[GEN Check Script] Adding a number of events check for pLHE

[DickyChant]

Recently we noticed that in production there could be requests mistakenly requesting more events than presented private LHE events. See [1] for an example.

Without a check we (GEN and PdmV) could not effectively identify this issue... We should brainstorm on how we could add this check.

1. The most straightforward way would be reading the whole files, though it would be very heavy workload
2. @vlimant proposed to skip events until the final one 1k event with a cmsDriver command.

More ideas are welcomed for sure.

[bbilin]

Hi,

I chose a random folder with many files: WJets sample at 8 TeV mcdbid: 5497.

The simple
[bbilin@lxplus9110 5497]$ grep "</event>" *lhe |wc -l

Returned the nevents =107350000 in a timely manner. This can hence indeed be included in the script to check nreq<= n available.

Cheers,

B.

[DickyChant]

I think one of the issue with this solution is however that we could have ".xz" compressed LHE files :(

[bbilin]

ah, at least vim was working on them, we can find a way out :)

[bbilin]

google helped. xzgrep works.

[DickyChant]

Look forward to a PR to close this then ;)

[agrohsje]

Just a reminder that vim created spurious copies in the past and screwed the system. Keep also in mind when running locally.

[bbilin]

yes, it causes problems in the event generation. We had to remove those swp files. (I am not using vim by default :) )

[DickyChant]

I forget to have [1]: https://its.cern.ch/jira/browse/HIGHPRIOREQ-825 ...

[Kiarendil]

Hi all,
the current check by @efeyazgan in #3817 xzgrep "</event>" /eos/cms/store/lhe/'+str(mcdbid)+'/*.lhe | wc -l does not allow to count the events within ".xz" compressed LHE files — obviously, it checks only ".lhe"
However, xzgrep "</event>" /eos/cms/store/lhe/'+str(mcdbid)+'/*.lhe* | wc -l works for them.
Thus, maybe such fix with an extra wild card is needed.

I have faced this issue with the validation failed for BPH-RunIII2024Summer24pLHEGS-00001.

While I fully support the need of such a check for the production, I'm not happy with the current implementation within request_fragment_check.py script. It works only before the validation step at McM side or if MC contact uses script manually (as they should always do, of course...).
The problem is MC contacts can change events number for "validation" and "defined" McM statuses after validation, and the current implementation could not help with it.
Personally as MC contact I almost always adjust event number after the validation, thus it is easy to miss this check.

Also, the current check could take up to 10-20 minutes for the large folders (it took so for me for e.g. 20022 and 5497 ids). I'm not sure it is OK for the fast validation check.

Thus I would suggest maybe to remove this check from the request_fragment_check.py script to not waste (sometimes big) time on this (however leave a warning for all pLHE PrepIDs to check event number with xzgrep "</event>" /eos/cms/store/lhe/'+str(mcdbid)+'/*.lhe* | wc -l command) and rather implement this check on the McM side during approval / injection.
E.g. McM have checks during the injection to avoid identical dataset names. It should also check this for pLHE samples, this would be much better rather than current check during the validation.

Best,
Kirill