(ITS#3993) slapd is using 99% of the cpu, hanging in sched_yield()

[r@sil.at]

Full_Name: Raimund Sacherer
Version: 2.2.23 (Debian: 2.2.23-8)
OS: Debian GNU/Linux
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (213.235.212.193)


Hello,

we are using the slapd version 2.2.23 (-8 Debian) for production use. We are
currently switching from an older version from bsd to debian, on our test system
(Linux Kernel: 2.6.8-1-k7) i encounterd the following problem:

i tried to test import ldiff data yesterday evening, it worked, i could connect
to the server and view/modify the data. 

i went home and today in the morning i tried to ssh on the box, now:

1. the load of this machine was about 29
2. 100% of the cpu was used
3. no process claimed using cpu

then i tried to restart the slapd because of changes in the slapd.conf file and
from i now can not tell if slapd stoped and started again, or if it's hanging
since the restart command.

strace -p <slapd-process> delivers the following output:

sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0


i googled a bit about the sched_yield and found that other people also have some
problems with sched_yield.


i am currently trying to use slapd 2.2.26 as it's called the 'stable' version,
we backport it from debian/testing, but i am not sure if this will solve the
problem, as i do not know if it's kernel-related or whatever.


best regards

raimund