*: stability fixes

Signed-off-by: Antonio Murdaca <runcom@redhat.com>

pkg/storage/runtime.go

@@ -345,11 +345,11 @@ func (r *runtimeService) RemovePodSandbox(idOrName string) error {
 }
 
 func (r *runtimeService) DeleteContainer(idOrName string) error {
-	container, err := r.storageImageServer.GetStore().Container(idOrName)
+	if idOrName == "" {
-	if err != nil {
-		if err == storage.ErrContainerUnknown {
 		return ErrInvalidContainerID
 	}
+	container, err := r.storageImageServer.GetStore().Container(idOrName)
+	if err != nil {
 		return err
 	}
 	err = r.storageImageServer.GetStore().DeleteContainer(container.ID)
@@ -403,11 +403,11 @@ func (r *runtimeService) StartContainer(idOrName string) (string, error) {
 }
 
 func (r *runtimeService) StopContainer(idOrName string) error {
-	container, err := r.storageImageServer.GetStore().Container(idOrName)
+	if idOrName == "" {
-	if err != nil {
-		if err == storage.ErrContainerUnknown {
 		return ErrInvalidContainerID
 	}
+	container, err := r.storageImageServer.GetStore().Container(idOrName)
+	if err != nil {
 		return err
 	}
 	err = r.storageImageServer.GetStore().Unmount(container.ID)

server/sandbox_remove.go

@@ -5,6 +5,8 @@ import (
 	"syscall"
 
 	"github.com/Sirupsen/logrus"
+	"github.com/containers/storage"
+	"github.com/docker/docker/pkg/mount"
 	"github.com/kubernetes-incubator/cri-o/oci"
 	"github.com/opencontainers/selinux/go-selinux/label"
 	"golang.org/x/net/context"
@@ -56,10 +58,11 @@ func (s *Server) RemovePodSandbox(ctx context.Context, req *pb.RemovePodSandboxR
 			continue
 		}
 
-		if err := s.storageRuntimeServer.StopContainer(c.ID()); err != nil {
+		if err := s.storageRuntimeServer.StopContainer(c.ID()); err != nil && err != storage.ErrContainerUnknown {
-			return nil, fmt.Errorf("failed to delete container %s in pod sandbox %s: %v", c.Name(), sb.id, err)
+			// assume container already umounted
+			logrus.Warnf("failed to stop container %s in pod sandbox %s: %v", c.Name(), sb.id, err)
 		}
-		if err := s.storageRuntimeServer.DeleteContainer(c.ID()); err != nil {
+		if err := s.storageRuntimeServer.DeleteContainer(c.ID()); err != nil && err != storage.ErrContainerUnknown {
 			return nil, fmt.Errorf("failed to delete container %s in pod sandbox %s: %v", c.Name(), sb.id, err)
 		}
 
@@ -76,10 +79,12 @@ func (s *Server) RemovePodSandbox(ctx context.Context, req *pb.RemovePodSandboxR
 
 	// unmount the shm for the pod
 	if sb.shmPath != "/dev/shm" {
+		if mounted, err := mount.Mounted(sb.shmPath); err == nil && mounted {
 			if err := syscall.Unmount(sb.shmPath, syscall.MNT_DETACH); err != nil {
 				return nil, err
 			}
 		}
+	}
 
 	if err := sb.netNsRemove(); err != nil {
 		return nil, fmt.Errorf("failed to remove networking namespace for sandbox %s: %v", sb.id, err)
@@ -89,7 +94,7 @@ func (s *Server) RemovePodSandbox(ctx context.Context, req *pb.RemovePodSandboxR
 
 	// Remove the files related to the sandbox
 	if err := s.storageRuntimeServer.StopContainer(sb.id); err != nil {
-		return nil, fmt.Errorf("failed to delete sandbox container in pod sandbox %s: %v", sb.id, err)
+		logrus.Warnf("failed to stop sandbox container in pod sandbox %s: %v", sb.id, err)
 	}
 	if err := s.storageRuntimeServer.RemovePodSandbox(sb.id); err != nil {
 		return nil, fmt.Errorf("failed to remove pod sandbox %s: %v", sb.id, err)

server/sandbox_run.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"regexp"
 	"strconv"
 	"strings"
 	"syscall"
@@ -66,6 +67,10 @@ func (s *Server) runContainer(container *oci.Container, cgroupParent string) err
 	return nil
 }
 
+var (
+	conflictRE = regexp.MustCompile(`already reserved for pod "([0-9a-z]+)"`)
+)
+
 // RunPodSandbox creates and runs a pod-level sandbox.
 func (s *Server) RunPodSandbox(ctx context.Context, req *pb.RunPodSandboxRequest) (resp *pb.RunPodSandboxResponse, err error) {
 	s.updateLock.RLock()
@@ -84,8 +89,23 @@ func (s *Server) RunPodSandbox(ctx context.Context, req *pb.RunPodSandboxRequest
 
 	id, name, err := s.generatePodIDandName(kubeName, namespace, attempt)
 	if err != nil {
+		if strings.Contains(err.Error(), "already reserved for pod") {
+			matches := conflictRE.FindStringSubmatch(err.Error())
+			if len(matches) != 2 {
 				return nil, err
 			}
+			dupID := matches[1]
+			if _, err := s.RemovePodSandbox(ctx, &pb.RemovePodSandboxRequest{PodSandboxId: dupID}); err != nil {
+				return nil, err
+			}
+			id, name, err = s.generatePodIDandName(kubeName, namespace, attempt)
+			if err != nil {
+				return nil, err
+			}
+		} else {
+			return nil, err
+		}
+	}
 	_, containerName, err := s.generateContainerIDandName(name, "infra", attempt)
 	if err != nil {
 		return nil, err