raft: use RawNode for node's event loop

It has always bugged me that any new feature essentially needed to be
tested twice due to the two ways in which apps can use raft (`*node` and
`*RawNode`). Due to upcoming testing work for joint consensus, now is a
good time to rectify this somewhat.

This commit removes most logic from `(*node).run` and uses `*RawNode`
internally. This simplifies the logic and also lead (via debugging) to
some insight on how the semantics of the approaches differ, which is now
documented in the comments.

raft/rawnode.go

@@ -37,38 +37,6 @@ type RawNode struct {
 	prevHardSt pb.HardState
 }
 
-func (rn *RawNode) newReady() Ready {
-	return newReady(rn.raft, rn.prevSoftSt, rn.prevHardSt)
-}
-
-func (rn *RawNode) commitReady(rd Ready) {
-	if rd.SoftState != nil {
-		rn.prevSoftSt = rd.SoftState
-	}
-	if !IsEmptyHardState(rd.HardState) {
-		rn.prevHardSt = rd.HardState
-	}
-
-	// If entries were applied (or a snapshot), update our cursor for
-	// the next Ready. Note that if the current HardState contains a
-	// new Commit index, this does not mean that we're also applying
-	// all of the new entries due to commit pagination by size.
-	if index := rd.appliedCursor(); index > 0 {
-		rn.raft.raftLog.appliedTo(index)
-	}
-
-	if len(rd.Entries) > 0 {
-		e := rd.Entries[len(rd.Entries)-1]
-		rn.raft.raftLog.stableTo(e.Index, e.Term)
-	}
-	if !IsEmptySnap(rd.Snapshot) {
-		rn.raft.raftLog.stableSnapTo(rd.Snapshot.Metadata.Index)
-	}
-	if len(rd.ReadStates) != 0 {
-		rn.raft.readStates = nil
-	}
-}
-
 // NewRawNode returns a new RawNode given configuration and a list of raft peers.
 func NewRawNode(config *Config, peers []Peer) (*RawNode, error) {
 	if config.ID == 0 {
@@ -78,27 +46,48 @@ func NewRawNode(config *Config, peers []Peer) (*RawNode, error) {
 	rn := &RawNode{
 		raft: r,
 	}
-	lastIndex, err := config.Storage.LastIndex()
+	if err := rn.init(peers); err != nil {
+		return nil, err
+	}
+	return rn, nil
+}
+
+func (rn *RawNode) init(peers []Peer) error {
+	r := rn.raft
+	lastIndex, err := rn.raft.raftLog.storage.LastIndex()
 	if err != nil {
-		panic(err) // TODO(bdarnell)
+		return err
 	}
 	// If the log is empty, this is a new RawNode (like StartNode); otherwise it's
 	// restoring an existing RawNode (like RestartNode).
 	// TODO(bdarnell): rethink RawNode initialization and whether the application needs
 	// to be able to tell us when it expects the RawNode to exist.
 	if lastIndex == 0 {
-		r.becomeFollower(1, None)
+		rn.raft.becomeFollower(1, None)
 		ents := make([]pb.Entry, len(peers))
 		for i, peer := range peers {
 			cc := pb.ConfChange{Type: pb.ConfChangeAddNode, NodeID: peer.ID, Context: peer.Context}
 			data, err := cc.Marshal()
 			if err != nil {
-				panic("unexpected marshal error")
+				return err
 			}
 
 			ents[i] = pb.Entry{Type: pb.EntryConfChange, Term: 1, Index: uint64(i + 1), Data: data}
 		}
-		r.raftLog.append(ents...)
+		rn.raft.raftLog.append(ents...)
+
+		// Now apply them, mainly so that the application can call Campaign
+		// immediately after StartNode in tests. Note that these nodes will
+		// be added to raft twice: here and when the application's Ready
+		// loop calls ApplyConfChange. The calls to addNode must come after
+		// all calls to raftLog.append so progress.next is set after these
+		// bootstrapping entries (it is an error if we try to append these
+		// entries since they have already been committed).
+		// We do not set raftLog.applied so the application will be able
+		// to observe all conf changes via Ready.CommittedEntries.
+		//
+		// TODO(bdarnell): These entries are still unstable; do we need to preserve
+		// the invariant that committed < unstable?
 		r.raftLog.committed = uint64(len(ents))
 		for _, peer := range peers {
 			r.applyConfChange(pb.ConfChange{NodeID: peer.ID, Type: pb.ConfChangeAddNode})
@@ -113,7 +102,7 @@ func NewRawNode(config *Config, peers []Peer) (*RawNode, error) {
 		rn.prevHardSt = r.hardState()
 	}
 
-	return rn, nil
+	return nil
 }
 
 // Tick advances the internal logical clock by a single tick.
@@ -182,14 +171,61 @@ func (rn *RawNode) Step(m pb.Message) error {
 	return ErrStepPeerNotFound
 }
 
-// Ready returns the current point-in-time state of this RawNode.
+// Ready returns the outstanding work that the application needs to handle. This
+// includes appending and applying entries or a snapshot, updating the HardState,
+// and sending messages. Ready() is a read-only operation, that is, it does not
+// require the caller to actually handle the result. Typically, a caller will
+// want to handle the Ready and must pass the Ready to Advance *after* having
+// done so. While a Ready is being handled, the RawNode must not be used for
+// operations that may alter its state. For example, it is illegal to call
+// Ready, followed by Step, followed by Advance.
 func (rn *RawNode) Ready() Ready {
 	rd := rn.newReady()
-	rn.raft.msgs = nil
-	rn.raft.reduceUncommittedSize(rd.CommittedEntries)
 	return rd
 }
 
+func (rn *RawNode) newReady() Ready {
+	return newReady(rn.raft, rn.prevSoftSt, rn.prevHardSt)
+}
+
+// acceptReady is called when the consumer of the RawNode has decided to go
+// ahead and handle a Ready. Nothing must alter the state of the RawNode between
+// this call and the prior call to Ready().
+func (rn *RawNode) acceptReady(rd Ready) {
+	if rd.SoftState != nil {
+		rn.prevSoftSt = rd.SoftState
+	}
+	if len(rd.ReadStates) != 0 {
+		rn.raft.readStates = nil
+	}
+	rn.raft.msgs = nil
+}
+
+// commitReady is called when the consumer of the RawNode has successfully
+// handled a Ready (having previously called acceptReady).
+func (rn *RawNode) commitReady(rd Ready) {
+	if !IsEmptyHardState(rd.HardState) {
+		rn.prevHardSt = rd.HardState
+	}
+
+	// If entries were applied (or a snapshot), update our cursor for
+	// the next Ready. Note that if the current HardState contains a
+	// new Commit index, this does not mean that we're also applying
+	// all of the new entries due to commit pagination by size.
+	if index := rd.appliedCursor(); index > 0 {
+		rn.raft.raftLog.appliedTo(index)
+	}
+	rn.raft.reduceUncommittedSize(rd.CommittedEntries)
+
+	if len(rd.Entries) > 0 {
+		e := rd.Entries[len(rd.Entries)-1]
+		rn.raft.raftLog.stableTo(e.Index, e.Term)
+	}
+	if !IsEmptySnap(rd.Snapshot) {
+		rn.raft.raftLog.stableSnapTo(rd.Snapshot.Metadata.Index)
+	}
+}
+
 // HasReady called when RawNode user need to check if any Ready pending.
 // Checking logic in this method should be consistent with Ready.containsUpdates().
 func (rn *RawNode) HasReady() bool {
@@ -215,6 +251,11 @@ func (rn *RawNode) HasReady() bool {
 // Advance notifies the RawNode that the application has applied and saved progress in the
 // last Ready results.
 func (rn *RawNode) Advance(rd Ready) {
+	// Advance combines accept and commit. Callers can't mutate the RawNode
+	// between the call to Ready and the matching call to Advance, or the work
+	// done in acceptReady will clobber potentially newer data that has not been
+	// emitted in a Ready yet.
+	rn.acceptReady(rd)
 	rn.commitReady(rd)
 }