Thank you!
I have raised a PR[1] to remove the experimental annotation and update the
documents to remove the experimental connector.
The whole idea is to promote using native-s3-fs. At the same time,
users have the flexibility to use any of the supported s3-fs.

Bests,
Samrat

[1] https://github.com/apache/flink/pull/29159


On Fri, Sep 11, 2026 at 7:02 PM Ferenc Csaky <[email protected]> wrote:

> Hi Samrat,
>
> Thanks for driving this, +1 to the suggestion, also taking into account G's
> argument, it makes sense IMO.
>
> Best,
> Ferenc
>
>
>
> On Friday, September 11th, 2026 at 13:57, Samrat Deb <
> [email protected]> wrote:
>
> > [update]
> >
> > I have raised a PR[1] for the changes.
> >
> > The change doesn't affect any other changes to S3 connectors.
> >
> > Bests,
> > Samrat
> > [1] https://github.com/apache/flink/pull/29159
> >
> > On Fri, Sep 11, 2026 at 3:29 PM Samrat Deb <[email protected]>
> wrote:
> >
> > > Thank you @[email protected] <[email protected]>
> > >
> > > Yes, This totally aligned with making native-s3-fs mainline along with
> > > existing connectors.
> > > Looking forward to hearing from the community to move forward.
> > >
> > > Bests,
> > > Samrat
> > >
> > >
> > > On Tue, Sep 8, 2026 at 2:22 PM Gabor Somogyi <
> [email protected]>
> > > wrote:
> > >
> > >> Hey Samrat,
> > >>
> > >> +1 to promote in short.
> > >>
> > >> I personally think that one to one feature parity was not a target.
> > >> Hadoop is a general purpose FS with huge amount of tuning
> possibilities
> > >> however the new native S3 connector is specific to our Flink needs
> > >> with smaller amount of tuning.
> > >>
> > >> The feature set which allows users to use it in production is there
> > >> both from data processing and checkpointing perspective.
> > >> It's used in many production pipelines already so from my point of
> view
> > >> we can promote it as the preferred way. It's important to say it out
> > >> that the old connectors must remain from code and documentation
> > >> perspective but as a fallback if something can't be done with the new
> > >> native s3 connector.
> > >>
> > >> After the promotion we should gather experience from the community,
> fill
> > >> those
> > >> gaps if there are, fix bugs and that is going to allow us to sunset
> the
> > >> old
> > >> connectors. I'm pretty sure that won't be lightning fast since we
> should
> > >> allow everybody time to do the migration and express the missing
> > >> parts/bugfixes.
> > >>
> > >> Happy to hear other voices.
> > >>
> > >> BR,
> > >> G
> > >>
> > >>
> > >> On Mon, Sep 7, 2026 at 7:25 AM Samrat Deb <[email protected]>
> wrote:
> > >>
> > >> > Hi All,
> > >> >
> > >> > flink-s3-fs-native (the AWS-SDK-v2, Hadoop-free S3 filesystem for
> s3://
> > >> and
> > >> > s3a://) landed experimentally in Flink 2.3[1]. Since then, the
> [Phase 2]
> > >> > Native S3 Filesystem FLINK-39109 [2] has closed 19 subtasks,
> closing the
> > >> > feature gap compared to flink-s3-fs-hadoop or flink-s3-fs-presto.
> > >> >
> > >> > Proposal
> > >> >
> > >> > 1. promote NativeS3FileSystem from Experimental to mainstream in an
> > >> > upcoming 2.4 release[3]. What is the community view and overall
> > >> consensus?
> > >> >
> > >> > 2. If you run flink-s3-fs-hadoop or flink-s3-fs-presto in
> production,
> > >> > please reply with anything you rely on that flink-s3-fs-native
> doesn't
> > >> > cover yet. I've checked all three plugins' sources for the above,
> but
> > >> > deployment-specific setups (custom credential providers, particular
> > >> > S3-compatible endpoints, etc.) may surface gaps I've missed.
> > >> >
> > >> > Bests,
> > >> > Samrat
> > >> >
> > >> > [1]
> > >> >
> > >> >
> > >>
> https://cwiki.apache.org/confluence/spaces/FLINK/pages/396790457/FLIP-555+Flink+Native+S3+FileSystem
> > >> > [2] https://issues.apache.org/jira/browse/FLINK-39109
> > >> > [3]
> https://lists.apache.org/thread/14vym623dh08rry7jsxj2sm6ty4m6cho
> > >> >
> > >>
> > >
> >
>

Reply via email to