GSoC 2025 Project Ideas

[davidmirror-ops]

1. Explore support for if/else blocks outside of conditionals. This would enable a more "Pythonic" way to author workflows.
   Skills: Python
   Mentor(s): @eapolinario
2. Enrich pyflyte run
   Enable CLI users to pass parameters like endpoint, authType and security config instead of relying purely on a local config file. This would be beneficial for CI/CD systems registering/running workflows on Flyte to craft commands/scripts to run execution without constraining to local file availability. Also it would help speed up the "time to first workflow" for users evaluating Flyte.
   Skills: click and rich_click Python libraries
   Mentor(s): @Future-Outlier
3. Revamp examples and tutorials
   Execute all examples and tutorials in the docs and update dependencies where needed, detect and fix errors, improve docstrings, and add examples.
   Skills: Python, Flyte, Flyte integrations
   Mentor(s): @davidmirror-ops
4. Automate dev environment setup
   Today, new contributors have to follow many manual steps before they have a working dev environment to start contributing. Automating this would streamline the process, especially for new contributors. Part of the project is to evaluate the implementation options (Codespaces, Gitpod, a VM, etc).
   Skills: Configuration Management, Git
   Mentor(s): @Future-Outlier
5. Implement auth token refresh in the UI
   By default, Identity Providers tend to emit short-lived access tokens. Extending token life is not implemented right now for the UI, but it is for the CLI. This project would help close the gap to enable Flyte users to configure token duration that is applied to the UI too. See [BUG] Flyte console refreshes often to re-authenticate #4817 for additional context.
   Skills: Go, OAuth protocol
   Mentor(s): @wild-endeavor / @eapolinario
6. Obstore integration with flytekit
   Get rid of s3fs, gcsfs, and abfs dependencies, and use obstore instead to read/write data. This would reduce the size of flytekit by around 50-100MBs, significantly reducing security vulnerabilities as about 10s of dependencies are eliminated.
   Skills: Python or Rust
   Mentor(s): @pingsutw

[kumare3]

We would love offer serverless credits for folks to sign on and do any of these projects

[pingsutw]

obstore integration with flytekit: remove s3fs, gcsfs, abfs dependencies, and obstore to read/write data instead.

[davidmirror-ops]

obstore integration with flytekit: remove s3fs, gcsfs, abfs dependencies, and obstore to read/write data instead.

This is great. Could you elaborate a bit on the impact/benefits this project would have on Flyte users and what skills are required from contributors? Is this the obstore repo?

[kumare3]

@davidmirror-ops it reduces size of flytekit by around 50-100MBs and makes security vulnz lower as about 10s of dependencies are eliminated. Moreover performance in our testing seems to be improved upto 2-3x

[pingsutw]

cc @machichima ^^^

[fg91]

Fix broken with_overrides, see this issue. TL;DR some task args like e.g. the pod template cannot be made overridable properly because for instance the pod template get's merged with other information like the image before being sent to the backend. This means that even if idl had a pod template override message, the backend couldn't apply it without overriding other data that is not part of the pod template. It would be really good to clean the entire with_overrides implementation up with the goal that all task args are overridable and that there is a unified way to handle overrides.

[davidmirror-ops]

@fg91 do you think this is closer to this change planned to go with the next release? #6118

[RRap0so]

Expand the capabilities of the cluster resource sync. Give it the ability to sync projects based in all clusters (I believe this exists) but also be able to assign custom KV that are inside the cluster resources or projects to specific things in a k8s objects to facilitate for example Workload Identity.

For eg:
A project

description: 'Component: flytesnacks'
domains:
- id: development
  name: development
- id: staging
  name: staging
- id: production
  name: production
id: flytesnacks
name: flytesnacks

The associated cluster resource

{"project":"flytesnacks","domain":"production","attributes":{"default_ksa_iam_gke_io":"flytesnacks@ flytesnacks.iam.gserviceaccount.com"}}

The default_ksa_iam_gke_io is then assigned to a ServiceAccount object under annotations to ensure GCP Workload identity works:

apiVersion: v1
kind: ServiceAccount
metadata:
  annotations:
    iam.gke.io/gcp-service-account: flytesnacks@flytesnacks.iam.gserviceaccount.com
  name: default
  namespace: flytesnacks-production

This should always listen to changes within the cluster resource objects under each project, and update accordingly. This change can be made via other systems or user-based flytectl

[davidmirror-ops]

Skills: Go, Kubernetes