Description
Is your feature request related to a problem? Please describe.
Yes. Basically, the current situation is that we need to first provision a node, wait for it to be ready, change the needed settings, and carefully restart pods after restarting the containerd service : this process takes time, impacting the overall provisioning time seen by the end-user. It is also fragile.
Optional: Set the appropriate GitHub label(s) to indicate the AKS feature or area your feature request relates to.
Describe the solution you'd like
Allow some control on containerd settings, so that directly at provisioning time, the node given has already the expected settings set. This results in no brittleness, low provisioning time as seen by the end-user.
Describe alternatives you've considered
We have attempted utilizing a daemonset. The daemonset approach is not viable as it results in:
- running pods being killed/restarted
- increasing the provisioning time (to the state where the pods can do the job needed)
Additional context
This change would allow us to support huge model for our big numerical solvers in Compute.