[HTCondor-users] Hierarchical groups redefined? Throw away your accounting database!

[Jeff Templon <templon@xxxxxxxxx>]

Yo

Just got through a thoroughly satisfying debugging session (how lovely it is, cursing at the software running on machines behind your screen), trying to understand why the quota / priority dance looked like a waltz even though the music was in 4/4 time.  One group was obviously way over quota and still the first being served.  After slicing and dicing the output of condor_userprio many different ways, and scrutinizing the NegotiatorLog tables of usage and such, it became clear that something was off in how the âWeighted in Useâ values of individual submitters was being summed to reach the group value, that group value being the thing that is used in the priority / quota dance.

Long story short: I once had subgroups for ATLAS, and then after seeing how confusing that was, I dropped them and just had a top level ATLAS group.  However, since the Accounting log file is now some transactional thing, those transactions of having subgroups WERE STILL IN THAT FILE and had some weird effect on how things were being added up.

The solution : condor_off on the central manager, remove the $(SPOOL)/Accountantnew.log file, then condor_on

Now the weighted in use values sum up to the correct group value, and the priority / quota dance behaves as it should.

Maybe worth a word in the documentation?

JT