LHC@home

After 35000 events all events throwing away. Job will stop after 10000 rejections.

From batch 161205_162337:ireid_crab_BPH-RunIISummer15GS-00046_BD job 7228

found this answer in vLHCathome:
https://lhcathome.cern.ch/vLHCathome/forum_thread.php?id=1834&postid=21434#21434

Maybe the 10,000 rejections can safely be reduced to e.g. 1,000, so the VM can start faster with the next job.
I also had such a case after only 3,000 events from 200,000 events.
The VM was 2 hours useless consuming CPU, before the job was aborted by itself.

Maybe the 10,000 rejections can safely be reduced to e.g. 1,000, so the VM can start faster with the next job.
I also had such a case after only 3,000 events from 200,000 events.
The VM was 2 hours useless consuming CPU, before the job was aborted by itself.

OK, I need to find out if that's a configurable parameter. Please remind me if I don't respond within a couple of days... :-)

---

Maybe the 10,000 rejections can safely be reduced to e.g. 1,000, so the VM can start faster with the next job.
I also had such a case after only 3,000 events from 200,000 events.
The VM was 2 hours useless consuming CPU, before the job was aborted by itself.

OK, I need to find out if that's a configurable parameter. Please remind me if I don't respond within a couple of days... :-)

After a bit of gwgling I found some code at http://svn.cern.ch/guest/evtgen/tags/R00-04-00/src/EvtGen.cpp:
int times=0;
do {
times+=1;
EvtStatus::initRejectFlag();
p->decay();
//ok then finish.
if ( EvtStatus::getRejectFlag()==0 ) {
times=0;
}
...
if ( times==10000) {
report(ERROR,"EvtGen") << "Your event has been rejected 10000 times! << endl;
report(ERROR,"EvtGen") << "Will now abort."<< endl;
::abort();
times=0;
}
} while (times);

So, if that is indeed the code we are using, then there doesn't seem to be an easy way of changing the limit.

[Edit] The exact same code is in the latest release :-( [/Edit]

---

...
if ( times==10000) {
report(ERROR,"EvtGen") << "Your event has been rejected 10000 times! << endl;
report(ERROR,"EvtGen") << "Will now abort."<< endl;
::abort();
times=0;
}
} while (times);

So, if that is indeed the code we are using, then there doesn't seem to be an easy way of changing the limit.

Yes, that's exactly the text, I've seen in the Running.log.
The easy way would be skipping 1 zero (in statement and text to report), but of course I've no idea what's the impact elsewhere.
The programmers maybe can tell why 10,000 times is set. I suppose it was rather random chosen. At least a lower value would decrease (almost) idle CPU-time.

btw: yesterday I had again such a task on the dev-project. Thrown away 10,000 times after about the 35,000th event out of 200,000.

I found 135 in one batch, so there's a bit over 1%. It's worth noting that this is in EvtGen (EventGenerator) which is separate from Pythia (but obviously used by it). It's included in CMSSW as an external, and as you can see from the code, there's no way we can make a change to it. Short of convincing the developers to make the change...

---