Flyte Console - Dataclasses & Pydantic models as inputs

[adartaud]

Hi,

After updating to version 1.14.5, I noticed a regression in the Flyte console regarding python dataclasses.

Prior to version 1.14.0, in the Flyte console, the input was correctly inferred and I could set each property of the dataclass with their appropriate types.

Now, after updating Flyte, dataclasses seem to only accept "raw json" as input (same as pydantic base models).

Is this intended ? Is there something I am missing ?

Regards,

[Future-Outlier]

Please use json str to do it!

[Future-Outlier]

according to this PR
flyteorg/flytekit#2792
You can see me create input like this.

In summary, you have to turn your dataclass/dict/pydantic basemodel input to a json str as input, then it will work.

[adartaud]

Thanks for your answer !

Indeed I know I can write my raw json this way, but the previous way of handling dataclasses in the UI seemed way more convenient to me. When writing a json as string in the ui, we have to know exactly what properties are needed and what values they accept.

When using dataclasses or pydantic models, there is no way in the current version to let the UI infer the inner types as this was done previously? If this isn't the case anymore, is this something that is planned in a future release?

Regards,

[Future-Outlier]

oh we are not going back to this, but you can use Union Console to solve your needs.

https://docs.union.ai/byoc/user-guide/core-concepts/workflows/launching-workflows#launching-workflows-from-the-ui

We support more features at Union, and of course, more stable

[Future-Outlier]

Please try it and mark me as the answer if it works cc @adartaud