Question about caching: Why mypy cache is under named caches and not the lmdb cache? Intended behaviour? I wanted to leverage the remote cache so we can get faster typechecks.

There's two layers of caching in Pants: • LMDB caches exact matches: re-invoking a process with the same files, args and env vars => serve from cache, don't actually execute it. • Named caches are designed to help with partial matches: invoking a process with (slightly) different files, but still able to benefit from previous work. Pants' Mypy uses both: • if you re-run on exactly the same files, mypy isn't even invoked • if you make a change and rerun, it's not an exact match, but mypy might still benefit from its work from a previous run, which pants tells it to store as a "named cache"

OK. Problem is that the mypy cache keeps growing like crazy. Is there any tips to optimize? Or any way to send the lmdb cache to the remote caching service?

(The contents of lmdb cache is what is cached by a remote cache, I’m going to guess you mean a named cache.) Not sure about optimising. The only general one I can think of is to preserve it usually but regularly clear it when it is too large, so most builds benefit but not all: https://www.pantsbuild.org/2.21/docs/using-pants/using-pants-in-ci “nuking the cache when too big” You could investigate mypy-specific optimisations, eg understand what is in the mypy cache and work out what can be deleted (etc.). This’ll be a balance of pants and mypy behaviour, but mostly mypy I suspect. Re named caches -> remote cache, I don’t think there’s a current possibility for that.