When to enable this option? When the templates of a cloud are weighted by cost, with the least expensive option carrying the highest weight. The highest weight is always prioritized, so the least expensive provisioning is always prioritized, and a more expensive template is only reached once the cheaper ones cannot supply an instance.

Changes what a template's Hot spare weight means when several AMI configurations in this cloud match the same label. Requires Round-robin templates matching the same label to be enabled; without it, templates are tried in configured order and weights are ignored.

Unticked (the default) — weighted capacity. A weight is a share of the provisioning requests, distributed by smooth weighted round-robin, the algorithm nginx uses to balance load across servers. A spot template with weight 3 against an on-demand template with weight 1 leads three requests out of four, and the on-demand template leads the fourth, evenly spaced rather than in bursts. Every weight keeps receiving a share of the work whether or not the heavier template has capacity, so the label runs on a deliberate mix of instance types.

Ticked — saturate the highest weight first. A weight becomes a rank. Templates sharing a weight form a band, and every request is offered to the highest band first; a lower band is only used once every template above it is at its instance cap or has been demoted for reporting insufficient capacity. Templates within a band still take turns leading, so a band spreads across the instance types and availability zones it contains instead of concentrating on one. Falling back costs nothing extra: the next band is tried inside the same provisioning request, and the demotion lasts a few minutes so following requests go straight to a template that can supply an instance.

Tick this when the weights express cost or preference rather than a mix — for example spot templates above their on-demand equivalents, so a label only pays the on-demand price once spot capacity is genuinely exhausted. Give equally-preferred templates the same weight to keep spreading over them. Leave it unticked when you want a predictable proportion of each instance type regardless of availability.

In both cases a weight of 0 keeps its meaning: the template is a last resort, used only when every other template matching the label is in a capacity cooldown.