Yes — you can set env vars from the

@kubernetes(...)

decorator, but the key is that you should pass them as

env=...

, not as arbitrary nested Kubernetes manifest fields. Here’s a minimal example:

python
from prefect import flow, get_run_logger
from prefect_kubernetes.experimental.decorators import kubernetes
import os


@flow
def child_flow():
    logger = get_run_logger()
    <http://logger.info|logger.info>(f"MY_ENV_VAR={os.getenv('MY_ENV_VAR')}")


@flow
def parent_flow():
    child_flow()


@kubernetes(
    work_pool="my-k8s-work-pool",
    env={
        "MY_ENV_VAR": "hello-from-decorator",
    },
)
@flow
def kubernetes_subflow():
    logger = get_run_logger()
    <http://logger.info|logger.info>(f"MY_ENV_VAR={os.getenv('MY_ENV_VAR')}")


if __name__ == "__main__":
    kubernetes_subflow()

A couple of important notes: * The decorator signature is

@kubernetes(work_pool=..., **job_variables)

, so

env={...}

is itself a job variable. * For Kubernetes workers, Prefect converts that

env

dict into the container env list in the generated Job spec. * This should behave similarly to setting env vars at the work-pool level, except it only applies to this decorated flow. You can also use dotted job variables syntax like this:

python
@kubernetes(
    work_pool="my-k8s-work-pool",
    **{
        "env.MY_ENV_VAR": "hello-from-decorator",
    },
)
@flow
def kubernetes_subflow():
    import os
    print(os.getenv("MY_ENV_VAR"))

If your current version is ignoring this, the most likely causes are: * you’re passing the variable under the wrong key (for example, trying to modify raw manifest fields instead of using

env

) * the flow is not actually being run with that decorated infrastructure path * you may be hitting a bug/version mismatch in

prefect-kubernetes

A good known-working pattern is:

python
@kubernetes(
    work_pool="my-k8s-work-pool",
    image="my-image:latest",
    namespace="default",
    env={"MY_ENV_VAR": "hello"},
)
@flow
def my_subflow():
    import os
    print(os.environ["MY_ENV_VAR"])

I verified the decorator accepts

**job_variables

, and

env

is the right job variable for Kubernetes worker configuration. If you want, I can also show you: * a full parent flow → subflow on Kubernetes Job example * how to inspect the rendered job configuration to confirm the env var is making it into the pod spec * or help debug the exact

job_variables

dict you’re currently using