I just found and fixed this issue for 2.4.3. Patch 45ea18c2103c44d0b5b0d4043090494d2d35a28b will fix it. It has to do with the start time of the job being after the start time of the loop. The patch is pretty simple and should work on any 2.4 system.
Danny On 09/05/12 09:33, Danny Auble wrote: > Lennart/Miguel, > > The chaos version Miguel pointed to is actually very old code. If you > point to the schedmd github version you will get the latest code which > looks the same as the snip he posted earlier. > > https://github.com/schedmd/slurm/blob/master/src/plugins/priority/multifactor/priority_multifactor.c#L463 > > > > Perhaps you could add some debug around there and print out variables > or an xassert if the factor turns to 1. > > Danny > > On 09/05/12 06:21, Lennart Karlsson wrote: >> On 09/05/2012 02:32 PM, Miguel Méndez wrote: >>> Sorry, that link is for code's latest version. It's a bit different >>> for the >>> one I sent you. >> Hi Miguel, >> >> Yes, thanks, it is easy to believe that the code sets >> the value to one, according to one of the "if" cases >> that you showed. >> >> I can not see any problems with the specifics of a >> "scontrol show job" on the jobs, that get the >> maximum value instead of the lowest. (Perhaps some >> values, like the begin value, are not always correctly >> initiated in some other part of the code?) >> >> I can not see anything about bug fixing of Age >> Priority calculations in the NEWS file of next >> version, 2.4.2. >> >> If I am very lucky, the changed code might indicate >> that someone already is working on fixing a >> bug that I am now suffering from? >> >> Cheers, >> -- Lennart Karlsson >> UPPMAX, Uppsala University, Sweden >> http://www.uppmax.uu.se >
