Brian Grant

Why is #kubernetes PodDisruptionBudget a separate object when Rollout Strategy is inlined into Deployment? You need both to be HA.

We can chat about this sometime. Use cases are different: disrupting yourself vs being disrupted. I would like minReadySeconds to be on Pod, though

I get that the consumers are different, but the producer of this info is always the workload operator, IME. So it makes for worse UX having to configure in two places.

One issue is that, like Service, PDB sometimes/often needs to span multiple controller instances. Also, it wouldn't reduce the object count by much. And there hasn't been much demand. Is it otherwise working well for you?

PDB works ok when you define it. There’s just too many things for K8s tenants need to know about to stay highly available. I wrote an internal article, but I should make it a blog. It’s like 10 things to configure.

It would be nice to be able to define a default PDB tho. I’m not exactly sure of the default behavior but I often have new tenants forget PDB and then have outages during the first node pool upgrade.

I vaguely recall a proposal for that. I'll look/inquire. In the meantime, the main focus is GA/stable: https://github.com/kubernetes/enhancements/blob/6f42c8560c2c36fde3730fcb0d882b2451cdc00d/keps/sig-apps/20190318-PodDisruptionBudget-graduation-to-stable.md …