[Bug] cuOpt Server crash - Internal Server Error 500 - {'cuopt_complete': -1}

[algomia]

Hi all!

I am new to cuOpt - so please be kind should any noobly issues arise here. I have experienced a strange bug/behavior, which sort of has the "look and feel" of a race condition. I am running cuOpt in a hosted Docker container (details below) and access it via API.

When I submit an input derived from an LP file (attached below) I get the following:

• cuOpt always finds an optimal solution (verified in the logs)
• Sometimes, the server returns a solution with a HTTP 200 message, and all is fine. That happens when the cuOpt logs report "{'cuopt_complete': 'Optimal'}".
• More frequently, the server crashes with a HTTP 500. The logs say "{'cuopt_complete': -1}" in that case. But strangely, the optimal solution was actually found. I have also noticed that the logs say "cuopt received signal 17".

Setup:

• Hosted at runpod
• GPU: A100 SXM x1
• Container image: nvidia/cuopt:latest-cuda12.9-py3.13
• Port: 5000
• cuOpt server: cuopt server version 26.04.000

Files:

• LP file:
  lowerbound.lp.standardized.zip
• Container log:
  container-logs.txt
• System log:
  system-logs.txt

Any help here is highly appreciated!

Cheers,
Alex

[akifcorduk]

Hi @algomia , thanks for sumitting the bug report! I have run the instance on both A100 and H100 machines on a loop for a long time with our binary and neither of the machines produced a crash. We have fixed many small bugs in the current release. Would you be so kind to rerun it with the 26.06 container once it released (very soon)?

[algomia]

Thanks @akifcorduk for the quick reply! Looking forward to trying it out ...