Versions of CoreDNS < 1.8.4 contained a bug that caused non-ready pod IPs to be returned when doing a DNS lookup for a headless k8s service.
CoreDNS v1.8.4 contained a fix for this, which caused the initContainers of our cilium-etcd pods to start failing, as the DNS lookups were no longer succeeding.
I think etcd-operator will need changing to set the publishNotReadyAddresses field in the k8s service spec.
Seeing as https://github.com/coreos/etcd-operator is no longer actively developed or maintained, would it make sense for Cilium to fork that repo and publish their own docker image (with the aforementioned etcd-operator fix)?
Versions of CoreDNS < 1.8.4 contained a bug that caused non-ready pod IPs to be returned when doing a DNS lookup for a headless k8s service.
CoreDNS v1.8.4 contained a fix for this, which caused the initContainers of our
cilium-etcd
pods to start failing, as the DNS lookups were no longer succeeding.I think etcd-operator will need changing to set the
publishNotReadyAddresses
field in the k8s service spec.Seeing as https://github.com/coreos/etcd-operator is no longer actively developed or maintained, would it make sense for Cilium to fork that repo and publish their own docker image (with the aforementioned
etcd-operator
fix)?