Nick Hill

@yuzisun do you have any thoughts/ideas about this?

Thanks @bherta, I agree with all of the above. I think I'm good with adding status to ServingRuntimes, and you make a great point that ClusterServingRuntimes don't need to share...

Thanks @bherta, I'm not really a fan of status conditions - on the modelmesh-serving side we actually just have more explicit status that's updated (see the [ModelStatus](https://github.com/kserve/kserve/blob/master/pkg/apis/serving/v1beta1/inference_service_status.go#L50) field), though I...

Thanks @bherta @gaodan-fang @lizzzcai > If there is spec validation error, this is something a user would see when trying to create the CRD instance, and the object won't actually...

> - When the new version of the model is available, what is the best option in ModelMesh to update the model without any service interruption? @MLHafizur if you just...

Hi @OvervCW, this is a very reasonable requirement and one I've also heard from others. Implementation is complicated by the fact that the same etcd secret is not only mounted...

I started to work on this, not sure when I'll have a chance to finish but will aim to push a WIP branch at least.

@lizzzcai FYI I have pushed my in-progress work on this here: https://github.com/kserve/modelmesh-serving/tree/separate-etcd-cert

@meijies it looks like this is a general model-mesh related issue. You opened it in the adapter repo which is one specific component, so I'll move it to the main...

I'll just note that serving runtime specific configuration should mostly be reflected in the ServingRuntime CRD itself. What we are talking about here is more cross-cutting configuration that applies to...