search

openforcefield / openff-qcsubmit

Automated tools for submitting molecules to QCFractal
https://openff-qcsubmit.readthedocs.io/en/latest/index.html
MIT License
26 stars 4 forks source link

QCSubmit CI is failing with QCF 0.54 #275

Closed j-wags closed 2 months ago

j-wags commented 3 months ago

Nightly CI has been failing for the last 3 days. Seems to be something with pydantic object parsing.

bennybp commented 3 months ago

This looks to be something with the new QCPortal. It's not fetching all the fields required for some reason. I don't see much wrong with qcsubmit. Let me dig into it a bit

bennybp commented 3 months ago

Ok I do see a few issues with the v0.54 release. The good news is that it is largely an artifact with how the new caching mechanism can result in caches shared between tests. So overall QCSubmit should largely work regardless of the failing tests.

Let me think of a way to more properly disable cache sharing during tests

(The entry_points errors from a few days ago are related to Python 3.9 compatibility with networkx which has already been fixed)

j-wags commented 3 months ago

Ahh, that makes sense. Do you think it'd work if we had the clients we create in our testing use a temporary directory for their cache_dir, or set the cache_max_size to 0?

(and not that it's relevant since the networkx issue is resolved, but since we follow NEP29 we're not going to be testing 3.9 any more)

bennybp commented 3 months ago

I've got a PR up that should fix these issues (https://github.com/MolSSI/QCFractal/pull/816). Would it be possible to test this somehow before I make a release?

If it works, I can do a quick v0.54.1 release

mattwthompson commented 3 months ago

I threw up an attempt at testing against that branch: https://github.com/openforcefield/openff-qcsubmit/pull/277

I'm less experienced with monorepo deployment so it might take a couple of iterations to get set up correctly

j-wags commented 3 months ago

At least one of the tests (openff/qcsubmit/_tests/results/test_filters.py::test_torsion_drive_record_filter_apply[result_filter0-expected_ids0]) fails even if it's run alone with QCP 0.54 (and it passes with 0.53), so I'm going to apply a metadata patch to the existing conda-forge packages.

j-wags commented 3 months ago

Hm, on second thought the test failures all seem to be due to test fixtures having problems, so I'll hold off on the patch.

j-wags commented 2 months ago

Resolved by #277