Replies: 3 comments 5 replies
-
In LHCb we extend the output upload of vanilla DIRAC by allowing a second replica (look for We also have resolution logic for where to upload the output for production jobs. However, I would recommend not to upload more than one replica from the job. I would suggest you upload one replicas, and then asynchronously (RMS/FTS machinery) replicate wherever you like. This is much more efficient and reliable in many aspects |
Beta Was this translation helpful? Give feedback.
-
Answer to your case: as of today it is not possible, this would require development. This discussion can be turned in an issue to be implemented later on. RMS can be fed by DIRAC transformations as @adriansev pointed out above, but not only, and in fact the way your case could be treated would be by inserting directly in RMS. DIRAC's code to handle this case was started long time ago but never fully finished: https://github.com/DIRACGrid/DIRAC/blob/integration/src/DIRAC/Workflow/Modules/UploadOutputs.py . Modifications to the Job APIs would also be necessary. I can turn this into an issue, and I can provide guidance for the implementation (I would not have time to do this myself). |
Beta Was this translation helpful? Give feedback.
-
Concerning the possible AugerDIRAC extension. If deemed necessary, it can be discussed how this can be installed in the DIRAC EGI service. This is not impossible, including Pilots |
Beta Was this translation helpful? Give feedback.
-
Would be possible to make setOutputData of job to read a construct like:
outputSE=replica:2,SE1,SE2,!SE3
which would mean: make 2 random replicas beside the explicitly requested SE1 and SE2 (so, a total of 4 replicas) but do not use SE3
or
outputSE=replica:3,!SE4
that would mean: just make 3 random replicas but do not use SE4
This would allow flexibility in data placement from the beginning of job production
Thanks a lot!
Beta Was this translation helpful? Give feedback.
All reactions