Describe the bug
The deployment API allows a function version with multiple GPU spec IDs to have separate autoscaling configurations. The function autoscaler produces one scaling decision per function version, while its invocation and worker metric inputs do not identify the GPU spec ID. Conflicting per-GPU-spec policies therefore cannot be mapped unambiguously to scaling decisions.
Expected behavior
Align the policy and metric identity with the scaling decision scope, or constrain the supported autoscaling configuration to one policy per function version.
Describe the bug
The deployment API allows a function version with multiple GPU spec IDs to have separate autoscaling configurations. The function autoscaler produces one scaling decision per function version, while its invocation and worker metric inputs do not identify the GPU spec ID. Conflicting per-GPU-spec policies therefore cannot be mapped unambiguously to scaling decisions.
Expected behavior
Align the policy and metric identity with the scaling decision scope, or constrain the supported autoscaling configuration to one policy per function version.