Re: [Condor-users] Problems on Mac OS X

[John Wheez <john@xxxxxxxxxx>]

dautret wrote:

> 1. I use Condor on G5' s cluster with Mac OS X. After few hours, 
> condor_master crashes on the submitter machine what leads the break of 
> the jobs. It seems that this pb occurs with many applications on mac 
> os X but no one in forum Mac has got an idea to solve this pb.
> This pb occurs with condor6.6.5 condor6.6.7 and condor6.7.2…
> So, has anyone use condor on mac os x and have you got these crashes ?


I run Condor 6.7.3 on a Power book G4 with 1 gig of memory and Condor 
seems very stable. It was also stable under 6.7.2.  I have it running 
jobs where a single node computes 400 processes in a row without a 
crash. My jobs are also fairly memory intensive. It is using the Shake 
program by Apple.
If you update to Condor 6.7.3 beware there is a slight bug and you will 
need to add this to your local config file or the classadd for OS type 
will be wrong:

OpSys="OSX"
STARTD_EXPRS=$(STARTD_EXPRS) OpSys


Here are some questions whihc might help debug the problem:

1) Do all programs cause crashes? Try to monitor teh ram usage on the 
machines..are they running out of memory or is the CPU activity suddenly 
always 100% which idicates some sort of crash....Perhaps the program you 
are running has some sort of bug in it which eventually causes the Mac's 
to crash. Like a memory leak..This was your error message  Exception: 
EXC_BAD_ACCESS (0x0001)    Codes: KERN_PROTECTION_FAILURE (0x0002) at 
0x00000000

2) Do all your G5's crash consistently? or is it a specific set of 
machines? Perhaps a few of your machines have bad memory chips..that 
happened to me on a fairly new computer and it caused teh system to 
crash many times.


> 2. Sometimes, only the manager machine crashes but not the submitter 
> machine… At this moment, condor stops… but when I launch again 
> condor_master on the manager machine, jobs restart although I have a 
> vanilla configuration…!
The jobs should not restart from the first process in the job 
cluster..they should continue from the  process # just before the crash 
i think. Is this what is happening or do your jobs restart from the 
first  process in the cluster of jobs.