Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Prefect Community

is there a way to edit the command on k8s jobs? it's `prefect execute flow-run` but I want to add a command before that

Hey <@U01MEG3ET6C>,
AFAIK not via the CLI, that command would simply run a flow against your backend API with the supplied `run_config` determining the k8s job. Looking through the <https://docs.prefect.io/api/latest/cli/run.html|CLI flags>, you may be able to parameterize your job config and select it with a flag, but this would likely be specific to your use case.

basically what is happening is we use `conda` to install our packages, but conda needs `conda activate &lt;env name&gt;` to run before `prefect` runs

in most of our docker containers we run `conda activate &lt;env&gt; &amp;&amp; &lt;command&gt;`

but it appears that prefect hardcodes the command and doesnt let us change it, even with a custom k8s job template :disappointed:

<https://github.com/PrefectHQ/prefect/blob/master/src/prefect/utilities/agent.py#L71>

hmm, I’m surprised this isn’t honored by the image nor the job template… I think we’re looking for something to the equivalent of the `on_start` from the old <https://docs.prefect.io/api/latest/environments/execution.html#kubernetesjobenvironment|KubernetesJobEnvironment> days, but I’m still looking for a  `run_config` solution - I must be missing something.

yeah, it looks like the steps are: prefect-agent (running in pod on k8s) picks up flow, it uses the user-provided job-template as a starting point, and then overlays some settings on top of it before sending to k8s to be created as a Job

so even if i set the args in my template, the prefect code overrides it

here's where: <https://github.com/PrefectHQ/prefect/blob/8c662ce5524aca9cb635e44784081a1701cfb5c3/src/prefect/agent/kubernetes/agent.py#L657>

Hey Dana - check out<https://github.com/PrefectHQ/prefect/issues/4128#issuecomment-780675602| this issue> on GH, in which Jim’s comment provides an example of using an entrypoint for the Dockerfile (in our case something like a `conda run` CLI) and an explanation for this behavior. I hope this may shed some light on this, but please feel free to comment there with your thoughts.