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

Solved: openldap+nscd crash linux box



Hello,

I've got a huge problem 70 days ago: a linux server which
was running openldap 1.2.9 with nss_ldap-108 and pam_ldap-51
was suddenly totally unusable (no more file descriptors).
Now I've found the reason, long time after :-)


I've worked on another problem yesterday: sometimes,
suddenly, slapd die. I didn't found any reason (The slapd
daemon is restarted every day).
So I've made a small script wich restart automatically slapd 
if it die.
After few tests, all seems to works. So I've restarted nscd too to increase speed.

And suddenly, CPU become more and more busy. I've got a load
average of more than 12 !
The reason was that LDAP died and nscd seems to doesn't
handle this case.

Here is the lines in /var/log/message:
	Aug  2 23:55:32 uweb1 nscd: nss_ldap: could not get LDAPresult - Can't contact  LDAP server
	Aug  2 23:55:32 uweb1 nscd: 20772: short write in cache_addpw: Broken pipe

I've found a bug report concerning this kind of problem in debian lists:
	http://cgi.debian.org/cgi-bin/bugreport.cgi?bug=3D58367


So may be it's safer to avoid using nscd with LDAP + nss_LDAP !

BTW, the slapd die isn't solved. I can't enable logs because the ldap base is very used. Is there another way to try to find what's happening ?

Manuel

--