Generalize backend_validate_gpu into a capability profile recorded when a backend joins/pools: hardware class (CPU generation, GPU model, RAM), a lightweight benchmark, and region. Lets the control plane schedule by verified capability rather than self-reported flags (a thin box can't claim a large class).
Acceptance: join records a capability profile; benchmark is bounded/non-disruptive; profile is queryable on the backend surface; self-reported class is reconciled against the measured profile.
Part of a control-plane capacity-pooling effort (internal tracking: FootprintAI/Containarium-cloud#491).
Generalize
backend_validate_gpuinto a capability profile recorded when a backend joins/pools: hardware class (CPU generation, GPU model, RAM), a lightweight benchmark, and region. Lets the control plane schedule by verified capability rather than self-reported flags (a thin box can't claim a large class).Acceptance: join records a capability profile; benchmark is bounded/non-disruptive; profile is queryable on the backend surface; self-reported class is reconciled against the measured profile.
Part of a control-plane capacity-pooling effort (internal tracking: FootprintAI/Containarium-cloud#491).