let monitor know we are restarting a container

pkg/compose/compose.go

@@ -70,6 +70,8 @@ type composeService struct {
 	clock          clockwork.Clock
 	maxConcurrency int
 	dryRun         bool
+
+	Monitor *monitor
 }
 
 // Close releases any connections/resources held by the underlying clients.

pkg/compose/monitor.go

@@ -35,15 +35,17 @@ type monitor struct {
 	api     client.APIClient
 	project string
 	// services tells us which service to consider and those we can ignore, maybe ran by a concurrent compose command
-	services  map[string]bool
-	listeners []api.ContainerEventListener
+	services   map[string]bool
+	listeners  []api.ContainerEventListener
+	restarting chan string
 }
 
 func newMonitor(api client.APIClient, project string) *monitor {
 	return &monitor{
-		api:      api,
-		project:  project,
-		services: map[string]bool{},
+		api:        api,
+		project:    project,
+		services:   map[string]bool{},
+		restarting: make(chan string),
 	}
 }
 
@@ -91,6 +93,8 @@ func (c *monitor) Start(ctx context.Context) error {
 		select {
 		case err := <-errCh:
 			return err
+		case ctr := <-c.restarting:
+			restarting.Add(ctr)
 		case event := <-evtCh:
 			if len(c.services) > 0 && !c.services[event.Actor.Attributes[api.ServiceLabel]] {
 				continue
@@ -100,6 +104,8 @@ func (c *monitor) Start(ctx context.Context) error {
 				return err
 			}
 
+			logrus.Debugf("event: Container %s %s", ctr.Name, event.Action)
+
 			switch event.Action {
 			case events.ActionCreate:
 				if len(c.services) == 0 || c.services[ctr.Labels[api.ServiceLabel]] {
@@ -140,7 +146,9 @@ func (c *monitor) Start(ctx context.Context) error {
 				// when a container is in restarting phase, and we stop the application (abort-on-container-exit)
 				// we won't get any additional start+die events, just this stop as a proof container is down
 				logrus.Debugf("container %s stopped", ctr.Name)
-				containers.Remove(ctr.ID)
+				if !restarting.Has(ctr.ID) {
+					containers.Remove(ctr.ID)
+				}
 			case events.ActionDie:
 				logrus.Debugf("container %s exited with code %d", ctr.Name, ctr.ExitCode)
 				inspect, err := c.api.ContainerInspect(ctx, event.Actor.ID)
@@ -165,7 +173,9 @@ func (c *monitor) Start(ctx context.Context) error {
 					for _, listener := range c.listeners {
 						listener(newContainerEvent(event.TimeNano, ctr, api.ContainerEventExited))
 					}
-					containers.Remove(ctr.ID)
+					if !restarting.Has(ctr.ID) {
+						containers.Remove(ctr.ID)
+					}
 				}
 			}
 		}
@@ -216,3 +226,7 @@ func (c *monitor) getContainerSummary(event events.Message) (*api.ContainerSumma
 func (c *monitor) withListener(listener api.ContainerEventListener) {
 	c.listeners = append(c.listeners, listener)
 }
+
+func (c *monitor) Restarting(id string) {
+	c.restarting <- id
+}

pkg/compose/restart.go

@@ -89,6 +89,11 @@ func (s *composeService) restart(ctx context.Context, projectName string, option
 				eventName := getContainerProgressName(ctr)
 				w.Event(progress.RestartingEvent(eventName))
 				timeout := utils.DurationSecondToInt(options.Timeout)
+				if s.Monitor != nil {
+					// Let monitor know we are restarting container, as there's no way to guess based on engine events
+					s.Monitor.Restarting(ctr.ID)
+				}
+
 				err = s.apiClient().ContainerRestart(ctx, ctr.ID, container.StopOptions{Timeout: timeout})
 				if err != nil {
 					return err

pkg/compose/up.go

@@ -271,6 +271,7 @@ func (s *composeService) Up(ctx context.Context, project *types.Project, options
 		})
 	})
 
+	s.Monitor = monitor
 	eg.Go(func() error {
 		err := monitor.Start(globalCtx)
 		// cancel the global context to terminate signal-handler goroutines