Skip to content

Fleet refresh: name budget in the unpacker, a doc nit, lint job speed (#196 follow-up) #209

Closed

Description

@scuttletoby

Follow-ups from the reviews of #196 (runner template refresh). None of them blocks; #196 is merged.

  1. Name total in the unpacker (.github/scripts/fork-fleet-launch.sh, the read loop near MAX_NAME / MAX_ENTRIES). Each name is capped at 4096 bytes, but up to 100000 entries are kept. 100000 symlinks with names and targets near the cap come from a ~3 MB download and peak at about 1.25 GB. Under the unit, MemoryMax=1G kills it and nothing is written (fails closed). By hand (host step 7), it briefly takes that much memory. Fix: keep a running total of len(m.name) + len(m.linkname) and refuse past a budget (e.g. 64 MiB), or lower MAX_ENTRIES. Red test: 100000 such entries under a 1 GiB address-space limit give MemoryError, not "refused:".
  2. Doc nit (.github/scripts/fork-fleet-refresh.service, header): it still says "a second refresh waits for the first (refresh.lock)". The lock is now the template's folder.
  3. Lint job speed (.github/workflows/fork-tests.yml, "Workflow lint and upstream drift"). It takes 4 to 6 minutes on main. Its cap was 5 minutes, which cancelled PRs that add tests; fork fleet: runner template refresh + daily timer (D7) #196 and fork fleet: desk link, desktop side (D3) #205 raised it to 10. Split the job or make it faster, so the cap is not the thing that keeps it green.

State question for any fix: what happens when this runs again, stops half-way, or meets existing, old or empty state?

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions