3 ms·
This is sorely needed. It causes service disruption for weeks because of weird upgrade schemes in GKE (or any other vendor) which works on top of Kubernetes. Th
by airocker 3y ago
This is sorely needed. It causes service disruption for weeks because of weird upgrade schemes in GKE (or any other vendor) which works on top of Kubernetes. There are too many options with non-intuitive defaults related to how the control plane and the node pools will be repaired or upgraded. Clusters get upgraded without our knowledge and breaks service arbitrarily. Plus, if you are doing some infrastructure level changes, you have to put in extreme effort to keep upgrading.
IMHO , infrastructure is too low level for frequent updates. Older versions need LTS.
- MuffinFlavored 3y agohow much of what you said is part of the complexity that comes with Kubernetes in general? at the end of the day it’s about cramming a ton of functionality into something like YAML is it not?
- airocker 3y agoOur cluster has reasonable complexity and yamls work well so far. Maybe we will see what you are saying if we scale further. It allows us to maintain a much more complex cluster with fewer developers. And the software is well tested/reliable. But not having LTS is really difficult. It is impossible to keep rewriting things. And all related components: terraform, helm, GKE etc also update too quickly for us to manage with a small team. We do some lower level infrastructure work so these problems bites us harder.