Merge pull request #698 from liubin/feature/146-add-cgroup-v2-for-agent

agent: add cgroup v2 support

src/runtime/containerd-shim-v2/wait.go

@@ -7,6 +7,7 @@ package containerdshim
 
 import (
 	"context"
+	"os"
 	"path"
 	"time"
 
@@ -15,6 +16,8 @@ import (
 	"github.com/containerd/containerd/mount"
 	"github.com/sirupsen/logrus"
 	"google.golang.org/grpc/codes"
+
+	"github.com/kata-containers/kata-containers/src/runtime/virtcontainers/pkg/oci"
 )
 
 func wait(s *service, c *container, execID string) (int32, error) {
@@ -146,12 +149,26 @@ func watchOOMEvents(ctx context.Context, s *service) {
 				// If the GetOOMEvent call is not implemented, then the agent is most likely an older version,
 				// stop attempting to get OOM events.
 				// for rust agent, the response code is not found
-				if isGRPCErrorCode(codes.NotFound, err) {
+				if isGRPCErrorCode(codes.NotFound, err) || err.Error() == "Dead agent" {
 					return
 				}
 				continue
 			}
 
+			// write oom file for CRI-O
+			if c, ok := s.containers[containerID]; ok && oci.IsCRIOContainerManager(c.spec) {
+				oomPath := path.Join(c.bundle, "oom")
+				shimLog.Infof("write oom file to notify CRI-O: %s", oomPath)
+
+				f, err := os.OpenFile(oomPath, os.O_CREATE, 0666)
+				if err != nil {
+					shimLog.WithError(err).Warnf("failed to write oom file %s", oomPath)
+				} else {
+					f.Close()
+				}
+			}
+
+			// publish event for containerd
 			s.send(&events.TaskOOM{
 				ContainerID: containerID,
 			})

src/runtime/virtcontainers/pkg/oci/utils.go

@@ -1036,3 +1036,13 @@ func GetOCIConfig(status vc.ContainerStatus) (specs.Spec, error) {
 
 	return *status.Spec, nil
 }
+
+// IsCRIOContainerManager check if a Pod is created from CRI-O
+func IsCRIOContainerManager(spec *specs.Spec) bool {
+	if val, ok := spec.Annotations[crioAnnotations.ContainerType]; ok {
+		if val == crioAnnotations.ContainerTypeSandbox || val == crioAnnotations.ContainerTypeContainer {
+			return true
+		}
+	}
+	return false
+}

src/runtime/virtcontainers/pkg/oci/utils_test.go

@@ -17,6 +17,7 @@ import (
 	"testing"
 
 	"github.com/cri-o/cri-o/pkg/annotations"
+	crioAnnotations "github.com/cri-o/cri-o/pkg/annotations"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/stretchr/testify/assert"
 	"golang.org/x/sys/unix"
@@ -866,3 +867,34 @@ func TestAddRuntimeAnnotations(t *testing.T) {
 	assert.Equal(config.NetworkConfig.DisableNewNetNs, true)
 	assert.Equal(config.NetworkConfig.InterworkingModel, vc.NetXConnectMacVtapModel)
 }
+
+func TestIsCRIOContainerManager(t *testing.T) {
+	assert := assert.New(t)
+
+	testCases := []struct {
+		annotations map[string]string
+		result      bool
+	}{
+		{
+			annotations: map[string]string{crioAnnotations.ContainerType: "abc"},
+			result:      false,
+		},
+		{
+			annotations: map[string]string{crioAnnotations.ContainerType: crioAnnotations.ContainerTypeSandbox},
+			result:      true,
+		},
+		{
+			annotations: map[string]string{crioAnnotations.ContainerType: crioAnnotations.ContainerTypeContainer},
+			result:      true,
+		},
+	}
+
+	for i := range testCases {
+		tc := testCases[i]
+		ocispec := specs.Spec{
+			Annotations: tc.annotations,
+		}
+		result := IsCRIOContainerManager(&ocispec)
+		assert.Equal(tc.result, result, "test case %d", (i + 1))
+	}
+}

src/runtime/virtcontainers/sandbox.go

@@ -1203,10 +1203,10 @@ func (s *Sandbox) CreateContainer(contConfig ContainerConfig) (VCContainer, erro
 		// Rollback if error happens.
 		if err != nil {
 			logger := s.Logger().WithFields(logrus.Fields{"container-id": c.id, "sandox-id": s.id, "rollback": true})
-			logger.Warning("Cleaning up partially created container")
+			logger.WithError(err).Error("Cleaning up partially created container")
 
 			if err2 := c.stop(true); err2 != nil {
-				logger.WithError(err2).Warning("Could not delete container")
+				logger.WithError(err2).Error("Could not delete container")
 			}
 
 			logger.Debug("Removing stopped container from sandbox store")
@@ -1894,9 +1894,11 @@ func (s *Sandbox) updateResources() error {
 		return err
 	}
 
+	s.Logger().Debugf("Request to hypervisor to update oldCPUs/newCPUs: %d/%d", oldCPUs, newCPUs)
 	// If the CPUs were increased, ask agent to online them
 	if oldCPUs < newCPUs {
 		vcpusAdded := newCPUs - oldCPUs
+		s.Logger().Debugf("Request to onlineCPUMem with %d CPUs", vcpusAdded)
 		if err := s.agent.onlineCPUMem(vcpusAdded, true); err != nil {
 			return err
 		}