some pod energy not reported about kepler

[rootfs]

I suspect kindnetd is in a different pid namespace (i.e. pid 18005, seen by kepler ebpf may be different when kepler userspace runtime reads from its /proc namespace).

[jichenjc]

I suspect kindnetd is in a different pid namespace (i.e. pid 18005, seen by kepler ebpf may be different when kepler userspace runtime reads from its /proc namespace).

ok, I might need some guide on how the pod energy is actually calculated
I know we have ebpf, we have cgroup data, what's the formula to do the calcuation?
e.g some record should be in cgroup data first then find corresponding record in ebpf?

@rootfs

[rootfs]

the current formula uses ratio based method. If enabled, there is also a model based dynamic power estimate

You'll find the energy calculation is done here on the pods

[rootfs]

some record should be in cgroup data first then find corresponding record in ebpf?

ebpf gets the info associated with the running process: cgroup id, pid, cpu runtime, perf counters, etc. This info is coupled with userspace info collected from sysfs and feed into the energy estimate formula to get the pod energy.

[jichenjc]

indeed, previously I used
BM => VM => docker (kind) => k8s
now I switched to
BM => VM => k8s

it's much better now, will check all the links you provided, thanks