I think this is fine. Two questions from an operations point of view:

1. how easy is it to revert this in case "weird things" start happening? Can use the "revert PR" feature? As long as we have a plan for this eventuality I'm happy. I think it is smart to prepare a bit instead of having to make it up on the fly while stressed because stuff is broken.
2. do the people who run the plausible instance know who to talk to on the scikit-learn side (Loic?) in case they notice something weird (e.g. load goes through the roof)?

I'd propose merging this in the European morning as that is presumably when the load is lowest, and we have a whole EU day + US day to watch it.

Reverting this PR seems like the easy fix if indeed things go badly. @stefanv seemed to say #25547 (comment) that even if the analytics server goes down, it will not prevent scikit-learn.org to load. IIUC that means that at worst we will not have any analytics data for Plausible for some period of time (potentially other projects will be affected too if the analytics server goes down completely not sure ...). @tupui in the Scientific Python Discord also indicated that the scikit-learn.org load is similar to the scipy.org and I scipy.org is already using the Scientific Python plausible instance.

About communication with the Scientific Python people, I guess we can let @stefanv and @jarrodmillman (who can let us know if there are better contact people for this) when we plan to merge this. I am fine being pinged on a github issue (or Scientific Python Discord) if something goes bad and the best short-term fix is to revert this PR.

@lesteve just FYI. As Stéfan indicated in the other issue/PR, there was an outage of the server at some point and I confirm that it did not impact at all the loading of the pages on SciPy (I actually only noticed the outage as I wanted to get some metrics 😅). You could also test that locally by pointing to a random URL.

I'd propose merging this in the European morning as that is presumably when the load is lowest, and we have a whole EU day + US day to watch it.

If you ping me on the Scientific Python discord before merging, I am happy to monitor the roll-out.

Do you want to propose a datetime @lesteve?

OK I have contacted Stefan by Discord and will let you know. I would say if today does not work, it is probably better to postpone until next week to avoid making such a change on a Friday 😉.

Keeping an eye on the server; thanks for coordinating!

So Plausible has been on for ~8 hours on scikit-learn.org/stable, everything seems to work 🎉.

Last time I heard from Stefan the Plausible server seemed to behave just fine.