[Leaplist] Problems w/ users under CentOS 5.3 ....

[William A. Mahaffey III]

On 09/03/09 23:59, Bryan J Smith wrote:
> On Wed, 2009-09-02 at 17:50 -0500, William A. Mahaffey III wrote:
>    
>> Soooooo tacky to self-reply, but for posterity here goes: the problem
>> here turned out to be that we were trying to give user-process too much
>> stack space.  This caused the login shell to fail, eventually screwing
>> the rest of the graphical login up.
>>      
> And this was because ... and here it comes ... (oh joy!) @-ppp
>
>    
>> We have had (for years) a custom initialization file in /etc/profile.d
>> with a line in it that said:
>> limit stackspace 512m
>> this worked AOK under SuSE 9.3 32-bit (as did 1536m), but not under
>> CentOS 5.3 32-bit (we have 3 GB of RAM on the machine).
>>      
> And what did I warn you about?  @-ppp (and big grin)
>
> Exact words ...
>
> "Reality:  you switched to a Fedora-based system from a SuSE-based
> system, and an older one at that.  Expect issues."
>
> And most, most _specifically_ ...
>
> "Fedora and SuSE based systems have different profile/bash settings"
>
> I could only assume your differences were in the user profiles, and not
> system-wide, especially after you replied ...
>
> "I *did* start from the pristine CentOS after the 1st barf"
>
> Actually, you did not.  You didn't merely just reuse only your home
> directories, you _modified_ the "prestine" CentOS root yourself.  Most
> specifically, the /etc/profile.d.
>
> Again, I warned you about that.  @-ppp
>    


The offending  /etc/profile.d file (*1* file, BTW, *not* the whole 
directory) was copied over from a 933 MHz PIII running FC5, where is has 
worked AOK for about 5 years now :-/ (w/ 1.5 GB of stackspace enabled on 
a box w/ 1.5 GB of RAM).


> So instead of having only user profile differences of SuSE into a
> Fedora-based system, you moved your system-wide differences of SuSE into
> a Fedora-based system.  You copied over your /etc/profile.d from a SuSE
> to a Fedora based system and assumed it would work.  It doesn't.  @-ppp
>
> This is exactly what I predicted would happen, only I took your comments
> that you only had legacy SuSE stuff in the user profiles at face value.
> I guess my mistake for assuming you wouldn't touch that "prestine"
> root. ;)
>
> [ "Real Life":  Be forward on all that you change.  I get paid the "big
> bucks" because I catch people when they don't.  Had I been the one that
> installed CentOS, and then you modified it, then come back to me, I
> would have caught this in 2 minutes.  Why?  Because I maintain
> configuration management on this and would have caught it in a diff.
> Again, my daily "real life" an why I get paid the "big bucks." ]
>
>    
>> When I commented that line out, the login went AOK, but we only have
>> 10MB of stackspace.  I seem to recall something about a PAE kernel
>> being needed under RedHat  to get a ~4 GB address space for user
>> processes, so I will look into  that tomorrow.
>>      
> You only get up to 3GiB per user process period using any 32-bit kernel.
> No PAE kernel on 32-bit can overcome that limit at all.
>
> Red Hat did offer a 4G/4G kernel model in Red Hat Enterprise Linux
> Release 3 and 4 called "hugemem."  But that is no longer offered in Red
> Hat Enterprise Linux Release 5, because Red Hat highly recommends
> Linux/x86-64 for more than 4GiB of RAM, and reduced the official amount
> supported to 16GiB on 32-bit (and that's with caveats).
>
> Why do you need such a huge stack size again?
>    

legacy in-house code needs it, stipulated & irrelevant ....

-- 

	William A. Mahaffey III

  ----------------------------------------------------------------------

	"The M1 Garand is without doubt the finest implement of war
	 ever devised by man."
                            -- Gen. George S. Patton Jr.


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.