Each Pod has a unique IP that is included in the algorithm used when forwarding requests. Actually, it’s not much of an algorithm. Requests will be sent to those Pods randomly. That randomness results in something similar to round-robin load balancing. If the number of Pods does not change, each will receive an approximately equal number of requests.

Random requests forwarding should be enough for most use cases. If it’s not, we’d need to resort to a third-party solution. However soon, when the newer Kubernetes versions get released, we’ll have an alternative to the iptables solution. We’ll be able to apply different types of load balancing algorithms like last connection, destination hashing, newer queue, and so on. Still, the current solution is based on iptables, and we’ll stick with it, for now.