[RESOLVED] MCR issue in ICA+Fix

Issues
1

Hey Jure,

I am running into a MATLAB issue with the 1.1.1 version during ICAFIX. I tried all the different matlab modes, using different versions (1.0.3, 1.0.4) and re-building containers, but had no luck. Would you happen to know what might be causing this?

Thanks a lot!

Deniz

Call:

qunex_container run_recipe \
  --batchfile="/batch.txt" \
  --steps="hcp_icafix" \
  --sessions="RXXXX_3TSESS1" \
  --sessionsfolder="/sessions" \
  --hcp_filename="userdefined" \
  --hcp_icafix_bolds="REST:rfMRI_REST_AP,rfMRI_REST_PA|TASK:tfMRI_PVT_RUN1_AP,tfMRI_PVT_RUN2_AP" \
  --bash_pre="module load singularity" \
  --overwrite="yes" \
  --container="/qunex_suite-1.1.1.sif" \
  --scheduler="SLURM,cpus-per-task=10,partition=blade,mem=48gb"

Logs:

LD_LIBRARY_PATH is .:/opt/matlab/R2022b/runtime/glnxa64:/opt/matlab/R2022b/bin/glnxa64:/opt/matlab/R2022b/sys/os/glnxa64:/opt/matlab/R2022b/sys/opengl/lib/glnxa64
MCR:mclmcr:MCLMCR_CTF_MISSING_PRODUCTS
Error: Unable to access required products file. The CTF archive file may be corrupt. Please recompile the application.
2

Hi!

This seems like an issue with the compiled matlab code for ICAFix. Could you provide two more things, the full comlog for the run (found in processing/logs/comlogs) and your batch.txt. I want to check if you have any additional parmeters there.

And you said you get the same error regardless of the hcp_matlab_mode? This is weird, we now use pyfix as the default which is much more robust and we have a bunch of users running ICAFix with 1.1.1 (or 1.1.0). Furthermore, if you set --hcp_matlab_mode="octave" the processing should not touch MATLAB/MCR.

Are you getting the same error regardless of settings.

Best, Jure

3

Hey Jure,

Thanks! I have attached the logs you have requested below. The weird thing is that the code successfully runs for a few of the parallel sessions, but fails for the majority as I described above. Assuming that it might be some binding issue to the ./.mcrCache, I looked into whether there are any commonalities in the nodes that the call fails, but can’t seem to find any.

As for the octave version, I have the same issue in certain parallel sessions, but this time I get the following error and then the call just gets stuck there. So far, my only solution has been to resubmit jobs numerous times :slight_smile:

error: package nan is not installed
error: called from
    load_packages at line 47 column 7
    pkg at line 632 column 7
    /home1/deniz/.octaverc at line 7 column 1

Best,

Deniz

error_hcp_icafix_TASK_R0205_3TSESS1_2025-04-02_13.43.36.686565.log (8.6 KB)
Log-hcp_icafix-2025-04-02_13.43.36.564190.log (5.9 KB)
qunex_container_2025-04-02_13.42.59.586109.txt (14.6 KB)

4

Hi!

  1. Octave: It seems like you are loading an Octave package that is not in the container. Can you delete the file /home1/deniz/.octaverc? Are you using Octave and have this setup for something?

  2. MATLAB: You can try downloading the latest HCP Pipelines from the HCP repo and then bind those over the ones in the container with the snippet below. The error suggest some MATLAB compilation issues and the only compiled MATLAB code here is from HCP Pipelines.

--bind="<location_of_downloaded_hcp_pipelines>:/opt/HCP/HCPpipelines"

Best, Jure

5

Thank you very much for these suggestions Jure! I am going through them one by one.

  1. I have deleted the /home1/deniz/.octaverc which I believe is generated by the Qunex call. It reappeared after I re-ran ICA+Fix with the octave option and I don’t have it setup or using it for anything else. There are no errors in the batch output anymore and ICA+Fix seemed to have finished successfully (continuing with refix and msmall now). If I am not mistaken the issue was related to running 0.99.2d and 1.1.1 on two separate calls for two separate datasets (I am using 1.1.1 on a multi-echo dataset also with octave). These two octave-based calls might be interfering with each other.
  2. I will give this a go after trying out the octave version (which seems to be working now).

Thanks a lot again for your all your help and support!

Best wishes,

Deniz

6

Hi,

Yes, it could be that your .octaverc was from 0.99.2d which used some different libraries. Glad it works now.

Best, Jure

1 Like