Merge pull request #1598 from Random-Liu/fix-load-task

Fix task load.

linux/task.go

@@ -5,6 +5,7 @@ package linux
 import (
 	"context"
 
+	"github.com/pkg/errors"
 	"google.golang.org/grpc"
 
 	"github.com/containerd/cgroups"
@@ -32,7 +33,8 @@ func newTask(id, namespace string, pid int, shim *client.Client, monitor runtime
 		cg  cgroups.Cgroup
 	)
 	if pid > 0 {
-		if cg, err = cgroups.Load(cgroups.V1, cgroups.PidPath(pid)); err != nil {
+		cg, err = cgroups.Load(cgroups.V1, cgroups.PidPath(pid))
+		if err != nil && err != cgroups.ErrCgroupDeleted {
 			return nil, err
 		}
 	}
@@ -253,6 +255,9 @@ func (t *Task) Process(ctx context.Context, id string) (runtime.Process, error)
 
 // Metrics returns runtime specific system level metric information for the task
 func (t *Task) Metrics(ctx context.Context) (interface{}, error) {
+	if t.cg == nil {
+		return nil, errors.Wrap(errdefs.ErrNotFound, "cgroup does not exist")
+	}
 	stats, err := t.cg.Stat(cgroups.IgnoreNotExist)
 	if err != nil {
 		return nil, err
@@ -261,8 +266,11 @@ func (t *Task) Metrics(ctx context.Context) (interface{}, error) {
 }
 
 // Cgroup returns the underlying cgroup for a linux task
-func (t *Task) Cgroup() cgroups.Cgroup {
-	return t.cg
+func (t *Task) Cgroup() (cgroups.Cgroup, error) {
+	if t.cg == nil {
+		return nil, errors.Wrap(errdefs.ErrNotFound, "cgroup does not exist")
+	}
+	return t.cg, nil
 }
 
 // Wait for the task to exit returning the status and timestamp

metrics/cgroups/cgroups.go

@@ -62,10 +62,14 @@ type cgroupsMonitor struct {
 func (m *cgroupsMonitor) Monitor(c runtime.Task) error {
 	info := c.Info()
 	t := c.(*linux.Task)
-	if err := m.collector.Add(info.ID, info.Namespace, t.Cgroup()); err != nil {
+	cg, err := t.Cgroup()
+	if err != nil {
 		return err
 	}
-	return m.oom.Add(info.ID, info.Namespace, t.Cgroup(), m.trigger)
+	if err := m.collector.Add(info.ID, info.Namespace, cg); err != nil {
+		return err
+	}
+	return m.oom.Add(info.ID, info.Namespace, cg, m.trigger)
 }
 
 func (m *cgroupsMonitor) Stop(c runtime.Task) error {

services/tasks/service.go

@@ -514,7 +514,9 @@ func getTasksMetrics(ctx context.Context, filter filters.Filter, tasks []runtime
 		collected := time.Now()
 		metrics, err := tk.Metrics(ctx)
 		if err != nil {
+			if !errdefs.IsNotFound(err) {
 				log.G(ctx).WithError(err).Errorf("collecting metrics for %s", tk.ID())
+			}
 			continue
 		}
 		data, err := typeurl.MarshalAny(metrics)

vendor.conf

@@ -1,7 +1,7 @@
 github.com/coreos/go-systemd 48702e0da86bd25e76cfef347e2adeb434a0d0a6
 github.com/containerd/go-runc b3c048c028ddd789c6f9510c597f8b9c62f25359
 github.com/containerd/console 84eeaae905fa414d03e07bcd6c8d3f19e7cf180e
-github.com/containerd/cgroups 5933ab4dc4f7caa3a73a1dc141bd11f42b5c9163
+github.com/containerd/cgroups 9c238e632e80d94f71a067c3deb9b34b1886ef18
 github.com/containerd/typeurl f6943554a7e7e88b3c14aad190bf05932da84788
 github.com/docker/go-metrics 8fd5772bf1584597834c6f7961a530f06cbfbb87
 github.com/docker/go-events 9461782956ad83b30282bf90e31fa6a70c255ba9

vendor/github.com/containerd/cgroups/cgroup.go

@@ -10,6 +10,7 @@ import (
 	"sync"
 
 	specs "github.com/opencontainers/runtime-spec/specs-go"
+	"github.com/pkg/errors"
 )
 
 // New returns a new control via the cgroup cgroups interface
@@ -39,6 +40,9 @@ func Load(hierarchy Hierarchy, path Path) (Cgroup, error) {
 	for _, s := range pathers(subsystems) {
 		p, err := path(s.Name())
 		if err != nil {
+			if os.IsNotExist(errors.Cause(err)) {
+				return nil, ErrCgroupDeleted
+			}
 			return nil, err
 		}
 		if _, err := os.Lstat(s.Path(p)); err != nil {