Yeah, it looks like a problem with the consistent initialization algorithm BrownFullBasicInit in OrdinaryDiffEq. If you do @show solve(prob, Rodas4())[1] you see:
(solve(prob, Rodas4()))[1] = [3.0, 1.0, -Inf]
Oops, for some reason the initializer diverged, and I can see it diverges in one step too which is odd. If you can open an issue on OrdinaryDiffEq.jl we can follow up there. @SCher 's answer is correct but we might as well fix this too.