Seems like I'd have to stick with 1.0 for now then. Unfortunately, it wouldn't make sense for us cost-wise to run the agent on ec2 and have it execute flows on the same instance, because we are running ML training flows. Would prefer to have a small ecs instance running the agent and have it launch large (+gpu) ECS tasks for training.
a
Anna Geller
06/17/2022, 12:33 PM
did you read the link I sent?
there is explanation how you can still get all the cost benefits of Fargate using EKS on Fargate
m
marque
06/17/2022, 12:36 PM
Yes, but our team also has similar doubts regarding kubernetes as Julian, we're pretty hesitant to jump over towards adopting kubernetes since we have a lot of existing Infra & app code set up for ecs. Doesn't help that there's no one in our team that has experience with kubernetes.
a
Anna Geller
06/17/2022, 12:41 PM
that makes sense, totally understandable - I'm planning on releasing a tutorial showing how you can still use ECS Fargate with Prefect 2.0 - if you want to get notified, subscribe to this tag on Discourse https://discourse.prefect.io/tag/ecs