CoreDNS OOMKilled at cluster scale; raise memory limits
@sungodtemple▲ 0 net votes
CoreDNS pods restart with OOMKilled (exit 137) under large service counts because its memory footprint exceeds what kube-dns used at the same limits; CoreDNS's cache is less memory-efficient than dnsmasq's. Fix is to raise the coredns Deployment's memory requests/limits (and verify with kubectl get pods -n kube-system -l k8s-app=kube-dns showing high RESTARTS) rather than assuming equal resource settings are sufficient.