mantle/system/nproc: account for page cache in cgroup available memory

[dustymabe]

The cgroup available memory calculation used memory.current (total
cgroup usage) directly, which includes page cache (file-backed memory).
Since page cache is reclaimable by the kernel under memory pressure, it
should not count as unavailable. This caused GetCurrentMemAvailableMiB()
to significantly underestimate available memory, making QEMU instance
scheduling overly conservative.

Read the "file" field from /sys/fs/cgroup/memory.stat, which reports
the page cache size in bytes, and subtract it from current usage before
computing available memory. The effective formula becomes:

available = limit - (current - page_cache)

This mirrors how /proc/meminfo computes MemAvailable by considering
reclaimable caches.

A new helper getCgroupMemoryStatField() is added for parsing individual
fields from memory.stat, returning 0 gracefully if the file or field is
absent.

Written-by: <anthropic/claude-opus-4.6>

---

Also a second commit in there to warn when any test initially gets delayed due to memory availability.

[gemini-code-assist]

Code Review

The pull request effectively addresses the issue of cgroup available memory calculation by accounting for page cache, which significantly improves the accuracy of GetCurrentMemAvailableMiB(). The new getCgroupMemoryStatField helper function is a well-encapsulated addition for parsing cgroup memory statistics. The changes in harness.go to introduce conditional warning logs for memory reservation waits also enhance the clarity and debuggability of the test harness.

[prestist]

 Running [/home/runner/golangci-lint-1.64.4-linux-amd64/golangci-lint run --out-format=github-actions --timeout=5m] in [] ...
  level=warning msg="[config_reader] The output format `github-actions` is deprecated, please use `colored-line-number`"
  Error: can't load config: the Go language version (go1.24) used to build golangci-lint is lower than the targeted Go version (1.25.0)
  Failed executing command with error: can't load config: the Go language version (go1.24) used to build golangci-lint is lower than the targeted Go version (1.25.0)

wait... hmm I thought that the golang-lint was managed by repo-templates for this repo. Seems to be out of sync with the go.mod build version.

[dustymabe]

wait... hmm I thought that the golang-lint was managed by repo-templates for this repo

it will be soon. I'm working to get a few PRs merged before I pivot back to #4468 where there will be a lot of cleanups done for golangci-lint.

I don't want to do those cleanups yet because I want to get #3989 merged before I do all the new golangci lint cleanups because I want to be able to backport to older branches with less pain.

I just dropped the existing linters for now in e117447 and rebased this PR.

[prestist]

LGTM - the page cache fix makes sense. curious though, how bad was the underestimation? was the page cache eating up a big chunk of memory.current and causing tests to sit around waiting for memory?

[dustymabe]

was the page cache eating up a big chunk of memory.current and causing tests to sit around waiting for memory?

yeah. basically there were processes (like kola itself) that ate up a bunch of memory (virt) and it never got reclaimed. The RSS wasn't high, but it was all just sitting in page cache I think.

Look at the logs from the x86_64 kola run in https://jenkins-fedora-coreos-pipeline.apps.ocp.fedoraproject.org/blue/organizations/jenkins/bump-lockfile/detail/bump-lockfile/567/pipeline/542 and you can see it

[dustymabe]

ok i had to adjust the strategy here to use inactive_file instead of file. incative_file is page cache that's currently not being used (i.e. used page cache can't actually be evicted so the earlier code was overestimating how many byte we could subtract from the reported current usage).

luckily CI on the previous version was failing and made me look into the calculation more.

[prestist]

LGTM