Metrics , Metrics , Metrics

[aojea]

We need to have good visibility specially about the latency of the operations, since the time to attach interfaces will impact the pod readiness.

Adding new metrics is really simple, see for reference #89

[aojea]

/assign @ngcxy

[gauravkghildiyal]

@aman0408 -- Can you look into this? Few things that come to my mind that might be good starting points

For some context, the kind of metrics we'll want to instrument are:

• Count of success/failure (success failures indicated through metric labels), latency distribution for:
  • RunPodSandbox:

    dranet/pkg/driver/nri_hooks.go

    Line 91 in 3b38ea6

    func (np *NetworkDriver) RunPodSandbox(ctx context.Context, pod *api.PodSandbox) error {

  • CreateContainer:

    dranet/pkg/driver/nri_hooks.go

    Line 63 in 3b38ea6

    func (np *NetworkDriver) CreateContainer(_ context.Context, pod *api.PodSandbox, ctr *api.Container) (*api.ContainerAdjustment, []*api.ContainerUpdate, error) {

  • StopPodSandbox:

    dranet/pkg/driver/nri_hooks.go

    Line 231 in 3b38ea6

    func (np *NetworkDriver) StopPodSandbox(ctx context.Context, pod *api.PodSandbox) error {

  • PrepareResourceClaim:

    dranet/pkg/driver/dra_hooks.go

    Line 79 in 3b38ea6

    func (np *NetworkDriver) PrepareResourceClaims(ctx context.Context, claims []*resourceapi.ResourceClaim) (map[types.UID]kubeletplugin.PrepareResult, error) {

  • UnprepareResourceClaim:

    dranet/pkg/driver/dra_hooks.go

    Line 323 in 3b38ea6

    func (np *NetworkDriver) UnprepareResourceClaims(ctx context.Context, claims []kubeletplugin.NamespacedObject) (map[types.UID]error, error) {

• Number of devices published, which had RDMA, which did not (maybe have a generic feature label and RDMA as the first feature value).
  • Implementation can be done within PublishResources:

    dranet/pkg/driver/dra_hooks.go

    Line 54 in 3b38ea6

    func (np *NetworkDriver) PublishResources(ctx context.Context) {

  • Use the dra.net/rdma attribute to determine the feature