As I advocated in previous threads I don't think we have the necessary
(human & CI) resources to maintain multiple release lines so from my
perspective when 4.2.0 goes out 4.1.0 should go automatically EOL; in
other words there won't be a 4.1.1 release. We can continue iterating
on this discussion but it's important to converge before we publish
the 4.2.0 so that there is a clear communication to our users.

Regarding HIVE-29145 and HIVE-29213, I posted some thoughts under the
respective tickets and I would encourage others to do that as well.

Regarding the addition of the get_table method, I don't fully
understand how it would help Spark if they don't upgrade to a
supported version but feel free to log a jira ticket and we can
iterate there to find a solution.

Deprecation and removal is a very important discussion but slightly
off scope in the current thread. Both contributors and reviewers
should carefully evaluate these aspects when raising/accepting PRs.
Any concrete ideas/initiatives to improve on this aspect are always
welcome so feel free to start/submit threads/docs/tickets/tests on
this topic.

Best,
Stamatis

On Sat, Oct 25, 2025 at 5:55 AM Butao Zhang <[email protected]> wrote:
>
> Thank you, Okumin, for taking on the responsibility of the next Hive release.
>
> I'd also like to discuss an issue here. Since Hive 4.0.1, Apache Spark has 
> been unable to query Iceberg tables in HMS. The specific issue discussion can 
> be found at [1]. This problem occurred because we removed the `get_table` 
> method in [2]. Spark relies on Hive 2.3.10, which depends on the `get_table` 
> method to access Iceberg tables in HMS. The compatibility issue emerged after 
> HIVE-26537.
>
> To resolve this, the ideal solution would be for Apache Spark to upgrade its 
> built-in Hive dependency from version 2.3.10 to 4.0.1 or 4.1.0. I've noticed 
> that there is an upgrade ticket [3] in the Spark community for this. However, 
> this upgrade appears to involve significant changes and may not be resolved 
> for quite some time.
>
> Given that Apache Spark is one of the most widely used engines with HMS and 
> its adoption would help promote the use of Hive 4, is it possible for us to 
> reintroduce the removed `get_table` method? I understand that restoring 
> deprecated code is generally not a reasonable approach, so I'd like to ask if 
> there are any more elegant solutions on the HMS side to address this issue?
>
> Additionally, I have a personal thought: when removing or deprecating public 
> methods in Hive, we should carefully consider the dependencies of downstream 
> components on these methods to avoid long-term upgrade difficulties for them. 
> For example, when working on federated catalog [4] related tasks, I needed to 
> add catalog parameters to many methods in `Hive.java`. However, since Spark 
> code depends on `Hive.java` to access Hive, I had to introduce many new 
> methods with catalog parameters, which may lead to significant code 
> duplication. But there was no alternative, as modifying existing methods 
> would cause compatibility issues for downstream components with newer Hive 
> versions. Currently, this incompatibility risk relies on individual prior 
> knowledge, but I hope we can develop better methods in the future to 
> proactively identify such compatibility issues.
>
> [1]https://github.com/apache/iceberg/issues/12878
> [2]https://issues.apache.org/jira/browse/HIVE-26537
> [3]https://issues.apache.org/jira/browse/SPARK-52408
> [4]https://issues.apache.org/jira/browse/HIVE-29177
>
> Thanks,
> Butao Zhang
>
> On 2025/10/24 09:03:28 Shohei Okumiya wrote:
> > Hi Stamatis,
> >
> > Thanks for initiating this discussion. I'd love to have the great
> > release cadence!
> >
> > 1. I don't have any objections about the timeline. We might want to
> > release 4.1.1 in the future. I feel it should not block us from
> > shipping the master branch anyway.
> > 2. I'm willing to try being a release manager someday, but I have
> > frequent hospital visits nowadays. I'm probably not the right person,
> > given my unstable availability.
> > 3. Except for issues that have already been discussed in mailing
> > lists, I am not aware of critical problems.
> >
> > Best,
> > Okumin
> >
> > On Thu, Oct 23, 2025 at 11:50 AM Butao Zhang <[email protected]> wrote:
> > >
> > > Should we release version 4.1.1? If we decide to proceed with 4.1.1, we 
> > > may need to identify valuable patches from recent commits for the 4.1 
> > > branch, such as fixes for correctness issues, performance optimizations, 
> > > and security problems.
> > >
> > > If we skip 4.1.1 and instead release 4.2.0, would this imply that we've 
> > > reached a default agreement to only publish major version releases based 
> > > on the master branch going forward?
> > >
> > > Additionally, I have two tickets that may need to be addressed in the 
> > > next new version: one is HIVE-29145 (Iceberg Rest server cannot work if 
> > > launched by non-standalone HMS), and the other is HIVE-29213 (HS2 reports 
> > > 'Failed to get primary keys' when using old beeline client).
> > >
> > > Thanks,
> > > Butao Zhang
> > >
> > >
> > > On 2025/10/22 08:48:19 Denys Kuzmenko wrote:
> > > > Hi Stamatis,
> > > >
> > > > Thanks for bringing this up.
> > > >
> > > > We already have enough items ready to ship, including JDK 21.
> > > > I'm not sure what’s causing the delay. I thought we already had a 
> > > > release manager, but looks like we don’t.
> > > >
> > > > Regards,
> > > > Denys
> > > >
> >

Reply via email to