On Oct 15, 2012, at 23:45 , John Nash wrote: > One of my students reported a crash while assembling a 5 mb prokaryote > genome. I got him to repeat it and the same thing happened. I have snipped > some of the 'head' and 'tail' of the assembly log, and can make the log > available to email back-channel (It's only 27 K). I have also attached the > manifest file. The log and checkpoint directories are intact - I can zip > them up and put them on a shared dropbox folder if needed. Stop the press! I.e., please keep the whole directory as it is, make a copy of it to run further tests if you want, but keep a complete copy archived somehow :-) > > The student has successfully assembled about 25 other bacterial genomes with > mira, so he pretty much knows his way around the process, therefore I would > discount user error. At the time, we had plenty of RAM and processors free > as we limit our jobs to avoid clogging the beast. We run a bit of a mira > factory so we are aware of clogging up our computer with mira jobs. Usually > we only run 2-3 mira jobs at a time on that server. So hopefully running too > many mira jobs is not the cause of the crash. > > I'll also get him to repeat the assembly with one processor, as I recall > something like this from the recent past - is this your elusive randomly > occuring SKIM bug? It's one of the two Heisenbugs which make me sweat (the other one being that something's completely broken on Mac for bigger projects). Let's see ... can you please test the following (after having made a complete copy): - in the run which failed, can you please run with the "-r" (resume) option. If it crashes, things are looking good. If not, I'd need the complete log of the run that crashed and would then send you a specially crafted version. - in parallel or if the above fails, re-running with one processor to see whether it crashes there would be nice. Short explanation of the above: I have the slight feeling that a part of the non-reproducability of the problem lies in memory allocation, so having a reproducible case would be really smashing. Best, Bastien -- You have received this mail because you are subscribed to the mira_talk mailing list. For information on how to subscribe or unsubscribe, please visit http://www.chevreux.org/mira_mailinglists.html