Hi all, question regarding deploying flows to work pools and where the information about infrastructure is being pulled from. My setup is currently an AWS ECS Service running a worker listening to a work pool (which creates the pool when it doesn't exist) in Prefect Cloud. I am using the

prefect

and

prefect_aws

libraries in Python to create an "anonymous"

ECSTask

infra block which is used in the deployment via

prefect.deployments.Deployments.build_from_flow

. When I run the flow via "quick run" the run uses the infrastructure block/definition in the work pool instead of the deployment (of which there is none/minimal due to it being created by the worker). When I manually update the work pool configuration with the infra details (cluster name, image, etc). it springs to life and works as expected. Is there a way to have it utilize the deployment infra configuration instead or am I missing something (which is totally possible!)?

Hey @Joey Allison! When you’re using a worker, it will use the configuration from the work pool and ignore the infra block on any deployments. You can think of a worker as an agent + an infra block.

Ah ha! Okay, is there a different setup for having a "serverless" ECSTask that utilizes the infra details of the deployment? Is that adviseable or is it "best practice" to utilize work-pools going forward?

We recommend using work pools and workers moving forward. Since you’re using

Deployment.build_from_flow

, you can use

infra_overrides

to define infra details for your deployment (equivalent to

job_variables

in the new

flow.deploy

method). That will allow you to override any of the variables configured in your work pools base job template.

Fantastic! I'll give that a go. Thank you for all your help!