> On 七月 27, 2016, 2:15 a.m., Guangya Liu wrote:
> > src/slave/containerizer/docker.cpp, line 1333
> > <https://reviews.apache.org/r/50123/diff/1/?file=1445669#file1445669line1333>
> >
> > 1) s/allocator.get().allocate(requested)/allocator->allocate(requested)
> >
> > 2) I think it is not right to use `Option` here as the
> > `allocator->allocate(requested)` is returning a `Future`.
> >
> > One proposal is as following:
> > ```
> > allocator->allocate(requested)
> > .then(defer(...));
> > ```
> >
> > You can also refer to
> > https://github.com/apache/mesos/blob/master/src/slave/containerizer/mesos/isolators/gpu/isolator.cpp#L356
> > to get some detail.
>
> Yubo Li wrote:
> `Future.get()` will block the host threading and wait for the
> sub-threading return its value. For docker containerizer, we need to wait
> `Option<set<Gpu>> allocated` return, and pass it to mesos-docker-executor,
> it's hard to follow the async calling style. For mesos containerizer, it uses
> cgroup to set device control directly, so the async call is fine.
Since here we must block the host threading to get GPU allocated results, The
logic fixed as:
+ if (requested > 0) {
+ // Some GPUs requested.
+ Future<set<Gpu>> future = allocator->allocate(requested);
+
+ // We block the code until the GPU allocation finished.
+ // For docker containerizer, we must wait for GPUs allocated
+ // completed, and pass allocated GPUs to mesos-docker-executor.
+ future.await();
+ if (!future.isReady()) {
+ return Failure("GPU allocator allocating GPU failed");
+ }
+
+ set<Gpu> allocated = future.get();
+
+ foreach (const Gpu& gpu, allocated) {
+ container->gpuAllocated.push_back(gpu);
+ }
The logic should be safe because I refer to:
https://github.com/liyubobj/mesos/blob/master/src/slave/containerizer/fetcher.cpp#L326
- Yubo
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/50123/#review143639
-----------------------------------------------------------
On 七月 18, 2016, 9:17 a.m., Yubo Li wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/50123/
> -----------------------------------------------------------
>
> (Updated 七月 18, 2016, 9:17 a.m.)
>
>
> Review request for mesos, Benjamin Mahler, Kevin Klues, and Rajat Phull.
>
>
> Bugs: MESOS-5795
> https://issues.apache.org/jira/browse/MESOS-5795
>
>
> Repository: mesos
>
>
> Description
> -------
>
> This added 'NvidiaGpuAllocator' to docker containerizer so that the
> docker containerizer can use it to allocate GPUs to the task with 'gpus'
> resource. Also, allocated GPUs will automatically deallocated after the
> job destroyed.
>
>
> Diffs
> -----
>
> src/slave/containerizer/docker.hpp 43ca4317d608b3b43dd7bd0d1b55c721e7364885
> src/slave/containerizer/docker.cpp f1ecf3b25d85597f6c3dcaa47968860ed119dbd5
>
> Diff: https://reviews.apache.org/r/50123/diff/
>
>
> Testing
> -------
>
> make check
>
>
> Thanks,
>
> Yubo Li
>
>
