Comment # 3 on bug 1206592 from Thorsten Kukuk

(In reply to Jan Engelhardt from comment #1)
> >Don't restart *.socket services
> 
> I concur. But what's up with Fedora?
> https://src.fedoraproject.org/rpms/sssd/blob/rawhide/f/sssd.spec

That's all not that easy, but I wondering, why Fedora restarts services, which
reports an error if you try to restart them?

Restarting sockets can make sense if you make regular changes to the socket
service files...
But the execessive "systemctl daemon-reload" our macros do are not good...

If systemctl try-restart sssd.service restarts all depending services, I think
it's enough to do only that any maybe socket units, which change often.

> You also observed three crashes over roughly three seconds. I would expect
> that rpm and contemporary NVMe harddisks (if used) are fast enough such that
> the race condition goes away in less than 3s.

I assume the following happend (according to journald):
- we update sssd from 2.7.4 to 2.8.2
- sssd-ldap and sssd-krbt-common are not updated yet, but still at 2.7.4!
- we restart sssd -> we see the crashes
- sssd crashed so often, that systemd disables it
- we update sssd-krb5-common
- we update sssd-ldap

From the config on the infra.opensuse.org machines I see, that the ldap plugin
is loaded. I bet that a sssd 2.8.2 doesn't work with 2.7.4 plugins (that there
are binary incompatibilities) and this is the crash we see.
That's what I mean with we need to make sure that all depending packages get
updated before we restart sssd. To avoid this crashes if there are still
incompatible plugins which are not yet updated.