Repository navigation
check-rd-presets.sh reports present presets as missing under load, blocking merges #4193
Description
hack/check-rd-presets.sh reports presets as missing when they are in the file, and the false failure blocks merges on unrelated PRs.
From #4185:
hack/check-rd-presets.sh: line 39: printf: write error: Broken pipe
FAIL: packages/system/mariadb-rd/cozyrds/mariadb.yaml resourcesPreset enum missing: c1.small
c1.small is in that file on main, and the script passes when run on the same head.
The check is printf '%s\n' "$enums" | grep -Fqx -- "$want" under set -euo pipefail. grep -q exits on the first match and closes the pipe while printf is still writing, printf fails with EPIPE, pipefail makes the pipeline non-zero, and ! reads that as not found. A value gets reported missing because it matched first. It is a race, so it fires under make -j4 in CI and not on a laptop.
The cost is not the red tick. Unit & controller tests is in needs for finalize, finalize gates e2e, so this turns the required E2E Tests context red on a PR that never failed a test. #4185, #4138 and #3539 are red this way right now. #4109, #4097, #3802 and #3594 are not, which is what a race looks like rather than a schema drift.
Fix is a here-string instead of the pipe. This is the only script under hack/ with pipefail and | grep -q.
PR here: #4194
Activity
fix(ci): stop check-rd-presets reporting present presets as missing
myasnikovdaniil commentedon Sep 10, 2026
Fixed in #4191, it carries your commit from #4142 with your authorship.
I lifted it out of your branch instead of waiting for it, because this one reddens every pull request in the repo while #4142 is a feature that needs its own review time, and a one line fix should not sit behind that. Your commit is self contained and applies cleanly on its own, so cherry-picking it was the whole of it, nothing rewritten. Asked you in #4142 to drop 99d16a2 from that branch so it does not land twice.
Numbers from the same window in case they help triage: 32 of the 48 unit-lane failures on pull requests between 09-04 and 09-10 were this, on 21 different branches. All 41
printf: write error: Broken pipelines name line 39, and the count of accused presets equals the count of broken pipes in the same job every time, which is what tells it apart from a truncated jq read.Guard is there too, list doubled past 4 MiB like your repro. It also pins
-xand-F, neither of which a fixture that just omits a value can see.kvaps commentedon Sep 10, 2026
Duplicate of #4178, which names the same line, gives the same analysis and predates this by a day. I filed without searching first, my mistake. The fix is #4194 and I have re-pointed it at #4178.