[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
nscd and large group problem...
From: |
John R. Hover |
Subject: |
nscd and large group problem... |
Date: |
Fri, 22 Mar 2002 17:24:43 +1200 |
User-agent: |
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.8) Gecko/20020212 |
Nscd, (or at least nscd on RedHat) fails upon an 'ls -l' on a directory
with files owned by a group with a large number of users (in this case
2800), or a getent group <Groupname> of same. This occurs regardless of
the suggested-size (or TTL, NTTL) in /etc/nscd.conf. The nscd processes
continue to run, but stop responding (one always becomes <defunct>)--
and the 'ls -l' or 'getent group <blah>' command never completes. By the
time nscd fails, it's memory usage has always climbed from the initial
1600k up to 4-5 MB. The exact point varies.
We encountered the problem using nsswitch and nss_ldap.so, and thought
for a long time that it had something to do with our LDAP setup. That's
why I've sent this to the nss_ldap list. Someone else may be having the
same confusion.
I've confirmed that it still happens if the group is local to
/etc/group-- which suggests it is a more fundamental problem with
glibc/nscd! It shouldn't be hard to reproduce.
A group with approx. 1400 users does *not* cause the problem.
A group with approx. 2800 users *does* cause the problem.
SYSTEM CONFIG
Numerous computer models (HP, Compaq, Dell) i686/386 architecture
OS: Linux Red Hat 7.2
Kernel: Several from Redhat default up to latest 2.4.9-31--didn't try a
custom compiled kernel.
Versions: glibc|nscd-2.2.4-13, 2.2.4-19.3, 2.2.4-20
( I downloaded and rebuilt the glibc-2.2.4-20 SRPM, installed all the
resultant RPMS. Same problem. ) For probably obvious reasons, I didn't
build glibc from source without Redhat patches, so I can't be sure this
isn't Redhat specific.
nss_ldap-172-2 : doesn't cause problem because it doesn't return member
list for groups with more than 23 users
nss_ldap-184 : latest version, which correctly returns all members of a
group. This causes the problem when nscd is on. It works fine when nscd
is off.
As I am not a programmer, I'm not really capable of getting any deeper
into strace or debugging efforts. I figured someone here would want to
reproduce it and see that it gets to the right person. I thought about
submitting something to bugzilla.redhat or www-gnats.gnu.org, but I
figured a "this fails" description without debugging info wouldn't be of
much use.
Does anyone know anywhere else I should CC this?
Cheers,
--john
--
John R. Hover
Unix/Linux System Analyst
Department of Computer Science
University of Auckland
- nscd and large group problem...,
John R. Hover <=