client-supplied ID for EvaluatorRequest?

[Tobin Baker <td...@cs.washington.edu>]

Hi, I need to allocate Evaluators for some conceptually different Tasks
(master/workers), possibly on the same node (e.g., in single-node dev
environment). I'd like to be able to supply my own identifier with the
EvaluatorRequest, so in the AllocatedEvaluator event handler I could
associate the AllocatedEvaluator instance with an EvaluatorRequest, and
assign an appropriate Context (and subsequently assign an appropriate Task
based on the Context ID), but I don't see a way to do that. (The motivation
is that I might submit multiple EvaluatorRequests for the same node, but
they might have different resource requirements, and trying to guess which
Context I should submit to an AllocatedEvaluator just based on the
resources it reports is a ridiculous hack and unsustainable.) Is there
something available that I'm missing?

Thanks,
Tobin

Re: client-supplied ID for EvaluatorRequest?

[Tobin Baker <td...@cs.washington.edu>]

OK, thanks, my workaround for now is just to request my "special" evaluator
first and block all further requests until I've gotten an
AllocatedEvaluator event for that first request. If I had more than 2
evaluator types it would be more difficult, but this is good enough for now.

On Thu, Jun 25, 2015 at 2:22 PM, Markus Weimer <ma...@weimo.de> wrote:

> Hi,
>
> > Is there something available that I'm missing?
>
> This is a very common feature request that is, unfortunately,
> conditioned on YARN to satisfy. So far, that hasn't happened :-(
>
>
> Markus
>

Re: client-supplied ID for EvaluatorRequest?

[Markus Weimer <ma...@weimo.de>]

Hi,

> Is there something available that I'm missing?

This is a very common feature request that is, unfortunately,
conditioned on YARN to satisfy. So far, that hasn't happened :-(


Markus