Github user klueska commented on a diff in the pull request:
https://github.com/apache/spark/pull/14644#discussion_r78012264
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosCoarseGrainedSchedulerBackend.scala
---
@@ -103,6 +103,7 @@ private[spark] class MesosCoarseGrainedSchedulerBackend(
private val stateLock = new ReentrantLock
val extraCoresPerExecutor = conf.getInt("spark.mesos.extra.cores", 0)
+ val maxGpus = conf.getInt("spark.mesos.gpus.max", 0)
--- End diff --
Yeah, it makes sense. The only additional question I had was if there's a
way to auto-discover the number of GPUs on the machine and do some math on
it inside the config. i.e. does the config allow bash syntax or something
so I can inspect `/dev/` and count the number of GPUs installed? I'm
picturing a scenario where I'd like to deploy the same config to a bunch of
hosts, but have the autodiscover this value themselves.
On Thu, Sep 8, 2016 at 3:17 PM Timothy Chen <[email protected]>
wrote:
> In
>
core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosCoarseGrainedSchedulerBackend.scala
> <https://github.com/apache/spark/pull/14644#discussion_r78002761>:
>
> > @@ -103,6 +103,7 @@ private[spark] class
MesosCoarseGrainedSchedulerBackend(
> > private val stateLock = new ReentrantLock
> >
> > val extraCoresPerExecutor = conf.getInt("spark.mesos.extra.cores", 0)
> > + val maxGpus = conf.getInt("spark.mesos.gpus.max", 0)
>
> My thoughts was that by only allowing a Boolean flag a spark job either
> uses all GPUs from a host or not, which it won't be able to have different
> GPu devices shared by different jobs. By specifying a limit at least there
> is ability to let a job specify how much GPUs it should grab per node.
> thoughts?
>
> â
> You are receiving this because you commented.
> Reply to this email directly, view it on GitHub
>
<https://github.com/apache/spark/pull/14644/files/2f658b40251ff75ab4695e4afc55487a6a6eb015#r78002761>,
> or mute the thread
>
<https://github.com/notifications/unsubscribe-auth/AAF4o5c68RBufSO5UF2DpuLtcbkspxClks5qoArjgaJpZM4JkJdn>
> .
>
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
