You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 3484a7c
Browse filesBrowse the repository at this point in the historyBrowse files
docs: record the standard runner cost stop and ci-spot handoff
Reword the runner-selection and timing assumptions introduced by d093560 (#290) and the documentation-only runner sentences introduced by 133f571 (#308). Keep the platform job limits recorded by 2394ba2 (#373).
Record the owner decision of 2026-10-18 after larger-runner costs exceeded US$3,000: every event uses free standard runners until the ci-spot pool is live. Describe the unconditional runner contract and remove obsolete larger-runner timing estimates while retaining measured standard-runner durations.
No local verification run, as required by the executor brief.
The `build` matrix pairs each image with its `larger_runner` and names its legs
911
-
explicitly, so they stay `build (<image>, <type>)`. `native-live.yml` and
912
-
`corpus-parity.yml`apply the test themselves; in a `workflow_call` the
913
-
`github`context is the caller's, so a fork's pull request through CI gets the
914
-
standard runner there too. The advisory `build` aggregate runs a few seconds of
915
-
shell and stays on the standard runner for every event. So does the `changes`
916
-
job, a two-commit checkout and one `git diff`.
903
+
The `build` matrix selects its standard image with `runs-on: ${{ matrix.os }}`
904
+
and names its four legs explicitly, so they stay `build (<image>, <type>)`.
905
+
`native-live.yml`and `corpus-parity.yml` use `ubuntu-24.04` directly, whether
906
+
called through CI or started independently. The advisory `build` aggregate
907
+
and the `changes` job also use `ubuntu-24.04` for every event.
917
908
918
909
Each `ci_verify.py` call passes `--jobs "$(getconf _NPROCESSORS_ONLN)"` and
919
910
`check_corpus_parity.sh`gets the same count as `JOBS`, so build and CTest
920
-
parallelism follow the runner: 16 on the larger Linux runner, 5 on the M2, 4
921
-
and 3 on the standard runners. The larger Linux runner is configured for at
922
-
most eight jobs at once, and one CI run starts seven there (preflight, the two
923
-
Ubuntu `build` legs, sanitizers, kernel-only, native-live and the parity
924
-
subset); a pull request that touches the corpus pin or the parity tooling also
925
-
starts the whole sweep there, an eighth. A documentation-only change starts
926
-
preflight alone there. A second run that overlaps them waits
927
-
for a runner. Larger runners bill the organization's Actions budget: with no
928
-
payment method or a spending limit of zero, their jobs wait for a runner that
929
-
never comes, since GitHub does not fall back to a standard one. The test keeps
930
-
a fork's pull request that leaves the workflows alone off the larger runners;
931
-
one that edits them runs its own copy, which preflight reports but does not
932
-
stop, so the repository's approval rule for workflow runs from outside
933
-
contributors is what bounds that case.
911
+
parallelism follow the runner: 4 on Ubuntu and 3 on macOS. A documentation-only
912
+
change starts preflight alone among the verification jobs.
934
913
935
914
`scripts/ci_preflight.py`(`ci-workflow-contract`) pins every job's runner, time
936
915
limit and parallelism in the three workflows a CI run starts, the `build`
937
916
strategy block included, and keeps every other workflow -- `docs.yml`,
938
-
`promote-baseline.yml`and any new one -- on standard runners. Only
939
-
`release.yml`is exempt, by name and while its `on:` block names only events no
940
-
fork can raise (a push, the schedule, a dispatch; a dispatch alone today). A
941
-
heavy job moved back to the standard runner, a test or matrix entry that lets a
942
-
fork's pull request onto a larger runner, a call to a workflow outside
917
+
`promote-baseline.yml`and any new one -- on standard runners. `release.yml`
918
+
keeps its existing standard-runner matrix and is exempt from the fixed runner
919
+
allowlist, by name and while its `on:` block names only events no fork can
920
+
raise (a push, the schedule, a dispatch; a dispatch alone today). The contract
921
+
refuses `pf-linux-x64-16`, any `*-xlarge` or `*-large` label, and any
922
+
`larger_runner`key in every workflow, including `release.yml`. An event-based
923
+
runner selection, a changed build matrix, a call to a workflow outside
943
924
`.github/workflows/`, a changed time limit, a `ci_verify.py` call without
944
925
exactly one core-count `--jobs` (counted over the whole job, however its
945
926
commands are chained, blocked or carried, with the flag and its value on one
@@ -989,17 +970,15 @@ that replaced its old limit, so the margin the removal buys stays visible. The
989
970
numbers are measured durations, not a two-times multiplier.
990
971
The standard-runner times are the eight CI runs from
991
972
main's 8cf3be58 (2026-09-24) to 0d76a099 (2026-09-25), pull requests included,
992
-
and the corpus-parity workflow's ten runs of 2026-09-22 to 2026-09-25. Until
993
-
the larger runners' own runs accumulate, the maintainers' x86-64 verification
994
-
hosts at 8 to 12 jobs stand in for the Linux one on parallel work, the
995
-
standard runner's speed stands in for it on serial work, and the 3-core M1
996
-
bounds the M2 from above.
973
+
and the corpus-parity workflow's ten runs of 2026-09-22 to 2026-09-25. The
974
+
maintainers' x86-64 verification-host timings at 8 to 12 jobs remain separate
975
+
measurements; the standard-runner timings describe the current runner policy.
997
976
998
977
| Job | Limit (min) | Measured basis |
999
978
| --- | --- | --- |
1000
979
| `preflight` | 360 (was 45) | 7.3-9.8 min on the standard runner through 1a0e7ea1; cancelled at 10.3 min at 0d76a099, 593 s into `verifier-tests`, which takes about 15 min there at that tree (above), so preflight about 16 min. 4.8-8.2 min on the verification hosts. Preflight supplies no stage deadline -- the old 2400 s stage bound is gone (owner rule, 2026-10-12); the native-live aggregate job names a stop at the platform's 360-minute maximum `HANG (re-run)` in its log; the check still fails. |
1001
-
| `build` | 360 (was 75) | A Release leg runs its full population on every event, a fork's pull request included: 30.3 min Ubuntu and 32.0 min macOS on the standard runners at 0d76a099, where the full Debug legs took 43.5 and 37.0 min. Pull-request Debug sets 17.2-17.7 min. On the larger Linux runner a full leg is its build plus the serial `test_ci_verify` row, 976-1288 s: about 30 min. The M1's 37.0 min bounds the M2. 11.1-18.4 min on the verification hosts. |
1002
-
| `sanitizers` | 360 (was 120; the verifier's CTest stage bound was 60 min full, 30 min PR) | A fork's pull-request set took 35.2 min on the standard runner, whose compile alone took 1277-1335 s (pull-request run 36139176292, and main at 0d76a099). Full runs now land only on the larger runner; on the standard runner they took 34.0-69.4 min (at 0d76a099 30 min of build and ABI providers, then 39 min of CTest, 1768 s of it the serial `test_ci_verify` row), so about 40 min there with a faster build, and 18.2-19.0 min on the verification hosts. |
980
+
| `build` | 360 (was 75) | A Release leg runs its full population on every event, a fork's pull request included: 30.3 min Ubuntu and 32.0 min macOS on the standard runners at 0d76a099, where the full Debug legs took 43.5 and 37.0 min. Pull-request Debug sets 17.2-17.7 min. 11.1-18.4 min on the verification hosts. |
981
+
| `sanitizers` | 360 (was 120; the verifier's CTest stage bound was 60 min full, 30 min PR) | A fork's pull-request set took 35.2 min on the standard runner, whose compile alone took 1277-1335 s (pull-request run 36139176292, and main at 0d76a099). Full runs on the standard runner took 34.0-69.4 min (at 0d76a099 30 min of build and ABI providers, then 39 min of CTest, 1768 s of it the serial `test_ci_verify` row), and 18.2-19.0 min on the verification hosts. |
1003
982
| `kernel-only` | 360 (was 60) | Every event runs the whole kernel set: 7.9-22.3 min on the standard runner, 6.5-9.8 min on the verification hosts. The hosted kernel-only run of 2026-10-12 lost its job to the old 1800 s stage bound (exit 124). |
1004
983
| `native-live (native)` | 360 (was 75; verification step 65) | Main runs 37398893123, 37461653250 and 37478400505 spent 35.40–40.65 min in native, including 25.32–29.53 min in full CTest. Their matching PRs took 2.39–5.67 min in native CTest with the existing slow exclusion. The longest measured `test_ci_verify` row was 1695 s there and 2178 s on a c5.4xlarge Spot host, cut off by a trial deadline. The verifier supplies no default row or stage deadline and the workflow supplies no step deadline: its old 3600 s row bound is what collided with the 2820 s stage deadline on main's push run 38217844683. These are observed GitHub and Spot timings; final-head cold proof remains required. |
1005
984
| `native-live (live-sanitizers)` | 360 (was 45; verification step 35) | The same three main runs spent 15.10–16.33 min in this profile after native. Its old per-row 900 s and stage 1200 s deadlines are gone; the verifier supplies no replacement, and the platform's job maximum catches a hang. |
0 commit comments