Extend the multi-backend (PeerPool) model so a daemon can advertise spare scheduling headroom to the control plane and withdraw it, driven by the existing idle / scale-down signal (a box that would scale down can instead offer its freed headroom for control-plane-directed scheduling). Today a backend's capacity is implicit; this makes available headroom an explicit, policy-bounded signal (window, excluded workload classes).
Acceptance: a backend can publish/withdraw a headroom advertisement; the control plane sees it via the existing backend surface; advertisement respects a local policy (time window / workload-class exclusions); withdrawing is idempotent.
Part of a control-plane capacity-pooling effort (internal tracking: FootprintAI/Containarium-cloud#491).
Extend the multi-backend (
PeerPool) model so a daemon can advertise spare scheduling headroom to the control plane and withdraw it, driven by the existing idle / scale-down signal (a box that would scale down can instead offer its freed headroom for control-plane-directed scheduling). Today a backend's capacity is implicit; this makes available headroom an explicit, policy-bounded signal (window, excluded workload classes).Acceptance: a backend can publish/withdraw a headroom advertisement; the control plane sees it via the existing backend surface; advertisement respects a local policy (time window / workload-class exclusions); withdrawing is idempotent.
Part of a control-plane capacity-pooling effort (internal tracking: FootprintAI/Containarium-cloud#491).