Markus Thömmes
Markus Thömmes
Yes I'd expect us to hit 100 pods requested (deployed is another story :sweat_smile: ) almost immediately.
The rest of this is not a good first issue anymore and requires a fair bit of caution. I'm happy to help people through it, but it's much more involved....
Ack, allowing the probes seems okay for sidecar containers.
FWIW, I toyed a bit with a PostStart hook executing our binary and that seems to perform in the same ballpark as containers without probes at all.
This has some great content on the tradeoffs we'd buy into when using lifecycle hooks: https://livebook.manning.com/book/kubernetes-in-action/chapter-17/64. In a nutshell: Error reporting seems worse, so we prolly need to adjust the...
Tripleposting :joy: I did investigate the behavior of the postStart hook briefly, and sadly, I was right above: The pod isn't getting any IP while the post-start hook is running,...
Thanks for the writeup @jonjohnsonjr. That makes a lot of sense to me!
/remove-lifecycle stale
In the same vain, we should likely switch the probes into a workqueue based solution (like net-XXX probers) to be able to more easily control the lifecycle of the respective...
Not sure we talk about the same thing @vagababov? What does the activator have to do with pod probing in the autoscaler? :thinking: