Headless on HPC cluster

1

Hello,

Great program and i’m just recently starting out with it.

Has anyone successfully installed/used/built shapeworks on a HPC cluster system?
I’ve used it on my Mac and setup parallel grooming/optimization python scripts for a multi-label spine segmentation. Works great, just slow since I’m processing each label separately due to spine curvature position issues. Now I want to run these on a HPC cluster to speed up processing in parallel.

However, I’m struggling to get it running/built.
The binary linux install has GLIBC errors (below).
Building from source has so far been hit with version/config issues in various packages that I’m working to resolve, especially since it is a non-gui system the vtk/qt packages need to be compiled.
Is there a docker/container out there that would work? I couldn’t seem to locate one.
Any help would be appreciated.
thanks!

basic system info:
Linux ln02 4.18.0-425.19.2.el8_7.x86_64
NAME=“Rocky Linux”
VERSION=“8.7 (Green Obsidian)”
ldd (GNU libc) 2.28
gcc (GCC) 9.3.0

for binary linux install:
Error message: /lib64/libm.so.6: version `GLIBC_2.29’ not found (required by […..]/ShapeWorks-v6.6.1-linux/bin/shapeworks_py.cpython-39-x86_64-linux-gnu.so)

2

fwiw, for anyone aiming to do the same, I was able to successfully:

  • create a docker container (ubuntu:22.04)
  • install shapeworks w/ the conda env
  • save the docker image (.tar) and convert to an apptainer (.sif) file
  • run this in headless mode through on the HPC cluster using the python api to groom.run() & optimize.run()

I get about an 10-fold overall speed improvement depending on the number of parallel processes, so well worth the effort up front for my situation. let me know if details are needed.

3

That’s great info for those who can use Docker/Apptainer on their HPC cluster.

Meanwhile, we’ve just now moved the Linux CI/CD build to use ManyLinux for GLIBC 2.28 compatibility, so it should now run out of the box on Rocky Linux 8. You can try it out using the development version of ShapeWorks:

4

excellent. It does install and python can at least import shapeworks and load an existing project out of the box. thanks! (so much easier).