👋 hullo! we're thinking through our infrastructure story with prefect right now, and are aiming to deploy to EKS. a question i have is, is there a standard model of # of agents to # of nodes in a cluster? Or should you be having a single agent per flow that you want to execute in the cluster? Also apologies in advance if there's already a recipe or something i should be looking at 🙂
10/21/2022, 10:50 PM
Hi Jai! There’s no right or wrong answer here, but probably the easiest way to think about things is having one agent per execution environment. The agent is then the intermediary to that environment and the work that is shipped to it.
One agent can service multiple work queues, so your work queues can be as granular as you like - I recommend grouping deployments into a small number of work queues so that if you ever need to pause work delivery, or limit how many things are running you can do that at the work queue level.
TLDR: work queues should represent multiple deployments that do similar things and can be paused / limited together, and agents represent the places those things run.