Hi guys I got this error OSError: [WinError 123] Incorrect filename, directory name, or volume label syntax: 'gs: when I tried to deploy this example in GKE, any clue?

Hi @aldemarhernandezg,I think we'll need more details about what steps you took to configure your ZenML deployment, stack and so on. Are you using your own ZenML server deployed in GKE ? What about the ZenML stack ? The information you provided only tells me that you're trying to create a directory in Windows that has a path that starts with gs:, which is obviously not a well formed directory path.

Hi, @stefan I'm using a cluster that I already exist on my account, and this is my stack

with a remote zenml server

I think is beacuse mlflow is trying to create mlflow folder but I'm using a gcp bucket for atifacts, I don't know if I'm right

I think you're right. If you're using a local MLflow tracker with a GCS artifact store in the same ZenML stack, MLflow will try to write to your local filesystem instead of the remote bucket.

The only scenarios officially supported by ZenML are those documented here:

as you can see, you can only combine the local MLflow tracker with a local artifact store

if you're using a remote Artifact Store, like GCS, the MLflow tracker needs to be deployed as a standalone service with its own storage back-end

I got it , do you know how to install mlflow on GKE ?

normally, I would direct you to our modular GCP recipe, but I don't think you can only select and install MLflow out of all the components that it installs. It also comes with its own GKE cluster and other resources:

you can probably use the helm chart to achieve the same:

can I use recipe with a existed cluster?

and for some reason this recipe uses kubeflow, I don't want to use it

that's what I mean, the recipe may be modular, but if I remember correctly, you still need to base it on something like GKE cluster and you can't provide your own

you can enable/disable most of those components, including Kubeflow: