You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@mesos.apache.org by Chun-Hung Hsiao <ch...@apache.org> on 2019/02/21 22:55:26 UTC
Re: Review Request 70016: Supported CUDA 10+ images that are based on
nvidia-container-runtime.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70016/
-----------------------------------------------------------
(Updated Feb. 21, 2019, 10:55 p.m.)
Review request for mesos, Gilbert Song, Jie Yu, and Kevin Klues.
Changes
-------
Addressed comments.
Summary (updated)
-----------------
Supported CUDA 10+ images that are based on nvidia-container-runtime.
Bugs: MESOS-9549
https://issues.apache.org/jira/browse/MESOS-9549
Repository: mesos
Description (updated)
-------
Nvidia's CUDA 10+ images are based on nvidia-container-runtime and thus
the runtime injection are from the images themselves. To adapt this
change, we adjusted the binaries and libraries and injected the `PATH`
and `LD_LIBRARY_PATH` environment variables in the `gpu/nvidia`
isolator.
Diffs (updated)
-----
src/slave/containerizer/mesos/isolators/gpu/isolator.cpp f39e7c3d1ccfe097116fe59b05c22fbb3f83b698
src/slave/containerizer/mesos/isolators/gpu/volume.hpp e71fe95234ff10c72cfaa4ad39591f70a531c383
src/slave/containerizer/mesos/isolators/gpu/volume.cpp 0d0d778d6a8467c1ac87286e75d47faf8243afa4
Diff: https://reviews.apache.org/r/70016/diff/2/
Changes: https://reviews.apache.org/r/70016/diff/1-2/
Testing
-------
`sudo make check`
More testing done later in chain.
Thanks,
Chun-Hung Hsiao
Re: Review Request 70016: Supported CUDA 10+ images that are based on
nvidia-container-runtime.
Posted by Chun-Hung Hsiao <ch...@apache.org>.
> On Feb. 25, 2019, 5:24 p.m., Jie Yu wrote:
> > src/slave/containerizer/mesos/isolators/gpu/volume.cpp
> > Lines 493 (patched)
> > <https://reviews.apache.org/r/70016/diff/5/?file=2126232#file2126232line502>
> >
> > Can you also add a TODO here about relying on `NVIDIA_DRIVER_CAPABILITIES` env var instead in the future?
After looking into the semantics of `NVIDIA_DRIVER_CAPABILITIES`, it seems better if we use the `NVIDIA_VISIBLE_DEVICES`:
`NVIDIA_VISIBLE_DEVICES`: void or empty or unset: nvidia-container-runtime will have the same behavior as runc.
`NVIDIA_DRIVER_CAPABILITIES`: empty or unset: use default driver capability: utility.
So the real control is `NVIDIA_VISIBLE_DEVICES`.
- Chun-Hung
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70016/#review213171
-----------------------------------------------------------
On Feb. 21, 2019, 11:03 p.m., Chun-Hung Hsiao wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70016/
> -----------------------------------------------------------
>
> (Updated Feb. 21, 2019, 11:03 p.m.)
>
>
> Review request for mesos, Gilbert Song, Jie Yu, and Kevin Klues.
>
>
> Bugs: MESOS-9549
> https://issues.apache.org/jira/browse/MESOS-9549
>
>
> Repository: mesos
>
>
> Description
> -------
>
> Nvidia's CUDA 10+ images are based on nvidia-container-runtime and thus
> the runtime injection are from the images themselves. To adapt this
> change, we adjusted the binaries and libraries and injected the `PATH`
> and `LD_LIBRARY_PATH` environment variables in the `gpu/nvidia`
> isolator.
>
>
> Diffs
> -----
>
> src/slave/containerizer/mesos/isolators/gpu/isolator.cpp f39e7c3d1ccfe097116fe59b05c22fbb3f83b698
> src/slave/containerizer/mesos/isolators/gpu/volume.hpp e71fe95234ff10c72cfaa4ad39591f70a531c383
> src/slave/containerizer/mesos/isolators/gpu/volume.cpp 0d0d778d6a8467c1ac87286e75d47faf8243afa4
>
>
> Diff: https://reviews.apache.org/r/70016/diff/5/
>
>
> Testing
> -------
>
> `sudo make check`
>
> More testing done later in chain.
>
>
> Thanks,
>
> Chun-Hung Hsiao
>
>
Re: Review Request 70016: Supported CUDA 10+ images that are based on
nvidia-container-runtime.
Posted by Jie Yu <yu...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70016/#review213171
-----------------------------------------------------------
Fix it, then Ship it!
src/slave/containerizer/mesos/isolators/gpu/volume.cpp
Lines 493 (patched)
<https://reviews.apache.org/r/70016/#comment299042>
Can you also add a TODO here about relying on `NVIDIA_DRIVER_CAPABILITIES` env var instead in the future?
- Jie Yu
On Feb. 21, 2019, 11:03 p.m., Chun-Hung Hsiao wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70016/
> -----------------------------------------------------------
>
> (Updated Feb. 21, 2019, 11:03 p.m.)
>
>
> Review request for mesos, Gilbert Song, Jie Yu, and Kevin Klues.
>
>
> Bugs: MESOS-9549
> https://issues.apache.org/jira/browse/MESOS-9549
>
>
> Repository: mesos
>
>
> Description
> -------
>
> Nvidia's CUDA 10+ images are based on nvidia-container-runtime and thus
> the runtime injection are from the images themselves. To adapt this
> change, we adjusted the binaries and libraries and injected the `PATH`
> and `LD_LIBRARY_PATH` environment variables in the `gpu/nvidia`
> isolator.
>
>
> Diffs
> -----
>
> src/slave/containerizer/mesos/isolators/gpu/isolator.cpp f39e7c3d1ccfe097116fe59b05c22fbb3f83b698
> src/slave/containerizer/mesos/isolators/gpu/volume.hpp e71fe95234ff10c72cfaa4ad39591f70a531c383
> src/slave/containerizer/mesos/isolators/gpu/volume.cpp 0d0d778d6a8467c1ac87286e75d47faf8243afa4
>
>
> Diff: https://reviews.apache.org/r/70016/diff/5/
>
>
> Testing
> -------
>
> `sudo make check`
>
> More testing done later in chain.
>
>
> Thanks,
>
> Chun-Hung Hsiao
>
>
Re: Review Request 70016: Supported CUDA 10+ images that are based on
nvidia-container-runtime.
Posted by Chun-Hung Hsiao <ch...@apache.org>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70016/
-----------------------------------------------------------
(Updated Feb. 21, 2019, 11:03 p.m.)
Review request for mesos, Gilbert Song, Jie Yu, and Kevin Klues.
Changes
-------
Updated the order of library path injection.
Bugs: MESOS-9549
https://issues.apache.org/jira/browse/MESOS-9549
Repository: mesos
Description
-------
Nvidia's CUDA 10+ images are based on nvidia-container-runtime and thus
the runtime injection are from the images themselves. To adapt this
change, we adjusted the binaries and libraries and injected the `PATH`
and `LD_LIBRARY_PATH` environment variables in the `gpu/nvidia`
isolator.
Diffs (updated)
-----
src/slave/containerizer/mesos/isolators/gpu/isolator.cpp f39e7c3d1ccfe097116fe59b05c22fbb3f83b698
src/slave/containerizer/mesos/isolators/gpu/volume.hpp e71fe95234ff10c72cfaa4ad39591f70a531c383
src/slave/containerizer/mesos/isolators/gpu/volume.cpp 0d0d778d6a8467c1ac87286e75d47faf8243afa4
Diff: https://reviews.apache.org/r/70016/diff/3/
Changes: https://reviews.apache.org/r/70016/diff/2-3/
Testing
-------
`sudo make check`
More testing done later in chain.
Thanks,
Chun-Hung Hsiao