Sponsors
Sponsor Products
RHEL/CentOS 7.6, Maya 2018, and NVIDIA
posted by Mike Rochefort  on Dec. 28, 2018, 1:35 p.m. (4 years, 11 months, 2 days ago)
6 Responses     0 Plus One's     0 Comments  
Hello,

Last night I upgraded my personal workstation to CentOS 7.6 (I use it as my testbed as well for desktop related things). Since then, Maya's Viewport 2.0 has
been impossible to activate. It seems like Maya isn't properly finding the NVIDIA driver, which may or may not be due to the switch to GLVND in 7.6. All other applications (Mari, Substance, Nuke, etc) seem to be finding things just fine. Has anyone else run into this issue or are you all playing it smart and not updating things?

Currently running 2018.2, also tried 2018.5 and no change. Driver 410.78 from elrepo.

Cheers,
Mike

Thread Tags:
  discuss-at-studiosysadmins 

Response from Mike Rochefort @ Jan. 24, 2019, 5:40 p.m.
I took a look through the spec file for systemd and you're right, it does run ldconfig in several spots. I made another container, fully updated it, then downgraded systemd and systemd-libs (libgudev didn't get installed from the base docker image). Everything still worked, so I think you've got the better answer here. The simple solutions often escape my memory.

Now it makes me curious how the library links/cache got out of whack during the upgrade to 7.6... I've never seen something like this happen before.

Cheers,
Mike


---- On Thu, 24 Jan 2019 06:40:07 -0500 James Pearson <james-p@moving-picture.com> wrote ----

I was just asking out of curiosity (and that we haven't updated to
CentOS 7.6 yet) - as I can't see how systemd would be involved here ...

I guess the update may have indirectly 'fixed' whatever the issue was -
e.g. the systemd-libs (and libgudev1) update will run ldconfig - which
may have helped ???

James

Michael Rochefort wrote:
>
> Hi James,
>
> Honestly, I haven't tried that yet (which really isn't good testing practice). I was just happy that things were actually working again. I'll give it a whirl when I get a chance.
>
> Cheers,
> Mike
>
>
> ---- On Mon, 21 Jan 2019 12:07:46 -0500 James Pearson <james-p@moving-picture.com> wrote ----
>
> Michael Rochefort wrote:
>>
>> Well, this was interesting. After a few days of further debugging with containers (and about to start testing 16 self-built libglvnd packages) everything started magically working today. The reason: systemd.
>>
>> In the past 24 hours RH/CentOS released updates for the following packages:
>>
>> systemd, systemd-libs, systemd-python, systemd-sysv, and libgudev1 which had the following in the changelog (the only two commits since RHEL 7.6's release):
>>
>> * Mon Jan 07 2019 Lukas Nykryn <lnykryn@redhat.com<mailto:lnykryn@redhat.com><mailto:lnykryn@redhat.com<mailto:lnykryn@redhat.com>>> - 219-62.2
>> - journald: do not store the iovec entry for process commandline on stack (#1657788)
>> - journald: set a limit on the number of fields (1k) (#1657792)
>> - journal-remote: set a limit on the number of fields in a message (#1657792)
>>
>> * Mon Nov 26 2018 Lukas Nykryn <lnykryn@redhat.com<mailto:lnykryn@redhat.com><mailto:lnykryn@redhat.com<mailto:lnykryn@redhat.com>>> - 219-62.1
>> - dhcp6: make sure we have enough space for the DHCP6 option header (CVE-2018-15688)
>>
>> Turns out something there was indirectly breaking libglvnd from functioning properly.
>
> Strange - are you able to reproduce the issue by downgrading systemd ?
>
> James Pearson
>
>
>
>
>
>
>
>
> To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe
>



0 Plus One's     0 Comments  
   

Response from James Pearson @ Jan. 24, 2019, 6:45 a.m.
I was just asking out of curiosity (and that we haven't updated to CentOS 7.6 yet) - as I can't see how systemd would be involved here ... I guess the update may have indirectly 'fixed' whatever the issue was - e.g. the systemd-libs (and libgudev1) update will run ldconfig - which may have helped ??? James Michael Rochefort wrote: > > Hi James, > > Honestly, I haven't tried that yet (which really isn't good testing practice). I was just happy that things were actually working again. I'll give it a whirl when I get a chance. > > Cheers, > Mike > > > ---- On Mon, 21 Jan 2019 12:07:46 -0500 James Pearson wrote ---- > > Michael Rochefort wrote: >> >> Well, this was interesting. After a few days of further debugging with containers (and about to start testing 16 self-built libglvnd packages) everything started magically working today. The reason: systemd. >> >> In the past 24 hours RH/CentOS released updates for the following packages: >> >> systemd, systemd-libs, systemd-python, systemd-sysv, and libgudev1 which had the following in the changelog (the only two commits since RHEL 7.6's release): >> >> * Mon Jan 07 2019 Lukas Nykryn >> - 219-62.2 >> - journald: do not store the iovec entry for process commandline on stack (#1657788) >> - journald: set a limit on the number of fields (1k) (#1657792) >> - journal-remote: set a limit on the number of fields in a message (#1657792) >> >> * Mon Nov 26 2018 Lukas Nykryn >> - 219-62.1 >> - dhcp6: make sure we have enough space for the DHCP6 option header (CVE-2018-15688) >> >> Turns out something there was indirectly breaking libglvnd from functioning properly. > > Strange - are you able to reproduce the issue by downgrading systemd ? > > James Pearson > > > > > > > > > To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe > To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

0 Plus One's     0 Comments  
   

Response from Mike Rochefort @ Jan. 23, 2019, 8:40 p.m.
Hi James,

Honestly, I haven't tried that yet (which really isn't good testing practice). I was just happy that things were actually working again. I'll give it a whirl when I get a chance.

Cheers,
Mike


---- On Mon, 21 Jan 2019 12:07:46 -0500 James Pearson <james-p@moving-picture.com> wrote ----

Michael Rochefort wrote:
>
> Well, this was interesting. After a few days of further debugging with containers (and about to start testing 16 self-built libglvnd packages) everything started magically working today. The reason: systemd.
>
> In the past 24 hours RH/CentOS released updates for the following packages:
>
> systemd, systemd-libs, systemd-python, systemd-sysv, and libgudev1 which had the following in the changelog (the only two commits since RHEL 7.6's release):
>
> * Mon Jan 07 2019 Lukas Nykryn <lnykryn@redhat.com<mailto:lnykryn@redhat.com>> - 219-62.2
> - journald: do not store the iovec entry for process commandline on stack (#1657788)
> - journald: set a limit on the number of fields (1k) (#1657792)
> - journal-remote: set a limit on the number of fields in a message (#1657792)
>
> * Mon Nov 26 2018 Lukas Nykryn <lnykryn@redhat.com<mailto:lnykryn@redhat.com>> - 219-62.1
> - dhcp6: make sure we have enough space for the DHCP6 option header (CVE-2018-15688)
>
> Turns out something there was indirectly breaking libglvnd from functioning properly.

Strange - are you able to reproduce the issue by downgrading systemd ?

James Pearson






0 Plus One's     0 Comments  
   

Response from James Pearson @ Jan. 21, 2019, 12:10 p.m.
Michael Rochefort wrote: > > Well, this was interesting. After a few days of further debugging with containers (and about to start testing 16 self-built libglvnd packages) everything started magically working today. The reason: systemd. > > In the past 24 hours RH/CentOS released updates for the following packages: > > systemd, systemd-libs, systemd-python, systemd-sysv, and libgudev1 which had the following in the changelog (the only two commits since RHEL 7.6's release): > > * Mon Jan 07 2019 Lukas Nykryn > - 219-62.2 > - journald: do not store the iovec entry for process commandline on stack (#1657788) > - journald: set a limit on the number of fields (1k) (#1657792) > - journal-remote: set a limit on the number of fields in a message (#1657792) > > * Mon Nov 26 2018 Lukas Nykryn > - 219-62.1 > - dhcp6: make sure we have enough space for the DHCP6 option header (CVE-2018-15688) > > Turns out something there was indirectly breaking libglvnd from functioning properly. Strange - are you able to reproduce the issue by downgrading systemd ? James Pearson To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

0 Plus One's     0 Comments  
   

Response from Mike Rochefort @ Jan. 17, 2019, 3:20 a.m.
Well, this was interesting. After a few days of further debugging with containers (and about to start testing 16 self-built libglvnd packages) everything started magically working today. The reason: systemd.

In the past 24 hours RH/CentOS released updates for the following packages:

systemd, systemd-libs, systemd-python, systemd-sysv, and libgudev1 which had the following in the changelog (the only two commits since RHEL 7.6's release):

* Mon Jan 07 2019 Lukas Nykryn <lnykryn@redhat.com> - 219-62.2
- journald: do not store the iovec entry for process commandline on stack (#1657788)
- journald: set a limit on the number of fields (1k) (#1657792)
- journal-remote: set a limit on the number of fields in a message (#1657792)

* Mon Nov 26 2018 Lukas Nykryn <lnykryn@redhat.com> - 219-62.1
- dhcp6: make sure we have enough space for the DHCP6 option header (CVE-2018-15688)

Turns out something there was indirectly breaking libglvnd from functioning properly.

Cheers,
Mike


---- On Wed, 02 Jan 2019 15:57:51 -0500 Michael Rochefort <mike@michaelrochefort.com> wrote ----

Alright, results:

Maya is not compatible with GLVND enabled drivers. If you are using the ElRepo NVIDIA driver, updating to 7.6 will nulify Maya's VP2 and color management. You will need to install the driver by hand and explicitly specify the GLVND disabling flags. This isn't ideal as it will overwrite system libraries, and any updates to the glvnd-* libraries could potentially overwrite the driver.

Cheers,
Mike


---- On Fri, 28 Dec 2018 13:30:32 -0500 Michael Rochefort <mike@michaelrochefort.com> wrote ----




To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe
Hello,

Last night I upgraded my personal workstation to CentOS 7.6 (I use it as my testbed as well for desktop related things). Since then, Maya's Viewport 2.0 has
been impossible to activate. It seems like Maya isn't properly finding the NVIDIA driver, which may or may not be due to the switch to GLVND in 7.6. All other applications (Mari, Substance, Nuke, etc) seem to be finding things just fine. Has anyone else run into this issue or are you all playing it smart and not updating things?

Currently running 2018.2, also tried 2018.5 and no change. Driver 410.78 from elrepo.

Cheers,
Mike


To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

0 Plus One's     0 Comments  
   

Response from Mike Rochefort @ Jan. 2, 2019, 4 p.m.
Alright, results:

Maya is not compatible with GLVND enabled drivers. If you are using the ElRepo NVIDIA driver, updating to 7.6 will nulify Maya's VP2 and color management. You will need to install the driver by hand and explicitly specify the GLVND disabling flags. This isn't ideal as it will overwrite system libraries, and any updates to the glvnd-* libraries could potentially overwrite the driver.

Cheers,
Mike


---- On Fri, 28 Dec 2018 13:30:32 -0500 Michael Rochefort <mike@michaelrochefort.com> wrote ----

Hello,

Last night I upgraded my personal workstation to CentOS 7.6 (I use it as my testbed as well for desktop related things). Since then, Maya's Viewport 2.0 has
been impossible to activate. It seems like Maya isn't properly finding the NVIDIA driver, which may or may not be due to the switch to GLVND in 7.6. All other applications (Mari, Substance, Nuke, etc) seem to be finding things just fine. Has anyone else run into this issue or are you all playing it smart and not updating things?

Currently running 2018.2, also tried 2018.5 and no change. Driver 410.78 from elrepo.

Cheers,
Mike


To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

0 Plus One's     0 Comments