Merge pull request #9132 from dcantah/shim-kill-1.7

fuweid · web-flow · commit 7a75a30c60e1 · 2023-09-30T11:42:07.000+08:00
[release/1.7] Handle unexpected shim kill events
diff --git a/pkg/cri/sbserver/events.go b/pkg/cri/sbserver/events.go
@@ -393,7 +393,7 @@ func handleContainerExit(ctx context.Context, e *eventtypes.TaskExit, cntr conta
 		},
 	)
 	if err != nil {
-		if !errdefs.IsNotFound(err) {
+		if !errdefs.IsNotFound(err) && !errdefs.IsUnavailable(err) {
 			return fmt.Errorf("failed to load task for container: %w", err)
 		}
 	} else {
diff --git a/plugins/sandbox/controller.go b/plugins/sandbox/controller.go
@@ -260,6 +260,12 @@ func (c *controllerLocal) Wait(ctx context.Context, sandboxID string) (sandbox.E
 
 func (c *controllerLocal) Status(ctx context.Context, sandboxID string, verbose bool) (sandbox.ControllerStatus, error) {
 	svc, err := c.getSandbox(ctx, sandboxID)
+	if errdefs.IsNotFound(err) {
+		return sandbox.ControllerStatus{
+			SandboxID: sandboxID,
+			ExitedAt:  time.Now(),
+		}, nil
+	}
 	if err != nil {
 		return sandbox.ControllerStatus{}, err
 	}
@@ -286,7 +292,7 @@ func (c *controllerLocal) Status(ctx context.Context, sandboxID string, verbose
 func (c *controllerLocal) getSandbox(ctx context.Context, id string) (runtimeAPI.TTRPCSandboxService, error) {
 	shim, err := c.shims.Get(ctx, id)
 	if err != nil {
-		return nil, errdefs.ErrNotFound
+		return nil, err
 	}
 
 	return sandbox.NewClient(shim.Client())
diff --git a/services/sandbox/controller_service.go b/services/sandbox/controller_service.go
@@ -144,17 +144,21 @@ func (s *controllerService) Status(ctx context.Context, req *api.ControllerStatu
 	if err != nil {
 		return &api.ControllerStatusResponse{}, errdefs.ToGRPC(err)
 	}
+	extra := &anypb.Any{}
+	if cstatus.Extra != nil {
+		extra = &anypb.Any{
+			TypeUrl: cstatus.Extra.GetTypeUrl(),
+			Value:   cstatus.Extra.GetValue(),
+		}
+	}
 	return &api.ControllerStatusResponse{
 		SandboxID: cstatus.SandboxID,
 		Pid:       cstatus.Pid,
 		State:     cstatus.State,
 		Info:      cstatus.Info,
 		CreatedAt: protobuf.ToTimestamp(cstatus.CreatedAt),
 		ExitedAt:  protobuf.ToTimestamp(cstatus.ExitedAt),
-		Extra: &anypb.Any{
-			TypeUrl: cstatus.Extra.GetTypeUrl(),
-			Value:   cstatus.Extra.GetValue(),
-		},
+		Extra:     extra,
 	}, nil
 }
 
diff --git a/services/tasks/local.go b/services/tasks/local.go
@@ -339,7 +339,7 @@ func getProcessState(ctx context.Context, p runtime.Process) (*task.Process, err
 
 	state, err := p.State(ctx)
 	if err != nil {
-		if errdefs.IsNotFound(err) {
+		if errdefs.IsNotFound(err) || errdefs.IsUnavailable(err) {
 			return nil, err
 		}
 		log.G(ctx).WithError(err).Errorf("get state for %s", p.ID())

Original file line number	Diff line number	Diff line change
`@@ -393,7 +393,7 @@ func handleContainerExit(ctx context.Context, e *eventtypes.TaskExit, cntr conta`
`393`	`393`	`},`
`394`	`394`	`)`
`395`	`395`	`if err != nil {`
`396`		`- if !errdefs.IsNotFound(err) {`
	`396`	`+ if !errdefs.IsNotFound(err) && !errdefs.IsUnavailable(err) {`
`397`	`397`	`return fmt.Errorf("failed to load task for container: %w", err)`
`398`	`398`	`}`
`399`	`399`	`} else {`
Original file line number	Diff line number	Diff line change
`@@ -339,7 +339,7 @@ func getProcessState(ctx context.Context, p runtime.Process) (*task.Process, err`
`339`	`339`
`340`	`340`	`state, err := p.State(ctx)`
`341`	`341`	`if err != nil {`
`342`		`- if errdefs.IsNotFound(err) {`
	`342`	`+ if errdefs.IsNotFound(err) \|\| errdefs.IsUnavailable(err) {`
`343`	`343`	`return nil, err`
`344`	`344`	`}`
`345`	`345`	`log.G(ctx).WithError(err).Errorf("get state for %s", p.ID())`