BenchmarkTools: How to use different random array on each run (esp. for mutating functions)?

What is the difference between samples and evaluations? I can I make sure there is only one eval/sample?