Hi! I'm trying to edit a field in the PodTemplate. All I wan't is to use the default values and use the labels argument.

pod_template=PodTemplate(labels={"spot": "false"})

That's what i'm trying to do, but I get the error

ModuleNotFoundError: No module named 'kubernetes'

Did anyone try to use just one argument in the PodTemplate?

hi! can you add

kubernetes

to your image? it is necessary for the pod template to work.

You mean add it in poetry?

are you using any image for your flyte task? if not, can you define an imagespec and assign the same to your task?

I'm not using a separate image for each task, I only have an image of the workflow

sorry, i don't follow. wdym by image of the workflow?

ok so adding kubernetes fixed it. I don't have a specific image for each task, I use poetry to manage the packages in my repo. I added kubernetes to poetry and that fixed it. All of the tasks in my workflow use the same image I guess if I don't define a specific one, right? In our CI/CD we create a image for each repo, and the workflow is registered using that image. for example, this is how I register a workflow if i'm doing it from my local computer.

pyflyte -c ./.config/local-config.yaml register -p dataengineer -d development --image normalized_roles=blah-blah-docker.dev/images/data-wf-normalized-roles:testing-out-spot-instance_50dfab2  flyte/workflows/normalized_roles_enrichment_wf.py

All of the tasks in my workflow use the same image I guess if I don't define a specific one, right?

that's right. understood. in that case, you'd have to add the package to your repo-specific image, which you did. 🙂

are we using Flyte in the wrong way? haha

this should be fine. if you've significantly different package requirements for tasks within a workflow, it might be beneficial to assign separate images to those tasks. also, with imagespec, you needn't worry about writing dockerfiles and such.

alright, thank you!