client: Handle sending/receiving in separate goroutines

Changes the TTRPC client logic so that sending and receiving with the
server are in completely independent goroutines, with shared state
guarded by a mutex. Previously, sending/receiving were tied together by
reliance on a coordinator goroutine. This led to issues where if the
server was not reading from the connection, the client could get stuck
sending a request, causing the client to not read responses from the
server. See [1] for more details.

The new design sets up separate sending/receiving goroutines. These
share state in the form of the set of active calls that have been made
to the server. This state is encapsulated in the callMap type and access
is guarded by a mutex.

The main event loop in `run` previously handled a lot of state
management for the client. Now that most state is tracked by the
callMap, it mostly exists to notice when the client is closed and take
appropriate action to clean up.

Also did some minor code cleanup. For instance, the code was previously
written to support multiple receiver goroutines, though this was not
actually used. I've removed this for now, since the code is simpler this
way, and it's easy to add back if we actually need it in the future.

[1] containerd#72

Signed-off-by: Kevin Parsons <[email protected]>

client.go

@@ -194,72 +194,131 @@ type message struct {
 	err error
 }
 
-type receiver struct {
-	wg       *sync.WaitGroup
-	messages chan *message
-	err      error
+// callMap provides access to a map of active calls, guarded by a mutex.
+type callMap struct {
+	m           sync.Mutex
+	activeCalls map[uint32]*callRequest
+	closeErr    error
 }
 
-func (r *receiver) run(ctx context.Context, c *channel) {
-	defer r.wg.Done()
+// newCallMap returns a new callMap with an empty set of active calls.
+func newCallMap() *callMap {
+	return &callMap{
+		activeCalls: make(map[uint32]*callRequest),
+	}
+}
 
-	for {
-		select {
-		case <-ctx.Done():
-			r.err = ctx.Err()
-			return
-		default:
-			mh, p, err := c.recv()
-			if err != nil {
-				_, ok := status.FromError(err)
-				if !ok {
-					// treat all errors that are not an rpc status as terminal.
-					// all others poison the connection.
-					r.err = filterCloseErr(err)
-					return
-				}
-			}
-			select {
-			case r.messages <- &message{
-				messageHeader: mh,
-				p:             p[:mh.Length],
-				err:           err,
-			}:
-			case <-ctx.Done():
-				r.err = ctx.Err()
-				return
-			}
-		}
+// set adds a call entry to the map with the given streamID key.
+func (cm *callMap) set(streamID uint32, cr *callRequest) error {
+	cm.m.Lock()
+	defer cm.m.Unlock()
+	if cm.closeErr != nil {
+		return cm.closeErr
 	}
+	cm.activeCalls[streamID] = cr
+	return nil
+}
+
+// get looks up the call entry for the given streamID key, then removes it
+// from the map and returns it.
+func (cm *callMap) get(streamID uint32) (cr *callRequest, ok bool, err error) {
+	cm.m.Lock()
+	defer cm.m.Unlock()
+	if cm.closeErr != nil {
+		return nil, false, cm.closeErr
+	}
+	cr, ok = cm.activeCalls[streamID]
+	if ok {
+		delete(cm.activeCalls, streamID)
+	}
+	return
+}
+
+// abort sends the given error to each active call, and clears the map.
+// Once abort has been called, any subsequent calls to the callMap will return the error passed to abort.
+func (cm *callMap) abort(err error) error {
+	cm.m.Lock()
+	defer cm.m.Unlock()
+	if cm.closeErr != nil {
+		return cm.closeErr
+	}
+	for streamID, call := range cm.activeCalls {
+		call.errs <- err
+		delete(cm.activeCalls, streamID)
+	}
+	cm.closeErr = err
+	return nil
 }
 
 func (c *Client) run() {
 	var (
-		streamID      uint32 = 1
-		waiters              = make(map[uint32]*callRequest)
-		calls                = c.calls
-		incoming             = make(chan *message)
-		receiversDone        = make(chan struct{})
-		wg            sync.WaitGroup
+		waiters      = newCallMap()
+		receiverDone = make(chan struct{})
 	)
 
-	// broadcast the shutdown error to the remaining waiters.
-	abortWaiters := func(wErr error) {
-		for _, waiter := range waiters {
-			waiter.errs <- wErr
+	// Sender goroutine
+	// Receives calls from dispatch, adds them to the set of active calls, and sends them
+	// to the server.
+	go func() {
+		var streamID uint32 = 1
+		for {
+			select {
+			case <-c.ctx.Done():
+				return
+			case call := <-c.calls:
+				id := streamID
+				streamID += 2 // enforce odd client initiated request ids
+				if err := waiters.set(id, call); err != nil {
+					call.errs <- err // errs is buffered so should not block.
+					continue
+				}
+				if err := c.send(id, messageTypeRequest, call.req); err != nil {
+					call.errs <- err // errs is buffered so should not block.
+					waiters.get(id)  // remove from waiters set
+				}
+			}
 		}
-	}
-	recv := &receiver{
-		wg:       &wg,
-		messages: incoming,
-	}
-	wg.Add(1)
+	}()
 
+	// Receiver goroutine
+	// Receives responses from the server, looks up the call info in the set of active calls,
+	// and notifies the caller of the response.
 	go func() {
-		wg.Wait()
-		close(receiversDone)
+		defer close(receiverDone)
+		for {
+			select {
+			case <-c.ctx.Done():
+				c.setError(c.ctx.Err())
+				return
+			default:
+				mh, p, err := c.channel.recv()
+				if err != nil {
+					_, ok := status.FromError(err)
+					if !ok {
+						// treat all errors that are not an rpc status as terminal.
+						// all others poison the connection.
+						c.setError(filterCloseErr(err))
+						return
+					}
+				}
+				msg := &message{
+					messageHeader: mh,
+					p:             p[:mh.Length],
+					err:           err,
+				}
+				call, ok, err := waiters.get(mh.StreamID)
+				if err != nil {
+					logrus.Errorf("ttrpc: failed to look up active call: %s", err)
+					continue
+				}
+				if !ok {
+					logrus.Errorf("ttrpc: received message for unknown channel %v", mh.StreamID)
+					continue
+				}
+				call.errs <- c.recv(call.resp, msg)
+			}
+		}
 	}()
-	go recv.run(c.ctx, c.channel)
 
 	defer func() {
 		c.conn.Close()
@@ -269,32 +328,14 @@ func (c *Client) run() {
 
 	for {
 		select {
-		case call := <-calls:
-			if err := c.send(streamID, messageTypeRequest, call.req); err != nil {
-				call.errs <- err
-				continue
-			}
-
-			waiters[streamID] = call
-			streamID += 2 // enforce odd client initiated request ids
-		case msg := <-incoming:
-			call, ok := waiters[msg.StreamID]
-			if !ok {
-				logrus.Errorf("ttrpc: received message for unknown channel %v", msg.StreamID)
-				continue
-			}
-
-			call.errs <- c.recv(call.resp, msg)
-			delete(waiters, msg.StreamID)
-		case <-receiversDone:
-			// all the receivers have exited
-			if recv.err != nil {
-				c.setError(recv.err)
-			}
+		case <-receiverDone:
+			// The receiver has exited.
 			// don't return out, let the close of the context trigger the abort of waiters
 			c.Close()
 		case <-c.ctx.Done():
-			abortWaiters(c.error())
+			// Abort all active calls. This will also prevent any new calls from being added
+			// to waiters.
+			waiters.abort(c.error())
 			return
 		}
 	}