batch_mutate atomic?

[Per Olesen]

Can I expect batch_mutate to work in what I would think of as an atomic 
operation?

That either all the mutations in the batch_mutate call are executed or none of 
them are? Or can some of them fail while some of them succeeds?