder();
for (String followerId : context.getPeerAddresses().keySet()) {
leaderId = context.getId();
- LOG.debug("Election:Leader has following peers: {}", getFollowerIds());
+ LOG.debug("{}: Election: Leader has following peers: {}", logName(), getFollowerIds());
minReplicationCount = getMajorityVoteCount(getFollowerIds().size());
// Upon election: send initial empty AppendEntries RPCs
// (heartbeat) to each server; repeat during idle periods to
// prevent election timeouts (§5.2)
- scheduleHeartBeat(new FiniteDuration(0, TimeUnit.SECONDS));
+ sendAppendEntries(0, false);
+
+ // It is important to schedule this heartbeat here
+ scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
}
/**
return followerToLog.keySet();
}
- private Optional<ByteString> getSnapshot() {
- return snapshot;
- }
-
@VisibleForTesting
void setSnapshot(Optional<ByteString> snapshot) {
this.snapshot = snapshot;
protected RaftActorBehavior handleAppendEntries(ActorRef sender,
AppendEntries appendEntries) {
- if(LOG.isDebugEnabled()) {
- LOG.debug(appendEntries.toString());
- }
+ LOG.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
return this;
}
protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
AppendEntriesReply appendEntriesReply) {
- if(! appendEntriesReply.isSuccess()) {
- if(LOG.isDebugEnabled()) {
- LOG.debug(appendEntriesReply.toString());
- }
+ if(LOG.isTraceEnabled()) {
+ LOG.trace("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
+ } else if(LOG.isDebugEnabled() && !appendEntriesReply.isSuccess()) {
+ LOG.debug("{}: handleAppendEntriesReply: {}", logName(), appendEntriesReply);
}
// Update the FollowerLogInformation
followerToLog.get(followerId);
if(followerLogInformation == null){
- LOG.error("Unknown follower {}", followerId);
+ LOG.error("{}: handleAppendEntriesReply - unknown follower {}", logName(), followerId);
return this;
}
+ if(followerLogInformation.timeSinceLastActivity() >
+ context.getConfigParams().getElectionTimeOutInterval().toMillis()) {
+ LOG.error("{} : handleAppendEntriesReply delayed beyond election timeout, " +
+ "appendEntriesReply : {}, timeSinceLastActivity : {}, lastApplied : {}, commitIndex : {}",
+ logName(), appendEntriesReply, followerLogInformation.timeSinceLastActivity(),
+ context.getLastApplied(), context.getCommitIndex());
+ }
+
followerLogInformation.markFollowerActive();
if (appendEntriesReply.isSuccess()) {
// Apply the change to the state machine
if (context.getCommitIndex() > context.getLastApplied()) {
+ LOG.debug("{}: handleAppendEntriesReply: applying to log - commitIndex: {}, lastAppliedIndex: {}",
+ logName(), context.getCommitIndex(), context.getLastApplied());
+
applyLogToStateMachine(context.getCommitIndex());
}
+ if (!context.isSnapshotCaptureInitiated()) {
+ purgeInMemoryLog();
+ }
+
+ //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
+ sendUpdatesToFollower(followerId, followerLogInformation, false, false);
return this;
}
+ private void purgeInMemoryLog() {
+ //find the lowest index across followers which has been replicated to all.
+ // lastApplied if there are no followers, so that we keep clearing the log for single-node
+ // we would delete the in-mem log from that index on, in-order to minimize mem usage
+ // we would also share this info thru AE with the followers so that they can delete their log entries as well.
+ long minReplicatedToAllIndex = followerToLog.isEmpty() ? context.getLastApplied() : Long.MAX_VALUE;
+ for (FollowerLogInformation info : followerToLog.values()) {
+ minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
+ }
+
+ super.performSnapshotWithoutCapture(minReplicatedToAllIndex);
+ }
+
@Override
protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
final Iterator<ClientRequestTracker> it = trackerList.iterator();
return this;
}
- @Override
- public RaftState state() {
- return RaftState.Leader;
- }
+ protected void beforeSendHeartbeat(){}
@Override
public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
// set currentTerm = T, convert to follower (§5.1)
// This applies to all RPC messages and responses
if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
+ LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {} - switching to Follower",
+ logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
+
context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
return switchBehavior(new Follower(context));
}
}
- try {
- if (message instanceof SendHeartBeat) {
- sendHeartBeat();
- return this;
-
- } else if(message instanceof InitiateInstallSnapshot) {
- installSnapshotIfNeeded();
+ if (message instanceof SendHeartBeat) {
+ beforeSendHeartbeat();
+ sendHeartBeat();
+ scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
+ return this;
- } else if(message instanceof SendInstallSnapshot) {
- // received from RaftActor
- setSnapshot(Optional.of(((SendInstallSnapshot) message).getSnapshot()));
- sendInstallSnapshot();
+ } else if(message instanceof SendInstallSnapshot) {
+ // received from RaftActor
+ setSnapshot(Optional.of(((SendInstallSnapshot) message).getSnapshot()));
+ sendInstallSnapshot();
- } else if (message instanceof Replicate) {
- replicate((Replicate) message);
+ } else if (message instanceof Replicate) {
+ replicate((Replicate) message);
- } else if (message instanceof InstallSnapshotReply){
- handleInstallSnapshotReply((InstallSnapshotReply) message);
+ } else if (message instanceof InstallSnapshotReply){
+ handleInstallSnapshotReply((InstallSnapshotReply) message);
- }
- } finally {
- scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
}
+
return super.handleMessage(sender, message);
}
private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
+ LOG.debug("{}: handleInstallSnapshotReply: {}", logName(), reply);
+
String followerId = reply.getFollowerId();
FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
+
+ if (followerToSnapshot == null) {
+ LOG.error("{}: FollowerId {} in InstallSnapshotReply not known to Leader",
+ logName(), followerId);
+ return;
+ }
+
FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
followerLogInformation.markFollowerActive();
- if (followerToSnapshot != null &&
- followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
-
+ if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
+ boolean wasLastChunk = false;
if (reply.isSuccess()) {
if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
//this was the last chunk reply
if(LOG.isDebugEnabled()) {
- LOG.debug("InstallSnapshotReply received, " +
- "last chunk received, Chunk:{}. Follower:{} Setting nextIndex:{}",
- reply.getChunkIndex(), followerId,
+ LOG.debug("{}: InstallSnapshotReply received, " +
+ "last chunk received, Chunk: {}. Follower: {} Setting nextIndex: {}",
+ logName(), reply.getChunkIndex(), followerId,
context.getReplicatedLog().getSnapshotIndex() + 1
);
}
context.getReplicatedLog().getSnapshotIndex() + 1);
mapFollowerToSnapshot.remove(followerId);
- if(LOG.isDebugEnabled()) {
- LOG.debug("followerToLog.get(followerId).getNextIndex()=" +
- followerToLog.get(followerId).getNextIndex());
- }
+ LOG.debug("{}: follower: {}, matchIndex set to {}, nextIndex set to {}",
+ logName(), followerId, followerLogInformation.getMatchIndex(),
+ followerLogInformation.getNextIndex());
if (mapFollowerToSnapshot.isEmpty()) {
// once there are no pending followers receiving snapshots
// we can remove snapshot from the memory
setSnapshot(Optional.<ByteString>absent());
}
+ wasLastChunk = true;
} else {
followerToSnapshot.markSendStatus(true);
}
} else {
- LOG.info("InstallSnapshotReply received, " +
- "sending snapshot chunk failed, Will retry, Chunk:{}",
- reply.getChunkIndex()
- );
+ LOG.info("{}: InstallSnapshotReply received sending snapshot chunk failed, Will retry, Chunk: {}",
+ logName(), reply.getChunkIndex());
followerToSnapshot.markSendStatus(false);
}
+ if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
+ ActorSelection followerActor = context.getPeerActorSelection(followerId);
+ if(followerActor != null) {
+ sendSnapshotChunk(followerActor, followerId);
+ }
+ }
+
} else {
- LOG.error("ERROR!!" +
- "FollowerId in InstallSnapshotReply not known to Leader" +
- " or Chunk Index in InstallSnapshotReply not matching {} != {}",
- followerToSnapshot.getChunkIndex(), reply.getChunkIndex()
- );
+ LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
+ logName(), reply.getChunkIndex(), followerId,
+ followerToSnapshot.getChunkIndex());
if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
// Since the Follower did not find this index to be valid we should reset the follower snapshot
private void replicate(Replicate replicate) {
long logIndex = replicate.getReplicatedLogEntry().getIndex();
- if(LOG.isDebugEnabled()) {
- LOG.debug("Replicate message {}", logIndex);
- }
+ LOG.debug("{}: Replicate message: identifier: {}, logIndex: {}", logName(),
+ replicate.getIdentifier(), logIndex);
// Create a tracker entry we will use this later to notify the
// client actor
context.setCommitIndex(logIndex);
applyLogToStateMachine(logIndex);
} else {
- sendAppendEntries();
+ sendAppendEntries(0, false);
}
}
- private void sendAppendEntries() {
+ private void sendAppendEntries(long timeSinceLastActivityInterval, boolean isHeartbeat) {
// Send an AppendEntries to all followers
for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
final String followerId = e.getKey();
- ActorSelection followerActor = context.getPeerActorSelection(followerId);
+ final FollowerLogInformation followerLogInformation = e.getValue();
+ // This checks helps not to send a repeat message to the follower
+ if(!followerLogInformation.isFollowerActive() ||
+ followerLogInformation.timeSinceLastActivity() >= timeSinceLastActivityInterval) {
+ sendUpdatesToFollower(followerId, followerLogInformation, true, isHeartbeat);
+ }
+ }
+ }
- if (followerActor != null) {
- FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
- long followerNextIndex = followerLogInformation.getNextIndex();
- boolean isFollowerActive = followerLogInformation.isFollowerActive();
-
- if (mapFollowerToSnapshot.get(followerId) != null) {
- // if install snapshot is in process , then sent next chunk if possible
- if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
- sendSnapshotChunk(followerActor, followerId);
- } else {
- // we send a heartbeat even if we have not received a reply for the last chunk
- sendAppendEntriesToFollower(followerActor, followerNextIndex,
- Collections.<ReplicatedLogEntry>emptyList());
- }
+ /**
+ *
+ * This method checks if any update needs to be sent to the given follower. This includes append log entries,
+ * sending next snapshot chunk, and initiating a snapshot.
+ * @return true if any update is sent, false otherwise
+ */
- } else {
- long leaderLastIndex = context.getReplicatedLog().lastIndex();
- long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
- final List<ReplicatedLogEntry> entries;
-
- if (isFollowerActive &&
- context.getReplicatedLog().isPresent(followerNextIndex)) {
- // FIXME : Sending one entry at a time
- entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
-
- } else if (isFollowerActive && followerNextIndex >= 0 &&
- leaderLastIndex >= followerNextIndex ) {
- // if the followers next index is not present in the leaders log, and
- // if the follower is just not starting and if leader's index is more than followers index
- // then snapshot should be sent
-
- if(LOG.isDebugEnabled()) {
- LOG.debug("InitiateInstallSnapshot to follower:{}," +
- "follower-nextIndex:{}, leader-snapshot-index:{}, " +
- "leader-last-index:{}", followerId,
- followerNextIndex, leaderSnapShotIndex, leaderLastIndex
- );
- }
- actor().tell(new InitiateInstallSnapshot(), actor());
-
- // we would want to sent AE as the capture snapshot might take time
- entries = Collections.<ReplicatedLogEntry>emptyList();
-
- } else {
- //we send an AppendEntries, even if the follower is inactive
- // in-order to update the followers timestamp, in case it becomes active again
- entries = Collections.<ReplicatedLogEntry>emptyList();
+ private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation,
+ boolean sendHeartbeat, boolean isHeartbeat) {
+
+ ActorSelection followerActor = context.getPeerActorSelection(followerId);
+ if (followerActor != null) {
+ long followerNextIndex = followerLogInformation.getNextIndex();
+ boolean isFollowerActive = followerLogInformation.isFollowerActive();
+
+ if (mapFollowerToSnapshot.get(followerId) != null) {
+ // if install snapshot is in process , then sent next chunk if possible
+ if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
+ sendSnapshotChunk(followerActor, followerId);
+ } else if(sendHeartbeat) {
+ // we send a heartbeat even if we have not received a reply for the last chunk
+ sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
+ Collections.<ReplicatedLogEntry>emptyList(), followerId);
+ }
+ } else {
+ long leaderLastIndex = context.getReplicatedLog().lastIndex();
+ long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
+
+ if(!isHeartbeat || LOG.isTraceEnabled()) {
+ LOG.debug("{}: Checking sendAppendEntries for follower {}, leaderLastIndex: {}, leaderSnapShotIndex: {}",
+ logName(), followerId, leaderLastIndex, leaderSnapShotIndex);
+ }
+
+ if (isFollowerActive && context.getReplicatedLog().isPresent(followerNextIndex)) {
+
+ LOG.debug("{}: sendAppendEntries: {} is present for follower {}", logName(),
+ followerNextIndex, followerId);
+
+ // FIXME : Sending one entry at a time
+ final List<ReplicatedLogEntry> entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
+
+ sendAppendEntriesToFollower(followerActor, followerNextIndex, entries, followerId);
+
+ } else if (isFollowerActive && followerNextIndex >= 0 &&
+ leaderLastIndex > followerNextIndex && !context.isSnapshotCaptureInitiated()) {
+ // if the followers next index is not present in the leaders log, and
+ // if the follower is just not starting and if leader's index is more than followers index
+ // then snapshot should be sent
+
+ if (LOG.isDebugEnabled()) {
+ LOG.debug(String.format("%s: InitiateInstallSnapshot to follower: %s," +
+ "follower-nextIndex: %d, leader-snapshot-index: %d, " +
+ "leader-last-index: %d", logName(), followerId,
+ followerNextIndex, leaderSnapShotIndex, leaderLastIndex));
}
- sendAppendEntriesToFollower(followerActor, followerNextIndex, entries);
+ // Send heartbeat to follower whenever install snapshot is initiated.
+ sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
+ Collections.<ReplicatedLogEntry>emptyList(), followerId);
+ initiateCaptureSnapshot(followerId, followerNextIndex);
+
+ } else if(sendHeartbeat) {
+ //we send an AppendEntries, even if the follower is inactive
+ // in-order to update the followers timestamp, in case it becomes active again
+ sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
+ Collections.<ReplicatedLogEntry>emptyList(), followerId);
}
+
}
}
}
private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
- List<ReplicatedLogEntry> entries) {
- followerActor.tell(
- new AppendEntries(currentTerm(), context.getId(),
- prevLogIndex(followerNextIndex),
- prevLogTerm(followerNextIndex), entries,
- context.getCommitIndex()).toSerializable(),
- actor()
- );
+ List<ReplicatedLogEntry> entries, String followerId) {
+ AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
+ prevLogIndex(followerNextIndex),
+ prevLogTerm(followerNextIndex), entries,
+ context.getCommitIndex(), super.getReplicatedToAllIndex());
+
+ if(!entries.isEmpty() || LOG.isTraceEnabled()) {
+ LOG.debug("{}: Sending AppendEntries to follower {}: {}", logName(), followerId,
+ appendEntries);
+ }
+
+ followerActor.tell(appendEntries.toSerializable(), actor());
}
/**
- * An installSnapshot is scheduled at a interval that is a multiple of
- * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
- * snapshots at every heartbeat.
- *
* Install Snapshot works as follows
- * 1. Leader sends a InitiateInstallSnapshot message to self
- * 2. Leader then initiates the capture snapshot by sending a CaptureSnapshot message to actor
- * 3. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
+ * 1. Leader initiates the capture snapshot by sending a CaptureSnapshot message to actor
+ * 2. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
* and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
- * 4. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
- * 5. On complete, Follower sends back a InstallSnapshotReply.
- * 6. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
+ * 3. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
+ * 4. On complete, Follower sends back a InstallSnapshotReply.
+ * 5. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
* and replenishes the memory by deleting the snapshot in Replicated log.
- *
+ * 6. If another follower requires a snapshot and a snapshot has been collected (via CaptureSnapshotReply)
+ * then send the existing snapshot in chunks to the follower.
+ * @param followerId
+ * @param followerNextIndex
*/
- private void installSnapshotIfNeeded() {
- for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
- final ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
-
- if (followerActor != null) {
- long nextIndex = e.getValue().getNextIndex();
-
- if (!context.getReplicatedLog().isPresent(nextIndex) &&
- context.getReplicatedLog().isInSnapshot(nextIndex)) {
- LOG.info("{} follower needs a snapshot install", e.getKey());
- if (snapshot.isPresent()) {
- // if a snapshot is present in the memory, most likely another install is in progress
- // no need to capture snapshot
- sendSnapshotChunk(followerActor, e.getKey());
-
- } else {
- initiateCaptureSnapshot();
- //we just need 1 follower who would need snapshot to be installed.
- // when we have the snapshot captured, we would again check (in SendInstallSnapshot)
- // who needs an install and send to all who need
- break;
- }
+ private void initiateCaptureSnapshot(String followerId, long followerNextIndex) {
+ if (!context.getReplicatedLog().isPresent(followerNextIndex) &&
+ context.getReplicatedLog().isInSnapshot(followerNextIndex)) {
+ if (snapshot.isPresent()) {
+ // if a snapshot is present in the memory, most likely another install is in progress
+ // no need to capture snapshot.
+ // This could happen if another follower needs an install when one is going on.
+ final ActorSelection followerActor = context.getPeerActorSelection(followerId);
+ sendSnapshotChunk(followerActor, followerId);
+
+ } else if (!context.isSnapshotCaptureInitiated()) {
+
+ LOG.info("{}: Initiating Snapshot Capture to Install Snapshot, Leader:{}", logName(), getLeaderId());
+ ReplicatedLogEntry lastAppliedEntry = context.getReplicatedLog().get(context.getLastApplied());
+ long lastAppliedIndex = -1;
+ long lastAppliedTerm = -1;
+
+ if (lastAppliedEntry != null) {
+ lastAppliedIndex = lastAppliedEntry.getIndex();
+ lastAppliedTerm = lastAppliedEntry.getTerm();
+ } else if (context.getReplicatedLog().getSnapshotIndex() > -1) {
+ lastAppliedIndex = context.getReplicatedLog().getSnapshotIndex();
+ lastAppliedTerm = context.getReplicatedLog().getSnapshotTerm();
}
+
+ boolean isInstallSnapshotInitiated = true;
+ long replicatedToAllIndex = super.getReplicatedToAllIndex();
+ ReplicatedLogEntry replicatedToAllEntry = context.getReplicatedLog().get(replicatedToAllIndex);
+ actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm,
+ (replicatedToAllEntry != null ? replicatedToAllEntry.getIndex() : -1),
+ (replicatedToAllEntry != null ? replicatedToAllEntry.getTerm() : -1),
+ isInstallSnapshotInitiated), actor());
+ context.setSnapshotCaptureInitiated(true);
}
}
}
- // on every install snapshot, we try to capture the snapshot.
- // Once a capture is going on, another one issued will get ignored by RaftActor.
- private void initiateCaptureSnapshot() {
- LOG.info("Initiating Snapshot Capture to Install Snapshot, Leader:{}", getLeaderId());
- ReplicatedLogEntry lastAppliedEntry = context.getReplicatedLog().get(context.getLastApplied());
- long lastAppliedIndex = -1;
- long lastAppliedTerm = -1;
-
- if (lastAppliedEntry != null) {
- lastAppliedIndex = lastAppliedEntry.getIndex();
- lastAppliedTerm = lastAppliedEntry.getTerm();
- } else if (context.getReplicatedLog().getSnapshotIndex() > -1) {
- lastAppliedIndex = context.getReplicatedLog().getSnapshotIndex();
- lastAppliedTerm = context.getReplicatedLog().getSnapshotTerm();
- }
-
- boolean isInstallSnapshotInitiated = true;
- actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(),
- lastAppliedIndex, lastAppliedTerm, isInstallSnapshotInitiated),
- actor());
- }
-
private void sendInstallSnapshot() {
+ LOG.debug("{}: sendInstallSnapshot", logName());
for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
try {
if (snapshot.isPresent()) {
+ ByteString nextSnapshotChunk = getNextSnapshotChunk(followerId,snapshot.get());
+
+ // Note: the previous call to getNextSnapshotChunk has the side-effect of adding
+ // followerId to the followerToSnapshot map.
+ FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
+
followerActor.tell(
new InstallSnapshot(currentTerm(), context.getId(),
context.getReplicatedLog().getSnapshotIndex(),
context.getReplicatedLog().getSnapshotTerm(),
- getNextSnapshotChunk(followerId,snapshot.get()),
- mapFollowerToSnapshot.get(followerId).incrementChunkIndex(),
- mapFollowerToSnapshot.get(followerId).getTotalChunks(),
- Optional.of(mapFollowerToSnapshot.get(followerId).getLastChunkHashCode())
+ nextSnapshotChunk,
+ followerToSnapshot.incrementChunkIndex(),
+ followerToSnapshot.getTotalChunks(),
+ Optional.of(followerToSnapshot.getLastChunkHashCode())
).toSerializable(),
actor()
);
- LOG.info("InstallSnapshot sent to follower {}, Chunk: {}/{}",
- followerActor.path(), mapFollowerToSnapshot.get(followerId).getChunkIndex(),
- mapFollowerToSnapshot.get(followerId).getTotalChunks());
+ LOG.info("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
+ logName(), followerActor.path(),
+ followerToSnapshot.getChunkIndex(),
+ followerToSnapshot.getTotalChunks());
}
} catch (IOException e) {
- LOG.error(e, "InstallSnapshot failed for Leader.");
+ LOG.error("{}: InstallSnapshot failed for Leader.", logName(), e);
}
}
mapFollowerToSnapshot.put(followerId, followerToSnapshot);
}
ByteString nextChunk = followerToSnapshot.getNextChunk();
- if (LOG.isDebugEnabled()) {
- LOG.debug("Leader's snapshot nextChunk size:{}", nextChunk.size());
- }
+
+ LOG.debug("{}: next snapshot chunk size for follower {}: {}", logName(), followerId, nextChunk.size());
+
return nextChunk;
}
private void sendHeartBeat() {
if (!followerToLog.isEmpty()) {
- sendAppendEntries();
+ LOG.trace("{}: Sending heartbeat", logName());
+ sendAppendEntries(context.getConfigParams().getHeartBeatInterval().toMillis(), true);
}
}
* snapshot chunks
*/
protected class FollowerToSnapshot {
- private ByteString snapshotBytes;
+ private final ByteString snapshotBytes;
private int offset = 0;
// the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
private int replyReceivedForOffset;
// if replyStatus is false, the previous chunk is attempted
private boolean replyStatus = false;
private int chunkIndex;
- private int totalChunks;
+ private final int totalChunks;
private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
if(LOG.isDebugEnabled()) {
- LOG.debug("Snapshot {} bytes, total chunks to send:{}",
- size, totalChunks);
+ LOG.debug("{}: Snapshot {} bytes, total chunks to send:{}",
+ logName(), size, totalChunks);
}
replyReceivedForOffset = -1;
chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
}
}
- if(LOG.isDebugEnabled()) {
- LOG.debug("length={}, offset={},size={}",
+
+ LOG.debug("{}: Next chunk: length={}, offset={},size={}", logName(),
snapshotLength, start, size);
- }
+
ByteString substring = getSnapshotBytes().substring(start, start + size);
nextChunkHashCode = substring.hashCode();
return substring;
Code Review / controller.git / blobdiff