CI: stop installing Node 20 on Node 20, and assert every leg's runtime #1604

Merged
jaylouisw merged 1 commit from ci/1116-0420 into main 2026-11-16 05:11:58 +00:00
Owner

What was slow / broken

A false red on a green tree. Run 3122 (PR #1564, ci.yml), job Test (Node 20) =
task 6834: the leg died 22 s in, before running a single check:

Unable to checkout 49933ea5288caeca8642d1e84afbd3f7d6820020:
remove /data/.cache/act/e5/877e7fc2f7e5000a2f22526584a2565cc2ae38cd26a9b1938dbca653b056cc/.gitignore: no such file or directory

Tasks 6833 (Test (Node 18)) and 6835 (Test (Node 22)) of the same run ran the
identical suite to completion (260 suites, 0 failures) - the two other legs are green, the
third is red, and the commit cannot merge. All three legs clone actions/setup-node@v4 into
the same act cache path within 5 s of each other (04:27:07 / 04:27:09 / 04:27:12).

The same clone is also pure wall clock. Task 6833: setup-node action clone at 04:27:14,
Node tarball download at 04:27:32 - ~25 s of a 229 s job - to install Node 20 on a job image
that already is node:20-bookworm (Start image=node:20-bookworm, first line of every log,
and why the changes job has never had a setup step).

The change

  • The Node 20 leg no longer runs actions/setup-node (if: matrix.node != 20). It uses the
    image's node, as the changes job always has.
  • Every leg now asserts node --version matches the version on its check name. That is what
    keeps the saving honest: the repo has twice shipped legs that reported coverage they never
    had (runs 1398/1401/1402), and this makes an image change fail loudly instead.

Baseline

job run / task duration outcome
Test (Node 18) 3122 / 6833 229 s (suite 173.3 s) success
Test (Node 20) 3122 / 6834 22 s failure - act action-cache collision
Test (Node 22) 3122 / 6835 207 s (suite 144.2 s) success

Queue context: run 3122's classify job ran at 03:00:41 and its legs only started at 04:27 -
86 minutes in the queue, with 27 ci.yml runs waiting at 04:30. Each leg costs ~230 s of
which ~25 s is now gone from one of the three.

Coverage

Unchanged. Node 18, 20 and 22 still each run the full suite, the five static checks, the
example build and the catalogue sweep; Node 20 is the only leg whose runtime now comes from
the image instead of a download, and it is asserted rather than assumed.

After

This branch's own run is subject to the documented dispatch defect (own branches do not
dispatch) and the queue is 27 runs deep; the after number will be the Test (Node 20) leg of
the first run that carries this change.

Still red, not mine to fix

spec-vs-description fails on main and therefore on every PR run examined - 3053/3055/3059/
3065/3067 all end 2 passed, 1 failed: 4 edgecore AS4610 models claim 2x 20G QSFP+ with no
port that can do 20G. That is catalogue data, not workflows, and PR #1564 appears to carry the
fix.

## What was slow / broken **A false red on a green tree.** Run 3122 (PR #1564, `ci.yml`), job `Test (Node 20)` = task **6834**: the leg died 22 s in, before running a single check: ``` Unable to checkout 49933ea5288caeca8642d1e84afbd3f7d6820020: remove /data/.cache/act/e5/877e7fc2f7e5000a2f22526584a2565cc2ae38cd26a9b1938dbca653b056cc/.gitignore: no such file or directory ``` Tasks **6833** (`Test (Node 18)`) and **6835** (`Test (Node 22)`) of the same run ran the identical suite to completion (260 suites, 0 failures) - the two other legs are green, the third is red, and the commit cannot merge. All three legs clone `actions/setup-node@v4` into the same act cache path within 5 s of each other (04:27:07 / 04:27:09 / 04:27:12). **The same clone is also pure wall clock.** Task 6833: setup-node action clone at 04:27:14, Node tarball download at 04:27:32 - ~25 s of a 229 s job - to install Node 20 on a job image that already is `node:20-bookworm` (`Start image=node:20-bookworm`, first line of every log, and why the `changes` job has never had a setup step). ## The change - The Node 20 leg no longer runs `actions/setup-node` (`if: matrix.node != 20`). It uses the image's node, as the `changes` job always has. - Every leg now asserts `node --version` matches the version on its check name. That is what keeps the saving honest: the repo has twice shipped legs that reported coverage they never had (runs 1398/1401/1402), and this makes an image change fail loudly instead. ## Baseline | job | run / task | duration | outcome | |---|---|---|---| | Test (Node 18) | 3122 / 6833 | 229 s (suite 173.3 s) | success | | Test (Node 20) | 3122 / 6834 | 22 s | **failure** - act action-cache collision | | Test (Node 22) | 3122 / 6835 | 207 s (suite 144.2 s) | success | Queue context: run 3122's classify job ran at 03:00:41 and its legs only started at 04:27 - **86 minutes** in the queue, with 27 `ci.yml` runs waiting at 04:30. Each leg costs ~230 s of which ~25 s is now gone from one of the three. ## Coverage Unchanged. Node 18, 20 and 22 still each run the full suite, the five static checks, the example build and the catalogue sweep; Node 20 is the only leg whose runtime now comes from the image instead of a download, and it is asserted rather than assumed. ## After This branch's own run is subject to the documented dispatch defect (own branches do not dispatch) and the queue is 27 runs deep; the after number will be the `Test (Node 20)` leg of the first run that carries this change. ## Still red, not mine to fix `spec-vs-description` fails on main and therefore on every PR run examined - 3053/3055/3059/ 3065/3067 all end `2 passed, 1 failed`: 4 edgecore AS4610 models claim `2x 20G QSFP+` with no port that can do 20G. That is catalogue data, not workflows, and PR #1564 appears to carry the fix.
CI: the Node 20 leg no longer downloads Node 20 over Node 20
Some checks failed
CI / Classify the change (pull_request) Successful in 34s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Failing after 11m48s
CI / Docker build (pull_request) Failing after 15m22s
CI / Server smoke test (pull_request) Failing after 15m26s
CI / Test (Node 22) (pull_request) Failing after 15m34s
CI / Test (Node 20) (pull_request) Failing after 15m37s
CI / Test (Node 18) (pull_request) Failing after 15m41s
9d16b4faa1
The runner image for ubuntu-latest is node:20-bookworm - the Start image= line in
every job log, and why the changes job has never needed a setup step. The Node 20
test leg was still cloning actions/setup-node and downloading the 20 tarball on top
of that: 25 s of a 229 s job in run 3122 task 6833, and three legs cloning the same
action into the same /data/.cache/act path within five seconds of each other.

That collision is not theoretical. In the same run, task 6834 (Test (Node 20)) died
22 s in with 'Unable to checkout 49933ea...: remove /data/.cache/act/e5/877e.../
.gitignore: no such file or directory' while tasks 6833 and 6835 ran the identical
260-suite suite to completion: a red run on a green tree, on PR #1564.

Two clones can now collide instead of three. Coverage is unchanged - Node 20 still
runs, on the image's node - and every leg now asserts the runtime it actually got
matches the name on the check, so an image change can never silently retitle a leg.
jaylouisw force-pushed ci/1116-0420 from 9d16b4faa1
Some checks failed
CI / Classify the change (pull_request) Successful in 34s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Failing after 11m48s
CI / Docker build (pull_request) Failing after 15m22s
CI / Server smoke test (pull_request) Failing after 15m26s
CI / Test (Node 22) (pull_request) Failing after 15m34s
CI / Test (Node 20) (pull_request) Failing after 15m37s
CI / Test (Node 18) (pull_request) Failing after 15m41s
to 3c66f9e897
Some checks failed
PR rebase / Rebase onto main and arm auto-merge (pull_request) Failing after 14m47s
CI / Classify the change (pull_request) Failing after 15m13s
CI / Test (Node 18) (pull_request) Has been cancelled
CI / Test (Node 20) (pull_request) Has been cancelled
CI / Test (Node 22) (pull_request) Has been cancelled
CI / Server smoke test (pull_request) Has been cancelled
CI / Docker build (pull_request) Has been cancelled
2026-11-16 04:58:09 +00:00
Compare
jaylouisw scheduled this pull request to auto merge when all checks succeed 2026-11-16 04:58:40 +00:00
jaylouisw scheduled this pull request to auto merge when all checks succeed 2026-11-16 05:11:36 +00:00
jaylouisw force-pushed ci/1116-0420 from 3c66f9e897
Some checks failed
PR rebase / Rebase onto main and arm auto-merge (pull_request) Failing after 14m47s
CI / Classify the change (pull_request) Failing after 15m13s
CI / Test (Node 18) (pull_request) Has been cancelled
CI / Test (Node 20) (pull_request) Has been cancelled
CI / Test (Node 22) (pull_request) Has been cancelled
CI / Server smoke test (pull_request) Has been cancelled
CI / Docker build (pull_request) Has been cancelled
to 8c69dcaa61
Some checks failed
CI / Classify the change (pull_request) Waiting to run
PR rebase / Rebase onto main and arm auto-merge (pull_request) Waiting to run
CI / Test (Node 18) (pull_request) Has been cancelled
CI / Test (Node 20) (pull_request) Has been cancelled
CI / Test (Node 22) (pull_request) Has been cancelled
CI / Server smoke test (pull_request) Has been cancelled
CI / Docker build (pull_request) Has been cancelled
2026-11-16 05:11:38 +00:00
Compare
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge!1604
No description provided.