Open shreyas-s-rao opened 1 month ago
can you also add what is the gap that you have seen which causes issues when using kupid/kyverno? Also can you adjust your description as these are not really schedulers.
The label was not added in #883 , as it breaks certain scheduling decisions when topology spread constraints are orthogonally added to the statefulset without going through the Etcd spec, especially during the scale-up scenario from 1 replica to 3 replicas. This is because, the single-replica pod will have cluster-size=1
, while the two new pods that come up upon scale-up, will have label cluster-size=3
, since pod-0 is rolled only after the two new pods come up.
How to categorize this issue?
/area usability ops-productivity /kind enhancement
What would you like to be added: Druid should add a new label to etcd pods to indicate the cluster size, or whether it's part of a single-node or multi-node cluster.
Why is this needed: This allows for better scheduling decisions by orthogonal schedulers like kupid or kyverno. The label can be named
druid.gardener.cloud/etcd-cluster-size: N
for example.