On 11 September 2017 at 14:28, Jakub Hrozek wrote:
On Mon, Sep 11, 2017 at 12:23:26PM +0100, John Beranek wrote:
> On 1 September 2017 at 15:54, Lukas Slebodnik wrote:
> >
> > On (01/09/17 09:33), William Edsall wrote:
> > >Had a few communications with Michal but we're still stuck.
> > >
> > >One issue is that we have dozens of domain controllers globally. A standard
> > >dns lookup could give me a domain controller overseas which will be slow,
> > >or maybe even a domain controller that isn't responding. As such, I
have
> > >been inserting ad_server = x into the sssd.conf to improve performance.
> > >
> > >I noticed that if I do not insert ad_server = x, I'm getting different
> > >results. My initial id request is very slow but seems to produce results.
> > >While searching, it seems to also be 'inserting' users into the
users hash
> > >table - almost as if it's searching and inserting our entire user
database?
> > >For example there are countless lines of the following:
> > >(Fri Sep 1 09:28:37 2017) [sssd[be[example.com]]]
> > >[sdap_nested_group_hash_insert] (0x4000): Inserting
> > >[CN=user_name,OU=bla,OU=bla Users,DC=dow,DC=com] into hash table [users]
> > >
> > >As my initial id request returns, it seems to return several chunks of my
> > >group ids at once as if it's processing them individually and searching
all
> > >users in that group (thus the above log entries).
> > >
> > >Not sure if this helps or just muds up the issue but it's strange
indeed.
> > >
> > You needn't hardcode ad_server. You can still rely on dns discovery.
> > I assume you use sites in AD. So you can "pin" sssd to your
local/nearest site
> > with option ad_site.
>
> I've got something to add to this, some behaviour we're seeing with
> CentOS 7 servers using sssd-ad.
>
> sssd-1.14.0-43.el7_3.18.x86_64
> sssd-ad-1.14.0-43.el7_3.18.x86_64
> sssd-client-1.14.0-43.el7_3.18.x86_64
> sssd-common-1.14.0-43.el7_3.18.x86_64
> sssd-common-pac-1.14.0-43.el7_3.18.x86_64
> sssd-ipa-1.14.0-43.el7_3.18.x86_64
> sssd-krb5-1.14.0-43.el7_3.18.x86_64
> sssd-krb5-common-1.14.0-43.el7_3.18.x86_64
> sssd-ldap-1.14.0-43.el7_3.18.x86_64
> sssd-proxy-1.14.0-43.el7_3.18.x86_64
>
> In our case we have some DCs which are located at a partner site, and
> are therefore inaccessible to clients on our standard LANs.
>
> When SSSD starts it will correctly determine there are 2 primary DCs
> (these are the ones for the site) and 7 backup DCs.
>
> However, what is happening from time to time is that for some reason
> I've not yet determined the connection(s) to the primary DC(s) are
> dropping, and then sssd attempts to connect to one of the DCs that are
> inaccessible.
>
> In what circumstances would sssd prefer a backup server to a primary server?
>
> I've got a chunk of log which I've anonymised:
>
>
https://paste.fedoraproject.org/paste/G69lC9COQfbnWFI~qtFLZw
The issue is here:
[snip]
So this is a known bug where locating the site (even though we know it
already) can contact the DCs outside that site.
In the meantime, until the fix is released, you can hardcode the site
using the 'ad_site' option.
Thank you Jakub, is there a ticket/BZ for this? I noticed after my
post that it's evident on CentOS 6 too.
Cheers,
John
--
John Beranek To generalise is to be an idiot.
http://redux.org.uk/ -- William Blake