fix bugs to trim container id in pod spec

pkg/metaserver/agent/metric/malachite/client/client.go

@@ -45,9 +45,11 @@ const (
 )
 
 type MalachiteClient struct {
+	// those fields are for testing
 	sync.RWMutex
+	urls             map[string]string
+	relativePathFunc *func(podUID, containerId string) (string, error)
 
-	urls    map[string]string
 	fetcher pod.PodFetcher
 }
 

pkg/metaserver/agent/metric/malachite/client/client_pod.go

@@ -25,6 +25,7 @@ import (
 	"github.com/kubewharf/katalyst-core/pkg/metaserver/agent/metric/malachite/types"
 	cgroupcm "github.com/kubewharf/katalyst-core/pkg/util/cgroup/common"
 	"github.com/kubewharf/katalyst-core/pkg/util/general"
+	"github.com/kubewharf/katalyst-core/pkg/util/native"
 )
 
 func (c *MalachiteClient) GetAllPodContainersStats(ctx context.Context) (map[string]map[string]*types.MalachiteCgroupInfo, error) {
@@ -55,18 +56,31 @@ func (c *MalachiteClient) GetPodStats(ctx context.Context, podUID string) (map[s
 
 	containersStats := make(map[string]*types.MalachiteCgroupInfo)
 	for _, containerStatus := range pod.Status.ContainerStatuses {
-		stats, err := c.GetPodContainerStats(podUID, containerStatus.ContainerID)
+		containerID := native.TrimContainerIDPrefix(containerStatus.ContainerID)
+		stats, err := c.GetPodContainerStats(podUID, containerID)
 		if err != nil {
 			general.Errorf("GetPodStats err %v", err)
 			continue
 		}
-		containersStats[containerStatus.ContainerID] = stats
+		containersStats[containerID] = stats
 	}
 	return containersStats, nil
 }
 
 func (c *MalachiteClient) GetPodContainerStats(podUID, containerID string) (*types.MalachiteCgroupInfo, error) {
-	cgroupPath := cgroupcm.GetContainerRelativeCgroupPath(podUID, containerID)
+	var cgroupPath string
+	var err error
+
+	// if relativePathFunc has been set, we should use it
+	if c.relativePathFunc != nil {
+		cgroupPath, err = (*c.relativePathFunc)(podUID, containerID)
+	} else {
+		cgroupPath, err = cgroupcm.GetContainerRelativeCgroupPath(podUID, containerID)
+	}
+	if err != nil {
+		return nil, fmt.Errorf("GetPodContainerStats %s/%v get-relative-path err %v", podUID, containerID, err)
+	}
+
 	containersStats, err := c.GetCgroupStats(cgroupPath)
 	if err != nil {
 		return nil, fmt.Errorf("GetPodContainerStats %s/%v get-status %v err %v", podUID, containerID, cgroupPath, err)

pkg/metaserver/agent/metric/malachite/client/client_pod_test.go

@@ -19,8 +19,10 @@ package client
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"net/http"
 	"net/http/httptest"
+	"path"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -30,11 +32,10 @@ import (
 
 	"github.com/kubewharf/katalyst-core/pkg/metaserver/agent/metric/malachite/types"
 	"github.com/kubewharf/katalyst-core/pkg/metaserver/agent/pod"
+	"github.com/kubewharf/katalyst-core/pkg/util/cgroup/common"
 )
 
 func TestGetPodContainerStats(t *testing.T) {
-	t.Parallel()
-
 	cgroupData := map[string]*types.MalachiteCgroupResponse{
 		"podp-uid1/p1-c-uid1": {
 			Status: 0,
@@ -96,11 +97,11 @@ func TestGetPodContainerStats(t *testing.T) {
 				ContainerStatuses: []v1.ContainerStatus{
 					{
 						Name:        "p1-c-name1",
-						ContainerID: "p1-c-uid1",
+						ContainerID: "docker://p1-c-uid1",
 					},
 					{
 						Name:        "p1-c-name2",
-						ContainerID: "p1-c-uid2",
+						ContainerID: "containerd://p1-c-uid2",
 					},
 				},
 			},
@@ -128,7 +129,7 @@ func TestGetPodContainerStats(t *testing.T) {
 				ContainerStatuses: []v1.ContainerStatus{
 					{
 						Name:        "p3-c-name1",
-						ContainerID: "p3-c-uid1",
+						ContainerID: "containerd://p3-c-uid1",
 					},
 				},
 			},
@@ -142,7 +143,7 @@ func TestGetPodContainerStats(t *testing.T) {
 				ContainerStatuses: []v1.ContainerStatus{
 					{
 						Name:        "p4-c-name1",
-						ContainerID: "p4-c-uid1",
+						ContainerID: "containerd://p4-c-uid1",
 					},
 				},
 			},
@@ -157,6 +158,15 @@ func TestGetPodContainerStats(t *testing.T) {
 
 	stats, err := malachiteClient.GetAllPodContainersStats(context.Background())
 	assert.NoError(t, err)
+	assert.Equal(t, 0, len(stats))
+
+	relativePathFunc := func(podUID, containerId string) (string, error) {
+		return path.Join(fmt.Sprintf("%s%s", common.PodCgroupPathPrefix, podUID), containerId), nil
+	}
+	malachiteClient.relativePathFunc = &relativePathFunc
+
+	stats, err = malachiteClient.GetAllPodContainersStats(context.Background())
+	assert.NoError(t, err)
 	assert.Equal(t, 3, len(stats))
 
 	assert.NotNil(t, stats["p-uid1"]["p1-c-uid1"].V1)

pkg/util/cgroup/common/path.go

@@ -113,6 +113,24 @@ func GetKubernetesAnyExistAbsCgroupPath(subsys, suffix string) (string, error) {
 	return "", fmt.Errorf("failed to find absolute path of suffix: %s, error: %v", suffix, utilerrors.NewAggregate(errs))
 }
 
+// GetKubernetesAnyExistRelativeCgroupPath returns any relative cgroup path that exists for kubernetes
+func GetKubernetesAnyExistRelativeCgroupPath(suffix string) (string, error) {
+	var errs []error
+
+	k8sCgroupPathLock.RLock()
+	defer k8sCgroupPathLock.RUnlock()
+
+	for _, cgPath := range k8sCgroupPathList.List() {
+		relativePath := path.Join(cgPath, suffix)
+		p := GetKubernetesAbsCgroupPath(DefaultSelectedSubsys, relativePath)
+		if general.IsPathExists(p) {
+			return relativePath, nil
+		}
+	}
+
+	return "", fmt.Errorf("failed to find relative path of suffix: %s, error: %v", suffix, utilerrors.NewAggregate(errs))
+}
+
 // GetPodAbsCgroupPath returns absolute cgroup path for pod level
 func GetPodAbsCgroupPath(subsys, podUID string) (string, error) {
 	return GetKubernetesAnyExistAbsCgroupPath(subsys, podUID)
@@ -124,8 +142,8 @@ func GetContainerAbsCgroupPath(subsys, podUID, containerId string) (string, erro
 }
 
 // GetContainerRelativeCgroupPath returns relative cgroup path for container level
-func GetContainerRelativeCgroupPath(podUID, containerId string) string {
-	return path.Join(fmt.Sprintf("%s%s", PodCgroupPathPrefix, podUID), containerId)
+func GetContainerRelativeCgroupPath(podUID, containerId string) (string, error) {
+	return GetKubernetesAnyExistRelativeCgroupPath(path.Join(fmt.Sprintf("%s%s", PodCgroupPathPrefix, podUID), containerId))
 }
 
 func IsContainerCgroupExist(podUID, containerID string) (bool, error) {