cgroupv2 support

[ehashman]

Closes #138.

Goals:

• don't get rid of cgroupv1 support
• add support for cgroupv2 for all existing metrics pulled out of cgroups
• update the direct cgroup walking logic, the ctrstats code paths, and the prometheus metrics
  • as of 09/04, this PR updates the cgroup walking logic and ctrstats code paths but not the prometheus metrics (yet)

Design:

Details can be found at #138 (comment)

Summary of changes:

• cgroup tree is now unified, no need to specify separate paths for cpu, cpuacct, memory, etc.
• CPU:
  • cpuacct is gone
  • All cpu metrics can be gathered from cpu.stat file
  • cgroupv2 cpu metrics are now in microseconds, not nanoseconds
• Memory:
  • An accurate measure of current memory usage is available at memory.current (unlike memory.usage_in_bytes on v1 which was an estimate)
  • We can now get all OOM information out of memory.events
  • If we need RSS, we can use anon field out of memory.stat file
• For UMC, we need to update both the cgroup-walking code (this PR) and the ctrstats code that uses the CRI library (possibly a follow-up PR)

[k8s-ci-robot]

Skipping CI for Draft Pull Request.
If you want CI signal for your change, please convert it to an actual PR.
You can still manually trigger a test run with /test all

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: ehashman

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [ehashman]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ehashman]

Taking this out of draft so CI runs.

I was able to get the basic cgroupv2 integration test (copied from the "basic" test) to mostly pass. There was a little bit of weirdness in the output values that I am not sure if I should just update the test case for or if there's an actual problem.

I noticed min.slice on cgroupv1 had a -1 value in the input samples. This did not work with cgroupv2 so I changed those results to 0s, but the test data isn't quite as expected. I'm not sure if there is a special meaning to the -1 CPU usage on cgroupv1.

Otherwise, this should not break any cgroupv1 functionality. I will work on ctrstats next but that should be more of a trivial change, so if folks want to take a look at this now, it should be ready for initial review.

I will also edit my first comment with the dictionary for updating v1 to v2.

[ehashman]

/retest-required

flake #126

[ehashman]

/triage accepted

[ehashman]

/retest-required

flake #126

[ehashman]

/retest-required