QSIPrep stuck at eddy - Docker (Windows)

[Anomandaris-Purake]

Summary

Hi. Absolut novice here. Need expert help!
While running QSIPrep the process stucks at executing "eddy"

Additional details

• QSIPrep version: 26.0.0
• Docker version: 4.53.0 (211793)
• Singularity version: ---

What were you trying to do?

Trying to preprocess MRI scans (HARDI). Windows 11.
Processor: Intel(R) Core(TM) Ultra 7 255U (2.00 GHz)
Installed RAM: 32.0 GB
System type 64-bit operating system, x64-based processor
WSL settings: processors 6, Memory size 26000 MB, swapfile size 16000 MB

What did you expect to happen?

Continue and finish the process.

What actually happened?

I believe it finishes TOPUP; then trying to start "eddy", but ends in an endless loop. It seems that there is an issue with how TOPUT and eddy interact with my data.

Reproducing the bug

Assumed it was an issue with how many threads were active or allocated memory, so I tested different combinations, and the issue persists even with just one thread. I deleted the fmap folder in BIDS structure; the problem persists. Tried to pull QSIPrep from a clean state in Docker (after purge); the same. I moved the temp folder; the same. Tried the process to another windows PC with GeForce GPU and enabled the cuda option; again stack at executing the "eddy". No .nii.gz files are created. It doesn't even crash, it goes straight to a loop. I even tried QSIPrep 1.1.1 and it was the same. What am I doing wrong? Or is this a bug or something?

Codes I used.

docker run --rm -it `
>>     -v "path\BIDS:/data:ro" `
>>     -v "path\output\QSIPrep:/out" `
>>     -v "path\output\QSIPrep\scratch:/work" `
>>     -v "path\license.txt:/opt/freesurfer/license.txt" `
>>     -v "path\eddy_config.json:/eddy_config.json" `
>>     pennlinc/qsiprep:26.0.0 `
>>     /data /out participant `
>>     -w /work `
>>     --participant-label pilot `
>>     --output-resolution 2.0 `
>>     --b0-to-t1w-transform Rigid `
>>     --unringing-method mrdegibbs `
>>     --nprocs 6 `
>>     --omp-nthreads 6 `
>>     --mem-mb 26000 `
>>     --verbose

or

docker run --rm -it `
>>     -v "path\BIDS:/data:ro" `
>>     -v "path\output\QSIPrep:/out" `
>>     -v "path\output\QSIPrep\scratch:/work" `
>>     -v "path\license.txt:/opt/freesurfer/license.txt" `
>>     -v "path\eddy_config.json:/eddy_config.json" `
>>     pennlinc/qsiprep:26.0.0 `
>>     /data /out participant `
>>     -w /work `
>>     --participant-label pilot `
>>     --output-resolution 2.0 `
>>     --b0-to-t1w-transform Rigid `
>>     --unringing-method mrdegibbs `
>>     --nprocs 6 `
>>     --omp-nthreads 6 `
>>     --mem-mb 26000 `
>>     --eddy-config /eddy_config.json `
>>     --verbose

or

docker run --rm -it `
>>     -v "path\BIDS:/data:ro" `
>>     -v "path\output\QSIPrep:/out" `
>>     -v "path\Pilot\scratch:/work" `
>>     -v "path\license.txt:/opt/freesurfer/license.txt" `
>>     pennlinc/qsiprep:26.0.0 `
>>     /data /out participant `
>>     -w /work `
>>     --participant-label pilot `
>>     --output-resolution 2.0 `
>>     --b0-to-t1w-transform Rigid `
>>     --unringing-method mrdegibbs `
>>     --nprocs 4 `
>>     --omp-nthreads 4 `
>>     --mem-mb 26000 `
>>     --fs-license-file /opt/freesurfer/license.txt `
>>     --separate-all-dwis `
>>     --distortion-group-merge average `
>>     --verbose

[tsalo]

Can you clarify how you concluded it got stuck in a loop? In my data, EDDY can take several hours, but it does eventually finish.

[Anomandaris-Purake]

Hi! Does it, in your case, produce any files that are slowly getting bigger? Something to verify that is actually working, even if it takes hours?
In my case, it created just some files, smaller than 10 kb each, that I'm guessing would be the commands it would run, and then nothing. I left it for maybe 12 hours or more, multiple times. Nothing changed; there was no evidence that it was working on something. CPU throttled at the same level, memory consumption stayed constantly high, without any fluctuations. It didn't crash to give me an error log, so I deduced it must have gotten stuck in a loop. Other than that, I don't have any evidence, though.

[tsalo]

It's possible that it does, but I run QSIPrep with a job scheduler so I just check the job status after it finishes. I'm pretty sure I had a couple of subjects in my last job that stayed on EDDY for ~12 hours, so maybe your run is just taking a while to finish. @mattcieslak WDYT?

[mattcieslak]

eddy run times can be VERY long, especially if you have a lot of data. If memory and cpu are staying high it's probably just working its way through. If you have a gpu it will finish much more quickly

[Anomandaris-Purake]

Thank you for your replies. So, I might have been impatient? I can try one more time and leave it for 24 hours, and see if it proceeds. I was really hoping to have some kind of evidence, a growing file maybe, to know that it will eventually finish... It's a leap of faith right now for me. I've been trying for 5 days, and nothing has worked.
Unfortunately, I don't have a GPU on this PC, got the same problem even with another PC with GPU. Would an external GPU also work?

[mattcieslak]

I didn't recognize originally that you're running WSL. You might be the first person I'm aware of that's tried that. This is uncharted territory

[Anomandaris-Purake]

I didn't recognize originally that you're running WSL. You might be the first person I'm aware of that's tried that. This is uncharted territory

Well, there's that :)
I'll keep a low profile for now. I did run QSIPrep about 2 months ago, but it wasn't the same version (can't remember which, probably not the latest). Didn't have any issues then. So, something must have changed since then.