THOR 0.9.7 fail QXcbConnection

[ghost]

Hi,
I am running THOR 0.9.7 in batch mode on a compute cluster (=headless servers) and my runs fail with the following message:

QXcbConnection: Could not connect to display

Is that coming from the parameter --report, i.e., does it try to open the generated HTML report?
+Peter

[fabio-t]

Hi Peter,

we are going to try to identify the culprit of this. We don't attempt to open the report, but it might be some system library is required to generate it.

Can you tell us the Linux distro and version you are running it on? I'll give it a try in a virtual machine.

It's a long shot, but can you prepend this before running THOR?

export DISPLAY=:0

In the next version the --report option will be renamed --no-report so that you can skip that step.

[ghost]

Hi Fabio,
thanks for the quick reply. The sys config is as follows:

Debian 7.11
Linux 3.18.38.1

Thor was installed via pip in a Conda environment:

name: thor
channels:
  - bioconda
  - conda-forge
dependencies:
  - python=2.7
  - numpy=1.11.0
  - scipy
  - scikit-learn
  - pysam
  - htseq
  - hmmlearn=20151031
  - mpmath
  - cython
  - pip
  - pip:
    - RGT

Don't get me wrong - I like tools that output a human readable report (well, haven't seen it so far, but I assume that ;-) ) - if that is indeed the source of the error, I hope there is a solution that lets one keep the report.

I'll try the export DISPLAY=:0 stuff and report back to you.

[fabio-t]

Thanks. It would also be helpful to get the full error message/stacktrace, if you have it.

[ghost]

From what I got back from the job scheduling system, there seems not to be more - it just says:

[some warnings about index file being older than alignment file]
QXcbConnection: Could not connect to display
Aborted

[fabio-t]

I just noticed that you are using 0.9.7. Can you update to 0.9.8? It might actually be fixed, if it's what I think.

[ghost]

Ok, the runs finish with THOR 0.9.8, just two failed with memory errors, but that is unrelated.