cgroup: refine the handling of error 'no cpu controller detected' (#5999

)

ref #5953

Signed-off-by: Sen Han <00hnes@gmail.com>

Co-authored-by: lhy1024 <admin@liudos.us>

pkg/cgroup/cgroup_cpu.go

@@ -21,6 +21,8 @@ import (
 	"github.com/pingcap/errors"
 )
 
+var errNoCPUControllerDetected = errors.New("no cpu controller detected")
+
 // Helper function for getCgroupCPU. Root is always "/", except in tests.
 func getCgroupCPUHelper(root string) (CPUUsage, error) {
 	path, err := detectControlPath(filepath.Join(root, procPathCGroup), "cpu,cpuacct")
@@ -30,7 +32,7 @@ func getCgroupCPUHelper(root string) (CPUUsage, error) {
 
 	// No CPU controller detected
 	if path == "" {
-		return CPUUsage{}, errors.New("no cpu controller detected")
+		return CPUUsage{}, errNoCPUControllerDetected
 	}
 
 	mount, ver, err := getCgroupDetails(filepath.Join(root, procPathMountInfo), path, "cpu,cpuacct")

pkg/cgroup/cgroup_cpu_test.go

@@ -12,18 +12,57 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-//go:build linux && enable_flaky_tests
+//go:build linux
 
 package cgroup
 
 import (
+	"fmt"
+	"regexp"
 	"runtime"
+	"strconv"
 	"sync"
+	"syscall"
 	"testing"
 
 	"github.com/stretchr/testify/require"
 )
 
+func checkKernelVersionNewerThan(t *testing.T, major, minor int) bool {
+	u := syscall.Utsname{}
+	err := syscall.Uname(&u)
+	require.NoError(t, err)
+	releaseBs := make([]byte, 0, len(u.Release))
+	for _, v := range u.Release {
+		if v == 0 {
+			break
+		}
+		releaseBs = append(releaseBs, byte(v))
+	}
+	releaseStr := string(releaseBs)
+	t.Log("kernel release string:", releaseStr)
+	versionInfoRE := regexp.MustCompile(`[0-9]+\.[0-9]+\.[0-9]+`)
+	kernelVerion := versionInfoRE.FindAllString(releaseStr, 1)
+	require.Equal(t, 1, len(kernelVerion), fmt.Sprintf("release str is %s", releaseStr))
+	kernelVersionPartRE := regexp.MustCompile(`[0-9]+`)
+	kernelVersionParts := kernelVersionPartRE.FindAllString(kernelVerion[0], -1)
+	require.Equal(t, 3, len(kernelVersionParts), fmt.Sprintf("kernel verion str is %s", kernelVerion[0]))
+	t.Logf("parsed kernel version parts: major %s, minor %s, patch %s",
+		kernelVersionParts[0], kernelVersionParts[1], kernelVersionParts[2])
+	mustConvInt := func(s string) int {
+		i, err := strconv.Atoi(s)
+		require.NoError(t, err, s)
+		return i
+	}
+	versionNewerThanFlag := false
+	if mustConvInt(kernelVersionParts[0]) > major {
+		versionNewerThanFlag = true
+	} else if mustConvInt(kernelVersionParts[0]) == major && mustConvInt(kernelVersionParts[1]) > minor {
+		versionNewerThanFlag = true
+	}
+	return versionNewerThanFlag
+}
+
 func TestGetCgroupCPU(t *testing.T) {
 	exit := make(chan struct{})
 	var wg sync.WaitGroup
@@ -42,9 +81,18 @@ func TestGetCgroupCPU(t *testing.T) {
 		}()
 	}
 	cpu, err := GetCgroupCPU()
-	require.NoError(t, err)
-	require.NotZero(t, cpu.Period)
-	require.Less(t, int64(1), cpu.Period)
+	if err == errNoCPUControllerDetected {
+		// for more information, please refer https://github.com/pingcap/tidb/pull/41347
+		if checkKernelVersionNewerThan(t, 4, 7) {
+			require.NoError(t, err, "linux version > v4.7 and err still happens")
+		} else {
+			t.Logf("the error '%s' is ignored because the kernel is too old", err)
+		}
+	} else {
+		require.NoError(t, err)
+		require.NotZero(t, cpu.Period)
+		require.Less(t, int64(1), cpu.Period)
+	}
 	close(exit)
 	wg.Wait()
 }