Repository navigation
Release DataFusion 55.0.0 (Jul / Aug 2026) #22393
Description
Activity
- changed the title
[-]Release DataFusion `55.0.0` (Jun 2026 / Jul 2026)[/-][+]Release DataFusion `55.0.0` (Jul 2026)[/+]on May 20, 2026 - changed the title
[-]Release DataFusion `55.0.0` (Jul 2026)[/-][+]Release DataFusion `55.0.0` (Jul / Aug2026)[/+]on Jun 22, 2026 - changed the title
[-]Release DataFusion `55.0.0` (Jul / Aug2026)[/-][+]Release DataFusion `55.0.0` (Jul / Aug 2026)[/+]on Jun 22, 2026 Updated the title to reflect my prediction we'll get this one done in late July / early August
Reacted by Shehab Amin- added a commit that references this issue
on Jul 14, 2026 #22988 should also be merged so we can announce complete MERGE INTO support all the way down to the TableProvider hook.
Reacted by Tim Saucer and Alexander HirnerI am thiniking we would start working on this release mid aug and then target release end of Aug?
Reacted by Matt Butrovich and Shehab AminI think @timsaucer mentioned he wanted to get this into the release
👋🏽 Highlighting some dynamic filtering related changes which would be nice to get in
Reacted by Tim SaucerReacted by RJ Barman47 remaining items
Ok, #24377 is ready for the backport of #24365 into
branch-55.I'm looking at #24195 which looks like it's probably important because without it we're releasing the
MERGE INTOwithout implementation for our memory table. 🤔 It has a merge conflict @wirybeaver but I'll start reviewing it.Ok, #24377 is ready for the backport of #24365 into
branch-55.I approved and merge this
I'm looking at #24195 which looks like it's probably important because without it we're releasing the
MERGE INTOwithout implementation for our memory table. 🤔 It has a merge conflict @wirybeaver but I'll start reviewing it.I think this is something we could do as a follow on (e.g. 55.1 or just in 56) as it is a net new feature
The rationale for including MERGE INTO support as I understand it, is to allow downstream consumers such as Iceberg to implement merge (they have their own TableProvider so they won't use the MemoryTable implementation anyways for testing
Reacted by Tim SaucerThanks. I’ll start a new release candidate this evening. I think all open issues are resolved.
RC3 release vote: https://lists.apache.org/thread/wf9yzfo57z8hxz0dovchghbwmkqtlzor
It looks like we are good to release. As I mentioned during the community call the week before last I’m traveling. Is any PMC member available to do the last parts - sending results, publishing crates, and merging in commits for version updates and changelogs?
@timsaucer @alamb Could we review and land #24429 before DataFusion 55.0.0? #22988 canonicalizes the target alias (
t) to its table name (target). This rejects target-correlated subqueries and source aliastarget. Recursive rewriting is unsafe because qualifiers are scope-local and may be shadowed; it also cannot distinguish targett.idfrom sourcetarget.idafter rewriting.#24429 instead stores the SQL-visible
target_qualifierseparately from provider identity and serializes it in protobuf, withtable_namefallback for existing payloads. Since MERGE has not been released, landing this now avoids publishing a proto representation that immediately needs correction.#24195 consumes this MERGE representation and schema, so it should rebase on #24429 and can remain the execution follow-up after the 55.0.0 release
It looks like we are good to release. As I mentioned during the community call the week before last I’m traveling. Is any PMC member available to do the last parts - sending results, publishing crates, and merging in commits for version updates and changelogs?
I can handle doing this later today if no one else volunteers
@timsaucer @alamb Could we review and land #24429 before DataFusion 55.0.0?
@wirybeaver -- unfortunately these features / fixes missed the 55.0.0 release (which is approved and will be released later)
Instead, we can consider these items for the 55.1.0 minor release. I added them to the tracking ticket
I filed some follow on tickets for the next releases
Reacted by Matt Butrovich and Kumar UjjawalThe release is now available!
https://lists.apache.org/thread/90okqtgtz3xwsyddpwhmvxrkkrd9h2sfI hit two small errors during publish
- Doc update: docs: Update dev/release/README.md for new crate
datafusion-proto-models#24472 usetweak: Error publishing datafusion-spark in datafusion 55.0.0 release "cannot find module or cratedatafusionin this scope" #24474
Here is a PR to port version and changelog to main:
The next release is tracked in these tickets:
- Doc update: docs: Update dev/release/README.md for new crate
The blog post is a bit late but now ready for review
- added a commit that references this issue
on Aug 26, 2026
Is your feature request related to a problem or challenge?
Previous release is #21080 (May, 2026) so next major release would be approximately Jul / Aug 2026
Release Coordinator: @timsaucer
Steps:
branch-55branch: https://github.com/apache/datafusion/tree/branch-5556.0.0(Oct/Nov 2026) #2446155.0.0release #24216Features to mention in the blog (if they make it)
file_row_indexUDF to query file-level row indexes from Parquet files #22604 from @AdamGSunnest_outerfunction for arrays. #22100ROW_NUMBER < 5/ TopK #6899 from @SubhamSinghalfilter_pushdown) by default #3463Tracking Epics
Bugs that should be fixed
Bugs requested but determined non blocking
get_fieldpredicate when the file needs schema adaptation (wrong results) #24109 from @adriangb -- not a regression in 55 (was in 54 too)GroupedHashAggregateStream::emitthrows ArrowError: column types must match schema types #24069 from @hsiang-c -- not a regression, seeBugs requested to fix
RepartitionExecwithholds all output from an unbounded input untilbatch_sizerows accumulate #24044 (probably needs a more rigorous treatment in [EPIC] Better support / documentation / tests for Streaming SQL Engines #24265)Community Wishlist
ExecutionPlanAPI withreplace_children#23903 (@JSOD11 )