Track container status, and add delete command to OCI interface

containerd/api/grpc/server/server.go

@@ -147,7 +147,17 @@ func (s *apiServer) State(ctx context.Context, r *types.StateRequest) (*types.St
 }
 
 func (s *apiServer) UpdateContainer(ctx context.Context, r *types.UpdateContainerRequest) (*types.UpdateContainerResponse, error) {
-	return nil, errors.New("UpdateContainer() not implemented yet")
+	glog.V(3).Infof("gRPC handle UpdateContainer")
+
+	if r.Status != supervisor.ContainerStateDeleted {
+		return nil, fmt.Errorf("UpdateContainer() status %s not supported", r.Status)
+	}
+
+	err := s.sv.DeleteContainer(r.Id)
+	if err != nil {
+		return nil, err
+	}
+	return &types.UpdateContainerResponse{}, nil
 }
 
 func (s *apiServer) UpdateProcess(ctx context.Context, r *types.UpdateProcessRequest) (*types.UpdateProcessResponse, error) {
@@ -230,7 +240,7 @@ func supervisorContainer2ApiContainer(c *supervisor.Container) *types.Container
 	return &types.Container{
 		Id:         c.Id,
 		BundlePath: c.BundlePath,
-		Status:     "running",
+		Status:     c.Status,
 		Runtime:    "runv",
 	}
 }

containerd/containerd.go

@@ -189,9 +189,9 @@ func namespaceShare(sv *supervisor.Supervisor, namespace, state string) {
 	events := sv.Events.Events(time.Time{})
 	containerCount := 0
 	for e := range events {
-		if e.Type == supervisor.EventContainerStart {
+		if e.Type == supervisor.EventContainerCreate {
 			containerCount++
-		} else if e.Type == supervisor.EventExit && e.PID == "init" {
+		} else if e.Type == supervisor.EventContainerDelete {
 			containerCount--
 			if containerCount == 0 {
 				syscall.Kill(0, syscall.SIGQUIT)

delete.go

@@ -0,0 +1,52 @@
+package main
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+
+	"github.com/hyperhq/runv/containerd/api/grpc/types"
+	"github.com/hyperhq/runv/supervisor"
+	"github.com/urfave/cli"
+	netcontext "golang.org/x/net/context"
+)
+
+var deleteCommand = cli.Command{
+	Name:      "delete",
+	Usage:     "delete a stopped container",
+	ArgsUsage: `<container-id>`,
+	Action: func(context *cli.Context) error {
+		container := context.Args().First()
+		if container == "" {
+			return cli.NewExitError("container id cannot be empty", -1)
+		}
+
+		containerPath := filepath.Join(context.GlobalString("root"), container)
+
+		if dir, err := os.Stat(containerPath); err != nil || !dir.IsDir() {
+			return fmt.Errorf("container %s does not exist", container)
+		}
+
+		api, err := getClient(filepath.Join(containerPath, "namespace/namespaced.sock"))
+		if err != nil {
+			return fmt.Errorf("failed to get client: %v", err)
+		}
+
+		_, err = api.GetServerVersion(netcontext.Background(), nil)
+		if err != nil {
+			// if we can't connect to the api, runv was killed before it could clean up the stopped containers
+			err := os.RemoveAll(containerPath)
+			if err != nil {
+				return fmt.Errorf("delete stale container %s failed, %v", container, err)
+			}
+			return nil
+		}
+
+		_, err = api.UpdateContainer(netcontext.Background(), &types.UpdateContainerRequest{Id: container, Status: supervisor.ContainerStateDeleted})
+		if err != nil {
+			return fmt.Errorf("delete container %s failed, %v", container, err)
+		}
+
+		return nil
+	},
+}

integration-test/kill_test.go

@@ -35,9 +35,9 @@ func (s *RunVSuite) TestKillKILL(c *check.C) {
 
 	timeout := true
 	for count := 0; count < 10; count++ {
-		out, exitCode := s.runvCommand(c, "list")
+		out, exitCode := s.runvCommand(c, "state", ctrName)
 		c.Assert(exitCode, checker.Equals, 0)
-		if !strings.Contains(out, ctrName) {
+		if !strings.Contains(out, "running") {
 			timeout = false
 			break
 		}

list.go

@@ -9,6 +9,7 @@ import (
 	"text/tabwriter"
 	"time"
 
+	"github.com/hyperhq/runv/supervisor"
 	"github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/urfave/cli"
 )
@@ -110,15 +111,22 @@ func getContainers(context *cli.Context) ([]containerState, error) {
 			if err != nil && !os.IsNotExist(err) {
 				return nil, fmt.Errorf("Stat file %s error: %s", stateFile, err.Error())
 			}
+
 			state, err := loadStateFile(stateFile)
 			if err != nil {
 				return nil, fmt.Errorf("Load state file %s failed: %s", stateFile, err.Error())
 			}
 
+			status := supervisor.ContainerStateStopped // if we can't connect to runv-containerd then the container is stopped
+			if c, err := getContainerApi(context, item.Name()); err == nil {
+				status = c.Status
+			}
+
+			// FIXME: refactor to get container state only via API
 			s = append(s, containerState{
 				ID:             state.ID,
 				InitProcessPid: state.Pid,
-				Status:         "running",
+				Status:         status,
 				Bundle:         state.Bundle,
 				Created:        fi.ModTime(),
 			})

main.go

@@ -162,6 +162,7 @@ func main() {
 		pauseCommand,
 		resumeCommand,
 		containerd.ContainerdCommand,
+		deleteCommand,
 	}
 	if err := app.Run(os.Args); err != nil {
 		fmt.Fprintf(os.Stderr, "%v", err)

state.go

@@ -9,6 +9,7 @@ import (
 	"path/filepath"
 	"time"
 
+	"github.com/hyperhq/runv/supervisor"
 	"github.com/urfave/cli"
 )
 
@@ -79,11 +80,17 @@ func getContainer(context *cli.Context, name string) (*cState, error) {
 		return nil, fmt.Errorf("Load state file %s failed: %s", stateFile, err.Error())
 	}
 
+	status := supervisor.ContainerStateStopped // if we can't connect to runv-containerd then the container is stopped
+	if c, err := getContainerApi(context, name); err == nil {
+		status = c.Status
+	}
+
+	// FIXME: refactor to get container state only via API
 	s := &cState{
 		Version:        state.Version,
 		ID:             state.ID,
 		InitProcessPid: state.Pid,
-		Status:         "running",
+		Status:         status,
 		Bundle:         state.Bundle,
 		Rootfs:         filepath.Join(state.Bundle, "rootfs"),
 		Created:        fi.ModTime(),

supervisor/container.go

@@ -20,8 +20,18 @@ import (
 	"github.com/opencontainers/runtime-spec/specs-go"
 )
 
+// container states
+const (
+	ContainerStateCreating = "creating"
+	ContainerStateCreated  = "created"
+	ContainerStateRunning  = "running"
+	ContainerStateStopped  = "stopped"
+	ContainerStateDeleted  = "deleted" // special state that will lead to deleting the container resources
+)
+
 type Container struct {
 	Id         string
+	Status     string
 	BundlePath string
 	Spec       *specs.Spec
 	Processes  map[string]*Process
@@ -50,6 +60,7 @@ func (c *Container) start(p *Process) error {
 	}
 
 	go func() {
+		c.Status = ContainerStateRunning
 		e := Event{
 			ID:        c.Id,
 			Type:      EventContainerStart,
@@ -58,6 +69,7 @@ func (c *Container) start(p *Process) error {
 		c.ownerPod.sv.Events.notifySubscribers(e)
 
 		exit, err := c.wait(p, res)
+		c.Status = ContainerStateStopped
 		e = Event{
 			ID:        c.Id,
 			Type:      EventExit,
@@ -75,6 +87,7 @@ func (c *Container) start(p *Process) error {
 }
 
 func (c *Container) create() error {
+	c.Status = ContainerStateCreating
 	glog.V(3).Infof("prepare hypervisor info")
 	config := api.ContainerDescriptionFromOCF(c.Id, c.Spec)
 
@@ -182,6 +195,14 @@ func (c *Container) create() error {
 		return err
 	}
 
+	c.Status = ContainerStateCreated
+	e := Event{
+		ID:        c.Id,
+		Type:      EventContainerCreate,
+		Timestamp: time.Now(),
+	}
+	c.ownerPod.sv.Events.notifySubscribers(e)
+
 	return nil
 }
 
@@ -354,6 +375,13 @@ func (c *Container) reap() {
 	utils.Umount(containerRoot)
 	os.RemoveAll(containerRoot)
 	os.RemoveAll(filepath.Join(c.ownerPod.sv.StateDir, c.Id))
+
+	e := Event{
+		ID:        c.Id,
+		Type:      EventContainerDelete,
+		Timestamp: time.Now(),
+	}
+	c.ownerPod.sv.Events.notifySubscribers(e)
 }
 
 func mountToRootfs(m *specs.Mount, rootfs, mountLabel string) error {

supervisor/events.go

@@ -12,9 +12,11 @@ import (
 )
 
 const (
-	EventExit           = "exit"
-	EventContainerStart = "start-container"
-	EventProcessStart   = "start-process"
+	EventExit            = "exit"
+	EventContainerCreate = "create-container"
+	EventContainerStart  = "start-container"
+	EventContainerDelete = "delete-container"
+	EventProcessStart    = "start-process"
 )
 
 var (

supervisor/supervisor.go

@@ -94,6 +94,26 @@ func (sv *Supervisor) StartContainer(container string, spec *specs.Spec) (c *Con
 	return nil, nil, fmt.Errorf("container %s is not found for StartContainer()", container)
 }
 
+func (sv *Supervisor) DeleteContainer(container string) error {
+	glog.Infof("delete container %s", container)
+	sv.Lock()
+	defer sv.Unlock()
+
+	if c, ok := sv.Containers[container]; ok {
+		c.reap()
+		delete(c.ownerPod.Containers, container)
+		delete(sv.Containers, container)
+
+		if len(c.ownerPod.Containers) == 0 {
+			c.ownerPod.reap()
+		}
+
+		return nil
+	}
+
+	return fmt.Errorf("Container %s not found", container)
+}
+
 func (sv *Supervisor) AddProcess(container, processId, stdin, stdout, stderr string, spec *specs.Process) (*Process, error) {
 	sv.Lock()
 	defer sv.Unlock()
@@ -164,14 +184,6 @@ func (sv *Supervisor) reap(container, processId string) {
 				// TODO: kill all the other existing processes in the same container
 			}
 		}
-		if len(c.Processes) == 0 {
-			c.reap()
-			delete(c.ownerPod.Containers, container)
-			delete(sv.Containers, container)
-		}
-		if len(c.ownerPod.Containers) == 0 {
-			c.ownerPod.reap()
-		}
 	}
 }