Github user koeninger commented on the pull request:
https://github.com/apache/spark/pull/9597#issuecomment-156457252
Well, the implementation of HasOffsetRanges.offsetRanges in KafkaRDD is
just the val offsetRanges provided on creation, so you're talking about a
fair amount of change. It's a little weird from my perspective to create
an object with a given set of topicpartitions, then when you ask for them
back, to get a different set.
I'm also not 100% sure what practical problem this is solving. Is this
actually sufficient to get dynamic allocation on a stream working
correctly? Also, if someone has particular partitions that are empty
frequently enough to actually get a resource benefit from dynamic
allocation, it sounds like something is wrong with their partitioning
scheme.
In terms of user expectations, if I have a job currently that is e.g.
reporting on the rate of messages per topicpartition, isn't it suddenly
going to stop reporting anything in cases where it would previously report
0?
On Thu, Nov 12, 2015 at 7:46 PM, Saisai Shao <[email protected]>
wrote:
> OK, get it. How about this, still keeping the mapping relation from offset
> Range to rdd partition, but filter out empty partition, so offset(i)
> still map to partitions(i), what do you think?
>
> â
> Reply to this email directly or view it on GitHub
> <https://github.com/apache/spark/pull/9597#issuecomment-156291634>.
>
---
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]
