Browse Source

fix: use evictionHard to reserve node memory to avoid systemd mess

pull/39/head
Stefan Reimer 10 months ago
parent
commit
5a38a038a1
  1. 11
      charts/kubeadm/templates/KubeletConfiguration.yaml

11
charts/kubeadm/templates/KubeletConfiguration.yaml

@ -20,13 +20,16 @@ tlsCipherSuites: [TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256,TLS_ECDHE_RSA_WITH_AES
featureGates: {{ include "kubeadm.featuregates" ( dict "return" "map" "platform" .Values.platform ) }}
# Minimal unit is 50m per pod
podsPerCore: 20
# cpuCFSQuotaPeriod: 10ms
# Basic OS on Ubuntu 20.04 incl. crio
systemReserved:
memory: 256Mi
#systemReserved:
# memory: 256Mi
# This should be dynamic based on number of maxpods and available cores
# https://cloud.google.com/kubernetes-engine/docs/concepts/cluster-architecture#memory_cpu
# Below are default minimal for 2 cores and minimum kubelet
kubeReserved:
cpu: 70m
memory: 128Mi
# cpuCFSQuotaPeriod: 10ms
# memory: 128Mi
# Lets use below to reserve memory for system processes as kubeReserved/sytemReserved doesnt go well with systemd it seems
evictionHard:
memory.available: "484Mi"

Loading…
Cancel
Save