[GE users] Hostname checking

Daire Byrne Daire.Byrne at framestore-cfc.com
Thu Nov 8 18:07:38 GMT 2007


    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]

Daniel,

----- "Daniel Templeton" <Dan.Templeton at Sun.COM> wrote:
> Yes, Grid Engine deals very poorly with dynamic DNS environments.
> First off, I seem to remember that at some point we modified the
> qmaster 
> to allow operations on host groups even if all the hosts aren't 
> resolvable at the time.  Guys?

This would be useful or at the very least some sort of flag to qconf to not do DNS lookups would be nice. These machines drop out of DNS regularly and SGE doesn't seem to care until you try to edit the hostlist. Really we would like hosts to add themselves to SGE on bootup so that they automatically maintain the hostlist themselves. There may be a race condition doing this though....

> What you're describing with the host_aliases file would work in 
> principle.  When the host_aliases file is in place, all hostname 
> resolution passes through the alias mapping first.  The mapping will 
> also apply to qconf because everything qconf knows comes from the 
> qmaster, and the qmaster does the mapping.  With such an alias file in
> place, the qmaster will automatically treat every reference to 
> sys1.dhcp.local as though it were sys1.static.local.  No attempt will
> ever be made to resolve sys1.dhcp.local.  If sys1.static.local is 
> unresolvable, the host is unresolvable, period.

Yea doesn't sound like that'll work then - the dhcp.local IPs are dynamic so mapping them to the static ones wouldn't work.

Thanks,

Daire


> Do note that you have to restart the qmaster for the mapping to take
> effect.
> 
> Daniel
> 
> Daire Byrne wrote:
> > Daniel,
> >
> > Thanks for the response. Neither suggestion sounds particularly
> practical. Basically Gridengine doesn't support dynamic DNS
> environments very well atm. We have 600+ desktop machines which drop
> in and out of DNS all the time. So if a machine in the hostlist is
> just turned off for a week and I try and edit the hostlist I need to
> find that machine and remove it first before I can add any more
> machines. A real pain when you have 600+ machines where any 10-20 can
> be unresolvable at any time.
> >
> > Do you think it'd be possible to use the host_aliases file? Could we
> have sys[1-999].dhcp.local as DYN-DNS hostnames and map them to static
> DNS entries like sys[1-999].static.local? The static hostnames would
> never be in use but at least they would always be resolvable.... Not
> sure that the host_aliases file is considered for qconf settings
> though. If it does would SGE use the IP of the dhcp.local hostname if
> it resolves but if not use the static.local IP?
> >
> > Cheers,
> >
> > Daire
> >
> >
> >
> > ----- "Daniel Templeton" <Dan.Templeton at Sun.COM> wrote:
> >   
> >> Daire,
> >>
> >> There is a super-secret hidden feature that might help you.  It's
> >> never 
> >> been tested as a tool in production, though, so be warned.
> >>
> >> First, let me say that the usual advice in your situation is to 
> >> temporarily add the host to the master's /etc/hosts file, do what
> you
> >>
> >> need to do, then remove the host.  Assuming that's not an option,
> >> there 
> >> is a magic environment variable called SGE_COMMLIB_DEBUG_RESOLVE. 
> If
> >>
> >> you set it to a list of host names, those hosts will always be 
> >> considered resolvable by the comm lib.  The danger is that the
> comm
> >> lib 
> >> gets an empty data structure back for those hosts' contact
> >> information.  
> >> For just removing a host from a hostgroup, that *shouldn't* pose a
> 
> >> problem, but no guarantees.  The worst that could happen is that
> the 
> >> qmaster could segfault.
> >>
> >> Daniel
> >>
> >> Daire Byrne wrote:
> >>     
> >>> Hi,
> >>>
> >>> Is there way to turn off Gridengine's hostname checking? We use
> >>>       
> >> dynamic DHCP for all our desktops and sometimes a host will be
> retired
> >> and so the DNS entry will automatically be removed after the lease
> >> times out. When we then try to modify a hostlist (e.g. to add some
> new
> >> hosts) it will complain that the "retired" hosts don't exist (no
> DNS
> >> entries). It is very annoying to have to remove the non-existent
> >> hostnames. Is there any hidden control over Gridengine's hostname
> >> resolution? I believe you can toggle on/off FQDN resolution at
> >> installation time - is that it? 
> >>     
> >>> I can't see how it is harmful to have non-existent hostnames in
> the
> >>>       
> >> hostlists. If there is no easy way to do this then I'm thinking
> we'll
> >> have to write a daemon to keep track of all currently active
> desktop
> >> machines and update the hostlists dynamically every 5 mins or so.
> >>     
> >>> Any help/thoughts appreciated.
> >>>
> >>> Daire
> >>>
> >>>
> >>>
> >>>       
> >>
> ---------------------------------------------------------------------
> >>     
> >>> To unsubscribe, e-mail:
> users-unsubscribe at gridengine.sunsource.net
> >>> For additional commands, e-mail:
> >>>       
> >> users-help at gridengine.sunsource.net
> >>     
> >>>   
> >>>       
> >>
> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> >> For additional commands, e-mail:
> users-help at gridengine.sunsource.net
> >>     
> >
> >
> ---------------------------------------------------------------------
> > To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> > For additional commands, e-mail:
> users-help at gridengine.sunsource.net
> >
> >   
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list