This message was deleted.
# rke2a
adamant-kite-43734
10/03/2024, 12:58 PMThis message was deleted.
c
creamy-pencil-82913
10/03/2024, 3:14 PMSounds like you didn't set requests and limits properly. Rke2 doesn't use a custom scheduler. The Kubernetes scheduler will just put pods wherever they fit based on the requests and limits with respect to the node resources, it does not actually look at realtime utilization.
p
powerful-librarian-10572
10/03/2024, 3:15 PMI didnt do any requests or limit. I just guessed kubes would want to balance out by itself either ram or pods, or both. (its not doing either on my cluster)
c
creamy-pencil-82913
10/03/2024, 3:16 PMlol no it absolutely will not
p
powerful-librarian-10572
10/03/2024, 3:16 PM🥲 well damn
powerful-librarian-10572
10/03/2024, 3:16 PMCuz like apart from my outlier node you can see that the ram usage is rather balanced
c
creamy-pencil-82913
10/03/2024, 3:16 PMYou need to explicitly tell it what resources should be available to your workloads, and it will schedule to nodes based on that.
creamy-pencil-82913
10/03/2024, 3:17 PMIt will not rebalance after the fact either, for that you can look at the descheduler project
p
powerful-librarian-10572
10/03/2024, 3:18 PMNo i don't necesiraly look for re-balancing, but at least stop scheduling new workloads on the currently most used node 🥲
powerful-librarian-10572
10/03/2024, 3:19 PMThe reason the node went down (showing on the graph) is because a new pod was scheduled on it, causing memory constraint
It had 4 others nodes it couldve gone to, but didnt
c
creamy-pencil-82913
10/03/2024, 5:29 PMyeah but you didn’t tell it that it had memory constraints. the scheduler does NOT look at metrics at all when scheduling. It just looks at node resources, and compares them to the pod requests/limits