Winbindd in HEAD: cannot deal with 15,000 users+

[David Shapiro]

Do you know a solaris version of the cool panic command you have listed
below that starts up a terminal?  I have gdb on my solaris box, but I do not
use gnome.
 
David

-----Original Message-----
From: Esh, Andrew [mailto:AEsh at tricord.com]
Sent: Tuesday, April 02, 2002 10:11 AM
To: 'David Shapiro'
Cc: samba-technical at samba.org
Subject: RE: Winbindd in HEAD: cannot deal with 15,000 users+



The "panic action" smb.conf parameter simply takes its argument, and does a
system command with it. On Linux, /bin is where the "sleep" executable is
stored, so "/bin/sleep 9000" will call sleep, and have it do nothing for two
and a half hours. This gives a person time to attach a debugger, and find
out what went wrong.

If you are having a problem on another platform, maybe "sleep" is stored
somewhere else. Another thing to check is whether your system command (as
Samba calls it) actually works. Maybe it doesn't. Maybe we need a fix for
your platform.

Here's how your "panic action" argument is handled: 

/******************************************************************* 
something really nasty happened - panic! 
********************************************************************/ 
void smb_panic(char *why) 
{ 
        char *cmd = lp_panic_action(); 
        if (cmd && *cmd) { 
                system(cmd); 
        } 
        DEBUG(0,("PANIC: %s\n", why)); 
        dbgflush(); 
        abort(); 
} 

Here's something Andrew Bartlett taught me: 

   panic action = "gnome-terminal --command 'gdb /proc/%d/exe %d'" 

This will pop up a new terminal window with gdb already positioned within
the panicked code. A simple "bt" tells me where the problem occurred.

-----Original Message----- 
From: David Shapiro [ mailto:David.Edward.Shapiro at btitele.com
<mailto:David.Edward.Shapiro at btitele.com> ] 
Sent: Tuesday, April 02, 2002 7:02 AM 
To: 'abartlet at samba.org'; Mike Papper 
Cc: samba-technical at samba.org; tridge at samba.org 
Subject: RE: Winbindd in HEAD: cannot deal with 15,000 users+ 


How do you set the panic action = /bin/sleep...? 

Is that a linux only thing? 

David 

-----Original Message----- 
From: abartlet at samba.org [ mailto:abartlet at samba.org
<mailto:abartlet at samba.org> ] 
Sent: Tuesday, April 02, 2002 12:33 AM 
To: Mike Papper 
Cc: samba-technical at samba.org; tridge at samba.org 
Subject: Re: Winbindd in HEAD: cannot deal with 15,000 users+ 


On Mon, Apr 01, 2002 at 03:49:05PM -0800, Mike Papper wrote: 
> I have run the latest winbindd from the samba-3.0-alpha17 branch. I ran it


> for the past 4 days and all went well. However, in the following 
situation, 
> it breaks down... If I have 15,000 users, its OK UNTIL I add or delete a 
user 
> from the set of 15,000 users on the PDC. It seems like any update to the 
PDC 
> causes winbbind to fetch all the info from the PDC (and not just the 
> updates). This seems to make winbind crash. When I bring winbind back up, 
it 
> runs OK until I once again add or remove a user from the PDC. 

The RPC protocols are not spectacular in this area - so often we need to 
refetch the lot :-( 

> I am using NT 4 service pack 5 as the PDC. Any thoughts on why this could 
be 
> happening? Or thoughts on what this debug output is saying? 
> 
> Also, I did send a bug report to samba at samba.org as suggested in the 
release 
> docs). Will this bug report get into the right hands? 

Could you set 'panic action = /bin/sleep 9000' and attach a debugger to the 
dead winbindd?  Or start it as winbindd -i inside gdb? 

Grab a 'bt full' and post that to the list and we will have a much better 
chance of finding the bug.  Recompiling --enable-debug will put some more 
info 
in the backtrace. 

Also check out the very latest HEAD CVS (assuming you are running from a 
alpha 
tarball) and double-check.  

Thanks, 

Andrew Bartlett