feat: Add support for PDBs on deployment and statefulset

[JorTurFer]

Fixes: #2136

[JorTurFer]

Sorry, I missed the bundle during the first try. I have updated the PR

[JorTurFer]

awesome, a missing change in the stage 🤦

[jaronoff97]

Overall this looks good! Thank you for updating the logic for version detection as well to be more generic. Could you also update one of the e2e to set the PDB and ensure it gets reconciled correctly and the selector works as expected?

[jaronoff97]

note for other reviewers, this is a paired down version of the main PDB spec because we don't want to allow the configuration of the selector

[JorTurFer]

Thanks for the quick review, tomorrow I'll address al the comments 😄

[JorTurFer]

I have addressed all the comments. I'm not 100% if I have done the e2e test correctly because it's the 1st time I have used kuttl but at least it works in local 🤣

[JorTurFer]

It looks like I have to review the e2e test to work on multiple cluster versions. Let me do it before triggering again the CI because the check will fail again. I'm on it

[JorTurFer]

@pavolloffay ,
What should I do? I mean, should I wait until #2145 is decided? Will it require too much time? I'm not in rush with the pdb, during next month is enough for me (the natural month, not just September xD), but if the discussion goes longer I'd prefer to merge it with support to both if It's possible

[pavolloffay]

What should I do?

Let's see how quickly goes the conversation in #2145 I am happy to open a PR next week to bump the version.

[JorTurFer]

I have just rebased main branch and updated this PR. PTAL when you have some time @pavolloffay @jaronoff97

[jaronoff97]

I think we may want to add something in the validating webhook to be sure that we don't attempt a bad config

[pavolloffay]

End-to-end tests / End-to-end tests (1.23, e2e-pdb) (pull_request) - failed

[JorTurFer]

I'm spinning up a cluster with that version to check it

[JorTurFer]

@pavolloffay , I've tested it locally with k8s 1.23 and (at least locally) the feature works:

The problem I have seen is that the deletion is producing timeouts

[JorTurFer]

I've noticed that if I use my own namespace for the tests, the k8s GC removes the namespaces more quickly and I can also validate the generated labels, so I've updated my test to use it's own namespace, and now 30 seconds of timeout is more than enough (the test needs a few seconds indeed, so far from 30 seconds, at least locally).

I say this because I see other e2e tests with > 150 seconds of timeouts and maybe the root cause it's something like this. The cause of failures was that, timeout deleting the namespace :(
I can see pdb events and a timeout at the end

[pavolloffay]

The PR looks good to me. But I am also interested in @jaronoff97 question whether we could do some validation of the PDB fields that we added to our CR.

[JorTurFer]

Which kind of validation? I mean, I added a check in webhooks to validate if both have been set, and the other validations are already done by the openapi spec (but I can be missing any important thing at some point)

[jaronoff97]

@JorTurFer thank you so much for your contribution 🎊🎉

[JorTurFer]

You're welcome!
Thank you for accepting it, I just did it because I need it 🤣