Is it possible to access /health/status and /metadata endpoints defined in seldon through ambassador

[divyadilip91]

Hi,

At present, I have noticed that /health/status and /metadata endpoints are accessible when the docker is run locally and tested
Eg: curl http://ipaddress/health/status
But when I try to access http://ambassadorexposedip/prefix/health/status, it shows 404 error.
So is /api/v0.1/predictions endpoint only accessible through ambassador?
Please do Reply

[ukclivecox]

Yes - we need to extend the external API to add status, health and metadata endpoints that are applicable per model component and for the entire graph. @RafalSkolasinski

[divyadilip91]

Thankyou @cliveseldon .So would this be applicable for the other endpoints as well such as /predict , /transform-input ,/transform-output , /route , /aggregate ?

[ukclivecox]

Was not thinking those as they are internal to the inference graph. But the ability to call any sub-part of the inference graph externally would be useful for testing.

[divyadilip91]

Yes true

[RafalSkolasinski]

I do believe that single model metadata endpoints are already available, for example

curl -s http://ambassador.mydomain.com/seldon/seldon/my-deployment/api/v1.0/metadata/my-model

[divyadilip91]

I do believe that single model metadata endpoints are already available, for example

curl -s http://ambassador.mydomain.com/seldon/seldon/my-deployment/api/v1.0/metadata/my-model

my-model is the name of the model used in the graph rite?
Is this available in seldon v1.1.0?

[RafalSkolasinski]

Yes, that's the name used in the graph. It should be available in v1.1.0 if you run the new golang orchestrator (default).

[RafalSkolasinski]

The single model metadata (via ambassador) are covered in the example: https://docs.seldon.io/projects/seldon-core/en/latest/examples/metadata.html

[divyadilip91]

@RafalSkolasinski Ok..so will this work if we have multinode graph?
Eg : a graph with a transformer and model.

[RafalSkolasinski]

This right now is querying the a single model at a time specified by the endpoint you're calling.
But yes, it will work with multi-node graphs but you have now to call each endpoint on your own.

We are currently working on providing a graph-level metadata in #1728 (including all models' metadata too) but this is still WIP with initial PR #1803 to land soon.

I guess the scope of the issue here would be to allow for sth like

curl -s http://ambassador.mydomain.com/seldon/seldon/my-deployment/api/v1.0/health/my-model

to work in similar manner as metadata.

[divyadilip91]

@RafalSkolasinski Yes

[divyadilip91]

Was not thinking those as they are internal to the inference graph. But the ability to call any sub-part of the inference graph externally would be useful for testing.

@RafalSkolasinski Is this possible to be included in future seldon version wherein we can call the subpart of inference graph using the ambassador endpoint?

[ukclivecox]

In v2 you can check status of individual models as well as for Pipelines get overall readiness.