I've seen this error twice on 6 node remote clusters (1 master + 5 datanodes) of different platforms:
Is it really a product failure though? It seems like we might be asserting an unreasonable degree of precision – perhaps we just need to tweak our test verification expectations?
tarmstrong, I'm randomly assigning to you because you were the last one to touch verify_agg_result, where the assertion is defined. Please reassign as you see fit.