Re: [Condor-users] Core files

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

Miskell, Craig wrote:
> > Kewley, J (John) wrote:
> > > Are you on Linux?
> > >
> > > If so, you may have a "limit" for corefiles.
> > >
> > > Try typing in "limit" and see what it says, it might be that.
> > If you are on unix/linux, and you indeed have
> >     CREATE_CORE_FILES = True
> > in condor_config, then you should get core files dropped into 
> > Condor's 
> > LOG directory.  Note that CREATE_CORE_FILES will 
> > automagically set the 
> > limit for you as John suggests above, but only if you start the 
> > condor_master as root.  IIRC, if you are starting the 
> > condor_master as a 
> > non-root, you may need to setup the limit beforehand since non-root 
> > users cannot push the limit past any predefined hard limit.
> I'm running on linux (Centos 4), condor 6.8.6.  The bash builtin ulimit
> reports "unlimited", and I've not configured any specific limits on the
> nodes, besides which condor_master starts as root (then child processes
> drop to the "condor" user).  
>
> My LOG dir is an NFS mount on the compute nodes (back to a central
> server, for ease of management).  I've double checked, and there's no
> core files in any of those directories; I wouldn't have thought being an
> NFS mount would stop core files being generated, but would welcome any
> suggestions to the contrary.  

Hmmm....  if you have root squash activated on the nfs mount on the 
volume with your LOG directory, that could be the culprit.

Dropping a core file still requires proper permissions.  The core file 
is being written as user "root" by the startd since you started the 
master as root -- but if root squash is on, then it won't have 
permission (unless you are keeping your LOG directory with world write 
access!!).

-Todd