Reproducible saltatlas runs?

[bwpriest]

I've noticed that there is some nondeterminism occuring within saltatlas when run on the same inputs on multiple ranks. This nondeterminism appears to disappear when run on one rank, so it seems likely that it arises in the partitioner. I haven't explored the source further. If the nondeterminism is the result of explicit randomness in the partitioner, can we expose the random seed so that it can be set for repeatability? If the source is instead some parallel nondeterminism, do we understand where it occurs?