[NO MRG] Retest gpuCI

[jakirkham]

Opening this PR to test gpuCI now that PR ( #315 ) has been merged

cc @GenevieveBuckley @charlesbluca

[charlesbluca]

rerun tests

[jakirkham]

Thanks Charles! 🙏

Is there typically a lag between when a CI fix is merged and when other PRs can get the fix?

[charlesbluca]

Yeah, currently the images are made on a nightly basis, so any changes to the environment files would only be reflected once the next scheduled run is complete; in this case, I manually triggered a rebuild of the images so that we could more immediately test the change.

Perhaps a good compromise between minimizing runtime and reflecting changes to the environment files would be to run mamba update -f $ENV at the beginning of each run?

[jakirkham]

Ah ok. Thanks for the context (and the assist)! 🙏

That seems reasonable. Would that go after this step or somewhere else?

dask-image/continuous_integration/gpuci/build.sh

Lines 36 to 38 in c62a503

	gpuci_logger "Activate conda env"
	. /opt/conda/etc/profile.d/conda.sh
	conda activate dask_image

[GenevieveBuckley]

Thank you @charlesbluca and @jakirkham for tying to figure this out!

[charlesbluca]

Would that go after this step or somewhere else?

I don't think it should matter, but I suppose updating before activating the environment wouldn't hurt - I can open up a PR trying this out to see if there's any immediately obvious issues we would run into

EDIT:

Opened #318 which seems to work well, if that seems like a sensible solution to the lag between environment updates being reflected on the GPU CI side happy to try this out for the other Dask projects where this has also occasionally been an issue

[jakirkham]

Toggling to retest with PR ( #318 ) merged

[jakirkham]

Seems like things are working now as PR ( #309 ) also passed

Closing this out

Thanks again Charles! 🙏