fix: backoff retry and timeouts in HNS restart

Signed-off-by: Evan Baker <[email protected]>

Author: rbtr

platform/os_windows.go

@@ -308,13 +308,15 @@ func restartHNS(ctx context.Context) error {
 		tryStopServiceFn(ctx, service),
 		retry.UntilSucceeded(),
 		retry.Context(ctx),
+		retry.DelayType(retry.BackOffDelay),
 	)
 	// Start the service again
 	log.Printf("Starting HNS service")
 	_ = retry.Do(
 		tryStartServiceFn(ctx, service),
 		retry.UntilSucceeded(),
 		retry.Context(ctx),
+		retry.DelayType(retry.BackOffDelay),
 	)
 	log.Printf("HNS service started")
 	return nil
@@ -330,6 +332,14 @@ func tryStartServiceFn(ctx context.Context, service managedService) func() error
 	shouldStart := func(state svc.State) bool {
 		return !(state == svc.Running || state == svc.StartPending)
 	}
+	var n, limit time.Duration = 0, 3
+	deadline := func(parent context.Context) (context.Context, context.CancelFunc) {
+		n++
+		if n > limit {
+			n = limit
+		}
+		return context.WithTimeout(parent, n*30*time.Second) //nolint:durationcheck // it's correct
+	}
 	return func() error {
 		status, err := service.Query()
 		if err != nil {
@@ -342,6 +352,8 @@ func tryStartServiceFn(ctx context.Context, service managedService) func() error
 			}
 		}
 		// Wait for the service to start
+		deadline, cancel := deadline(ctx)
+		defer cancel()
 		ticker := time.NewTicker(500 * time.Millisecond) //nolint:gomnd // 500ms
 		defer ticker.Stop()
 		for {
@@ -354,8 +366,8 @@ func tryStartServiceFn(ctx context.Context, service managedService) func() error
 				break
 			}
 			select {
-			case <-ctx.Done():
-				return errors.Wrap(ctx.Err(), "context cancelled")
+			case <-deadline.Done():
+				return errors.Wrap(deadline.Err(), "context cancelled")
 			case <-ticker.C:
 			}
 		}
@@ -367,6 +379,14 @@ func tryStopServiceFn(ctx context.Context, service managedService) func() error
 	shouldStop := func(state svc.State) bool {
 		return !(state == svc.Stopped || state == svc.StopPending)
 	}
+	var n, limit time.Duration = 0, 3
+	deadline := func(parent context.Context) (context.Context, context.CancelFunc) {
+		n++
+		if n > limit {
+			n = limit
+		}
+		return context.WithTimeout(parent, n*30*time.Second) //nolint:durationcheck // it's correct
+	}
 	return func() error {
 		status, err := service.Query()
 		if err != nil {
@@ -379,6 +399,8 @@ func tryStopServiceFn(ctx context.Context, service managedService) func() error
 			}
 		}
 		// Wait for the service to stop
+		deadline, cancel := deadline(ctx)
+		defer cancel()
 		ticker := time.NewTicker(500 * time.Millisecond) //nolint:gomnd // 500ms
 		defer ticker.Stop()
 		for {
@@ -391,8 +413,8 @@ func tryStopServiceFn(ctx context.Context, service managedService) func() error
 				break
 			}
 			select {
-			case <-ctx.Done():
-				return errors.Wrap(ctx.Err(), "context cancelled")
+			case <-deadline.Done():
+				return errors.Wrap(deadline.Err(), "context cancelled")
 			case <-ticker.C:
 			}
 		}