-
-
Notifications
You must be signed in to change notification settings - Fork 155
Description
Describe the issue:
The library makes a strong assumption that ClusterDomain is the same for any cluster:
dask-kubernetes/dask_kubernetes/operator/controller/controller.py
Lines 176 to 178 in 1daa175
scheduler_env = { | |
"name": "DASK_SCHEDULER_ADDRESS", | |
"value": f"tcp://{cluster_name}-scheduler.{namespace}.svc.cluster.local:8786", |
and passes DASK_SCHEDULER_ADDRESS
that always ends with .cluster.local
. However, this part can be configured in CoreDNS to a different domain. It is useful when multiple Kubernetes clusters are set up and require unique DNS records and the ability to communicate with each other.
However, the pod tries to reach out to any other pod in the same cluster and the same domain, we don't have to specify the FQN of the cluster domain:
https://kubernetes.io/docs/concepts/services-networking/dns-pod-service/#:~:text=In%20summary%2C%20a%20Pod%20in%20the%20test%20namespace%20can%20successfully%20resolve%20either%20data.prod%20or%20data.prod.svc.cluster.local
Minimal Complete Verifiable Example:
Change the ClusterDomain in CoreDNS configuration to something different:
https://kubernetes.io/docs/tasks/administer-cluster/dns-custom-nameservers/
Environment:
- Dask version: 2024.12.1
- Python version: 3.11
- Operating System: WSL
- Install method (conda, pip, source): poetry