We currently prefer IPV4 addresses when establishing a TCP connection. This causes connection establishment delays when a network only supports IPV6 (see this issue for an example). We previously considered using happy eyeballs in RUST-257, but decided on the IPV4-first approach for simplicity and consistency with other drivers (comment, discussion). We should revisit this decision to improve connection times for use cases like the one in the linked issue. We should investigate whether there's already a suitable implementation of this in the ecosystem, e.g. the happy_eyeballs crate.