EPP as a container in decode deployment

[asm582]

In the current stack, EPP is used as a deployment with its service. Should EPP be a container in the decode deployment in an attempt to launch fewer objects for each model service CR?

[sriumcp]

@lionelvillard @nilig

Thoughts? The idea is to run EPP for any base-model as part of the model sidecars rather than as a separate service. It might simplify the overall architecture for a given model.

[lionelvillard]

You can't. They don't have the same cardinality. 1 EPP for xPyD.

[asm582]

Thanks, is that for the same model family?

[asm582]

Should EPP exist if there are zero D pods?