Different mathcing results on two machines

I'm using OpenSfM on different machines, one is a workstation with Ubuntu and the other is a server with Oracle Linux.
After running a set of 176 images, I get successful results on the workstation, but always get failed matching results on the server (0 robust matches for all images and consequently 0 tracks and 0 reconstructions).

Successful results on the workstation: (e.g.)
2021-09-03 12:44:19,093 DEBUG: Matching G0030109.JPG and G0030114.JPG. Matcher: FLANN (symmetric) T-desc: 0.547 T-robust: 0.014 T-total: 0.561 Matches: 139 Robust: 76 Success: True

Failed results on the server: (e.g.)
2021-09-02 23:43:34,546 DEBUG: Matching G0030109.JPG and G0030114.JPG. Matcher: FLANN (symmetric) T-desc: 1.122 T-robust: 0.492 T-total: 1.614 Matches: 132 Robust: 0 Success: False

I'm using exactly same config.yaml and I've confirmed the data is fine since I got successful results on the workstation and other SfM softwares (e.g. COLMAP).
Also, I've confirmed the extracted SIFT features are same (exactly same size of features.npz) on both machines.

What would be the causes of the inconsistent matching results?

1. Different version of Python dependencies (e.g. Numpy, opencv-python) on different machines?
2. Different version of OpenSfM? It was installed on the server by our lab administrator, but I've tested older and newer versions on the workstation and all got successful results.
3. Any other possible reason?

Compare the Python, OpenCV, and OpenSfM versions on both machines, then review the matching logs for the same image pairs. Check whether the identical config.yaml and features.npz produce different matcher behavior. Done means identifying the environment difference that causes zero robust matches or documenting a reproducible cause.