- As user vlapipe, either copy from spool, or download from the archive, a calibrated MS to /lustre/aoc/cluster/pipeline/dsoc-prod/test_imaging/<JobID>/working (e.g. /lustre/aoc/cluster/pipeline/dsoc-prod/test_imaging/19B-172_2019_12_01_T23_25_29.441/working)
- Copy casa_imagingpipescript.py from /lustre/aoc/cluster/pipeline/dsoc-prod/test_imaging/scripts into the working directory, and edit it to change the MS name.
- Copy run_casa.sh from the scripts directory into the working directory and change the working directory (#PBS -d). Make sure it is executable.
- Login to nmpost-master, cd back to the working directory, then, as the vlapipe user, type qsub run_casa.sh
- Review the weblog in products and check the output image in the viewer.
- Make a qa_notes.html file with observations on how well the imaging went if there is anything to add on top of what's in the weblog.
- Copy the products directory, including the weblog to /lustre/aoc/cluster/pipeline/dsoc-prod/image-qa/<JobID>
For pipeline testing of the upcoming production version:
- As user vlapipe, either copy from dsoc-test/spool, or download from the archive, a calibrated MS to /lustre/aoc/cluster/pipeline/dsoc-test/test_imaging/<JobID>/working (e.g. /lustre/aoc/cluster/pipeline/dsoc-test/test_imaging/19B-172_2019_12_01_T23_25_29.441/working)
- Copy casa_imagingpipescript.py from /lustre/aoc/cluster/pipeline/dsoc-test/test_imaging/scripts into the working directory, and edit it to change the MS name.
- Copy run_casa.sh from the scripts directory into the working directory and change the working directory (#PBS -d and in the xvfb command line). Make sure it is executable
- Follow steps 4-7 above, but copy the products directory to /lustre/aoc/cluster/pipeline/dsoc-test/image-qa/<JobID>
{"serverDuration": 96, "requestCorrelationId": "d2753d2e692dc2c4"}
2 Comments
John Tobin
Do you only plan to run the imaging jobs single-threaded rather than using mpicasa? I suppose single threaded would be overall more efficient in terms of memory and throughput.
James Robnett [X]
I'd suggest serial at first and then see what the run time is. If it's sub 24 hours probably leave it serial, if it's > 24 hours parallelize. Runtime is likely to be highly variable.