testing out 2.15.0a0 - with the new pytest batchin...
# development
s
testing out 2.15.0a0 - with the new pytest batching enabled in the hairiest subtree of our monorepo,
./pants test
of that subtree now runs nearly as fast as raw
pytest
🎉
🚀 3
đź‘– 2
❤️ 3
âť— 2
“nearly as fast” => 62 minutes in pants, 59 minutes in pytest
h
Great work Dan!!
s
I do see a lot of “Resolve transitive targets” in the output of
./pants test
now
h
throw in (remote) caching and
--changed-since
, and hopefully Pants will be faster in most cases
s
wondering if https://github.com/pantsbuild/pants/pull/17394 will help
test
as well as
lint
👍 1
w
it absolutely will
part of why i wanted to pick it up recently. because TT calculation will become a more obvious long pole with remote execution, since one client will be computing them for all tests
👍 1
h
Incredible work!
h
One of the most exciting things imo is you took a vague idea we've talking about wanting to do for 2 years, and turned it into something concrete and useful đź’ś
w
“nearly as fast” => 62 minutes in pants, 59 minutes in pytest
was this with any batching enabled, or was it all tests in the subdirectory in one batch?
s
all tests in subdirectory as one batch
I think pants will probably be faster if I set the batch_size so that we end up with 2-4 batches
but wanted to do the apples-to-apples comparison first
👍 1
following up because it’s so exciting - if I just leave
[test].batch_size
at the default of 128, running all our big tangled django tests takes ~20 minutes locally, which is 3x faster than a vanilla
pytest
of the same dir. so we now have a very attractive lure when trying to get more of our engineers off our old tooling and onto using pants full-time 🎉
❤️ 3
🎉 1