we are passing `--python-setup-resolver-cache-dir`...
# general
we are passing
to a directory that is shared between the 8 CI worker processes running on each box and is not cleaned out between runs, but maybe that option changed?
Circling back to this detail,
is still used so even though we're falling through Pants
reads and continuing to resolve, the resolve itself should be hitting the resolver cache ultimately used by pex. It may just be that the overhead still left is large compared to the prior python stack in <1.3.0. That older stack avoided re-building python chroots - and re-resolving, using
. The new stack does not use that option. Were you perchance also sdetting that option to a global dir in CI?
we nuked the chroot cache between CI runs because without separate requirements/source pexes they’d be invalidated all the time anyway, but we didn’t see pex creation taking quite this long