Invalid MOI TerminationStatus in v26.02

[mtanneau]

In a nutshell

While trying to wrap the latest cuOpt release (v26.02), I encountered an issue with the test_air05 test.
This test passes with cuopt 25.12, but errors with v26.02 (see details below).

[EDIT]: this is likely coming from an upstream bug being triggered in this specific instance. Upstream issue opened.

Root cause

cuOpt --> MOI status code conversion

I tracked the issue to the MOI-level retrieval of the TerminationStatusCode here:

cuOpt.jl/src/MOI_wrapper.jl

Lines 288 to 290 in aa06a3a

	function MOI.get(model::Optimizer, ::MOI.TerminationStatus)
	return _TerminationStatusMap[model.termination_status][1]
	end

and more specifically here:

cuOpt.jl/src/MOI_wrapper.jl

Lines 284 to 285 in aa06a3a

	CUOPT_TERIMINATION_STATUS_FEASIBLE_FOUND =>
	(MOI.OTHER_LIMIT, "cuOptModelStatusFeasible"),

Note that:

• cuOpt's return status FEASIBLE_FOUND gets mapped to MOI.OtherLimit
• (see solver log below) cuOpt returns a lower bound (2.6383013888888891e+04) which is higher than the optimal value (2.6373999999999985e+04)

cuOpt termination status

I also noted the following difference at the cuOpt level:

• in v25.12, after the call to MOI.optimize!(model), model.termination_status evaluates to 1
• in v26.02, after the call to MOI.optimize!(model), model.termination_status evaluates to 8

Steps to reproduce (v26.02)

1. Make sure you have cuOpt v26.02 installed
2. Run test_air05 test from the test suite
3. You should obtain a log and stracktrace similar to the ones reproduced below:

cuOpt log (note that I activated logging by setting log_to_console to `true)

Setting parameter time_limit to 6.000000e+01
Setting parameter log_to_console to true
cuOpt version: 26.2.0, git hash: f73da24d, host arch: aarch64, device archs: 75-real,80-real,86-real,90a-real,100f-real,120a-real,120
CPU: Unknown, threads (physical/logical): 1/20, RAM: 105.63 GiB
CUDA 13.1, device: NVIDIA GB10 (ID 0), VRAM: 119.70 GiB
CUDA device UUID: 43bce240-d7d6-65e5-ca7a-e351bab5de7a

Solving a problem with 426 constraints, 7195 variables (7195 integers), and 52121 nonzeros
Problem scaling:
Objective coefficents range:          [4e+01, 3e+03]
Constraint matrix coefficients range: [1e+00, 1e+00]
Constraint rhs / bounds range:        [0e+00, 1e+00]
Variable bounds range:                [0e+00, 1e+00]

Original problem: 426 constraints, 7195 variables, 52121 nonzeros
Calling Papilo presolver (git hash 741a2b9c)
Presolve status: reduced the problem
Presolve removed: 90 constraints, 1116 variables, 16171 nonzeros
Presolved problem: 336 constraints, 6079 variables, 35950 nonzeros
Objective function is integral
Papilo presolve time: 0.42
Objective offset 9167.000000 scaling_factor 1.000000
Model fingerprint: 0x3257a27
Running presolve!
Unused variables detected, eliminating them! Unused var count 4
After cuOpt presolve: 335 constraints, 6075 variables, objective offset 9167.000000.
cuOpt presolve time: 4.49

Solving LP root relaxation in concurrent mode
Skipping column scaling
Dual Simplex Phase 1
Dual feasible solution found.
Dual Simplex Phase 2
 Iter     Objective           Num Inf.  Sum Inf.     Perturb  Time
    0 -8.1936000000000000e+04     277 2.31020000e+04 0.00e+00 5.26
    1 -8.0926000000000000e+04     266 1.48790000e+04 0.00e+00 5.26
 1000 -5.7784953825729302e+04     127 1.13852044e+02 0.00e+00 5.33


Root relaxation solution found in 1194 iterations and 0.10s by Dual Simplex
Root relaxation objective +2.58776093e+04


 | Explored | Unexplored |    Objective    |     Bound     | IntInf | Depth | Iter/Node |   Gap    |  Time  |
           0            0             +inf    +2.588508e+04      228      0   1.2e+03        -        5.58
Gomory    cuts : 1
MIR       cuts : 0
Knapsack  cuts : 0
Strong CG cuts : 0
Cut pool size  : 211
Size with cuts : 336 constraints, 6411 variables, 1250070204 nonzeros
Strong branching using 19 threads and 228 fractional variables
H                            +4.551700e+04    +2.588508e+04                                43.1%      5.93
Strong branching completed in 0.40s
Exploring the B&B tree using 19 threads

 | Explored | Unexplored |    Objective    |     Bound     | IntInf | Depth | Iter/Node |   Gap    |  Time  |
H                            +4.354400e+04    +2.591578e+04                                40.5%      7.76
B         31           27    +2.645600e+04    +2.591578e+04        0     19   1.3e+02       2.0%      7.76
D         49           43    +2.643900e+04    +2.591578e+04        0     21   1.2e+02       2.0%      7.86
D        166          108    +2.641400e+04    +2.604586e+04        0     33   9.8e+01       1.4%      8.14
B        226          134    +2.640200e+04    +2.612955e+04        0     15   9.4e+01       1.0%      8.37
D        322          140    +2.637500e+04    +2.622476e+04        0     15   9.5e+01       0.6%      8.60
B        380          120    +2.637400e+04    +2.628093e+04        0     18   9.2e+01       0.4%      8.71
         510           27    +2.637400e+04    +2.638301e+04      166      8   7.8e+01       0.0%      8.86
Explored 510 nodes in 8.86s.
Absolute Gap -9.013889e+00 Objective 2.6373999999999985e+04 Lower Bound 2.6383013888888891e+04
Optimal solution found.
Solution objective: 26374.000000 , relative_mip_gap 0.000342 solution_bound 26383.013889 presolve_time 4.913286 total_solve_time 9.023263 max constraint violation 0.000000 max int violation 0.000000 max var bounds violation 0.000000 nodes 510 simplex_iterations 39853

test failure stack trace:

test_air05: Test Failed at ~/github/cuOpt.jl/test/MOI_wrapper.jl:84
  Expression: MOI.get(model, MOI.TerminationStatus()) == MOI.OPTIMAL
   Evaluated: MathOptInterface.OTHER_LIMIT == MathOptInterface.OPTIMAL

Stacktrace:
 [1] macro expansion
   @ ~/.julia/juliaup/julia-1.12.4+0.aarch64.linux.gnu/share/julia/stdlib/v1.12/Test/src/Test.jl:680 [inlined]
 [2] test_air05()
   @ Main.TestMOIWrapper ~/github/cuOpt.jl/test/MOI_wrapper.jl:84
 [3] macro expansion
   @ ~/github/cuOpt.jl/test/MOI_wrapper.jl:27 [inlined]
 [4] macro expansion
   @ ~/.julia/juliaup/julia-1.12.4+0.aarch64.linux.gnu/share/julia/stdlib/v1.12/Test/src/Test.jl:1776 [inlined]
 [5] runtests()
   @ Main.TestMOIWrapper ~/github/cuOpt.jl/test/MOI_wrapper.jl:27

More specifically, model.termination_status evaluates to 8, which yields an MOI.OTHER_LIMIT termination status and the test fails.

Additional logs (v25.12)

For completeness, log from the same problem, using cuOpt 25.12

Setting parameter time_limit to 6.000000e+01
Setting parameter log_to_console to true
cuOpt version: 25.12.0, git hash: d97ff6b, host arch: aarch64, device archs: 75-real,80-real,86-real,90a-real,100f-real,120a-real,120
CPU: Unknown, threads (physical/logical): 1/20, RAM: 104.57 GiB
CUDA 13.0, device: NVIDIA GB10 (ID 0), VRAM: 119.70 GiB
CUDA device UUID: 43bce240-d7d6-65e5-ca7a-e351bab5de7a

Solving a problem with 426 constraints, 7195 variables (7195 integers), and 52121 nonzeros
Problem scaling:
Objective coefficents range:          [4e+01, 3e+03]
Constraint matrix coefficients range: [1e+00, 1e+00]
Constraint rhs / bounds range:        [0e+00, 1e+00]
Variable bounds range:                [0e+00, 1e+00]

Original problem: 426 constraints, 7195 variables, 52121 nonzeros
Calling Papilo presolver
Presolve status: reduced the problem
Presolve removed: 90 constraints, 1116 variables, 16171 nonzeros
Presolved problem: 336 constraints, 6079 variables, 35950 nonzeros
Objective function is integral
Papilo presolve time: 0.426261
Objective offset 9167.000000 scaling_factor 1.000000
Running presolve!
Unused variables detected, eliminating them! Unused var count 2
After trivial presolve: 335 constraints, 6077 variables, objective offset 9167.000000.
Using 19 CPU threads for B&B
Solving LP root relaxation
Scaling matrix. Maximum column norm 1.000000e+00, minimum column norm 8.965530e-02
Dual Simplex Phase 1
Dual feasible solution found.
Dual Simplex Phase 2
 Iter     Objective           Num Inf.  Sum Inf.     Perturb  Time
    1 -8.1100000000000000e+04     272 1.84122651e+02 0.00e+00 0.01


Root relaxation solution found in 990 iterations and 0.10s
Root relaxation objective +2.58776093e+04


Strong branching using 19 threads and 222 fractional variables
Exploring the B&B tree using 19 threads (best-first = 4, diving = 15)
 | Explored | Unexplored |    Objective    |     Bound     | Depth | Iter/Node |   Gap    |  Time  |
D        92           94    +2.637400e+04    +2.615126e+04        2   1.6e+02      0.8%      0.79
B       347          157    +2.637400e+04    +2.629885e+04       16   8.9e+01      0.3%      1.32
Explored 532 nodes in 1.52s.
Absolute Gap 2.634867e+00 Objective 2.6373999999999989e+04 Lower Bound 2.6371365133232783e+04
Optimal solution found within relative MIP gap tolerance (1.0e-04)
Post-solve status: succeeded
Solution objective: 26374.000000 , relative_mip_gap 0.000100 solution_bound 26371.365133 presolve_time 0.698856 total_solve_time 3.687329 max constraint violation 0.000000 max int violation 0.000000 max var bounds violation 0.000000 nodes 532 simplex_iterations 37434

In that version, model.termination_status is 1, which yields an MOI.OPTIMAL termination status and the test passes.

[mlubin]

Are you sure this is an issue in the MOI wrapper and not in the solver?

[mtanneau]

Not 100% sure. AFAICT, there's a change in the underlying (raw) solver termination status between 25.12 and 26.02.
But the MOI-level conversion from "Feasible solution found" to "Other limit" feels weird.

[mtanneau]

There is definitely an issue with cuopt for that specific instance, will file an issue upstream
--> NVIDIA/cuopt#855

[odow]

Perhaps it should be MOI.LOCALLY_SOLVED? I don't know under what conditions it returns CUOPT_TERIMINATION_STATUS_FEASIBLE_FOUND.

[mlubin]

Upstream issue is resolved