Hi team, I had a few contributions to point out. The first two were already merged in LinkedIn codebase and running for a while now. The final contribution has an approach we use internally, but might not be suitable for open source. Maybe a topic to discuss on Thursdays contributor meeting if I can make it. 1. Refresh token support for DeviceFlow auth - we had to do this for Azure AD authentication. It should solve this open issue. 2. Allow for Histogram support in StopWatch. I expanded the StopWatch class in promutils such that latency/duration metrics have the option to use Histograms, alongside an exponential histogram generator function, such that users who want more accurate latency metrics across multiple flyte instances can rely on histograms for their monitoring set up. 3. Expand

limit-namespace

such that flytepropeller can listen to a specific set of namespaces. At the moment we can only specify one namespace for flytepropeller to listen to, or all namespaces. This was a necessary ask internally from our ML teams that I could go into more detail about. I made some changes to allow for this internally, but I don't think it follows the best conventions and would like some high level feedback.

@witty-evening-52399 thanks for sharing, it'd be great to see those contributions coming in! Also, you're so welcome to join the contributors sync on Thursday Nov 7, 11:00am PT, direct Zoom. Import the calendar

Thanks @average-finland-92144, will try and join!

Do I understand it correctly that for device flow in contrast to pkce, auth tokens are not cached 😧

Yea, I believe so - https://github.com/flyteorg/flytekit/blob/4bef16127c62f04ad34b3d868439db47afe51d8e/flytekit/clients/auth/authenticator.py#L311 . In our internal fork, we've added this logic and has been working for 8 months now. We use DeviceFlow with Microsoft Azure Ad.