Hi all, I’m new to metaflow as well. Does anyone h...
# ask-metaflow
h
Hi all, I’m new to metaflow as well. Does anyone have any insights on generating cards when flows are running in a kubernetes cluster? I’ve tried adding the card decorator to my flow but when when I do
python dummy.py card view start
the pathspec is not found.
1
h
hey! do you have any logs of the error message.
are u running
python dummy.py card view start
after the flow finishes execution ? Since cards are run time artifacts, you need to ensure that the step a
@card
is added to finishes execution; After which the
python dummy.py card view <stepname>
should work.
h
Hi! thanks for responding.
Yes I am running it after it finishes running. Looks like the relevant log error is
Copy code
>Card render failed with error :
>Object not found:
>Task('HelloFlow/1690412901605615/hello/2') does not exist
h
I see, you are running a local metadata with kuberentes. Short story is that cards are not compatible on kuberentes with local metadata provider. You need a metadata service to generate cards on kubernetes.
h
Ah ok. I figured it was something like that. Can I find examples of setting that up in the docs?
h
You need a remote metadata provider for cards to get generated on kuberentes. The metaflow service helps with this
🙌 1
h
Thank you! I’ll read up and see what I can get cooking 🙂
h
Basically here are the changes you need to make : 1. Deploy a metadata service that can be accessed by your cluster 2. Reference the address to this service in your
METAFLOW_SERVICE_URL
in your configuration file This much should get you started with cards getting generated on kubernetes and they also showing up on local. Ideally outside deploying the service, there should be no change in the code. Only in the metaflow config file to make everything work as expected.
h
Perfect, this has been extremely helpful. Thank you again.