Re: [gdm-list] gdm-binary CPU usage

[Brian Cameron <Brian Cameron Sun COM>]

Brandon Leger wrote:
>     Please make sure you have the latest Solaris 10 update installed?  If
>
>     not, then I'd recommend updating to the latest to make sure you have
>
>     all the latest bug fixes.
>
>     Could you run "pstack (pid)" of the process eating CPU so we can see
>     what the process(es) are doing when it is consuming so much CPU?
>
>     Do you have the face browser turned on?  If so, and if you have
>     IncludeAll turned on, and a lot of users on your system, then GDM
>     can consume a lot of CPU trying to get face images for all users.
>
>     Aside from the above, I haven't heard of any common issues that cause
>     GDM on Solaris 10 to consume a lot of CPU.
>
> Hi Brian,
>  
>     First thank you for the quick response I really do appreciate it. We 
> ended up rebooting the server as it had become unusable for our 
> customers and post re-boot all login issues and GDM-binary CPU usage 
> issues were resolved and we currently have 24 users logged in through 
> xdremote using exceed on a windows box. With that said since the 
> offending processes are no longer chewing through our CPU cycles i cant 
> run the pstack however given another week or two it will likely occur 
> again and I will most definatly run it then and let you all know what 
> ive found. We are running Update 5 on the server and will likely install 
> update 6 soon and see if that helps. 

Sounds good.

> Additionally I had a look in the 
> system log and saw this error before we rebooted 
> "gdm_slave_xioerror_handler: Fatal X error - Restarting <ip 
> address>:0.0" I am going to try to get debug turned on in the config 
> file and do a safe-restart of the process tonight so we have a little 
> more information next time we get a rouge gdm-binary process. 

That error normally shows up if users hit Ctrl-Alt-Backspace to kill
the Xserver.  You might want to disable that "feature" when running
in a Sun Ray environment since it can cause GDM to stop managing a
display.  Sometimes GDM doesn't recover cleanly when users kill
sessions this wsy, forcing you to kill the hung GDM slave daemon
process by hand to get it to restart.  I believe you can turn off
this feature in the Xorg config.

> Additionally we tried a test on one of our other boxes with GDM where we 
> enabled the gdm2-login service from the local console and disabled 
> cde-login, connected to the box through an XDMCP connection like our 
> users. Once that was done we did a ps -ef | grep gdm-binary and looked 
> at the PID and PPID to determine which process was the master and which 
> were slave, we identified the gdm-master process as being the PPID of 
> the other gdm-binary processes and attempted to kill the slave 
> gdm-binary process in hopes that we could just kill that one process 
> without affecting all the connections to the box unfourtunatly it didnt 
> matter which slave/child process we killed it brought down all 
> connections to the box, so our only recourse to fix the CPU usage issues 
> is to kick everyone off by killing the slave gdm-binary process or 
> rebooting the box, both of which arent good :/
>  
> I will reply back once i can get the pstack done on the gdm-binary once 
> it spikes the CPU again, it may be another week or so before I can get 
> to it.

Let me know, and the extra debug output should be useful.

Brian