Skip to content

Weighted / canary routing across model versions #50

Description

@spron-in

Problem. There's no way to split traffic across multiple llm Instances (e.g. two versions of a model) behind one endpoint for canary, blue-green, or A/B rollouts.

Impact. Safe model upgrades require external routing; today an upgrade is effectively in-place.

Current workaround. None.

Upstream. Router.Route Group + Weight (Gateway API backendRef weight semantics; members sharing a group split traffic).

Scope. Expose a routing group + weight so several Instances can front one endpoint with weighted traffic.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

llmIssues related to LLM Inference Service

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions