[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Bug#245014: slapd stops working when machine becomes CPU bound (ITS#3149)



Hi James, 

On Tue, May 18, 2004 at 08:30:26PM -0400, James Blackwell wrote:
> > I suggest you try to isolate the problem.  I suggest you start by
> > using OpenLDAP clients (e.g., ldapsearch(1)) to reproduce the behavior
> > instead of involving non-OpenLDAP clients.
> 
> root@neptune:~# time ldapsearch -x |wc -l
> 2949
> 
> real    0m0.098s
> user    0m0.015s
> sys     0m0.009s
> 
> root@neptune:~# ./main & time ldapsearch -x | wc -l
> [1] 20258
> 2949
> 
> real    1m7.241s
> user    0m0.006s
> sys     0m0.002s

Sorry, but that does not really proof anything - the machine just gets
very slow. I tried to reproduce your problem with kernel 2.6.6 on 
Debian unstable running slapd 2.1.29-2 but it works just fine...

> Here is an is an strace of when ldapsearch -x gets stuck the first time:
> 
> select(1024, [3], [], NULL, NULL)       = 1 (in [3])
> read(3, "0j\2\1\2de\4", 8)              = 8
> read(3, "!ou=ArchProjects,dc=gnuarch,dc=o"..., 100) = 100
> write(1, "\n", 1)                       = 1
> write(1, "# ArchProjects, gnuarch.org\n", 28# ArchProjects, gnuarch.org
> ) = 28
> write(1, "dn: ou=ArchProjects,dc=gnuarch,d"..., 38dn:
> ou=ArchProjects,dc=gnuarch,dc=org) = 38
> write(1, "ou: ArchProjects\n", 17ou: ArchProjects)      = 17
> write(1, "objectClass: top\n", 17objectClass: top)      = 17
> write(1, "objectClass: organizationalUnit\n", 32objectClass:
> organizationalUnit) = 32
> select(1024, [3], [], NULL, NULL

That shows nothing as well. It just does not get more data at that
point. How long it blocks is not shown and that's going on there stays
completely in the dark :(

> Other software appears to run as expected, unless pam is involved, in
> which case that software blocks too .

So basically slapd blocks somewhere for unknown reason it seems. Can you
verify that? Question is what slapd is waiting for at that point.
Perhaps you have a problem with your I/O system or something. No idea, I
just can't explain. slapd should run equally well under any CPU load
(apart from the fact that the system as a whole will slow down...).

Greetings

	Torsten