Gurobi not working in v9.15

[quentinplsrd]

I have the Gurobi 13 license, and it works fine in my python when I use it with gurobipy
But there is an issue when I try to solve a problem with mathopt and the gurobi solver.

Here is my environment, built with uv

uv init gurobi_test_uv
cd gurobi_test_uv
uv add ortools gurobipy numpy

And here is my python code

import ortools
from ortools.math_opt.python import mathopt
import gurobipy as gp
import os

print("GUROBI_HOME:", os.environ["GUROBI_HOME"])
print("GRB_LICENSE_FILE:", os.environ["GRB_LICENSE_FILE"])
print("ortools:", ortools.__version__)

print("gurobipy:", gp.gurobi.version())
print("\n\nSolving dummy model with gurobipy")
m = gp.Model()
x = m.addVar()
m.setObjective(x)
m.optimize()
print("status:", m.Status)

print("\n\nSolving dummy model with mathopt and gurobi")
model = mathopt.Model()
x = model.add_variable(ub=1.0)
model.maximize(x)
result = mathopt.solve(model, mathopt.SolverType.GUROBI, 
                       params=mathopt.SolveParameters(enable_output=True))

And the output:

GUROBI_HOME: C:\gurobi1300\win64
GRB_LICENSE_FILE: C:\Users\qploussard\gurobi.lic
ortools: 9.15.6755
gurobipy: (13, 0, 1)


Solving dummy model with gurobipy
Gurobi Optimizer version 13.0.1 build v13.0.1rc0 (win64 - Windows 11+.0 (26200.2))

CPU model: 13th Gen Intel(R) Core(TM) i9-13900H, instruction set [SSE2|AVX|AVX2]
Thread count: 14 physical cores, 20 logical processors, using up to 20 threads

Optimize a model with 0 rows, 1 columns and 0 nonzeros (Min)
Model fingerprint: 0x84abb838
Model has 1 linear objective coefficients
Coefficient statistics:
  Matrix range     [0e+00, 0e+00]
  Objective range  [1e+00, 1e+00]
  Bounds range     [0e+00, 0e+00]
  RHS range        [0e+00, 0e+00]

Presolve removed 0 rows and 1 columns
Presolve time: 0.00s
Presolve: All rows and columns removed
Iteration    Objective       Primal Inf.    Dual Inf.      Time
       0    0.0000000e+00   0.000000e+00   0.000000e+00      0s

Solved in 0 iterations and 0.00 seconds (0.00 work units)
Optimal objective  0.000000000e+00
status: 2


Solving dummy model with mathopt and gurobi
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
File c:\users\qploussard\gurobi_test_uv\main.py:22
     20 x = model.add_variable(ub=1.0)
     21 model.maximize(x)
---> 22 result = mathopt.solve(model, mathopt.SolverType.GUROBI, 
     23                        params=mathopt.SolveParameters(enable_output=True))

File ~\gurobi_test_uv\.venv\Lib\site-packages\ortools\math_opt\python\solve.py:92, in solve(opt_model, solver_type, params, model_params, msg_cb, callback_reg, cb, streamable_init_args, remove_names, interrupter)
     90 # Solve
     91 try:
---> 92     proto_result = solver.solve(
     93         model_proto,
     94         solver_type.value,
     95         streamable_init_args.to_proto(),
     96         params.to_proto(),
     97         model_params.to_proto(),
     98         msg_cb,
     99         callback_reg.to_proto(),
    100         proto_cb,
    101         interrupter.pybind_interrupter if interrupter is not None else None,
    102     )
    103 except StatusNotOk as e:
    104     raise _status_not_ok_to_exception(e) from None

RuntimeError: bad function call

Any idea what the issue is?

[quentinplsrd]

I can confirm that the most recent combination of ortools and gurobi that works is ortools 9.14.6206 with gurobi 12.0.3

ortools 9.15.6206 does not seem to work with gurobi 13.0.1, 13.0.0, or 12.0.3, even when I specify the correct path for os.environ["GUROBI_HOME"]

[lperron]

If your code is not too big, can you try with ortools.linear_solver.python.model_builder ?

[quentinplsrd]

Thanks @lperron
I tried this code but it made the kernel die

from ortools.linear_solver.python import model_builder
model = model_builder.Model()
x = model.new_var(lb=0.0, ub=1.0, is_integer=False, name='x')
model.maximize(x)
model_builder.Solver('GUROBI').solve(model)

[quentinplsrd]

Actually, in a python environment with ortools 9.15.6755, this simple code is enough to make the kernel die...

from ortools.linear_solver.python import model_builder
model_builder.Solver("gurobi").solver_is_supported()

But everything works fine with ortools 9.14.6206

[jnation3406]

I can second that I can't get any Gurobi version (11.0.1, 12.0.2, or 13.0.0) to work with ortools v9.15.6755. I am running in python, and always receive a segmentation fault on trying to call Solver.CreateSolver('GUROBI'). All paths and libraries are linked correctly, and work fine with ortools v9.14 and below.

Simplest example to cause seg fault:

from ortools.linear_solver import pywraplp as p

p.Solver.CreateSolver('GUROBI')

I tried running in gdb and got this at the end of the backtrace:

#0  0x00007fffe30282c0 in ?? ()
#1  0x00007fffe84b7637 in operations_research::GetGurobiEnv() () from /.../lib64/python3.12/site-packages/ortools/linear_solver/../../ortools/.libs/libortools.so.9
#2  0x00007fffe84b79d3 in operations_research::GurobiIsCorrectlyInstalled() () from /.../lib64/python3.12/site-packages/ortools/linear_solver/../../ortools/.libs/libortools.so.9

[Mizux]

1. Concerning gurobi auto detect.

v9.15 can't detect gurobi 13.0.1

or-tools/ortools/third_party_solvers/gurobi_environment.cc

Lines 340 to 341 in 551ad10

	const std::vector<absl::string_view> kGurobiVersions = {
	"1300",

main should be able to:

or-tools/ortools/third_party_solvers/gurobi_environment.cc

Lines 341 to 342 in 5cf2345

	const std::vector<absl::string_view> kGurobiVersions = {
	"1301", "1300",

2. concerning the segfault, unfortunately we don't have any windows gurobi lib to test against so it's difficult to debug/reproduce

3. If I build a 10.0RC1 win64 python wheel, will you be able to test it ?

[jnation3406]

Sorry if this wasn't clear, I only have access to develop on linux. If you can produce a linux64 wheel for ortools 10.0RC1 I would be happy to test locally. I was considering trying to build it myself to test but haven't taken the plunge yet.