Dear devs
I've recently encountered some strange issues when running the htseq-count and deseq2 tools installed from the toolshed onto our production instance. In both cases there have been problems with a couple of the conda dependency installations.
Specifically:
- For htseq-count, runs failed with an error message about a missing libbz2.so.1.0.
- For deseq2, runs failed silently first with the same "missing libbz2.so.1.0" message, then when this was fixed, with a new message about an undefined symbol in libreadline.so.6.
In both cases I seemed to be able to fix the problems manually by activating the appropriate conda environments (under .../tool_dependencies/_conda/envs) and reinstalling the affected packages from different sources:
- htseq-count: in __samtools@1.3.1, reinstalled bzip2 1.0.2 from 'conda-forge' rather than 'defaults'
- deseq2: in __r-getopt@1.20.0, did the same for bzip2 and also got readline 6.3 from 'conda-forge' rather than 6.2 from 'defaults'
However: deseq2 installed into a local Galaxy instance doesn't seem to have this problem, so I'm guessing it's something peculiar to the setup of the production instance.
Can anyone suggest what might be happening here (e.g. some issue with the conda config for the production instance?), and how it might be avoided in future?
Thanks for your help,
Best wishes
Peter
(PS if it helps: the production instance is running Scientific Linux 6.8, and has recently been updated from v16.07 to v17.05. Both the tools were installed prior to the upgrade.)