koordlet: introduce Accelerators feature gate for GPU related features (

#393) Signed-off-by: Jason Liu <jasonliu747@gmail.com>
koordinator-sh · Jul 25, 2022 · 779ac80 · 779ac80
1 parent 47e7189
commit 779ac80
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 3 deletions.
diff --git a/pkg/features/koordlet_features.go b/pkg/features/koordlet_features.go
@@ -48,6 +48,10 @@ const (
 
 	// CgroupReconcile reconciles qos config for resources like cpu, memory, disk, etc.
 	CgroupReconcile featuregate.Feature = "CgroupReconcile"
+
+	// Accelerators enables GPU related feature in koordlet.
+	// Only Nvidia GPUs are supported as of v0.6.
+	Accelerators featuregate.Feature = "Accelerators"
 )
 
 func init() {
@@ -68,5 +72,6 @@ var (
 		CPUBurst:               {Default: false, PreRelease: featuregate.Alpha},
 		RdtResctrl:             {Default: false, PreRelease: featuregate.Alpha},
 		CgroupReconcile:        {Default: false, PreRelease: featuregate.Alpha},
+		Accelerators:           {Default: false, PreRelease: featuregate.Alpha},
 	}
 )
diff --git a/pkg/koordlet/metricsadvisor/collector.go b/pkg/koordlet/metricsadvisor/collector.go
@@ -129,8 +129,6 @@ func (c *collector) Run(stopCh <-chan struct{}) error {
 	}
 
 	go wait.Until(func() {
-
-		// collect gpu metrics.
 		c.collectGPUUsage()
 		c.collectNodeResUsed()
 		// add sync metaService cache check before collect pod information

diff --git a/pkg/koordlet/metricsadvisor/collector_gpu.go b/pkg/koordlet/metricsadvisor/collector_gpu.go
@@ -24,12 +24,13 @@ import (
 	"time"
 
 	"github.com/NVIDIA/go-nvml/pkg/nvml"
-	"github.com/koordinator-sh/koordinator/pkg/util"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/resource"
 	"k8s.io/klog/v2"
 
+	"github.com/koordinator-sh/koordinator/pkg/features"
 	"github.com/koordinator-sh/koordinator/pkg/koordlet/metriccache"
+	"github.com/koordinator-sh/koordinator/pkg/util"
 )
 
 type GPUDeviceManager interface {
@@ -82,6 +83,9 @@ type device struct {
 
 // initGPUDeviceManager will not retry if init fails,
 func initGPUDeviceManager() GPUDeviceManager {
+	if !features.DefaultKoordletFeatureGate.Enabled(features.Accelerators) {
+		return &dummyDeviceManager{}
+	}
 	if ret := nvml.Init(); ret != nvml.SUCCESS {
 		if ret == nvml.ERROR_LIBRARY_NOT_FOUND {
 			klog.Warning("nvml init failed, library not found")