mxm commented on PR #762:
URL:
https://github.com/apache/flink-kubernetes-operator/pull/762#issuecomment-1926743545
> The memory tuning logic looks good 👍
>
> I think we should unify the storing / application of parallelism and other
overrides. In this PR we introduced a bunch of new methods for storing config
overrides. But the next time we add other tuning logic such as ephemeralStorage
or whatever that doesn't translate to a config we would have to introduce again
a similar set of new methods everywhere.
>
> Maybe we could have a dedicated class for the overrides applied by the
autoscaler:
>
> ```
> class AutoscalerOverrides{
> Map<JobVertexId, Integer> parallelism;
> Configuration configOverrides;
> ...other stuff later
> }
> ```
>
> This way we could have all the related logic / utilities for ser/de
collected there.
>
> This would make the storage / retrieval logic simpler and would be very
easily extensible. Wdyt @mxm @1996fanrui ?
I had thought about that but I was a bit concerned about touching the
parallelism overrides because any breaking changes there can have a dramatic
effect on any deployed autoscaled applications. The newly added Configuration
at least will cover all configuration-related changes. Those would also include
the parallelism overrides themselves. So I understand the idea to combine those
two and potentially other types of overrides. Let me look into it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]