Skip to content

fix(ci): stop the cache janitor deleting Dylint's compile cache and verdicts - #2071

Merged
zackees merged 2 commits into
mainfrom
perf/dylint-cache
Nov 18, 2026
Merged

zackees merged 2 commits into
mainfrom
perf/dylint-cache

Conversation

@zackees

@zackees zackees commented Nov 18, 2026

Copy link
Copy Markdown
Member

Problem

Dylint is slow (~16 min for "Run dylint over workspace" on main) because its nightly-compiler build cache is never reused. Every main run logs build-cache MISS (no match) on restore and hits=0 misses=496 hit_rate=0.0%. The cache is saved; the end-of-queue ci_lint cache janitor deletes it afterwards:

  • CACHE-001 "undeclared family": [cache.family.compile] used via = "setup-soldr:build-cache". The ci_lint linter still maps that to setup-soldr-buildcache-v2-, but setup-soldr writes -v3-. So every live compile cache, the Dylint one included, was swept as undeclared.
  • CACHE-005 "poisoned": soldr verdict records (the ledger that lets soldr dylint skip a run whose inputs match a passing run) are ~450 B by design, which is under the 1 KB poison floor. So the skip fast path was wiped too.

Fix

Validation

  • ci_lint cache audit against the live FastLED/fbuild cache listing with the new pin:
    • CACHE-001: 15 → 8. The remaining 8 are ~200 B targetcache-thin stubs.
    • CACHE-005: 46 → 9.
  • ci_lint precheck --local (with PyYAML): identical findings before and after.
  • render_workflows.py --check passes; ci.test_fractional_workflows and ci.test_local_dylint_gate pass.

The upstream ci.yml setup-soldr:build-cache shape still names -v2-; that should be fixed in zackees/ci.yml separately.

…erdicts

Every Dylint run restored no build cache and got 0% zccache hits
(hits=0 misses=496 on each main run), so the nightly workspace check
recompiled from scratch for ~16 minutes. The cache was saved fine; the
end-of-queue `ci_lint cache janitor` then deleted it:

- CACHE-001 "undeclared family": `[cache.family.compile]` used
  `via = "setup-soldr:build-cache"`, which ci_lint still resolves to the
  `setup-soldr-buildcache-v2-` prefix. setup-soldr writes `-v3-`, so
  every live compile cache (Dylint's included) was classified as
  undeclared and swept. Declare the literal `-v3-` prefix instead.
- CACHE-005 "poisoned": soldr verdict records (the ledger that lets
  `soldr dylint` skip a run whose inputs match a passing one) are ~450 B
  by design, below the 1 KB poison floor. Declare `min = "0B"` on the
  verdict family and bump the linter pin to 635fb12, the first ci.yml
  revision where a declared `min` replaces the floor (#417).

Live audit against FastLED/fbuild with the new pin: CACHE-001 15 -> 8
(all remaining are tiny target-cache stubs), CACHE-005 46 -> 9.
`ci_lint precheck` findings are unchanged.
@coderabbitai

coderabbitai Bot commented Nov 18, 2026 •

Copy link
Copy Markdown

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • Trigger review
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

ci-minimal.yml's queue-cache-janitor (the one that swept the Dylint
build cache) checks out its own pin, 953c4e1, which the first commit
missed. Move it to 635fb12 (38 commits ahead, same lineage) so it reads
the fixed compile/verdict families.
@zackees
zackees added this pull request to the merge queue Nov 18, 2026
Merged via the queue into main with commit f98106b Nov 18, 2026
30 checks passed
@fastled-project-sync fastled-project-sync Bot moved this to Triage in FastLED Tracker Nov 19, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Triage

Development

Successfully merging this pull request may close these issues.

1 participant