Arvados

Update at 7812de8 - branch 22073-salt-v3006-downgrade

After some googling trying to understand what "Recursive requisite found" even means, I've found an open 8 months old Github issue that seems to indicate that salt v3007 has some kind of bug.
Doing some git archeology I've found that we upgraded to this version in our story about adding support to Ubuntu 20.04 (#21363), the specific commit is 08a4a99 and its message suggests that v3007 is needed because of Ubuntu 24.04.
Just in case, I've first tried the current installer with salt v3007 on AWS with Ubuntu 20.04 AMIs and it failed with the same message. Then, following the advice given in the Github issue, I tried deploying on Ubuntu 20.04 from scratch with salt v3006 and it worked OK so I think it's safe to use this version until v3008 comes out with a proper fix.

On a technical level, I have no problem downgrading. But we need to figure out how we want to document and message this.

My rough understanding is that we say we "support a distribution" if you can run the multi-node Salt installer on that distribution and get a working cluster. Based on your diagnosis, we can't say that for Ubuntu 24.04, not because there's anything missing on our end, but because of this Salt bug. The only version of Salt that works on Ubuntu 24.04 is 3007, and then this bug in it means our installer doesn't work as intended.

So: are we taking Ubuntu 24.04 support out of the release notes? Are we removing it from the list of distributions with packages (e.g., doc/_includes/_supportedlinux.liquid)? These are the kinds of things we should discuss and decide before we merge.

Sorry, I meant to write that I tried Ubuntu Noble 24.04 with salt v3006 (I think it was 3006.9) and it worked fine.

Here's the successful Jenkins run: test-provision-ubuntu2404: #18

Ubuntu 24.04 support was added in Salt 3006.9, released July 29. Git log says I was working on Ubuntu 24.04 support in early June, so that's what changed between now and then to let 3006 work again.

So sure, great. It would be good to have a run of the full test-provision suite to make sure there are no regressions on any other distros (I'm not expecting any, but still), and if they all pass I think we're good.

Running the entire test-provision pipeline: test-provision: #948