The manually triggered Integration workflow's macOS amd64 job is now too close to its timeout budget and should be made faster and less variable.
Evidence from exact-head run 31716454220:
- Last recent successful macOS job (run 31303214737): 25m36s total.
- Attempt 1: the application-network-policy suite exhausted its shared 5-minute context.
- Attempt 2: all tests progressed successfully, but the 40-minute job limit cancelled final staging cleanup after 40m35s.
- Attempt 3: the same application-network-policy shared-context timeout reproduced.
- Controlled-session and OmegaFlow conformance are Linux-only and did not add macOS work.
The largest serial costs are Colima setup, private-isolation tests, application-sandbox tests, and the CLI/persistent-install smoke.
Follow-up goals:
- Profile the macOS phases and remove avoidable repeated builds/work.
- Evaluate splitting sandbox and persistent-install coverage while accounting for duplicated Colima setup.
- Keep equivalent macOS coverage and improve failure isolation.
- Establish enough headroom that routine runner variance does not consume the job budget.
The immediate timeout-budget correction is intentionally handled separately from this performance work.
The manually triggered Integration workflow's macOS amd64 job is now too close to its timeout budget and should be made faster and less variable.
Evidence from exact-head run 31716454220:
The largest serial costs are Colima setup, private-isolation tests, application-sandbox tests, and the CLI/persistent-install smoke.
Follow-up goals:
The immediate timeout-budget correction is intentionally handled separately from this performance work.