[SSSD] [PATCHES] Support for ghost users

Simo Sorce simo at redhat.com
Wed Apr 25 14:00:47 UTC 2012


On Wed, 2012-04-25 at 15:41 +0200, Jan Zelený wrote:
> > Stephen Gallagher <sgallagh at redhat.com> wrote:
> > > On Mon, 2012-04-23 at 16:22 +0200, Jan Zelený wrote:
> > > > Hi,
> > > > I'm sending a patch set that removes support for fake user entries and
> > > > add
> > > 
> > > > ghost attribute instead:
> > > Jan, could you run a few performance tests with large groups so we can
> > > see what we got for improvement?
> > 
> > Sure, no problem. I will also include better commit messages (discussed
> > with Simo on IRC) and also a sysdb upgrade function is missing.
> > 
> > Jan
> 
> Ok, here is the updated patch set. So far I've run only a simple test:
> 
> - IPA server
> - large-group with 3k users
> - flat membership structure
> - log level 10 (not sure how much impact would it have to turn this off)

log level 10 causes a lot of writes to the log file, can be quite a
large impact, can you redo the test turning log level down to 0 ?

> - user testuser who is not part of the large-group
> 
> The testing user is queried for establishing connection to the server and 
> taking the connection time out of the equation.
> 
> 1. service sssd stop
> 2. rm -f /var/lib/sss/db* /var/log/sssd/*
> 3. git checkout master
> 4. reinstall sssd
> 5. service sssd start
> 6. getent group testuser
> 7. time getent group large-group
> 
> real    0m8.198s
> user    0m0.002s
> sys     0m0.006s
> 
> 8. service sssd stop
> 9. rm -f /var/lib/sss/db* /var/log/sssd/*
> 10. git checkout ghosts
> 11. reinstall sssd
> 12. service sssd start
> 13. getent group testuser
> 14. time getent group large-group
> 
> real    0m4.441s
> user    0m0.005s
> sys     0m0.004s

So I assume the first test is w/o the patch, and the second w/ the
patch ? :)

> If you have any other scenarios, I'll be happy to run them for you but please 
> don't ask me to add another 3k users to the LDAP server ;-)

ok, can you add 3k groups? (j/k)

Sound good from what I can see, if log level 0 confirms or exceeds
cutting time in a half, this will be a good improvement.

Simo.

-- 
Simo Sorce * Red Hat, Inc * New York




More information about the sssd-devel mailing list