Hey all, we are discussing the integration of Flyt...
# ask-the-community
Hey all, we are discussing the integration of Flyte in our group at the moment. Reading the documentation and slack I have the impression that typically, Flyte is deployed on its dedicated cluster. Is it possible to deploy it on an existing cluster which runs other applications? Are there solid ways to separate resources? I saw the
keyword in the
, is that an option?
Can just say from our own experience, we're running mixed clusters for Flyte as well without any major issues.
Definitely works, you could use node selectors as mentioned to keep them on separate nodes if you need to, however keep in mind that large Flyte runs could potentially create some strain on the k8s API server/etcd (creating larger amount of CRDs and increasing the number of API calls to k8s to keep track of the state)
Yep, we've got Flyte & Kubeflow in the same EKS cluster right now as we migrate
Best ways to separate resources are 1) via an appropriate choice of namespaces and 2) using nodeSelector/taints and tolerations if you want to force Flyte to schedule on different nodes. You may also be able to get away with letting the Flyte service components schedule on any node and limit actual task pods to a certain instance group/node group, but we've found with our autoscaler we get the best perf when we place as few constraints as possible on a k8s workload
Thanks a lot for your answers, that clarifies how we could run it 🙏