Finishing and redoing bad files, Yet ~16M

PbPb13 and PbPb150 2018 v1r21p3-el9 productions 1M each for Andrey:

consumes 10-20GB of memory, example is at:

/afs/cern.ch/user/n/na61mc/work/private/YehorBondar/logs/epospbpb150logs.10944134_3456.log

/afs/cern.ch/user/n/na61mc/work/private/YehorBondar/logs/epospbpb13logs.10935416_0375.log

/afs/cern.ch/user/n/na61mc/work/private/YehorBondar/logs/pp_158_11_10912009_93.log

• I am running 20 times 15 events (PbPb150) or 10 times 160 events (PbPb13, 1GB per iteration was the goal); each time removing shared memory by ipcs -m | awk '{if(($3=="na61mc") && ($6==0)) print $2}' | xargs -I {} ipcrm -m {}
• /usr/bin/time -v ShineOffline for 32bit shows 1-2GB memory usage, measuring ds memory is not possible(?): grep "Maximum resident set size (kbytes):" ~/work/private/YehorBondar/logs/epospbpb150logs.10947974_0012.out
• valgrind --tool=massif  doesn't work, there are no 32bit versions of valgrind tools on htcondor machines (but on lxplus there are)
• some jobs failed with std::bad_alloc
• As opposite, PbPb30 on CentOS7 uses ~2.5GGB, example at /afs/cern.ch/user/n/na61mc/work/private/YehorBondar/logs/epospbpb30logs.10948142_30001.log