Ah. What's going on here is that, for some reason, Prefect stopped respecting the

job_variables: image

parameter and is just using the default image

Can you share the relevant snippets of the

prefect.yaml

file? What version of Prefect CLI are you using?

prefect version

@Brendan Dalpe 🫡

pull:
- prefect.deployments.steps.set_working_directory:
    directory: /opt/prefect/salad

deployments:
- name: Binana
  requires: prefect-gcp prefect-docker prefect-ray
  version: null
  tags: []
  description: null
  schedule: null
  entrypoint: ./src/binanaflow.py:binanaflow
  parameters: {}
  work_pool:
    name: "GCP Cloud Run, Prod."
  work_queue_name: null
  job_variables:
    image: "{{ PREFECT_IMAGE }}"

This is the relevant part of my

prefect.yaml

(PREFECT_IMAGE is set to an image stored on our artifact repo).

root@cf9d2ee6c5c3:/opt/prefect# prefect --version
3.4.8

Ah, move

job_variables

under

work_pool

like this:

pull:
- prefect.deployments.steps.set_working_directory:
    directory: /opt/prefect/salad

deployments:
- name: Binana
  requires: prefect-gcp prefect-docker prefect-ray
  entrypoint: ./src/binanaflow.py:binanaflow
  work_pool:
    name: "GCP Cloud Run, Prod."
    job_variables:
      image: "{{ PREFECT_IMAGE }}"

... 😂 No way. Let me try that now

Take a look at reference here: https://docs.prefect.io/v3/how-to-guides/deployments/prefect-yaml#templating-options

We're installing them @ runtime, yeah. If this works, it'd frustratingly explain why a couple of our other deployments are working fine, and why this one suddenly stopped working

Try this:

pull:
- prefect.deployments.steps.set_working_directory:
    directory: /opt/prefect/salad

deployments:
- name: Binana
  requires: prefect-gcp prefect-docker prefect-ray
  entrypoint: ./src/binanaflow.py:binanaflow
  work_pool:
    name: "GCP Cloud Run, Prod."
    job_variables:
      image: "{{ PREFECT_IMAGE }}"
      env:
        EXTRA_PIP_PACKAGES: "prefect-gcp prefect-docker prefect-ray"

Oh! Good call that'll serve to make the container a lot smaller (there are a couple of different deployments that use the same base image)

Happy to help 🙂