enhance: add enable MSR flag to allow MSR model (#537)

Author: jichenjc

cmd/exporter.go

@@ -59,6 +59,7 @@ var (
 	cpuProfile                   = flag.String("cpuprofile", "", "dump cpu profile to a file")
 	memProfile                   = flag.String("memprofile", "", "dump mem profile to a file")
 	profileDuration              = flag.Int("profile-duration", 60, "duration in seconds")
+	enabledMSR                   = flag.Bool("enable-msr", false, "whether MSR is allowed to obtain energy data")
 )
 
 func healthProbe(w http.ResponseWriter, req *http.Request) {
@@ -151,6 +152,7 @@ func main() {
 	config.SetEnabledEBPFCgroupID(*enabledEBPFCgroupID)
 	config.SetEnabledHardwareCounterMetrics(*exposeHardwareCounterMetrics)
 	config.SetEnabledGPU(*enableGPU)
+	config.EnabledMSR = *enabledMSR
 
 	cgroup.SetSliceHandler()
 

pkg/bpfassets/attacher/bcc_attacher.go

@@ -21,10 +21,11 @@ package attacher
 
 import (
 	"fmt"
-	"golang.org/x/sys/unix"
 	"runtime"
 	"strconv"
 
+	"golang.org/x/sys/unix"
+
 	assets "github.com/sustainable-computing-io/kepler/pkg/bpfassets"
 	"github.com/sustainable-computing-io/kepler/pkg/config"
 

pkg/config/config.go

@@ -58,6 +58,8 @@ const (
 var (
 	modelServerService = fmt.Sprintf("kepler-model-server.%s.svc.cluster.local", KeplerNamespace)
 
+	EnabledMSR = false
+
 	KeplerNamespace              = getConfig("KELPER_NAMESPACE", defaultNamespace)
 	EnabledEBPFCgroupID          = getBoolConfig("ENABLE_EBPF_CGROUPID", true)
 	ExposeHardwareCounterMetrics = getBoolConfig("EXPOSE_HW_COUNTER_METRICS", true)

pkg/power/accelerator/gpu.go

@@ -21,6 +21,7 @@ package accelerator
 
 import (
 	"fmt"
+
 	accelerator_source "github.com/sustainable-computing-io/kepler/pkg/power/accelerator/source"
 	"k8s.io/klog/v2"
 )

pkg/power/components/power.go

@@ -19,6 +19,7 @@ package components
 import (
 	"k8s.io/klog/v2"
 
+	"github.com/sustainable-computing-io/kepler/pkg/config"
 	"github.com/sustainable-computing-io/kepler/pkg/power/components/source"
 )
 
@@ -45,15 +46,14 @@ var (
 	msrImpl                          = &source.PowerMSR{}
 	apmXgeneSysfsImpl                = &source.ApmXgeneSysfs{}
 	powerImpl         powerInterface = sysfsImpl
-	useMSR                           = false // it looks MSR on kvm or hyper-v is not working
 )
 
 func InitPowerImpl() {
 	if sysfsImpl.IsSystemCollectionSupported() /*&& false*/ {
 		klog.V(1).Infoln("use sysfs to obtain power")
 		powerImpl = sysfsImpl
 	} else {
-		if msrImpl.IsSystemCollectionSupported() && useMSR {
+		if msrImpl.IsSystemCollectionSupported() && config.EnabledMSR {
 			klog.V(1).Infoln("use MSR to obtain power")
 			powerImpl = msrImpl
 		} else {

pkg/power/components/source/rapl_msr_util.go

@@ -31,6 +31,8 @@ import (
 	"strings"
 	"syscall"
 	"unsafe"
+
+	"k8s.io/klog/v2"
 )
 
 const (
@@ -142,16 +144,19 @@ func ReadMSR(packageID int, msr int64) (uint64, error) {
 
 func InitUnits() error {
 	if err := mapPackageAndCore(); err != nil {
+		klog.V(1).Info(err)
 		return err
 	}
 	if err := OpenAllMSR(); err != nil {
+		klog.V(1).Info(err)
 		return err
 	}
 	cpuEnergyUnits = make([]float64, len(packageList))
 	dramEnergyUnits = make([]float64, len(packageList))
 	for i := 0; i < len(packageList); {
 		result, err := ReadMSR(i, msrRaplPowerUnit)
 		if err != nil {
+			klog.V(1).Info(err)
 			return fmt.Errorf("failed to read power unit: %v", err)
 		}
 		powerUnits = math.Pow(0.5, float64((result & 0xf)))