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

Re: Tweaking "threads"

--On Thursday, September 01, 2005 12:42 AM -0500 Ben Beuchler <insyte@gmail.com> wrote:

The saga continues...

As you may recall, I was experiencing some timeouts to my slapd
server.  After much advice from several members of the list, I tweaked
my DB_CONFIG and slapd-bdb settings.  This seemed to help, but the
problem returned when our new students stopped in for registration.
I saw several hundred entries like these in the logs:

Aug 27 15:46:46 swizzle slapd[22099]: connection_input: conn=2199
deferring operation: binding
Aug 27 15:55:21 swizzle slapd[22099]: connection_input: conn=2673
deferring operation: binding
Aug 27 15:56:00 swizzle slapd[22099]: connection_input: conn=1655
deferring operation: too many executing
Aug 27 15:56:02 swizzle slapd[22099]: connection_input: conn=1655
deferring operation: too many executing

Feeling extremely clever, I began increasing the value of "threads"
until the problem went away.  I ended up setting it to 100.  Watching
the number of threads in the process table (I'm running linux, so
they're visible) I seem to set at betwenn 65 and 85.

Doing some reading today on the OpenLDAP site, it sounds like this is
ridiculously high.  The FAQ implies that it's rare for anyone to need
more than the default 16 threads.

Our LDAP directory contains around 1400 entries and is used as
authentication for around 300 OS X workstations and 1100 email
accounts.  This results in just shy of 200,000 individual binds in a
24 hour period.  Is it silly for me to think that I may need that many
threads?  Or is it more likely that I didn't really fix that timeout
problem and, as a result, am hanging on to threads much longer than I

I'm suspicious that individual LDAP sessions may be sticking around a
lot longer than they should, as I am seeing a *lot* more "BIND" log
entries than I am "UNBIND".  For example:

swizzle:/var/log# grep ' BIND' slapd.log|wc -l
swizzle:/var/log# grep ' UNBIND' slapd.log|wc -l

Thanks for any input you may have!


Current DB_CONFIG settings:

swizzle:/var/log# egrep -v '(^#|^$)' /var/lib/ldap/DB_CONFIG
set_cachesize   0 8388608 0
set_lg_bsize    524288
set_lk_max_objects      5000
set_lk_max_locks        5000
set_lk_max_lockers      5000

DB settings from slapd.conf:
backend     bdb
checkpoint 512 30
cachesize = 2000
idlcachesize = 6000

I suggest you investigate the "idletimeout" directive. ;)


Quanah Gibson-Mount
Principal Software Developer
ITSS/Shared Services
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html

"These censorship operations against schools and libraries are stronger
than ever in the present religio-political climate. They often focus on
fantasy and sf books, which foster that deadly enemy to bigotry and blind
faith, the imagination." -- Ursula K. Le Guin