Non-deterministic optimization model building

[joaquimg]

I am running the main.jl script in this PR https://github.com/jump-dev/open-energy-modeling-benchmarks/pull/38

Every time I run it, I get a different mps file.

Rows are certainly and non-deterministic order:

This might be due to many reasons:

• Some bug
• Iterating in the Julia Dict (or some other collection that does not preserve ordering). OrderedDict from OrderedCollection is slower for some operations but preserves the order.

Even if the only issue is order, the models might have heavily varying solution times and might be difficult to debug.

Deterministic behavior would be ideal here.

[Tasqu]

There are a couple of places I think might cause this:

1. Our use of Dicts in storing JuMP variable and constraint references, as well as other stuff like temporal and stochastic structure information. Essentially anything we store in m.ext[:spineopt]. However, I'm not sure if/how this stuff impacts the JuMP model m generation directly.
2. Could potentially be caused by the indexing functions we use? I'm not sure if the order of entities is consistent when it is fetched from the database.