Are you monitoring the size of slapd in memory?
We are running OL 2.1.25 + BDB 4.2.52 with back_bdb and simple
bind only (no SASL or Kerberos) under RedHat Linux 9 (soon to be
migrated to RHES 3.1). In our setup, conducting searches against
the directory causes slapd to increase its use of RAM without
ever releasing it back to the operating system. In this context,
I've observed that if slapd isn't monitored (e.g., via cron) and
restarted when it gets too large, then it just quits.
Under test conditions, I've compiled OL and the major dependent libs
(OpenSSL, Cyrus SASL, BDB) from scratch, eliminating the RedHat MIT
Kerberos libs, which are known to have problems, but I get the same
result. I've replicated this with OL 2.1.30 also. I believe that
it's a leaky RedHat library among the ones that I haven't built
from scratch; no one (that I'm aware of) running OL on any other
platform has reported this issue. Fortunately, our data store
is small compared to others', and the rate of growth under our
use conditions generally requires restarting only once or twice
a week. (I haven't tried to isolate the leaky library since our
running platform is a moving target.)