[Samba] winbind 100% CPU usage

Mark Nienberg mnlists at tippingstructural.com
Mon Mar 26 22:12:45 UTC 2018


On Wed, Mar 21, 2018 at 3:50 AM, Michael Wandel via samba <
samba at lists.samba.org> wrote:

> Hey,
>
> we are using rhel7 with samba and winbind packages from redhat.
>
> it is a fileserver joined to samba 4 domain.
>
> os: rhel7.4
>
> samba-winbind-4.6.2-12.el7_4.x86_64
> samba-winbind-clients-4.6.2-12.el7_4.x86_64
> samba-winbind-modules-4.6.2-12.el7_4.x86_64
>
> approx. 350 clients
> approx. 10000 open files
>
> see attached stripped smb.conf
>
> the following problem exist: after a time (several hours) the winbindd
> process takes 100% cpu and after a while the nss is disfunctional. Next
> the system is unusable.
>


This sounds like it might be the same problem we are having. We had a
perfectly functioning domain member file server running samba 4.5.5 on
CentOS 7.3. When we did an in-place upgrade to samba 4.6.2 and CentOS 7.4
we had high CPU load and the file server performance was unacceptable. Our
installation is much smaller than yours, only about 35 clients. The system
did not become unusable, but it was slow. We reverted to a snapshot of our
previous configuration and everything went back to normal.

When we get some time we intend to break up the upgrade so that first we
update CentOS without updating samba to see if the problem occurs. And we
will probably stop the winbind daemon and run winbind in the foreground for
a while to see if that gives any clues. One suggestion is that there may be
lots of winbind queries for names that do not exist, and if so then
configuring a larger negative winbind cache may help.


More information about the samba mailing list