You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@mesos.apache.org by Kevin Klues <kl...@gmail.com> on 2016/04/05 01:41:39 UTC
Review Request 45715: Added support to grant access to /dev/nvidiactl
in Nvidia GPU isolator.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45715/
-----------------------------------------------------------
Review request for mesos, Ben Mahler, Rob Todd, and Vikrama Ditya.
Bugs: MESOS-5115
https://issues.apache.org/jira/browse/MESOS-5115
Repository: mesos
Description
-------
Previously, calls to 'nvidia-smi' would fail inside a container even
if access to a GPU had been granted. Moreover, access to
/dev/nvidiactl is actually required for a container to do anything
useful with a GPU even if it has access to it.
This patch explicitly grants/revokes access to /dev/nvidiactl as GPUs
are added and removed from a container in the Nvidia GPU isolator.
Diffs
-----
src/slave/containerizer/mesos/isolators/cgroups/devices/gpus/nvidia.cpp b0f58035c7c819b42e5f249fadd97312f9e3ac7b
Diff: https://reviews.apache.org/r/45715/diff/
Testing
-------
Thanks,
Kevin Klues
Re: Review Request 45715: Fixed access to /dev/nvidia{ctl,
-uvm} in Nvidia GPU isolator.
Posted by Kevin Klues <kl...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45715/
-----------------------------------------------------------
(Updated April 6, 2016, 12:38 a.m.)
Review request for mesos, Ben Mahler, Rob Todd, and Vikrama Ditya.
Changes
-------
Addressed all of bmahler's comments. Also determined we needed to add access to /dev/nvidia-uvm in addition to /dev/nvidiactl.
Summary (updated)
-----------------
Fixed access to /dev/nvidia{ctl,-uvm} in Nvidia GPU isolator.
Bugs: MESOS-5115
https://issues.apache.org/jira/browse/MESOS-5115
Repository: mesos
Description (updated)
-------
Previously, calls to 'nvidia-smi' would fail inside a container even
if access to a GPU had been granted. Moreover, access to
/dev/nvidiactl is actually required for a container to do anything
useful with a GPU even if it has access to it.
This patch explicitly grants/revokes access to /dev/nvidiactl and
/dev/nvidia-uvm as GPUs are added and removed from a container in the
Nvidia GPU isolator.
Diffs (updated)
-----
src/slave/containerizer/mesos/isolators/cgroups/devices/gpus/nvidia.cpp b0f58035c7c819b42e5f249fadd97312f9e3ac7b
Diff: https://reviews.apache.org/r/45715/diff/
Testing (updated)
-------
Test in subsequent commit.
Thanks,
Kevin Klues
Re: Review Request 45715: Added support to grant access to
/dev/nvidiactl in Nvidia GPU isolator.
Posted by Ben Mahler <be...@gmail.com>.
> On April 5, 2016, 11:53 p.m., Ben Mahler wrote:
> >
Do we also need the uvm device?
- Ben
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45715/#review127255
-----------------------------------------------------------
On April 4, 2016, 11:41 p.m., Kevin Klues wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/45715/
> -----------------------------------------------------------
>
> (Updated April 4, 2016, 11:41 p.m.)
>
>
> Review request for mesos, Ben Mahler, Rob Todd, and Vikrama Ditya.
>
>
> Bugs: MESOS-5115
> https://issues.apache.org/jira/browse/MESOS-5115
>
>
> Repository: mesos
>
>
> Description
> -------
>
> Previously, calls to 'nvidia-smi' would fail inside a container even
> if access to a GPU had been granted. Moreover, access to
> /dev/nvidiactl is actually required for a container to do anything
> useful with a GPU even if it has access to it.
>
> This patch explicitly grants/revokes access to /dev/nvidiactl as GPUs
> are added and removed from a container in the Nvidia GPU isolator.
>
>
> Diffs
> -----
>
> src/slave/containerizer/mesos/isolators/cgroups/devices/gpus/nvidia.cpp b0f58035c7c819b42e5f249fadd97312f9e3ac7b
>
> Diff: https://reviews.apache.org/r/45715/diff/
>
>
> Testing
> -------
>
>
> Thanks,
>
> Kevin Klues
>
>
Re: Review Request 45715: Added support to grant access to
/dev/nvidiactl in Nvidia GPU isolator.
Posted by Ben Mahler <be...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45715/#review127255
-----------------------------------------------------------
Fix it, then Ship it!
src/slave/containerizer/mesos/isolators/cgroups/devices/gpus/nvidia.cpp (lines 65 - 78)
<https://reviews.apache.org/r/45715/#comment190468>
Could we avoid the static non-POD?
- Ben Mahler
On April 4, 2016, 11:41 p.m., Kevin Klues wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/45715/
> -----------------------------------------------------------
>
> (Updated April 4, 2016, 11:41 p.m.)
>
>
> Review request for mesos, Ben Mahler, Rob Todd, and Vikrama Ditya.
>
>
> Bugs: MESOS-5115
> https://issues.apache.org/jira/browse/MESOS-5115
>
>
> Repository: mesos
>
>
> Description
> -------
>
> Previously, calls to 'nvidia-smi' would fail inside a container even
> if access to a GPU had been granted. Moreover, access to
> /dev/nvidiactl is actually required for a container to do anything
> useful with a GPU even if it has access to it.
>
> This patch explicitly grants/revokes access to /dev/nvidiactl as GPUs
> are added and removed from a container in the Nvidia GPU isolator.
>
>
> Diffs
> -----
>
> src/slave/containerizer/mesos/isolators/cgroups/devices/gpus/nvidia.cpp b0f58035c7c819b42e5f249fadd97312f9e3ac7b
>
> Diff: https://reviews.apache.org/r/45715/diff/
>
>
> Testing
> -------
>
>
> Thanks,
>
> Kevin Klues
>
>