Original description: While there is a function to randomly select, it is only used for single host selection here. This has implications for both connection reuse and read hedging. It means that if we have available connections to all hosts, we will always choose the first one in order if there is an available connection, essentially prioritizing that host. If we are hedging reads, we will also chose the second one in order for our hedge, which prioritizes yet another host and makes that host the one considered the "hedge" for our statistics. We should randomize the multiple host selection here as well.