Hi Chris,
Since you implemented a custom executor, it's really up to you how you want
to launch and make sure the task uses the right resources.
Essentially you'll need to launch your multiple executors with the right
resources flags (-c, -m).
We don't expose any API to get the current Task's cgroup root, but you
essentially can look it up yourself by examining /proc/{pid} where pid is
your current running executor's pid, and you can find out the current
cgroup root that this process is under and setup cgroup_parent with it.
Hopefully that makes sense.
Tim
On Sat, Jun 20, 2015 at 8:06 AM, Christoph Heer <christ...@thelabmill.de>
wrote:
> Thanks for your answer Alex. Thank you for the link to new docker executor
> it's very interesting and I see some great benefits for my usage of Mesos.
> But I think my previous explanation wasn't good enough.
>
>
> My scheduler starts a task which will be handle by a custom executor. The
> executor could run inside of a docker container but to keep the use-case
> more general the executor runs normally in the slave user-space.
> To process an single task the executor must start multiple Docker
> containers. Is it possible to make sure that these new docker containers
> are only able to use the assigned task resources?
>
> As I understood Mesos creates for every task a new cgroup to isolate
> resources like CPU and Memory. Would it be possible to get the name/path of
> the cgroup for the current task and start the required Docker containers
> with the "--cgroup-parent" option?
>
>
> Hopefully this explanation of my use-case is better.
>
>
> Best regards,
> Christoph
>
>
>
> > On 18 Jun 2015, at 16:52, Alex Rukletsov <a...@mesosphere.com> wrote:
> >
> > Christoph,
> >
> > I'm not sure I got your use case correctly, feel free to elaborate or
> refine your question. From what I've understood, you may want to write your
> own executor, that is able to handle multiple tasks (which can be docker
> containers). You may want to get inspiration from docker executor, which
> will appear in Mesos 0.23.
> >
> > On Wed, Jun 17, 2015 at 10:52 PM, Christoph Heer <
> christ...@thelabmill.de> wrote:
> > Hi,
> >
> > at the moment I'm implementing a framework for Mesos. My scheduler
> starts one task which spawns a container with an executor which is able to
> handle and run the task. This pattern allows that the code/executable of
> the executor mustn't present on the slave node. In my use-case the tasks
> requires to run multiple Docker container which would be started by the
> executor who also has access on the Docker daemon but this would break the
> resource isolation of Mesos.
> >
> > Have someone an advice to handle this types of tasks which requires
> additional containers? It could be also possible that a executor builds a
> Docker image which also requires to spawn additional contains in the
> context/resource space of one single task.
> >
> >
> > Thanks and regards
> > Christoph
> >
> >
>
>
