Bug: Execution hangs after running HiGHS solve.

[christiansegercrantz]

Details for the issue

What did you do?

I tried running the HiGHS solver (using pulp.apis.HiGHS, highspy == 1.9.0 (fa40bdf)) on my project with the latest version of PulP from Github. The execution hangs after the solver has completed its work and never progresses forward in the code. I am unsure what information to provide to easier debug it, so please do ask for more. The solver used the following parameters:

solver = pulp.apis.HiGHS(
     gapAbs=100,
     timeLimit=60 * 5,
     threads=6
)
model.solve(solver)

What did you expect to see?

That the execution is completed and the model.solveoperation completed.

What did you see instead?

The solver hangs indefinately.

Useful extra information

The info below often helps, please fill it out if you're able to. :)

What operating system are you using?

• Windows: ( version: ___ )
• Linux: ( distro: ___ )
• Mac OS: macOS Sequoia 15.3 (24D60)
• Other: ___

I'm using python version:

• 3.7
• 3.8
• 3.9
• 3.10
• 3.11
• Other: [email protected]

I installed PuLP via:

• pypi (python -m pip install pulp)
• github (python -m pip install -U git+https://github.com/coin-or/pulp)
• Other: rye add pulp --git https://github.com/coin-or/pulp

Did you also

• Tried out the latest github version: https://github.com/coin-or/pulp
• Searched for an existing similar issue: https://github.com/coin-or/pulp/issues?utf8=%E2%9C%93&q=is%3Aissue%20

[pchtsp]

can you share the model in some of the supported formats?
https://coin-or.github.io/pulp/guides/how_to_export_models.html

If it's a very big model, then it may be something wrong with post-processing, or memory, or anything, really.
Does it work correctly with a small instance of the model?

Also, if you put a shorter timeLimit, does it work? Do you see any log or output from the solver?

[christiansegercrantz]

I doubt it's the size since It takes approx 4 seconds to solve. Additionally, the solver prints the solving report indicating it's ready:

Solving report

Solving report Status Optimal Primal bound 505757

Dual bound 505757

Gap 0% (tolerance: 0.0198%)

P-D integral 0.757847382884

Solution status feasible

                505757 (objective)

                0 (bound viol.)

                0 (int. viol.)

                0 (row viol.)

Timing 4.04 (total)

                0.00 (presolve)

                0.00 (solve)

                0.00 (postsolve)

Max sub-MIP depth 1

Nodes 1

Repair LPs 0 (0 feasible; 0 iterations)

LP iterations 530 (total)

                0 (strong br.)

                21 (separation)

                17 (heuristics)

Unfortunately I can't provide the model since it contains data not to be shared. But I can provide more information if needed

[pchtsp]

unfortunately, without a test instance it's going to be too hard to debug. If you can come up with a non-sensitive dataset, please share it. You can make up your own data or anonymize the data in your dataset.

[christiansegercrantz]

I'm working on getting you a version but meanwhile it seems the problem might be caused by the duals since whenever I interrupt the solver after it hangs this is the line: constraint.pi = solution.row_dual[constraint.index] in highs_api.py:498

[christiansegercrantz]

I think I found the culprit. It seems that this loop 1527a4f#diff-c1280fbe0804a30c1ff2760dd4265b6211c720b2b03801fab43d4dd4d56c5e29R488 is very slow (1000 constraints/11 seconds) and when you have a problem, like me, with ~0.5 mil constraints that makes it take approx 1.7h to loop through it. Can you try if this is enough information?

Edit: I will make a PR, I solved the problem most likely.