/* * Copyright (c) 2014 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.behaviors; import akka.actor.ActorRef; import akka.actor.ActorSelection; import akka.actor.Address; import akka.cluster.Cluster; import akka.cluster.ClusterEvent.CurrentClusterState; import akka.cluster.Member; import akka.cluster.MemberStatus; import com.google.common.annotations.VisibleForTesting; import com.google.common.base.Stopwatch; import java.io.IOException; import java.util.ArrayList; import java.util.List; import java.util.Optional; import java.util.Set; import java.util.concurrent.TimeUnit; import java.util.concurrent.atomic.AtomicBoolean; import java.util.function.Consumer; import org.eclipse.jdt.annotation.Nullable; import org.opendaylight.controller.cluster.messaging.MessageAssembler; import org.opendaylight.controller.cluster.raft.RaftActorContext; import org.opendaylight.controller.cluster.raft.RaftState; import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry; import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot; import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout; import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow; import org.opendaylight.controller.cluster.raft.messages.AppendEntries; import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply; import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot; import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply; import org.opendaylight.controller.cluster.raft.messages.RaftRPC; import org.opendaylight.controller.cluster.raft.messages.RequestVote; import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply; import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload; import org.opendaylight.controller.cluster.raft.persisted.Snapshot; /** * The behavior of a RaftActor in the Follower raft state. * */ public class Follower extends AbstractRaftActorBehavior { private static final long MAX_ELECTION_TIMEOUT_FACTOR = 18; private final SyncStatusTracker initialSyncStatusTracker; private final MessageAssembler appendEntriesMessageAssembler; private final Stopwatch lastLeaderMessageTimer = Stopwatch.createStarted(); private SnapshotTracker snapshotTracker = null; private String leaderId; private short leaderPayloadVersion; public Follower(final RaftActorContext context) { this(context, null, (short)-1); } public Follower(final RaftActorContext context, final String initialLeaderId, final short initialLeaderPayloadVersion) { super(context, RaftState.Follower); this.leaderId = initialLeaderId; this.leaderPayloadVersion = initialLeaderPayloadVersion; initialSyncStatusTracker = new SyncStatusTracker(context.getActor(), getId(), context.getConfigParams() .getSyncIndexThreshold()); appendEntriesMessageAssembler = MessageAssembler.builder().logContext(logName()) .fileBackedStreamFactory(context.getFileBackedOutputStreamFactory()) .assembledMessageCallback((message, sender) -> handleMessage(sender, message)).build(); if (context.getPeerIds().isEmpty() && getLeaderId() == null) { actor().tell(TimeoutNow.INSTANCE, actor()); } else { scheduleElection(electionDuration()); } } @Override public final String getLeaderId() { return leaderId; } @VisibleForTesting protected final void setLeaderId(final @Nullable String leaderId) { this.leaderId = leaderId; } @Override public short getLeaderPayloadVersion() { return leaderPayloadVersion; } @VisibleForTesting protected final void setLeaderPayloadVersion(final short leaderPayloadVersion) { this.leaderPayloadVersion = leaderPayloadVersion; } private void restartLastLeaderMessageTimer() { if (lastLeaderMessageTimer.isRunning()) { lastLeaderMessageTimer.reset(); } lastLeaderMessageTimer.start(); } private boolean isLogEntryPresent(final long index) { if (context.getReplicatedLog().isInSnapshot(index)) { return true; } ReplicatedLogEntry entry = context.getReplicatedLog().get(index); return entry != null; } private void updateInitialSyncStatus(final long currentLeaderCommit, final String newLeaderId) { initialSyncStatusTracker.update(newLeaderId, currentLeaderCommit, context.getCommitIndex()); } @Override protected RaftActorBehavior handleAppendEntries(final ActorRef sender, final AppendEntries appendEntries) { int numLogEntries = appendEntries.getEntries().size(); if (log.isTraceEnabled()) { log.trace("{}: handleAppendEntries: {}", logName(), appendEntries); } else if (log.isDebugEnabled() && numLogEntries > 0) { log.debug("{}: handleAppendEntries: {}", logName(), appendEntries); } if (snapshotTracker != null && !snapshotTracker.getLeaderId().equals(appendEntries.getLeaderId())) { log.debug("{}: snapshot install is in progress but the prior snapshot leaderId {} does not match the " + "AppendEntries leaderId {}", logName(), snapshotTracker.getLeaderId(), appendEntries.getLeaderId()); closeSnapshotTracker(); } if (snapshotTracker != null || context.getSnapshotManager().isApplying()) { // if snapshot install is in progress, follower should just acknowledge append entries with a reply. AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true, lastIndex(), lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(), appendEntries.getLeaderRaftVersion()); log.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply); sender.tell(reply, actor()); return this; } // If we got here then we do appear to be talking to the leader leaderId = appendEntries.getLeaderId(); leaderPayloadVersion = appendEntries.getPayloadVersion(); if (appendEntries.getLeaderAddress().isPresent()) { final String address = appendEntries.getLeaderAddress().get(); log.debug("New leader address: {}", address); context.setPeerAddress(leaderId, address); context.getConfigParams().getPeerAddressResolver().setResolved(leaderId, address); } // First check if the logs are in sync or not if (isOutOfSync(appendEntries, sender)) { updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId()); return this; } if (!processNewEntries(appendEntries, sender)) { updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId()); return this; } long lastIndex = lastIndex(); long prevCommitIndex = context.getCommitIndex(); // If leaderCommit > commitIndex, set commitIndex = min(leaderCommit, index of last new entry) if (appendEntries.getLeaderCommit() > prevCommitIndex) { context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex)); } if (prevCommitIndex != context.getCommitIndex()) { log.debug("{}: Commit index set to {}", logName(), context.getCommitIndex()); } AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true, lastIndex, lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(), appendEntries.getLeaderRaftVersion()); if (log.isTraceEnabled()) { log.trace("{}: handleAppendEntries returning : {}", logName(), reply); } else if (log.isDebugEnabled() && numLogEntries > 0) { log.debug("{}: handleAppendEntries returning : {}", logName(), reply); } // Reply to the leader before applying any previous state so as not to hold up leader consensus. sender.tell(reply, actor()); updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId()); // If leaderCommit > lastApplied, increment lastApplied and apply log[lastApplied] to state machine (§5.3). // lastApplied can be equal to lastIndex. if (appendEntries.getLeaderCommit() > context.getLastApplied() && context.getLastApplied() < lastIndex) { if (log.isDebugEnabled()) { log.debug("{}: applyLogToStateMachine, appendEntries.getLeaderCommit(): {}, " + "context.getLastApplied(): {}, lastIndex(): {}", logName(), appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex); } applyLogToStateMachine(appendEntries.getLeaderCommit()); } if (!context.getSnapshotManager().isCapturing()) { super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex()); } appendEntriesMessageAssembler.checkExpiredAssembledMessageState(); return this; } private boolean processNewEntries(final AppendEntries appendEntries, final ActorRef sender) { int numLogEntries = appendEntries.getEntries().size(); if (numLogEntries == 0) { return true; } log.debug("{}: Number of entries to be appended = {}", logName(), numLogEntries); long lastIndex = lastIndex(); int addEntriesFrom = 0; // First check for conflicting entries. If an existing entry conflicts with a new one (same index but different // term), delete the existing entry and all that follow it (§5.3) if (context.getReplicatedLog().size() > 0) { // Find the entry up until the one that is not in the follower's log for (int i = 0;i < numLogEntries; i++, addEntriesFrom++) { ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i); if (!isLogEntryPresent(matchEntry.getIndex())) { // newEntry not found in the log break; } long existingEntryTerm = getLogEntryTerm(matchEntry.getIndex()); log.debug("{}: matchEntry {} is present: existingEntryTerm: {}", logName(), matchEntry, existingEntryTerm); // existingEntryTerm == -1 means it's in the snapshot and not in the log. We don't know // what the term was so we'll assume it matches. if (existingEntryTerm == -1 || existingEntryTerm == matchEntry.getTerm()) { continue; } if (!context.getRaftPolicy().applyModificationToStateBeforeConsensus()) { log.info("{}: Removing entries from log starting at {}, commitIndex: {}, lastApplied: {}", logName(), matchEntry.getIndex(), context.getCommitIndex(), context.getLastApplied()); // Entries do not match so remove all subsequent entries but only if the existing entries haven't // been applied to the state yet. if (matchEntry.getIndex() <= context.getLastApplied() || !context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex())) { // Could not remove the entries - this means the matchEntry index must be in the // snapshot and not the log. In this case the prior entries are part of the state // so we must send back a reply to force a snapshot to completely re-sync the // follower's log and state. log.info("{}: Could not remove entries - sending reply to force snapshot", logName()); sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex, lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(), appendEntries.getLeaderRaftVersion()), actor()); return false; } break; } else { sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex, lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(), appendEntries.getLeaderRaftVersion()), actor()); return false; } } } lastIndex = lastIndex(); log.debug("{}: After cleanup, lastIndex: {}, entries to be added from: {}", logName(), lastIndex, addEntriesFrom); // When persistence successfully completes for each new log entry appended, we need to determine if we // should capture a snapshot to compact the persisted log. shouldCaptureSnapshot tracks whether or not // one of the log entries has exceeded the log size threshold whereby a snapshot should be taken. However // we don't initiate the snapshot at that log entry but rather after the last log entry has been persisted. // This is done because subsequent log entries after the one that tripped the threshold may have been // applied to the state already, as the persistence callback occurs async, and we want those entries // purged from the persisted log as well. final AtomicBoolean shouldCaptureSnapshot = new AtomicBoolean(false); final Consumer appendAndPersistCallback = logEntry -> { final List entries = appendEntries.getEntries(); final ReplicatedLogEntry lastEntryToAppend = entries.get(entries.size() - 1); if (shouldCaptureSnapshot.get() && logEntry == lastEntryToAppend) { context.getSnapshotManager().capture(context.getReplicatedLog().last(), getReplicatedToAllIndex()); } }; // Append any new entries not already in the log for (int i = addEntriesFrom; i < numLogEntries; i++) { ReplicatedLogEntry entry = appendEntries.getEntries().get(i); log.debug("{}: Append entry to log {}", logName(), entry.getData()); context.getReplicatedLog().appendAndPersist(entry, appendAndPersistCallback, false); shouldCaptureSnapshot.compareAndSet(false, context.getReplicatedLog().shouldCaptureSnapshot(entry.getIndex())); if (entry.getData() instanceof ServerConfigurationPayload) { context.updatePeerIds((ServerConfigurationPayload)entry.getData()); } } log.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size()); return true; } private boolean isOutOfSync(final AppendEntries appendEntries, final ActorRef sender) { final long lastIndex = lastIndex(); if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) { // The follower's log is out of sync because the leader does have an entry at prevLogIndex and this // follower has no entries in it's log. log.info("{}: The followers log is empty and the senders prevLogIndex is {}", logName(), appendEntries.getPrevLogIndex()); sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion()); return true; } if (lastIndex > -1) { if (isLogEntryPresent(appendEntries.getPrevLogIndex())) { final long leadersPrevLogTermInFollowersLogOrSnapshot = getLogEntryOrSnapshotTerm(appendEntries.getPrevLogIndex()); if (leadersPrevLogTermInFollowersLogOrSnapshot != appendEntries.getPrevLogTerm()) { // The follower's log is out of sync because the Leader's prevLogIndex entry does exist // in the follower's log or snapshot but it has a different term. log.info("{}: The prevLogIndex {} was found in the log but the term {} is not equal to the append " + "entries prevLogTerm {} - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(), appendEntries.getPrevLogIndex(), leadersPrevLogTermInFollowersLogOrSnapshot, appendEntries.getPrevLogTerm(), lastIndex, context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm()); sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion()); return true; } } else if (appendEntries.getPrevLogIndex() != -1) { // The follower's log is out of sync because the Leader's prevLogIndex entry was not found in it's log log.info("{}: The log is not empty but the prevLogIndex {} was not found in it - lastIndex: {}, " + "snapshotIndex: {}, snapshotTerm: {}", logName(), appendEntries.getPrevLogIndex(), lastIndex, context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm()); sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion()); return true; } } if (appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1 && appendEntries.getReplicatedToAllIndex() != -1) { if (!isLogEntryPresent(appendEntries.getReplicatedToAllIndex())) { // This append entry comes from a leader who has it's log aggressively trimmed and so does not have // the previous entry in it's in-memory journal log.info("{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the " + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(), appendEntries.getReplicatedToAllIndex(), lastIndex, context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm()); sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion()); return true; } final List entries = appendEntries.getEntries(); if (entries.size() > 0 && !isLogEntryPresent(entries.get(0).getIndex() - 1)) { log.info("{}: Cannot append entries because the calculated previousIndex {} was not found in the " + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(), entries.get(0).getIndex() - 1, lastIndex, context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm()); sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion()); return true; } } return false; } private void sendOutOfSyncAppendEntriesReply(final ActorRef sender, final boolean forceInstallSnapshot, final short leaderRaftVersion) { // We found that the log was out of sync so just send a negative reply. final AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex(), lastTerm(), context.getPayloadVersion(), forceInstallSnapshot, needsLeaderAddress(), leaderRaftVersion); log.info("{}: Follower is out-of-sync so sending negative reply: {}", logName(), reply); sender.tell(reply, actor()); } private boolean needsLeaderAddress() { return context.getPeerAddress(leaderId) == null; } @Override protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender, final AppendEntriesReply appendEntriesReply) { return this; } @Override protected RaftActorBehavior handleRequestVoteReply(final ActorRef sender, final RequestVoteReply requestVoteReply) { return this; } @Override public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) { if (message instanceof ElectionTimeout || message instanceof TimeoutNow) { return handleElectionTimeout(message); } if (appendEntriesMessageAssembler.handleMessage(message, actor())) { return this; } if (!(message instanceof RaftRPC)) { // The rest of the processing requires the message to be a RaftRPC return null; } final RaftRPC rpc = (RaftRPC) message; // If RPC request or response contains term T > currentTerm: // set currentTerm = T, convert to follower (§5.1) // This applies to all RPC messages and responses if (rpc.getTerm() > context.getTermInformation().getCurrentTerm() && shouldUpdateTerm(rpc)) { log.info("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term", logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm()); context.getTermInformation().updateAndPersist(rpc.getTerm(), null); } if (rpc instanceof InstallSnapshot) { handleInstallSnapshot(sender, (InstallSnapshot) rpc); restartLastLeaderMessageTimer(); scheduleElection(electionDuration()); return this; } if (!(rpc instanceof RequestVote) || canGrantVote((RequestVote) rpc)) { restartLastLeaderMessageTimer(); scheduleElection(electionDuration()); } return super.handleMessage(sender, rpc); } private RaftActorBehavior handleElectionTimeout(final Object message) { // If the message is ElectionTimeout, verify we haven't actually seen a message from the leader // during the election timeout interval. It may that the election timer expired b/c this actor // was busy and messages got delayed, in which case leader messages would be backed up in the // queue but would be processed before the ElectionTimeout message and thus would restart the // lastLeaderMessageTimer. long lastLeaderMessageInterval = lastLeaderMessageTimer.elapsed(TimeUnit.MILLISECONDS); long electionTimeoutInMillis = context.getConfigParams().getElectionTimeOutInterval().toMillis(); boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning() || lastLeaderMessageInterval >= electionTimeoutInMillis; if (canStartElection()) { if (message instanceof TimeoutNow) { log.debug("{}: Received TimeoutNow - switching to Candidate", logName()); return internalSwitchBehavior(RaftState.Candidate); } else if (noLeaderMessageReceived) { // Check the cluster state to see if the leader is known to be up before we go to Candidate. // However if we haven't heard from the leader in a long time even though the cluster state // indicates it's up then something is wrong - leader might be stuck indefinitely - so switch // to Candidate, long maxElectionTimeout = electionTimeoutInMillis * MAX_ELECTION_TIMEOUT_FACTOR; if (isLeaderAvailabilityKnown() && lastLeaderMessageInterval < maxElectionTimeout) { log.debug("{}: Received ElectionTimeout but leader appears to be available", logName()); scheduleElection(electionDuration()); } else { log.debug("{}: Received ElectionTimeout - switching to Candidate", logName()); return internalSwitchBehavior(RaftState.Candidate); } } else { log.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout {}", logName(), lastLeaderMessageInterval, context.getConfigParams().getElectionTimeOutInterval()); scheduleElection(electionDuration()); } } else if (message instanceof ElectionTimeout) { if (noLeaderMessageReceived) { setLeaderId(null); } scheduleElection(electionDuration()); } return this; } private boolean isLeaderAvailabilityKnown() { if (leaderId == null) { return false; } Optional cluster = context.getCluster(); if (!cluster.isPresent()) { return false; } ActorSelection leaderActor = context.getPeerActorSelection(leaderId); if (leaderActor == null) { return false; } Address leaderAddress = leaderActor.anchorPath().address(); CurrentClusterState state = cluster.get().state(); Set unreachable = state.getUnreachable(); log.debug("{}: Checking for leader {} in the cluster unreachable set {}", logName(), leaderAddress, unreachable); for (Member m: unreachable) { if (leaderAddress.equals(m.address())) { log.info("{}: Leader {} is unreachable", logName(), leaderAddress); return false; } } for (Member m: state.getMembers()) { if (leaderAddress.equals(m.address())) { if (m.status() == MemberStatus.up() || m.status() == MemberStatus.weaklyUp()) { log.debug("{}: Leader {} cluster status is {} - leader is available", logName(), leaderAddress, m.status()); return true; } else { log.debug("{}: Leader {} cluster status is {} - leader is unavailable", logName(), leaderAddress, m.status()); return false; } } } log.debug("{}: Leader {} not found in the cluster member set", logName(), leaderAddress); return false; } private void handleInstallSnapshot(final ActorRef sender, final InstallSnapshot installSnapshot) { log.debug("{}: handleInstallSnapshot: {}", logName(), installSnapshot); leaderId = installSnapshot.getLeaderId(); if (snapshotTracker == null) { snapshotTracker = new SnapshotTracker(log, installSnapshot.getTotalChunks(), installSnapshot.getLeaderId(), context); } updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId()); try { final InstallSnapshotReply reply = new InstallSnapshotReply( currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true); if (snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(), installSnapshot.getLastChunkHashCode())) { log.info("{}: Snapshot installed from leader: {}", logName(), installSnapshot.getLeaderId()); Snapshot snapshot = Snapshot.create( context.getSnapshotManager().convertSnapshot(snapshotTracker.getSnapshotBytes()), new ArrayList<>(), installSnapshot.getLastIncludedIndex(), installSnapshot.getLastIncludedTerm(), installSnapshot.getLastIncludedIndex(), installSnapshot.getLastIncludedTerm(), context.getTermInformation().getCurrentTerm(), context.getTermInformation().getVotedFor(), installSnapshot.getServerConfig().orElse(null)); ApplySnapshot.Callback applySnapshotCallback = new ApplySnapshot.Callback() { @Override public void onSuccess() { log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply); sender.tell(reply, actor()); } @Override public void onFailure() { sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor()); } }; actor().tell(new ApplySnapshot(snapshot, applySnapshotCallback), actor()); closeSnapshotTracker(); } else { log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply); sender.tell(reply, actor()); } } catch (IOException e) { log.debug("{}: Exception in InstallSnapshot of follower", logName(), e); sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor()); closeSnapshotTracker(); } } private void closeSnapshotTracker() { if (snapshotTracker != null) { snapshotTracker.close(); snapshotTracker = null; } } @Override public void close() { closeSnapshotTracker(); stopElection(); appendEntriesMessageAssembler.close(); } @VisibleForTesting SnapshotTracker getSnapshotTracker() { return snapshotTracker; } }