Jupyter alternatives

[onacrame]

In many organisations notebooks are not preferred for production and conventional IDEs and scripts dominate production. How can we accommodate framework agnostic development environments in this project ala VS Code etc.

[bernardolk]

Hello @onacrame, using the notebooks is pretty much optional. In our organization, we code everything in pyCharm and connect to our OpenMLOps cluster using the exposed endpoints (for auth and interacting with Prefect, for instance). We provide a Jupyter deployment as a means to quickly test your deployment and do experiments, and it is also very popular among scientists and researchers.

[onacrame]

Hello @onacrame, using the notebooks is pretty much optional. In our organization, we code everything in pyCharm and connect to our OpenMLOps cluster using the exposed endpoints (for auth and interacting with Prefect, for instance). We provide a Jupyter deployment as a means to quickly test your deployment and do experiments, and it is also very popular among scientists and researchers.

Thanks for that. Do you have an example of how this would work. In my mind I was thinking even for prototyping we might use VS Code to build an object oriented set of scripts and modules. Ideally we want to develop in containerised environments whether it be in Jupyter or VSCode or RStudio etc.

In KubeFlow 1.3 I know they've added the concept of code servers so that you can access VS Code or R Studio essentially in the browser giving that optionality to easily switch development environments

In VS Code there is an add-in Okteto that will allow remote VS Code development in a cluster.

What's the mechanism to IDE to the cluster to enable dev work there?

[bernardolk]

Honestly, we do not connect any debugger to the deployed code in the cluster, we write tests and connect all library clients to services running in the cluster, but we don't actually debug the code running inside the cluster apart from checking logs, dashboards, etc. Usually, we build images and deploy the code as a Kubernetes deployment.

[bernardolk]

The key part is that you can connect to your OpenMLOps deployed services (Prefect, Dask, etc) as long as they have exposed endpoints (exposed to outside the cluster), then you can do any client/backend dev work from anywhere you want

[onacrame]

Understood, thanks.

[NhatAnh]

In our organization, we code everything in pyCharm and connect to our OpenMLOps cluster using the exposed endpoints (for auth and interacting with Prefect, for instance).

Hi @bernardolk , I would be very interested in how to do this. Do you have a tutorial/guide, or could you write up something to guide us about this? Thanks

[bernardolk]

@NhatAnh everything in the tutorial that we use jupyter for can be coded in pyCharm, you will find an example of connecting to services (like Prefect) there, for instance on the line that says: prefect_url = f"https://prefect.{domain}/graphql" we define the connection URL for our prefect graphql endpoint. Here is the link:

Basic OML tutorial