I was reading through the scheduling code and figured I'd offer a little optimization when selecting the highest priority host. I wouldn't argue that this is strictly needed, but I think it's a little better and not overly complicated. There aren't added tests because the output is the same and already tested.

• We can avoid a full sort of the list and find the top priority hosts with a single pass.
• We can avoid generating []string garbage from the top host names.
• We can avoid locking and generating a random number when there's one top host.
• We can avoid modulo bias in the random number.