[m15] containerstdout: fd-hygiene lsof golden missing — MILESTONES claim unsubstantiated

[trilamsr]

Summary

`MILESTONES.md` line 372 (M15 §File-handle hygiene) is checked ☑:

File-handle hygiene: ≤2× pod-count open fds steady-state; closed within 30s of Pod `Terminated`; verified by `lsof` golden.

The lsof golden test does not exist. `grep -r lsof components/receivers/containerstdout/` returns nothing. `TestFailure_FdHygieneAfterPodTermination` (failure_modes_test.go) is still a Phase-14-deferred SKIP.

Risk

Tailer leaks open fds on pod-deletion races (e.g. event handler executes before file close completes). Without a measured assertion, regression goes silent until kubelet inode pressure surfaces it.

Fix

Convert `TestFailure_FdHygieneAfterPodTermination` to a falsifying test:

• Snapshot `/proc/self/fd` count before pod-discovery
• Create N pods via fake informer, write log files, wait for tailers
• Delete pods, wait ≤30s
• Assert fd count returns to baseline + small slack
• Use `unix.Readdirnames("/proc/self/fd")` or shell out to `lsof -p $$`

Linux-only; gate with build tag if macOS dev requires.

Affected files

• `components/receivers/containerstdout/failure_modes_test.go` (currently SKIP)
• New `fd_hygiene_linux_test.go` (build tag)

Why load-bearing

MILESTONES.md ☑ misrepresents fd-leak posture. Either implement or flip to ☐.

Origin

PR #158 self-review.

[trilamsr]

Outdated by RFC-0013 distribution-first pivot (docs/rfcs/0013-distro-first-pivot.md §2 adoption matrix). containerstdout receiver replaced by upstream filelogreceiver + container stanza + file_storage extension; cursor persistence, fd hygiene, goleak, and failure-mode coverage all become upstream concerns. If pilot deployment surfaces before v0.2.0 cut, reopen with re-evaluation trigger.