https://docs.flyte.org/en/latest/deployment/deployment/multicluster.html#control-plane-configuration in a multi cluster deployment, would each data plane need its own docker registry and s3 bucket?

Why do you need a multi cluster deployment- we recommend this for advanced usecases only

I'm not suuuper deep into the multi cluster setup yet at the company I work, but we have a bucket for each plane. It e.g. stores execution caches from the workflows. We don't have anything specific regarding docker registry besides the one shipped with AWS EKS clusters.

@hallowed-dog-74273 regarding the S3 bucket, and more specifically the metadata store it can be shared or per data plane cluster. In terms of the container registry I think having a separate registry per cluster would complicate things (depending on the workflow registration pattern you use)

@freezing-airport-6809 i will have to run one cluster on-prem, one cluster in cloud, and potentially a dev and prod cluster in cloud

I think there are different layers where the workload placement can be controlled Using cluster execution labels, you can instruct a combination of project/domain to run on a specific K8s cluster (not sure if your workstations run K8s?)

yeah we would run k8s on those workstations

for more granularity, you can make use of the GPU selectors work that hopefully will land on the docs soon https://github.com/flyteorg/flyte/pull/4172 I think depending on the GPU drivers you use, different keys will be exposed that can be used to configure affinities

docs here - https://docs.flyte.org/en/latest/api/flytekit/extras.accelerators.html#specifying-accelerators