sbserver bug fixing

Signed-off-by: Maksym Pavlenko <pavlenko.maksym@gmail.com>

pkg/cri/sbserver/restart.go

@@ -30,6 +30,7 @@ import (
 	"github.com/containerd/containerd/errdefs"
 	containerdimages "github.com/containerd/containerd/images"
 	"github.com/containerd/containerd/log"
+	criconfig "github.com/containerd/containerd/pkg/cri/config"
 	"github.com/containerd/containerd/pkg/cri/sbserver/podsandbox"
 	"github.com/containerd/containerd/platforms"
 	"github.com/containerd/typeurl"
@@ -92,14 +93,39 @@ func (c *criService) recover(ctx context.Context) error {
 		if _, err := c.sandboxStore.Get(sbx.ID); err == nil {
 			continue
 		}
+
 		metadata := sandboxstore.Metadata{}
 		err := sbx.GetExtension(podsandbox.MetadataKey, &metadata)
 		if err != nil {
 			return fmt.Errorf("failed to get metadata for stored sandbox %q: %w", sbx.ID, err)
 		}
-		sb := sandboxstore.NewSandbox(metadata, sandboxstore.Status{State: sandboxstore.StateUnknown})
+
+		var (
+			state      = sandboxstore.StateUnknown
+			controller = c.sandboxControllers[criconfig.ModeShim]
+		)
+
+		status, err := controller.Status(ctx, sbx.ID, false)
+		if err != nil {
+			log.G(ctx).WithError(err).Error("failed to recover sandbox state")
+			if errdefs.IsNotFound(err) {
+				state = sandboxstore.StateNotReady
+			}
+		} else {
+			if code, ok := runtime.PodSandboxState_value[status.State]; ok {
+				if code == int32(runtime.PodSandboxState_SANDBOX_READY) {
+					state = sandboxstore.StateReady
+				} else if code == int32(runtime.PodSandboxState_SANDBOX_NOTREADY) {
+					state = sandboxstore.StateNotReady
+				}
+			}
+		}
+
+		sb := sandboxstore.NewSandbox(metadata, sandboxstore.Status{State: state})
+
 		// Load network namespace.
 		sb.NetNS = getNetNS(&metadata)
+
 		if err := c.sandboxStore.Add(sb); err != nil {
 			return fmt.Errorf("failed to add stored sandbox %q to store: %w", sbx.ID, err)
 		}
@@ -467,6 +493,9 @@ func getNetNS(meta *sandboxstore.Metadata) *netns.NetNS {
 	if goruntime.GOOS == "windows" && meta.Config.GetWindows().GetSecurityContext().GetHostProcess() {
 		return nil
 	}
+	if goruntime.GOOS == "darwin" {
+		return nil
+	}
 	return netns.LoadNetNS(meta.NetNSPath)
 }
 

pkg/cri/sbserver/sandbox_remove.go

@@ -85,7 +85,8 @@ func (c *criService) RemovePodSandbox(ctx context.Context, r *runtime.RemovePodS
 	if err != nil {
 		return nil, fmt.Errorf("failed to get sandbox controller: %w", err)
 	}
-	if _, err := controller.Delete(ctx, id); err != nil {
+
+	if _, err := controller.Delete(ctx, id); err != nil && !errdefs.IsNotFound(err) {
 		return nil, fmt.Errorf("failed to delete sandbox %q: %w", id, err)
 	}
 

pkg/cri/sbserver/sandbox_status.go

@@ -43,7 +43,7 @@ func (c *criService) PodSandboxStatus(ctx context.Context, r *runtime.PodSandbox
 		return nil, fmt.Errorf("failed to get sandbox controller: %w", err)
 	}
 
-	statusResponse, err := controller.Status(ctx, r.GetPodSandboxId(), r.GetVerbose())
+	statusResponse, err := controller.Status(ctx, sandbox.ID, r.GetVerbose())
 	if err != nil {
 		return nil, fmt.Errorf("failed to query controller status: %w", err)
 	}

pkg/cri/sbserver/sandbox_stop.go

@@ -64,13 +64,18 @@ func (c *criService) stopPodSandbox(ctx context.Context, sandbox sandboxstore.Sa
 		}
 	}
 
-	// Use sandbox controller to stop sandbox
-	controller, err := c.getSandboxController(sandbox.Config, sandbox.RuntimeHandler)
-	if err != nil {
-		return fmt.Errorf("failed to get sandbox controller: %w", err)
-	}
-	if _, err := controller.Stop(ctx, id); err != nil {
-		return fmt.Errorf("failed to stop sandbox %q: %w", id, err)
+	// Only stop sandbox container when it's running or unknown.
+	state := sandbox.Status.Get().State
+	if state == sandboxstore.StateReady || state == sandboxstore.StateUnknown {
+		// Use sandbox controller to stop sandbox
+		controller, err := c.getSandboxController(sandbox.Config, sandbox.RuntimeHandler)
+		if err != nil {
+			return fmt.Errorf("failed to get sandbox controller: %w", err)
+		}
+
+		if _, err := controller.Stop(ctx, id); err != nil {
+			return fmt.Errorf("failed to stop sandbox %q: %w", id, err)
+		}
 	}
 
 	sandboxRuntimeStopTimer.WithValues(sandbox.RuntimeHandler).UpdateSince(stop)

services/sandbox/controller_local.go

@@ -156,7 +156,7 @@ func (c *controllerLocal) Stop(ctx context.Context, in *api.ControllerStopReques
 
 func (c *controllerLocal) Delete(ctx context.Context, in *api.ControllerDeleteRequest, opts ...grpc.CallOption) (*api.ControllerDeleteResponse, error) {
 	if err := c.shims.Delete(ctx, in.SandboxID); err != nil {
-		return nil, fmt.Errorf("failed to delete sandbox shim: %w", err)
+		return nil, errdefs.ToGRPC(fmt.Errorf("failed to delete sandbox shim: %w", err))
 	}
 
 	return &api.ControllerDeleteResponse{}, nil
@@ -185,7 +185,7 @@ func (c *controllerLocal) Wait(ctx context.Context, in *api.ControllerWaitReques
 func (c *controllerLocal) Status(ctx context.Context, in *api.ControllerStatusRequest, opts ...grpc.CallOption) (*api.ControllerStatusResponse, error) {
 	svc, err := c.getSandbox(ctx, in.SandboxID)
 	if err != nil {
-		return nil, err
+		return nil, errdefs.ToGRPC(err)
 	}
 
 	resp, err := svc.SandboxStatus(ctx, &runtimeAPI.SandboxStatusRequest{SandboxID: in.SandboxID})