Github user koeninger commented on the pull request:
https://github.com/apache/spark/pull/3798#issuecomment-71131380
Point is, it's up to client code to commit, so that it can implement
exactly-once semantics if necessary. Committing automatically at the end
of compute would get you something like at-most-once semantics.
Regarding the spikiness, in practice, I think you're just going to end up
doing what you do with any other streaming job, namely tuning the batch
size down until it's just comfortably above the processing time.
On Thu, Jan 22, 2015 at 6:02 PM, Hari Shreedharan <[email protected]>
wrote:
> Say we start pulling down info in real time, wouldn't it be possible to
> say get me only "n" messages -- that should take care of the second point.
>
> I am not sure how the ending offset part is a problem. Wouldn't it make
> sense to do the commits at the end of the compute call? (You don't
actually
> commit the offsets to Kafka anyway, correct?)
>
> â
> Reply to this email directly or view it on GitHub
> <https://github.com/apache/spark/pull/3798#issuecomment-71124795>.
>
---
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]
