On large Rust repositories, it is easy to exceed GitHub's default cache limit of 10GB. I've recently been working on optimising our CI pipeline over at rust-libp2p. Here is what I landed on:

Our CI fans out testing into one job per crate. Applying this action naively completely busts the cache limit of Github because we would be saving and restoring 20+ caches per workflow run.

To combat this, I created a "cache factory" workflow that only runs on master and compiles the entire workspace: https://github.com/libp2p/rust-libp2p/blob/master/.github/workflows/cache-factory.yml. It is a bit more complicated in our case because we have different caches for the MSRV check and the actual testing but you get the overall idea.

More importantly, we still have other workflows (clippy, rustdoc, cross-compile to wasm etc) that always run on the entire workspace. Those use this action as is. That still creates a lot of caches, hence I am now adding

with:
  save-if: ${{ github.ref == 'refs/head/master' }}

to all cache invocations: libp2p/rust-libp2p#3249

Do you have any thoughts on that? I think that could actually be a safe default to add this action.