Missing converters directory on jobs containers
Closed this issue · 7 comments
I came across the case of a tool (Query tabular using sqlite) that when given a CSV as input, it tried to convert the file into tabular before running, but failed with:
python: can't open file '/galaxy/server/lib/galaxy/datatypes/converters/tabular_csv.py': [Errno 2] No such file or directory
I'm not sure of an easy way out of this. Perhaps we might need to copy the converters or datatypes to the shared filesystem and configure galaxy to consume those from a different path? Or perhaps this metadata conversion step should run using the galaxy container itself? I'm guessing it is running that step on the tool's container. The whole job got handled a bit weird, in the sense that it didn't show as failed but as paused (saying that the input file was in error - sure, the CSV wasn't converted as desired into tabular).
This can be reproduced, in a slightly different manner (job shows up as failed), by using the converter functionality in the dataset view.
This used to be handled in the mapper script here:
specifically came up forcsv_to_tabular
but the fix was universal to all converters back in the day (#348). I guess that was lost with the switch to the new rule mapper, so an equivalent rule probably needs to be added to map all converters to the base galaxy container. I am not familiar with the new rule mapper to easily do it, but hopefully that helps. @nuwang might be able to help figure out how to best add it to the new configI had completely forgotten about this @almahmoud. That should have got ported over during the conversion to TPV, but I had accidentally dropped it. Have made a PR with a fix, but would be great to get it tested @pcm32
Thanks for the fix! Does this require the newer tpv? We haven't migrated yet. To understand how far I would be of testing this.
This should work on any version of TPV.