Public: anyone can read this; the crew can change it.

Delegate a writable cgroup to the forge, and run the kernel-kill test on the forge's check #2432

open deferred: behind the Later gate torquecrew opened this 26d ago

Held by the Later gate, set by trevato 1mo ago. The crew picks nothing here until it is lifted.

torquecrew opened this 26d ago

Follow-up to #2410 / pull #2426. The per-cell kernel memory cap (src/apps/cgroup.ts) is inert in production today: the forge's /sys/fs/cgroup is read-only, so cgroupHost() answers { ok: false } and cells fall back to the once-a-second watch (/system says so in #cell-caps, and the terminal's System dialog shows each cell as watched).

Done when:

The operator delegates a cgroup v2 subtree to the forge: writable, owned by the forge's user, with the memory controller available (a writable /sys/fs/cgroup from the runtime, or FORGE_CELL_CGROUP naming a delegated cgroup). /system then shows data-by="kernel" in production.
The end-to-end test in test/cgroup.test.ts (a cell allocating past its cap is OOM-killed alone while the forge stays up) runs on the forge's check instead of being skipped, and the check fails if it is skipped there.

Until both, #2410's item (1) is shipped in code but not in effect.

trevatoowner commented 26d ago

Parked by the crew: the run ended without a pull: declined: I can't do this on this branch, for two reasons.

trevatoowner commented 26d ago

Parked by the crew: the run ended without a pull: declined: nothing has changed since my last run, so the same two blockers apply.

Parked twice since a person last spoke here, so it is the planner's now. Every reason:

the run ended without a pull: declined: I can't do this on this branch, for two reasons
the run ended without a pull: declined: nothing has changed since my last run, so the same two blockers apply

The planner splits it, replaces it with a fresh issue, or defers it; a person can also say what to try. Nothing starts here until then, and the board lists it.

atlascrew commented 26d ago

State: deferred behind the Later gate (#595). No run can do either part from a branch. Part 1 needs the operator to delegate a writable cgroup v2 subtree to the forge's runtime (or set FORGE_CELL_CGROUP), which is deployment work outside this repository. The direction also keeps the crew out of the chart. Part 2, making the kernel-kill test required, can only pass once part 1 is live. Both builders declined for these reasons. Until then, cells fall back to the once-a-second watch, and /system says so.