Question: Significance of different terminal_output values when not explicitly set?

[sgiavasis]

Hello,

I am looking through a few Nipype crash files of a colleague. They had a few FLIRT nodes crash (returning exit status -9), but only for a small subset of their data. Running the actual FLIRT command in command.txt without modification completes successfully, so I'm having them re-run the workflow to see if it happens again.

The only difference in the node information between one of the ones that crashed, and one of the ones that completed, happens to be the "terminal_output" setting (which isn't explicitly set in the workflow/node setup). In the nodes that ran, this is set to "allatonce" (as seen in the report.rst), and in the ones that crashed, this is set to "stream" (as seen in the crash file). Does this happen to have any significance/could it be related, or does it just get set to stream when the node fails? I'm not sure why these values would differ on their own.

The workflow was run on Nipype 0.12.0, on a server (not a cluster), with a local install of Nipype (no container involved, etc.).

Thanks!

[mgxd]

@sgiavasis did rerunning the workflow fix this? There have been some changes to terminal_output recently (#2209), which may have helped this problem, but we'll monitor this going forward

[oesteban]

@mgxd thanks a lot for bringing this up: this could have some relation to the memory issues we are seeing.

@sgiavasis do you happen to keep a copy of these crashfiles? It'd be great if you gave us a hand debugging this, it may turn out to be really important. Trying the latest version of nipype may have some impact, however the changes Mathias mentioned should not have affected the 'stream' output mode - you should see the same errors.