Skip to content

Latest commit

 

History

History
39 lines (31 loc) · 2.79 KB

TODO.md

File metadata and controls

39 lines (31 loc) · 2.79 KB

TODO

Containerd/Kubelet

Max Pods

  • Modify the max-pods calculation
    • One pod on a large instance should not carve out enough mem/cpu as if that instance will host the max-pods when its only intended to host a single pod (Spark workloads)
    • When utilizing GPUs/accelerators, max-pods is a combination of the number of GPUs attached to the instance and any concurrency configuration (time-slicing, MiG, etc.)
    • What algorithm is appropriate for allowing users to maximize the number of pods on an instance (i.e - can they schedule 737 pods on an instance safely and without resource contention?)
    • What tests can be used or constructed to test/validate these changes?
    • Karpenter

Wasi

(More from the aspect of - how does supporting this plus other similar methods affect the design of eksnode?)

"Warning!! The following files are not covered by default logrotate settings ensure they match site policy"\n"/etc/logrotate.d/dnf, /etc/logrotate.d/btmp, /etc/logrotate.d/wtmp, /etc/logrotate.d/aide, /etc/logrotate.d/firewalld"