union.ai/docs/…/overview#… Hey guys, we are thinking about using flyte2, but wanted to understand deployment options better. We don't currently have Kubernetes, so I wanted to know what are available options are? Are there any options similar to a docker compose option, or would binary be the best option + what are the downsides (e.g. can we deploy without taking down running jobs? etc...) Context: We use Dagster for our data pipelines

Hi Kevin, flyte2 depends on k8s for now. But if you would like to try it out, you can just run

flyte start devbox

to start a k3d cluster without any additional k8s settings. Here is the doc to teach you how. union.ai/docs/…/running-devbox

Hi Alex. Ah I see. How bad would it be if we used k3d on a single VM with flyte2 for production. (obvs not ideal, but these are due to legacy infra constraints - ideally we might still like to use this in the meantime until that all gets sorted out is why)

Hi Alex. Ah I see. How bad would it be if we used k3d on a single VM with flyte2 for production.

Thanks for letting me know your thoughts. The execution performance mainly depends on how much computing resources you assigned to your k3d cluster I would say. Flyte will assign a k8s pod for each of your training task, which means the more resources your cluster have, the more tasks you can run in parallel. Hope it helps.

Ok, perfect. Cheers!

Feel free to let us know if you need help

Thx!

@clean-diamond-78188 one option is Union. We offer BYOC which needs no k8s management

Hey, thanks for message. We are testing running it on AWS EKS now, so kubernetes may not be a problem. Should go well hopefully

Hey @clean-diamond-78188. Feel free to use this skill, it will help you to deploy on EKS super easy

cheers!