[SSSD] IPA client randomly lose memory of users

Jakub Hrozek jhrozek at redhat.com
Sun Dec 2 16:40:32 UTC 2012


On Sat, Dec 01, 2012 at 09:59:44PM -0500, Qing Chang wrote:
> 
> On 30/11/2012 5:39 PM, Simo Sorce wrote:
> >On Fri, 2012-11-30 at 17:21 -0500, Qing Chang wrote:
> >>my dovecot IMAP server would randomly lose memory of users, as an example:
> >>
> >>Samba/NFS server knows this user:
> >>[root at smb2 shassan]# getent passwd bqiang
> >>bqiang:*:47105:471:Beiping Qiang:/home2/bqiang:/bin/tcsh
> >>
> >>But dovecot server does not:
> >>[root at dovecot2 ~]# getent passwd bqiang
> >>
> >>Only when I apply this:
> >>[root at dovecot2 ~]# \rm /var/lib/sss/db/cache_sri.utoronto.ca.ldb
> >>[root at dovecot2 ~]# service sssd restart
> >Is it necessary to remove the ldb file, or does a simple restart 'fix'
> >things ?
> I could try that next time around when a user calls me, thought to clear the cache to be
> sure...
> >
> >>It gets it:
> >>[root at dovecot2 ~]# getent passwd bqiang
> >>bqiang:*:47105:471:Beiping Qiang:/home2/bqiang:/bin/tcsh
> >>
> >>So far I have to deal with this for three users. It's quite possible
> >>that there are more than 3 that are affected, they were just patient
> >>enough to wait until dovecot "recovers its memory".
> >>
> >>=====
> >>sssd.x86_64                      1.8.0-32.el6              @rhel-x86_64-server-6
> >>sssd-client.x86_64               1.8.0-32.el6              @rhel-x86_64-server-6
> >>=====
> >>
> >>Your help is much appreciated.
> >Is it possible for you to raise the log level, or would that slow down
> >your production environment too much ?
> >
> >In that case do you have any hints on how we could reproduce this
> >issue ? Have you notice if it happens periodically, or after specific
> >events ?
> I have asked the users, who had this problem recurring, to call me, I'll turn on debug
> log at that time.
> 
> BTW, to troubleshoot an unrelated problem, I set the debug level to 0xFFF0, is this
> overkill?
> =====
> [nss]
> debug_level = 0xFFF0
> 
> [pam]
> debug_level = 0xFFF0

I wouldn't say overkill, it may produce more info than necessary and
then it might be harder to analyze the logs simply because of the size.
I usually use debug_level 8, 9 is to verbose for my taste. But it really
depends on the particular problem. In general, it is better to have more
information than less :-)

What is the nature of this second problem? Maybe it would be nice to
also include the debug_level directive in the domain section to see what
the back end is doing..



More information about the sssd-devel mailing list