Cross-Zone Load Balancing
Whether a load balancer node distributes traffic to targets in all zones or only its own — a setting that affects both balance and data transfer cost.
A load balancer has a node in each zone. With cross-zone enabled, each node distributes across every healthy target regardless of zone. Disabled, each node sends only to targets in its own zone.
The trade is direct. Cross-zone gives even distribution — important when zones hold different numbers of targets, which happens routinely after a scaling event or a zonal failure. Without it, a zone with two targets receives the same share of traffic as a zone with ten, and those two are overloaded.
Disabling it gives lower latency (traffic stays in-zone) and avoids cross-zone data transfer charges, which at high volume is a meaningful line item.
Defaults differ by load balancer type and provider, which is itself a trap — teams assume behaviour that does not apply to the type they are using.
The usual right answer: enable it, and solve the cost problem by keeping target counts balanced across zones. Uneven load is an availability risk; cross-zone transfer is a bill. But if targets are reliably balanced and volume is very high, disabling it is a legitimate optimisation — made deliberately, with the understanding that a zonal capacity imbalance now becomes a hot zone.