Closed alex-dabija closed 2 years ago
Empty cluster:
aws-node: 30m 60mb
calico-node: 250m 150mb
cert-exporter 50m 50mb
csi-volume: 10m 50mb
kiam: 15m 100mb
kiam-wetchdog: 200m 200mb
kubeproxy: 35m 100mb
net-exporter: 50m 75mb
node-exporter: 50m 75mb
---------------------------------------------
690m 860mb
Some apps seem to consume much less than the reserved resources specially kube-state-metrics
In deu01:
aws-node: 30m 150mb
calico-node: 50m 450mb
cert-exporter 50m 50mb
csi-volume: 10m 100mb
kiam: 15m 100mb
kiam-wetchdog: 200m 200mb
kubeproxy: 1000m 200mb
net-exporter: 50m 75mb
node-exporter: 50m 75mb
---------------------------------------------
1455m 1400mb
4 CPU cores, 16 GB RAM is the recommended size, which seems to be fine, although I think 2xlarge could improve resource usage.
Reserved resources in xlarge 17.25% 5.3% Reserved resources in 2xlarge 8.62%% 2.68%
Story
-As a cluster admin, I want the smallest node size to have enough resources to run a few workloads which are not deployed by default on a new cluster in order to have a better user experience and not wonder why the pods are not running.
Background
Recent AWS releases have more default applications running on a new cluster (e.g.
vertical-pod-autoscaller
,kiam-watchdog
, etc.). Unfortunately, we haven't evaluated if the smallest node size still makes sense.This issue became visible during recent incidents where on some nodes there weren't enough resources to run the
kiam-agent
and just a few customer workloads.Tasks