Github user ryan-williams commented on the pull request:
https://github.com/apache/spark/pull/3525#issuecomment-65022899
sure, will file JIRA.
To answer what I think is the spirit of your question, whether a spark user
wants more or less than 7% seems like something that should be configurable. I
have definitely wanted to set this value to closer to 10% of the executor
memory, or higher, to make sure that that wasn't my problem, when debugging
some of my own Spark jobs. I can't swear that I've seen yarn kill my executors
for exceeding the 7%, but I've definitely felt like I wanted to configure it.
I can see that it's unusual to have two params that both affect this value,
but it seems like the right tradeoff, since the fraction more closely models
what we want, but by an accident of history we have this other value and we
don't want to break backwards-compatibility.
Everyone I've heard talk about how to think about the overhead (which is
mostly you, admittedly) talks about it in terms of the fraction of the executor
memory. If you think the best course is to deprecate the
absolute-amount-of-memory one as well, I'm open to that, but I don't think the
status quo (1. have a hard-coded value for the logical param that we care
about, and 2. have a param that sets the value we want indirectly) makes much
sense.
---
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]
