LeastConnections

Least connections is a load-balancing algorithm used by reverse proxies and application delivery controllers. It routes each new client connection to the backend server that has the fewest active connections at the moment the request arrives. The idea is that a server with fewer active connections is likely to become available sooner, reducing overall wait time.

Implementation typically involves the load balancer maintaining a per-backend count of active connections, updated when connections are opened and closed. Some systems support weighted least connections, where servers have assigned capacities and the algorithm selects the least-loaded server within those weights. When multiple servers share the same minimal connection count, ties are resolved by randomness or by a secondary method such as round robin.

Commonly used in HTTP and TCP load balancing, least connections is supported by many popular proxies and load balancers. For example, Nginx provides a least_conn option in upstream configurations, while HAProxy offers balance leastconn. LVS and other load-balancing solutions may also implement this method or variants of it.

Advantages include better handling of long-lived connections and dynamic workloads, as well as responsiveness to real-time load rather than relying solely on static distribution. Disadvantages include the potential to ignore non-connection-based resource usage (CPU, memory), possible skew when connection durations vary greatly, and the need for accurate tracking of active connections. Weights and health checks are commonly used to mitigate these issues and adapt the algorithm to heterogeneous server pools.