The policy uses the first fetch of worker info list as the base, and visits each of them in a
round-robin manner in the subsequent calls. The policy doesn't assume the list of worker info
in the subsequent calls has the same order from the first, and it will skip the workers that
are no longer active.