generated from kubernetes/kubernetes-template-project
-
Notifications
You must be signed in to change notification settings - Fork 182
Open
Labels
needs-triageIndicates an issue or PR lacks a `triage/foo` label and requires one.Indicates an issue or PR lacks a `triage/foo` label and requires one.triage/acceptedIndicates an issue or PR is ready to be actively worked on.Indicates an issue or PR is ready to be actively worked on.
Description
As discussed with @robscott :
The inference pool (and the endpoint picker) may return a diverse set of endpoints, and one desired behavior is that the Gateway validates that the returned endpoints are part of a well known admin configuration.
As an example:
- Gateway Admin may want to allow just endpoints that belong to the cluster (eg.: inside the Pod CIDR, or part of a created endpoint/endpointslice)
- Gateway Admin may want to allow just endpoints that belong to the same namespace of the InferencePool used by the HTTPRoute
- Gateway Admin may want to allow well known endpoints like external pool names/IPs
This way, a conformance test that does the following steps is desired:
- Deploy a generic InferencePool/EndpointPicker that returns the endpoints based on generic requests (eg.: some header)
- Deploy a generic echo backend
- Given the following requests, the actions should be taken:
- A request to the gateway, passing the endpoint of an existing backend on the same namespace should be accepted and forwarded to the right endpoint by the gateway
- A request to the gateway, passing the endpoint of an existing backend on a different namespace should be denied / return an error (unless there is a referencegrant for a inferencepool on a different namespace)
- A request to the gateway, passing the endpoint of an external resource should be accepted depending on ImplementationSpecific behavior (to be defined)
robscott
Metadata
Metadata
Assignees
Labels
needs-triageIndicates an issue or PR lacks a `triage/foo` label and requires one.Indicates an issue or PR lacks a `triage/foo` label and requires one.triage/acceptedIndicates an issue or PR is ready to be actively worked on.Indicates an issue or PR is ready to be actively worked on.