Skip to content

acceptance: skip pydabs_1000_tasks in migrate + continue_293 invariants#5062

Merged
pietern merged 1 commit into
mainfrom
test-time-debug
May 7, 2026
Merged

acceptance: skip pydabs_1000_tasks in migrate + continue_293 invariants#5062
pietern merged 1 commit into
mainfrom
test-time-debug

Conversation

@pietern
Copy link
Copy Markdown
Contributor

@pietern pietern commented Apr 22, 2026

Drop job_pydabs_1000_tasks from the migrate and continue_293 invariants — they are the top-2 slowest acceptance tests on */direct jobs. no_drift still runs the same 1000-task config every PR, so the deploy + plan path stays covered at scale.

We do lose some coverage: scale-specific bugs in the terraform→direct state migration path (migrate) or in v0.293.0-state back-compat (continue_293) wouldn't be caught at 1000 tasks anymore. Both invariants still run their full set of other configs every PR. Given how heavy these two tests are, the trade-off is worth it.

The real fix is to make the 1000-task case fast in principle, but that's a separate effort. This PR buys back PR-CI time today; speeding up the underlying path is what makes the dropped coverage cheap to restore later.

Measured savings vs main:

  • linux/direct: 8m47s → 5m20s
  • macos/direct: 11m9s → 8m19s
  • windows/direct: 25m44s → 24m15s

This pull request and its description were written by Isaac.

Comment thread libs/patchwheel/patch_test.go Outdated
Comment thread acceptance/bundle/invariant/migrate/test.toml
The 1000-task scale case is still exercised by `no_drift`; running the
same scale in `migrate` and `continue_293` adds dominant time in PR CI
without incremental coverage. Measured against a recent main build:

  linux/direct      migrate=228s   continue_293=98s    (no_drift kept=94s)
  windows/direct    migrate=422s   continue_293=143s   (no_drift kept=109s)

These two excluded tests are the top-2 slowest tests on linux/direct and
the top-2 on windows/direct, so this is the high-leverage cut.

Co-authored-by: Isaac
@pietern pietern force-pushed the test-time-debug branch from 17773bf to 0d443d6 Compare May 7, 2026 09:24
@pietern pietern temporarily deployed to test-trigger-is May 7, 2026 09:24 — with GitHub Actions Inactive
@pietern pietern temporarily deployed to test-trigger-is May 7, 2026 09:24 — with GitHub Actions Inactive
@pietern pietern changed the title ci: trim PR/merge-queue test time acceptance: skip pydabs_1000_tasks in migrate + continue_293 invariants May 7, 2026
@pietern pietern requested a review from denik May 7, 2026 09:55
@pietern pietern marked this pull request as ready for review May 7, 2026 09:57
@pietern pietern enabled auto-merge May 7, 2026 09:57
@pietern pietern disabled auto-merge May 7, 2026 12:42
@pietern pietern merged commit ef90f53 into main May 7, 2026
32 of 33 checks passed
@pietern pietern deleted the test-time-debug branch May 7, 2026 12:53
denik pushed a commit that referenced this pull request May 20, 2026
…ts (#5062)

Drop `job_pydabs_1000_tasks` from the `migrate` and `continue_293`
invariants — they are the top-2 slowest acceptance tests on `*/direct`
jobs. `no_drift` still runs the same 1000-task config every PR, so the
deploy + plan path stays covered at scale.

We do lose some coverage: scale-specific bugs in the terraform→direct
state migration path (`migrate`) or in v0.293.0-state back-compat
(`continue_293`) wouldn't be caught at 1000 tasks anymore. Both
invariants still run their full set of other configs every PR. Given how
heavy these two tests are, the trade-off is worth it.

The real fix is to make the 1000-task case fast in principle, but that's
a separate effort. This PR buys back PR-CI time today; speeding up the
underlying path is what makes the dropped coverage cheap to restore
later.

Measured savings vs main:

- linux/direct: 8m47s → 5m20s
- macos/direct: 11m9s → 8m19s
- windows/direct: 25m44s → 24m15s

This pull request and its description were written by Isaac.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants