Repository navigation
fleet-schedule-registry: systemd timer rows drop the OnCalendar time component — every-minute/hourly timers read as daily at midnight #5313
Description
Finding (from PR review)
Codex (chatgpt-codex-connector[bot]) raised this P2 finding on merged PR #5166 and it was never fixed or replied to:
- PR: fleet: unified schedule registry — Muse cron + OpenCode + Jules (#4988) #5166 — "fleet: unified schedule registry — Muse cron + OpenCode + Jules (implement: unified fleet schedule registry — Muse cron + OpenCode timers + Jules schedules in one record (from #2024) #4988)"
- Comment: fleet: unified schedule registry — Muse cron + OpenCode + Jules (#4988) #5166 (comment) (2026-11-03T21:43:42Z, on
doc/fleet-schedule-registry.md:243) - Essence: "Preserve the time component of systemd calendar schedules. The calendar-based rows omit the time component, changing the documented cadence rather than merely abbreviating it. For example,
laptop-autocommitis configured with*-*-* *:*:00innixos/home/laptop-autocommit.nix:51(every minute), but these rows show*-*-*, whichsystemd-analyze calendarnormalizes to*-*-* 00:00:00(daily at midnight). The same loss affects the hourly worktree reaper and 15-minute quota gatherer elsewhere in this table. Since this registry is explicitly an input to the placement audit, these entries misrepresent execution frequency and workload. Preserve the complete calendar expressions when collecting the timer specs."
No-response evidence
- The inline thread contains only the Codex comment; no reply followed it (reviewThreads check 2026-11-04: thread unresolved, sole author chatgpt-codex-connector).
- PR fleet: unified schedule registry — Muse cron + OpenCode + Jules (#4988) #5166 merged at 2026-11-03T21:44:28Z, under a minute after the review was submitted — no fix commit could have followed the comment on the PR, and the defect is present in the merged result on main.
Still live on current main (verified 2026-11-04 by gh api raw read)
doc/fleet-schedule-registry.md, section 2b (laptop systemd user timers):
- Lines 225–261: essentially every timer row records its cadence as
`OnCalendar=*-*-*`with no time component (e.g. line 240laptop-autocommit-hatch-personal.timer, line 254quota-gatherer.timer, line 261worktree-reaper.timer). - The section header claims "Cadence is the unit's own
OnCalendar/OnUnitActiveUSecspec, read live fromsystemctl --user show" — the recorded specs are truncated, not the units' own.
Meanwhile the source of truth disagrees:nixos/home/laptop-autocommit.nixsetsTimer.OnCalendar = "*-*-* *:*:00"("Autocommit ${name} every minute") — daily-at-midnight in the registry vs every minute in reality.
Acceptance criteria
- Every systemd-timer row in
doc/fleet-schedule-registry.mdrecords the timer's complete calendar expression (date + time components, e.g.*-*-* *:*:00), re-read from the live units / unit files, so the registry's documented cadence matches what systemd will actually run. - Or: the finding is resolved on this issue by a documented decision, with the reasoning recorded.
- The registry's maintenance/verification note is updated so a future refresh does not re-truncate the specs.
Activity
dezren39 commentedon Nov 4, 2026
🤖 Agent note · filed
dezren39 commentedon Nov 4, 2026
Launched and bound queue task 9264 to OpenCode session ses_e5e5a8f3609ffedFjEtvG5e3Pt via opencode-route/workhorse (build). The task claim remains owned by repo-worker-code.
dezren39 commentedon Nov 4, 2026
Triage verdict: confirmed on current
main(2026-11-04)Fresh raw GitHub API reads confirm the finding is still live:
doc/fleet-schedule-registry.mdSurface 2b still showsOnCalendar=*-*-*for the named laptop-autocommit, quota-gatherer, and worktree-reaper timers.nixos/home/laptop-autocommit.nixsetsTimer.OnCalendar = "*-*-* *:*:00"(every minute).nixos/home/worktree-reaper.nixsetsTimer.OnCalendar = "hourly".pm-stack/aggregator/systemd/quota-gatherer.timersetsOnCalendar=*:0/15(every 15 minutes).Routed a scoped, docs-only
small_fixpacket as queue task 9317, using dedupe keyrepo-health:v1:fix:developing-today/code:issue:5313. It covers correcting the Surface 2b cadences and updating the maintenance note so future refreshes preserve the complete effective schedule. Timer behavior and source definitions are explicitly out of scope. This triage did not reply to or resolve the source PR review thread.dezren39 commentedon Nov 4, 2026
🤖 Agent note · picked up
dezren39 commentedon Nov 4, 2026
🤖 Agent note · done
systemctl --user showTimersCalendar on the laptop (pm-digest rows fromnixos/services/pm-stack/default.nix); maintenance/verification notes updated so future refreshes preserve full specs; Codex review completed with no findings; CodeRabbit check passed.