ci: enable micromamba env cache

[jandom]

The micromamba environment creation takes about 5 minutes and is more than the actual build step (2 mins for buliding docs). Here is an example run

https://github.com/MDAnalysis/UserGuide/actions/runs/5622842883/job/15236245154

This is not ideal – developers want to see a crisp output of their work.

I'm not sure how to enable this feature, or how well it works, so using this branch to experiment. If it works well, maybe we can consider porting to MDA core?

Update Here are some preliminary results showing the impact on the build speed, with and without caching

Workflow Name	Caching on/off	Setup Duration	Job Duration
mda_gh_ci	Off	4m 39s	6m 16s
mda_gh_ci	On	21s	2m 6s
mda_gh_ci_tests	Off	6m 29s	16m 39s
mda_gh_ci_tests	On	27s	15m 26s

[IAlibay]

maybe we can consider porting to MDA core?

We can't unfortunately. I already looked into this some time ago. Because of the size of the matrix the amount of data stored becomes excessive. Our build times are tiny so it's not a big deal tbh.

[jandom]

That makes sense – yeah I can see how the build matrix can inflate the artifact size – but it's still a nice improvement to this repo?

[IAlibay]

That makes sense – yeah I can see how the build matrix can inflate the artifact size – but it's still a nice improvement to this repo?

I think so, but I'm trying to properly look at the impact, I might need to punt a complete look to the weekend.

1. There's a branch, develop precendence here that we need to consider. So the improvements only exist if CI ran on develop earlier in the day, or if you ran CI in the PR earlier in the same day.
2. There's a cache clash I'm going to quickly write a review comment about now.
3. We can't change the cache time limit or enforce an override. So it looks as if your PR suddenly requires an upstream fix, you'll need to wait a day until the cache invalidates?

[IAlibay]

There's a branch, develop precendence here that we need to consider. So the improvements only exist if CI ran on develop earlier in the day, or if you ran CI in the PR earlier in the same day.

From initial thoughts, I don't really care about this too much. At worse you get no improvement, at best it helps. It doesn't look like we'd get a regression so that's fine.

We can't change the cache time limit or enforce an override. So it looks as if your PR suddenly requires an upstream fix, you'll need to wait a day until the cache invalidates?

This I care about a lot more. A lot of the userguide is "emergency patch fixes". I am concerned it might slow us down in those cases (unless I'm missing something that allows us to invalidate the cache easily).

[jandom]

I've addressed the cache clash, so that should be fine – good catch!

Very wise re emergency invalidation but I think we can just wipe the caches here? https://github.com/MDAnalysis/UserGuide/actions/caches Optional tho, could be just more faff than it's worth – I'm 50/50 in my head right now

[IAlibay]

I've addressed the cache clash, so that should be fine – good catch!

Very wise re emergency invalidation but I think we can just wipe the caches here? https://github.com/MDAnalysis/UserGuide/actions/caches Optional tho, could be just more faff than it's worth – I'm 50/50 in my head right now

Yeah (edit: deleting) cache manually might be enough. I don't really mind tbh, @MDAnalysis/coredevs does anyone have strong feelings here?

[RMeli]

No strong feelings, but I think that if the cache can be removed manually, then it's OK.

[jandom]

hi there many thanks for all the reviews and the merge (although that one was unexpected)