I’m trying to understand how

KubernetesAgent

(code here) uses the environment variables used in the k8s deployment, particularly

IMAGE_PULL_POLICY

, to build the podspec for a flow run. It looks like if a flow specifies

run_config=KubernetesRun(…)

at all, then

IMAGE_PULL_POLICY

from the deployment env is ignored? Is that right?

I believe that is right. RunConfig will take precedence over the agent env variables

It’s a little surprising to me because if

IMAGE_PULL_POLICY

is specified in the k8s deployment and the

KubernetesRun()

object doesn’t specify

image_pull_policy

, then it should fall back on the deployment env. But that’s not the case I think, right?

Ah ok I see what you are saying. Yes they should be field by field. Are you supplying a job template too with KubernetesRun?

I think

container["env"] = container_env

doesn’t matter for creating the podspec, as those are environment variables and

imagePullPolicy

is under

containers:

in the podspec. But it is actually working like I wanted it to. my flow passes:

run_config=KubernetesRun(
    image="<http://registry.digitalocean.com/tdc-container-registry/prefect-flow|registry.digitalocean.com/tdc-container-registry/prefect-flow>",
    labels=["any"],
  ),

and my prefect-agent deployment is configured:

- name: IMAGE_PULL_POLICY
          value: 'Always'

and I see on a running job:

imagePullPolicy: Always

so, I guess it is actually working as intended. I don’t totally see how from reading the code 😅