Skip to content

Latest commit

 

History

History
34 lines (22 loc) · 1.58 KB

caching.md

File metadata and controls

34 lines (22 loc) · 1.58 KB
Raw

Caching Semantics

Invoked from a GitHub workflow

We cache the contents of ~/.cache/trunk/ if inputs.cache === true.

Note that the GitHub Actions cache only caches results on a per-branch/per-PR basis, so the first time Trunk Check runs on a given branch, it will be uncached. Subsequent runs will read from cache.

Invoked inside the .trunk repo

We only cache the contents of ~/.cache/trunk/tools/ruby/ (Ruby is the only tool install that is painfully slow, since it has to be built from scratch on every single install).

Although we could cache all of ~/.cache/trunk/, doing so would yield no benefit:

  • GitHub starts evicting cache entries once the size of all GitHub Actions cache entries for a repo exceeds 10GiB (documentation).

  • We expect the cache size to be dominated by ~/.cache/trunk/tools/, which for all tools in a repo usually reaches into multiple GiB, even compressed.

  • Since Trunk Check will run for all repos in a given GitHub org in workflows in your-org/.trunk, all such repos end up sharing the .trunk GitHub Actions cache, eventually the sizes of each repo's cache entry will cause them to stomp on each other.

Experimentally, installing all tools from scratch (except Ruby) only takes ~1 min, so we don't believe there will be any noticeable benefit to caching the contents of the tools cache.

We don't expect caching linter results to be a significant improvement either.