[GE users] Effects of changing network static ip and dns name

Andy Schwierskott andy.schwierskott at sun.com
Fri Sep 14 15:46:50 BST 2007


what hostname is configured in the "act_qmaster" name file? Is it the
external name or the internal name? Do you use SGE host aliasing

If it's the internal name and a change is made then qmaster would not update
the act_qmaster file during runtime and clients could not reach qmaster -
that would almost certainly cause a problem.

However shutting down qmaster and restart it would just cause a very short
outage of your qmaster service - running and pending jobs would not be
affect - just a qsub and other q* commands would fail for that short period.
The bigger risk here might be to find the right time when to restart qmaster
- you hardly have a guarantee whne a DNS changes is propagated to the client
machine - DNS caches and local nscd caches will delay actual propagation.
Using the SGE "gethostbyname" utility of course would answer that question.
If qmaster does not come up when you try to restart it it's likely the case
the the current name does not match the name in the act_qmaster file - in
this case just update the act_qmaster file manually and rerun the sgemaster
script wiht the current hostname (probably shortend by the domain suffix) if
ignore_fqdn is set to true.


On Fri, 14 Sep 2007, John Hearns wrote:

> On Fri, 2007-09-14 at 06:55 -0700, Brett W Grant wrote:
>> I have a private network set up with SGE v6.1.  The entire cluster is
>> connected to one nic as a private LAN.  The head node, where qmaster
>> is running, has a second nic that connects to the company's LAN.  The
>> only thing this second nic is used for is remote access, the grid runs
>> just fine whether it sees the LAN or not.  This second nic was
>> assigned an ip address through dhcp, which I requested that the
>> company change it to a static ip so I could use it more as a server.
> Sounds OK.
>>   Anyway, when they assigned the static ip, the dns name also
>> changed.
> Again sounds OK.
> You can change the setup now, no need to wait.
> In fact, why do you need to change any setup?
> As you say, SGE is sensitive to hostnames (no names, no packdrill but
> half an hour in a cold machine room yesterday taught me this.
> It is also sensitive to some bozo changing the ownership of act_qmaster
> to root rather than sge. Ahem.)
> So all you need to do is make sure that the master host resolves its own
> name to the INTERNAL interface. Which you already have done.
> Call the EXTERNAL interface anything you like within its own /etc/hosts
> file. In fact, you don't really need any entry for it at all.
> For external traffic coming in, those hosts will use the DNS to resolve
> the name and will get the external IP address.
> Making this clear - make changes and use the gethostbyname and gethostip
> utilities to make sure the machine itself always resolves to the
> internal address.
> ---------------------------------------------------------------------
> 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