Easy way to determine constraints and/or variables leading to failed solving?

[lfreese]

Hi,

I'm solving an optimization over time (1 year by hour, so 8759 runs of the model) by submitting it to a slurm, and during a given number of my timesteps (5001-6000), I get:
Parallel mode: deterministic, using up to 32 threads for concurrent optimization. Row 'c1' infeasible, all entries at implied bounds. Presolve time = 11.16 sec. (3193.35 ticks) ERROR: LoadError: The model was not solved correctly.

I'm wondering if there is a simple way within JuMP to figure out what constraints or variables is causing the inability to solve the model (eg: what timestep, which of my 4 constraints)?

I've looked at the documentation and haven't found something that would specifically work in the context of slurm submissions, but might be mistaken.

[odow]

In general:

At present, there is limited support for the accessing the Irreducible Infeasible Subset (IIS) in some solvers: #1035

We also have a GSOC project for this: https://github.com/jump-dev/GSOC2020/blob/master/ideas-list.md#feasibility-and-optimality-checker

For your specific case:

what timestep

Is each hour a different model? Impossible to say without your code.

which of my 4 constraints

From Row 'c1' infeasible, all entries at implied bounds, it looks like the first constraint.

If you want detailed advice, it might be better to post on https://discourse.julialang.org/c/domain/opt/13. Make sure to read the first post of https://discourse.julialang.org/t/psa-make-it-easier-to-help-you/14757 and provide a minimal working example.

For Gurobi, you can use

using JuMP
using Gurobi
using Test

model = direct_model(Gurobi.Optimizer())
@variable(model, x >= 1)
@constraint(model, c, x <= 0)
optimize!(model)
@test termination_status(model) == MOI.INFEASIBLE
Gurobi.compute_conflict(backend(model))
MOI.get(model, Gurobi.ConstraintConflictStatus(), c)
MOI.get(model, Gurobi.ConstraintConflictStatus(), LowerBoundRef(x))

[lfreese]

Thanks! I'll post on the site there. It does seem to be the first constraint based on the c1, so that makes it clearer.