Resource profiler callback seemingly not working

[oesteban]

Summary

We need to give some love to MultiProc and its resource profiler.

Actual behavior

I see two problems in the current implementation of the resource profiler.

1. The resource calculation only happens to interfaces derived from CommandLine (https://github.com/nipy/nipype/blob/master/nipype/interfaces/base.py#L1543-L1544).

2. Even for those, when I get to the runtime log I set for the callback, I always see the post-callback entries like:

callback:DEBUG {"name": "GenerateReport", "id": "GenerateReport.a0", "estimated_memory_gb": 1, "num_threads": 1, "finish": "2017-05-05 17:23:41.020064", "runtime_threads": "Unknown", "runtime_memory_gb": "Unknown"}

So both runtime_threads and runtime_memory_gb are "Unknown". This could be a consequence of the containerized execution.

Expected behavior

These should be measurements, at least for CommandLine-derived interfaces.

How to replicate the behavior

docker run -i -v /data/bids-dir:/data:ro -v $SCRATCH:/scratch -w /scratch poldracklab/mriqc:latest /data/ out/ participant --testing --verbose-reports --profile --n_procs 8 --ants-nthreads 4 --ica

[chrisgorgo]

Was profile_runtime set to true in the config file (it has to be in the file not set programmatically). See nipreps/mriqc#459 (comment)

[ccraddock]

If we could let's differentiate between multiprocessor and the resource profiler ;). We haven't given much attention to the profiler since Satra disabled it after noticing some odd behavior. But last I knew it was implemented for all interfaces. Maybe a merge wasn't complete or something. Unfortunately Danny has moved on.

…

Sent from my iPhone

On May 5, 2017, at 8:31 PM, Oscar Esteban ***@***.***> wrote: Summary We need to give some love to MultiProc. Actual behavior I see two problems in the current implementation. The resource calculation only happens to interfaces derived from CommandLine (https://github.com/nipy/nipype/blob/master/nipype/interfaces/base.py#L1543-L1544). Even for those, when I get to the runtime log I set for the callback, I always see the post-callback entries like: callback:DEBUG {"name": "GenerateReport", "id": "GenerateReport.a0", "estimated_memory_gb": 1, "num_threads": 1, "finish": "2017-05-05 17:23:41.020064", "runtime_threads": "Unknown", "runtime_memory_gb": "Unknown"} So both runtime_threads and runtime_memory_gb are "Unknown". This could be a consequence of the containerized execution. Expected behavior These should be measurements, at least for CommandLine-derived interfaces. How to replicate the behavior docker run -i -v /data/bids-dir:/data:ro -v $SCRATCH:/scratch -w /scratch poldracklab/mriqc:latest /data/ out/ participant --testing --verbose-reports --profile --n_procs 8 --ants-nthreads 4 --ica — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[oesteban]

@ccraddock thanks for pointing out. I've changed a few things on the issue.

@chrisfilo will check ASAP

[oesteban]

@chrisfilo: seems like adding the option did not change anything (https://circleci.com/gh/oesteban/mriqc/1017)

[oesteban]

The callback is still available, but it is not necessary to get profiling reports after #2200.