Merge pull request #939 from ijc25/reconnect-shim-event-stream

Reconnect to shim event stream after containerd restart

linux/runtime.go

@@ -257,6 +257,10 @@ func (r *Runtime) loadContainer(path string) (*Task, error) {
 		return nil, err
 	}
 
+	if err = r.handleEvents(s); err != nil {
+		return nil, err
+	}
+
 	data, err := ioutil.ReadFile(filepath.Join(path, configFilename))
 	if err != nil {
 		return nil, err

linux/shim/service.go

@@ -37,6 +37,8 @@ type Service struct {
 	mu            sync.Mutex
 	processes     map[int]process
 	events        chan *task.Event
+	eventsMu      sync.Mutex
+	deferredEvent *task.Event
 	execID        int
 }
 
@@ -145,12 +147,27 @@ func (s *Service) Pty(ctx context.Context, r *shimapi.PtyRequest) (*google_proto
 }
 
 func (s *Service) Events(r *shimapi.EventsRequest, stream shimapi.Shim_EventsServer) error {
-	for e := range s.events {
-		if err := stream.Send(e); err != nil {
+	s.eventsMu.Lock()
+	defer s.eventsMu.Unlock()
+
+	if s.deferredEvent != nil {
+		if err := stream.Send(s.deferredEvent); err != nil {
 			return err
 		}
+		s.deferredEvent = nil
+	}
+
+	for {
+		select {
+		case e := <-s.events:
+			if err := stream.Send(e); err != nil {
+				s.deferredEvent = e
+				return err
+			}
+		case <-stream.Context().Done():
+			return stream.Context().Err()
+		}
 	}
-	return nil
 }
 
 func (s *Service) State(ctx context.Context, r *shimapi.StateRequest) (*shimapi.StateResponse, error) {