Travis Johnson

I found a couple issues in the gRPC repo that look relevant: - https://github.com/grpc/grpc/issues/31570 - https://github.com/grpc/grpc/issues/30984 From the [later comments on the second issue](https://github.com/grpc/grpc/issues/31570#issuecomment-1397761508) it sounds like it could be...

Hmm, I'm not sure about this behavior being the default. As stated, ModelMesh is intended to handle downtime in etcd by continuing to serve requests for existing models (adding models...

Hello, thanks for pointing this out! I think this is actually just a documentation problem In the adapter code that processes the storage configuration, `bucket` is supported and takes priority...

Hi @AHB102, as noted in the issue description, using "bucket" should be the standard to match with KServe, so that is already decided. In this ModelMesh repo, the documentation and...

A couple different ways that you can determine the inputs and outputs of the exported ONNX model: 1. Use the [`ModelMetadata`](https://kserve.github.io/website/0.9/modelserving/inference_api/#server-metadata_1) gRPC API to query Triton for info about the...

I think this next issue is that you are using `input0.contents.int_contents.extend()` instead of `input0.contents.int64_contents.extend()` (notice int vs int64). `int_contents` is for 32-bit integers ([ref](https://github.com/kserve/kserve/blob/547996a4dcd125d564e0b5d8b9f00603e7599a26/docs/predict-api/v2/grpc_predict_v2.proto#L294)).

This issue looks a bit tougher 🤔 The `Failed to open new connection` errors to port 8080 indicate a communication issue between the model mesh containers (port 8080 is what...

Hmm. It seems that you had the model loaded in and could use the ModelMetadata call, but then, when trying an inference call, the Triton container crashed and restarted? It...

Hmm, interesting that the model worked before and cannot load at all now. Could you try to load one of the sample models (or another model that you know has...

Hello @MLHafizur. Do you have any updates on this issue? Are you still experiencing the "Nowhere available to load" errors?