Hey Dennis! Thanks for this Flip, this is pretty confusing part of the autoscaler logic especially with the hardcoded logic.
I agree with the new behaviour and proposed new defaults. Maybe one comment regarding the config keys. I think the old key: job.autoscaler.scaling.key-group.partitions.adjust.mode is pretty bad, in terms of naming/expressivity. Maybe this is a good chance to actually change the name to something like job.autoscaler.scaling.key-group-alignment.mode or job.autoscaler.scaling.key-group.parallelism-adjustment.mode In any case we should go head with this FLIP to the vote and we can agree on the config naming in the meantime. Gyula On Mon, May 18, 2026 at 2:36 PM Dennis-Mircea Ciupitu < [email protected]> wrote: > Hi all, > > Bringing this thread back up after several weeks of quiet. > > The FLIP proposes composable alignment modes for the autoscaler so users > can opt into the strategy that fits their job rather than the single fixed > rule applied today (FLINK-32119 and FLINK-32120 are concrete instances of > the resulting friction). Full motivation, alternatives, and design details > are in the FLIP doc. > > Design doc: > > https://docs.google.com/document/d/1IbvKspt5jjjqiXl_CeHgI8EBn8LP-Pw0t7LqF85yFRw/edit?usp=sharing > Draft implementation: > https://github.com/apache/flink-kubernetes-operator/pull/1088 > > @Marton Balassi, @Gabor Somogyi, and @Peter Huang, based on your vote on > [1] & [2], I would really value your read here. > @Maximilian Michels, you opened FLINK-32119 and FLINK-32120, would value > your read on whether the composable-modes direction matches what you had in > mind when filing those. > > Thanks, > Dennis > > [1] https://lists.apache.org/thread/zffx5r6fjkjldjjxr8oqtkwsobtwc6c3 > [2] https://lists.apache.org/thread/pkn8rmdfzfq5vl8smj8kw03jns9gq8ty > > On Thu, Apr 23, 2026 at 2:23 PM Dennis-Mircea Ciupitu < > [email protected]> wrote: > > > Hi David, > > > > Thanks for the feedback, I addressed them in the FLIP. > > > > Please have a look once again and let me know what you think. > > > > Best regards, > > Dennis > > > > > > On Thu, Apr 23, 2026 at 11:17 AM David Radley <[email protected]> > > wrote: > > > >> Hi there, > >> It would be useful in the motivation to explicitly describe the value > of > >> the different options and the use cases that would drive the need for > each > >> of these permutations and all best practises. Can the autoscaler work > out > >> automatically which of these to use? > >> Kind regards, David. > >> > >> From: Dennis-Mircea Ciupitu <[email protected]> > >> Date: Wednesday, 15 April 2026 at 08:14 > >> To: [email protected] <[email protected]> > >> Subject: [EXTERNAL] [DISCUSS] FLIP-XXX: Composable Parallelism Alignment > >> Modes for Flink Autoscaler > >> > >> Hi all, > >> > >> [Sorry for reposting. I noticed the title in the previous message wasn’t > >> correct, so I’m sharing the updated version here.] > >> > >> I’d like to start a discussion on FLIP-XXX: Composable Parallelism > >> Alignment Modes for Flink Autoscaler [1]. > >> > >> I’ve also opened a draft PR with a reference implementation to make the > >> proposal concrete and easier to review [2]. > >> > >> Feedback and suggestions are very welcome. > >> > >> Best regards, > >> Dennis > >> > >> [1] > >> > >> > https://docs.google.com/document/d/1IbvKspt5jjjqiXl_CeHgI8EBn8LP-Pw0t7LqF85yFRw/edit?usp=sharing > >> [2] https://github.com/apache/flink-kubernetes-operator/pull/1088 > >> > >> Unless otherwise stated above: > >> > >> IBM United Kingdom Limited > >> Registered in England and Wales with number 741598 > >> Registered office: Building C, IBM Hursley Office, Hursley Park Road, > >> Winchester, Hampshire SO21 2JN > >> > > >
