km-kube-agent Helm chart: make resource requests and priorityClassName configurable
complete
Olivier de Régis
Hi,
In km-kube-agent 2.3.0 (the latest version), container resources are hard-coded in the templates. There is no value to override them and no priorityClassName:
- DaemonSet agent: requests 100m CPU / 256Mi
- Cluster agent: requests 150m / 384Mi
- Fleet manager: requests 100m / 128Mi
On our production EKS cluster (2 vCPU nodes), 2 of the 8 DaemonSet pods have been Pending since install with "Insufficient cpu". With no priority class, they can't preempt anything either, so those nodes are never monitored.
The DaemonSet agent actually uses 2-3m CPU and ~90Mi.
Could you expose in values.yaml, with the current values as defaults:
- resources for each container (DaemonSet agent and its init container, cluster agent, fleet manager)
- priorityClassName for the DaemonSet, and ideally for the Deployments too
Thanks!
A

Amitava Saha
updated the status to
complete
This has been released
A

Amitava Saha
Sure Olivier, we'll add this to our next release in coming week.