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
ci(release): the half-hourly nightly check stops paying for an 8 vCPU runner - #28304
The release workflow runs on a schedule twice an hour (cron: "8,38 * * * *"), 48 times a day, and every run starts resolve_commit on blacksmith-8vcpu-ubuntu-2404. That job does a sparse checkout of .github/scripts and a few GitHub API calls, and most runs stop there with has_changes=false. So the workflow pays for 48 short 8 vCPU jobs a day, on days with no commits too, for work that needs one core.
Change
resolve_commit runs on blacksmith-2vcpu-ubuntu-2404, the runner the repo already uses for its light jobs, such as changes in ci.yml and deploy_web and deploy_marketing in this workflow. Preflight and the build jobs keep their runners.
Scope
A fix for wasted CI spend, not a product bug: 48 short 8 vCPU jobs a day for work that needs one core. No issue or discussion: this is an obvious one-line runner-size fix that only lowers runner spend. It changes no product behavior and no job output. The 8 vCPU label was never chosen for this work: it came over from the old check_changes job when #10410 renamed it to resolve_commit. t3dotgg's merged #33141 lists the runner size among the things it leaves unchanged and points to this PR for it: "and the runner size (that is #28304)".
Verification
No product runtime path: the evidence is upstream's own job timings and a 2-CPU stand-in.
"Resolve release commit" in the 30 newest completed scheduled runs on main (2026-11-11 16:40 to 2026-11-12 07:24 UTC), from the public Actions API:
Jobs
Job time
Longest step
Went past this job
30, all success on 8 vCPU
median 4 s, range 3 to 7 s
the script step, 5 s
2 of 30
Stand-in, my own Linux machine limited to 2 CPUs, not the 2 vCPU runner: the checkout's git commands and the step's script at 4e0b528, three times, read-only, took 2.0 to 2.2 s per run, each has_changes=false, the same skip upstream logged.
Not checked:
The job on blacksmith-2vcpu-ubuntu-2404 itself. It runs only in this repository, so only a maintainer can run it.
Blacksmith billing for this job.
Steps, stand-in method and timing output
Because of the 6-hour release gap, most scheduled runs stop at resolve_commit. The closest runner match is changes in ci.yml, which also does a sparse checkout and runs one script.
On every trigger the job only makes API calls: assertReleaseSource and shouldReleaseNightly on schedule, resolveLatestNightlyCommit and assertCommitOnDefaultBranch for a manual stable release, and the source check alone for tag pushes and other manual runs. Nothing in it installs, builds or compiles, and it keeps its 5-minute timeout.
A GitHub-hosted ubuntu-24.04 runner would cost nothing on a public repo, but the repo's other light jobs run on Blacksmith 2 vCPU, so this follows that.
Steps: Set up job, Set up runner, Checkout (sparse, .github/scripts), Resolve release commit (github-script, GitHub API reads only), Post Checkout, Complete runner, Complete job. Each step's median is 1 s or less.
Stand-in. I ran the checkout's git commands and the step's script at 4e0b528 three times, read-only, under taskset -c 0,1 and a 4 GiB cap: 2.0 to 2.2 s per run, each has_changes=false, the same skip upstream logged. Differences from the runner: Node 22 with @actions/github 6.0.1 in place of actions/github-script on Node 24, no retries, and no runner setup, which needs the Blacksmith image.
jobs: 30 ("Resolve release commit", scheduled runs on main)
window: 2026-11-11T16:40:26Z to 2026-11-12T07:24:14Z
runs-on labels: blacksmith-8vcpu-ubuntu-2404; conclusions: success
job duration (started_at to completed_at, s): median 4, min 3, max 7
step (median / max seconds across jobs; GitHub records step times to the second):
1 Set up job: 0 / 2
2 Set up runner: 0 / 1
3 Checkout: 1 / 1
4 Resolve release commit: 1 / 5
7 Post Checkout: 0 / 1
8 Complete runner: 0 / 0
9 Complete job: 0 / 0
longest single step: "Resolve release commit" 5 s
limits: nproc=2 cpus_allowed=0-1 memory.max=4294967296 node=v22.22.1 git=2.53.0
candidate sha (upstream main when run): 4e0b52806be83e66f80839e1e372398307346425
run 1 checkout: 4e0b528 files=32049
run 1 resolve: Nightly v0.1.10-nightly.20261112.4564 was published less than six hours ago. Skipping.
run 1 resolve: output has_changes=false
run 1 resolve: output ref=4e0b52806be83e66f80839e1e372398307346425
run 1 seconds: checkout 1.10, resolve 1.12, both 2.22
run 2 checkout: 4e0b528 files=32049
run 2 resolve: Nightly v0.1.10-nightly.20261112.4564 was published less than six hours ago. Skipping.
run 2 resolve: output has_changes=false
run 2 resolve: output ref=4e0b52806be83e66f80839e1e372398307346425
run 2 seconds: checkout 1.08, resolve 0.91, both 1.99
run 3 checkout: 4e0b528 files=32049
run 3 resolve: Nightly v0.1.10-nightly.20261112.4564 was published less than six hours ago. Skipping.
run 3 resolve: output has_changes=false
run 3 resolve: output ref=4e0b52806be83e66f80839e1e372398307346425
run 3 seconds: checkout 1.09, resolve 0.91, both 2.00
rc=0
2026-11-12T07:37:18Z
Prior PRs
Theo Browne's #7284 (closed) right-sized six runners in one bundle, three scale-ups and three trims, and was closed because the bundle as a whole added spend. It did not touch this job, then named check_changes. This PR changes one runner and only lowers spend.
Macroscope's review found this PR approvable — Moves the lightweight release-resolution job to a smaller runner while leaving builds, publishing and release safeguards unchanged. The limited operational downside is a delayed or failed gate, recoverable by restoring the runner and rerunning.
You can add or adjust custom eligibility rules. Learn more.
No actionable comments were generated in the recent review.
Walkthrough
The resolve_commit job in the release workflow now uses blacksmith-2vcpu-ubuntu-2404 instead of blacksmith-8vcpu-ubuntu-2404. Its timeout and release-commit resolution steps are unchanged.
The release commit resolver uses a smaller runner. No actionable merge-blocking risk is established.
Pre-merge checks: 4
Check name
Status
Explanation
Title check
The title clearly identifies the main change: reducing the release workflow runner from 8 vCPUs to 2 vCPUs. It is concise and uses a conventional commit format.
Linked Issues check
Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check
Check skipped because no linked issues were found for this pull request.
Description check
The description follows the repository template and fills every required section. Problem states that resolve_commit runs twice an hour on blacksmith-8vcpu-ubuntu-2404 and does light API work. Change …
Full details: Description check
Explanation
The description follows the repository template and fills every required section. Problem states that resolve_commit runs twice an hour on blacksmith-8vcpu-ubuntu-2404 and does light API work. Change explains the switch to blacksmith-2vcpu-ubuntu-2404 and names the light-job runner the repo already uses. Scope gives a bug-fix rationale and explains why no issue is linked. Verification reports 30 upstream job timings (median 4 s, max 7 s), a 2-CPU local stand-in run three times, and what was not checked. The author also names the model used, as the template asks. The Verification section is detailed and the Scope section gives a clear reason for having no issue. The template's conventional commit title is not visible in the description body, so the title cannot be confirmed from the provided context.
Finishing Touches
Generate unit tests (beta)
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
size:XS0-9 changed lines (additions + deletions).vouch:unvouchedAuthor is not yet trusted in the VOUCHED list.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
github-actions
Bot
added
vouch:unvouched
![macroscopeapp[bot]](https://avatars.githubusercontent.com/in/900172?s=60&v=4){kind=small}
Problem
The release workflow runs on a schedule twice an hour (
cron: "8,38 * * * *"), 48 times a day, and every run startsresolve_commitonblacksmith-8vcpu-ubuntu-2404. That job does a sparse checkout of.github/scriptsand a few GitHub API calls, and most runs stop there withhas_changes=false. So the workflow pays for 48 short 8 vCPU jobs a day, on days with no commits too, for work that needs one core.Change
resolve_commitruns onblacksmith-2vcpu-ubuntu-2404, the runner the repo already uses for its light jobs, such aschangesinci.ymlanddeploy_webanddeploy_marketingin this workflow. Preflight and the build jobs keep their runners.Scope
A fix for wasted CI spend, not a product bug: 48 short 8 vCPU jobs a day for work that needs one core. No issue or discussion: this is an obvious one-line runner-size fix that only lowers runner spend. It changes no product behavior and no job output. The 8 vCPU label was never chosen for this work: it came over from the old
check_changesjob when #10410 renamed it toresolve_commit. t3dotgg's merged #33141 lists the runner size among the things it leaves unchanged and points to this PR for it: "and the runner size (that is #28304)".Verification
No product runtime path: the evidence is upstream's own job timings and a 2-CPU stand-in.
"Resolve release commit" in the 30 newest completed scheduled runs on
main(2026-11-11 16:40 to 2026-11-12 07:24 UTC), from the public Actions API:successon 8 vCPUStand-in, my own Linux machine limited to 2 CPUs, not the 2 vCPU runner: the checkout's git commands and the step's script at
4e0b528, three times, read-only, took 2.0 to 2.2 s per run, eachhas_changes=false, the same skip upstream logged.Not checked:
blacksmith-2vcpu-ubuntu-2404itself. It runs only in this repository, so only a maintainer can run it.Steps, stand-in method and timing output
Because of the 6-hour release gap, most scheduled runs stop at
resolve_commit. The closest runner match ischangesinci.yml, which also does a sparse checkout and runs one script.On every trigger the job only makes API calls:
assertReleaseSourceandshouldReleaseNightlyon schedule,resolveLatestNightlyCommitandassertCommitOnDefaultBranchfor a manual stable release, and the source check alone for tag pushes and other manual runs. Nothing in it installs, builds or compiles, and it keeps its 5-minute timeout.A GitHub-hosted
ubuntu-24.04runner would cost nothing on a public repo, but the repo's other light jobs run on Blacksmith 2 vCPU, so this follows that.Steps: Set up job, Set up runner, Checkout (sparse,
.github/scripts), Resolve release commit (github-script, GitHub API reads only), Post Checkout, Complete runner, Complete job. Each step's median is 1 s or less.Stand-in. I ran the checkout's git commands and the step's script at
4e0b528three times, read-only, undertaskset -c 0,1and a 4 GiB cap: 2.0 to 2.2 s per run, eachhas_changes=false, the same skip upstream logged. Differences from the runner: Node 22 with@actions/github6.0.1 in place ofactions/github-scripton Node 24, no retries, and no runner setup, which needs the Blacksmith image.Prior PRs
Theo Browne's #7284 (closed) right-sized six runners in one bundle, three scale-ups and three trims, and was closed because the bundle as a whole added spend. It did not touch this job, then named
check_changes. This PR changes one runner and only lowers spend.Generated with Claude Opus 5.5 in T3 Code Nightly