util/eventbus: adjust worker goroutine management helpers

This makes the helpers closer in behavior to cancelable contexts
and taskgroup.Single, and makes the worker code use a more normal
and easier to reason about context.Context for shutdown.

Updates #15160

Signed-off-by: David Anderson <dave@tailscale.com>

util/eventbus/bus.go

@@ -15,8 +15,8 @@ import (
 // Bus is an event bus that distributes published events to interested
 // subscribers.
 type Bus struct {
+	router   *worker
 	write    chan any
-	stop     goroutineShutdownControl
 	snapshot chan chan []any
 
 	topicsMu sync.Mutex // guards everything below.
@@ -30,15 +30,13 @@ type Bus struct {
 // New returns a new bus. Use [PublisherOf] to make event publishers,
 // and [Bus.Queue] and [Subscribe] to make event subscribers.
 func New() *Bus {
-	stopCtl, stopWorker := newGoroutineShutdown()
 	ret := &Bus{
 		write:    make(chan any),
-		stop:     stopCtl,
 		snapshot: make(chan chan []any),
 		topics:   map[reflect.Type][]*subscribeState{},
 		clients:  set.Set[*Client]{},
 	}
-	go ret.pump(stopWorker)
+	ret.router = runWorker(ret.pump)
 	return ret
 }
 
@@ -67,7 +65,7 @@ func (b *Bus) Client(name string) *Client {
 // Close blocks until the bus is fully shut down. The bus is
 // permanently unusable after closing.
 func (b *Bus) Close() {
-	b.stop.StopAndWait()
+	b.router.StopAndWait()
 
 	var clients set.Set[*Client]
 	b.topicsMu.Lock()
@@ -79,8 +77,7 @@ func (b *Bus) Close() {
 	}
 }
 
-func (b *Bus) pump(stop goroutineShutdownWorker) {
+func (b *Bus) pump(ctx context.Context) {
-	defer stop.Done()
 	var vals queue
 	acceptCh := func() chan any {
 		if vals.Full() {
@@ -102,13 +99,13 @@ func (b *Bus) pump(stop goroutineShutdownWorker) {
 					select {
 					case d.write <- val:
 						break deliverOne
-					case <-d.stop.WaitChan():
+					case <-d.closed():
 						// Queue closed, don't block but continue
 						// delivering to others.
 						break deliverOne
 					case in := <-acceptCh():
 						vals.Add(in)
-					case <-stop.Stop():
+					case <-ctx.Done():
 						return
 					case ch := <-b.snapshot:
 						ch <- vals.Snapshot()
@@ -122,7 +119,7 @@ func (b *Bus) pump(stop goroutineShutdownWorker) {
 		// resuming.
 		for vals.Empty() {
 			select {
-			case <-stop.Stop():
+			case <-ctx.Done():
 				return
 			case val := <-b.write:
 				vals.Add(val)
@@ -168,59 +165,89 @@ func (b *Bus) unsubscribe(t reflect.Type, q *subscribeState) {
 	b.topics[t] = slices.Delete(slices.Clone(b.topics[t]), i, i+1)
 }
 
-func newGoroutineShutdown() (goroutineShutdownControl, goroutineShutdownWorker) {
+// A worker runs a worker goroutine and helps coordinate its shutdown.
-	ctx, cancel := context.WithCancel(context.Background())
+type worker struct {
+	ctx     context.Context
+	stop    context.CancelFunc
+	stopped chan struct{}
+}
 
-	ctl := goroutineShutdownControl{
+// runWorker creates a worker goroutine running fn. The context passed
-		startShutdown:    cancel,
+// to fn is canceled by [worker.Stop].
-		shutdownFinished: make(chan struct{}),
+func runWorker(fn func(context.Context)) *worker {
+	ctx, stop := context.WithCancel(context.Background())
+	ret := &worker{
+		ctx:     ctx,
+		stop:    stop,
+		stopped: make(chan struct{}),
 	}
-	work := goroutineShutdownWorker{
+	go ret.run(fn)
-		startShutdown:    ctx.Done(),
+	return ret
-		shutdownFinished: ctl.shutdownFinished,
 }
 
-	return ctl, work
+func (w *worker) run(fn func(context.Context)) {
+	defer close(w.stopped)
+	fn(w.ctx)
 }
 
-// goroutineShutdownControl is a helper type to manage the shutdown of
+// Stop signals the worker goroutine to shut down.
-// a worker goroutine. The worker goroutine should use the
+func (w *worker) Stop() { w.stop() }
-// goroutineShutdownWorker related to this controller.
-type goroutineShutdownControl struct {
-	startShutdown    context.CancelFunc
-	shutdownFinished chan struct{}
-}
 
-func (ctl *goroutineShutdownControl) Stop() {
+// Done returns a channel that is closed when the worker goroutine
-	ctl.startShutdown()
+// exits.
-}
+func (w *worker) Done() <-chan struct{} { return w.stopped }
 
-func (ctl *goroutineShutdownControl) Wait() {
+// Wait waits until the worker goroutine has exited.
-	<-ctl.shutdownFinished
+func (w *worker) Wait() { <-w.stopped }
-}
 
-func (ctl *goroutineShutdownControl) WaitChan() <-chan struct{} {
+// StopAndWait signals the worker goroutine to shut down, then waits
-	return ctl.shutdownFinished
+// for it to exit.
+func (w *worker) StopAndWait() {
+	w.stop()
+	<-w.stopped
 }
 
-func (ctl *goroutineShutdownControl) StopAndWait() {
+// stopFlag is a value that can be watched for a notification. The
-	ctl.Stop()
+// zero value is ready for use.
-	ctl.Wait()
+//
+// The flag is notified by running [stopFlag.Stop]. Stop can be called
+// multiple times. Upon the first call to Stop, [stopFlag.Done] is
+// closed, all pending [stopFlag.Wait] calls return, and future Wait
+// calls return immediately.
+//
+// A stopFlag can only notify once, and is intended for use as a
+// one-way shutdown signal that's lighter than a cancellable
+// context.Context.
+type stopFlag struct {
+	// guards the lazy construction of stopped, and the value of
+	// alreadyStopped.
+	mu             sync.Mutex
+	stopped        chan struct{}
+	alreadyStopped bool
+}
+
+func (s *stopFlag) Stop() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.alreadyStopped {
+		return
 	}
-
+	s.alreadyStopped = true
-// goroutineShutdownWorker is a helper type for a worker goroutine to
+	if s.stopped == nil {
-// be notified that it should shut down, and to report that shutdown
+		s.stopped = make(chan struct{})
-// has completed. The notification is triggered by the related
+	}
-// goroutineShutdownControl.
+	close(s.stopped)
-type goroutineShutdownWorker struct {
-	startShutdown    <-chan struct{}
-	shutdownFinished chan struct{}
 }
 
-func (work *goroutineShutdownWorker) Stop() <-chan struct{} {
+func (s *stopFlag) Done() <-chan struct{} {
-	return work.startShutdown
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.stopped == nil {
+		s.stopped = make(chan struct{})
+	}
+	return s.stopped
 }
 
-func (work *goroutineShutdownWorker) Done() {
+func (s *stopFlag) Wait() {
-	close(work.shutdownFinished)
+	<-s.Done()
 }

util/eventbus/publish.go

@@ -4,7 +4,6 @@
 package eventbus
 
 import (
-	"context"
 	"reflect"
 )
 
@@ -18,16 +17,12 @@ type publisher interface {
 // A Publisher publishes typed events on a bus.
 type Publisher[T any] struct {
 	client *Client
-	stopCtx context.Context
+	stop   stopFlag
-	stop    context.CancelFunc
 }
 
 func newPublisher[T any](c *Client) *Publisher[T] {
-	ctx, cancel := context.WithCancel(context.Background())
 	ret := &Publisher[T]{
 		client: c,
-		stopCtx: ctx,
-		stop:    cancel,
 	}
 	c.addPublisher(ret)
 	return ret
@@ -39,7 +34,7 @@ func newPublisher[T any](c *Client) *Publisher[T] {
 func (p *Publisher[T]) Close() {
 	// Just unblocks any active calls to Publish, no other
 	// synchronization needed.
-	p.stop()
+	p.stop.Stop()
 	p.client.deletePublisher(p)
 }
 
@@ -52,14 +47,14 @@ func (p *Publisher[T]) Publish(v T) {
 	// Check for just a stopped publisher or bus before trying to
 	// write, so that once closed Publish consistently does nothing.
 	select {
-	case <-p.stopCtx.Done():
+	case <-p.stop.Done():
 		return
 	default:
 	}
 
 	select {
 	case p.client.publish() <- v:
-	case <-p.stopCtx.Done():
+	case <-p.stop.Done():
 	}
 }
 

util/eventbus/subscribe.go

@@ -27,7 +27,7 @@ type subscriber interface {
 	// processing other potential sources of wakeups, which is how we end
 	// up at this awkward type signature and sharing of internal state
 	// through dispatch.
-	dispatch(vals *queue, stop goroutineShutdownWorker, acceptCh func() chan any) bool
+	dispatch(ctx context.Context, vals *queue, acceptCh func() chan any) bool
 	Close()
 }
 
@@ -35,8 +35,8 @@ type subscriber interface {
 type subscribeState struct {
 	client *Client
 
+	dispatcher *worker
 	write      chan any
-	stop     goroutineShutdownControl
 	snapshot   chan chan []any
 
 	outputsMu sync.Mutex
@@ -44,20 +44,17 @@ type subscribeState struct {
 }
 
 func newSubscribeState(c *Client) *subscribeState {
-	stopCtl, stopWorker := newGoroutineShutdown()
 	ret := &subscribeState{
 		client:   c,
 		write:    make(chan any),
-		stop:     stopCtl,
 		snapshot: make(chan chan []any),
 		outputs:  map[reflect.Type]subscriber{},
 	}
-	go ret.pump(stopWorker)
+	ret.dispatcher = runWorker(ret.pump)
 	return ret
 }
 
-func (q *subscribeState) pump(stop goroutineShutdownWorker) {
+func (q *subscribeState) pump(ctx context.Context) {
-	defer stop.Done()
 	var vals queue
 	acceptCh := func() chan any {
 		if vals.Full() {
@@ -74,7 +71,7 @@ func (q *subscribeState) pump(stop goroutineShutdownWorker) {
 				vals.Drop()
 				continue
 			}
-			if !sub.dispatch(&vals, stop, acceptCh) {
+			if !sub.dispatch(ctx, &vals, acceptCh) {
 				return
 			}
 		} else {
@@ -85,7 +82,7 @@ func (q *subscribeState) pump(stop goroutineShutdownWorker) {
 			select {
 			case val := <-q.write:
 				vals.Add(val)
-			case <-stop.Stop():
+			case <-ctx.Done():
 				return
 			case ch := <-q.snapshot:
 				ch <- vals.Snapshot()
@@ -120,7 +117,7 @@ func (q *subscribeState) subscriberFor(val any) subscriber {
 // Close closes the subscribeState. Implicitly closes all Subscribers
 // linked to this state, and any pending events are discarded.
 func (s *subscribeState) close() {
-	s.stop.StopAndWait()
+	s.dispatcher.StopAndWait()
 
 	var subs map[reflect.Type]subscriber
 	s.outputsMu.Lock()
@@ -131,10 +128,13 @@ func (s *subscribeState) close() {
 	}
 }
 
+func (s *subscribeState) closed() <-chan struct{} {
+	return s.dispatcher.Done()
+}
+
 // A Subscriber delivers one type of event from a [Client].
 type Subscriber[T any] struct {
-	doneCtx context.Context
+	stop stopFlag
-	done    context.CancelFunc
 	recv *subscribeState
 	read chan T
 }
@@ -142,10 +142,7 @@ type Subscriber[T any] struct {
 func newSubscriber[T any](r *subscribeState) *Subscriber[T] {
 	t := reflect.TypeFor[T]()
 
-	ctx, cancel := context.WithCancel(context.Background())
 	ret := &Subscriber[T]{
-		doneCtx: ctx,
-		done:    cancel,
 		recv: r,
 		read: make(chan T),
 	}
@@ -158,7 +155,7 @@ func (s *Subscriber[T]) subscribeType() reflect.Type {
 	return reflect.TypeFor[T]()
 }
 
-func (s *Subscriber[T]) dispatch(vals *queue, stop goroutineShutdownWorker, acceptCh func() chan any) bool {
+func (s *Subscriber[T]) dispatch(ctx context.Context, vals *queue, acceptCh func() chan any) bool {
 	t := vals.Peek().(T)
 	for {
 		// Keep the cases in this select in sync with subscribeState.pump
@@ -170,7 +167,7 @@ func (s *Subscriber[T]) dispatch(vals *queue, stop goroutineShutdownWorker, acce
 			return true
 		case val := <-acceptCh():
 			vals.Add(val)
-		case <-stop.Stop():
+		case <-ctx.Done():
 			return false
 		case ch := <-s.recv.snapshot:
 			ch <- vals.Snapshot()
@@ -187,13 +184,13 @@ func (s *Subscriber[T]) Events() <-chan T {
 // Done returns a channel that is closed when the subscriber is
 // closed.
 func (s *Subscriber[T]) Done() <-chan struct{} {
-	return s.doneCtx.Done()
+	return s.stop.Done()
 }
 
 // Close closes the Subscriber, indicating the caller no longer wishes
 // to receive this event type. After Close, receives on
 // [Subscriber.Events] block for ever.
 func (s *Subscriber[T]) Close() {
-	s.done() // unblock receivers
+	s.stop.Stop() // unblock receivers
 	s.recv.deleteSubscriber(reflect.TypeFor[T]())
 }