mongoc_cluster_t's have a try_sendv and a sendv method to actually write bytes for commands. For some reason we use try_sendv after the cluster is born, so we don't reconnect if select fails.

That doesn't make any sense. We should just use regular cluster_sendv() in all cases.