Add Flink Community Update for Q2 2026

[sofie204]

Flink Community Update for Q2 2026 (draft)

First edition in the quarterly format agreed on dev@
(https://lists.apache.org/thread/bt9f3hwyl9dckn9kmk0tgkbqbsx4fw1b).
Covers April 1 – June 30, 2026.

Known gaps in this draft (in progress):

• Two Web UI screenshots (Applications view, Rescales tab)
• Announce-thread links for Gustavo de Morais, Hao Li, David Anderson
• 2.1.3 / 1.20.5 announcement links
• Publish date in filename/front matter to be finalized at merge
• Final wording pass ongoing

Questions for reviewers:

1. FLIP-587's RESULT thread tallies a single binding vote is the
   acceptance complete, or am I misreading the thread?
2. Anything from the quarter that happened off-list I should cover?

[davidradl]

my AI says

1. Typos and Language
   No spelling errors found. The writing is clean. A few punctuation and minor issues:

"A stuck exactly-once Kafka transaction used to force a choice between a downstream outage and data loss at broker timeout, fixable only with knowledge of producer fencing, FLIP-572 turns that improvised process..." — double space before "FLIP-572", and this is a comma splice joining two independent clauses. Should be a period or semicolon after "producer fencing".

"The release also brings cross-language actions (Java actions inside Python agents and vice versa) and a durable reconciler for recovering in-flight external side effects after failure recovery," — trailing comma at end of sentence with no continuation. Should be a period.

"The community also welcomed four new committers this quarter:" — reads fine, but the list items lack terminal punctuation (fine for lists; noting for consistency since some prose sentences nearby do use periods).

"Congratulations to all" — missing full stop.

"more info found in FLIPs being discussed" — awkward passive construction. "See FLIPs being discussed below" or "discussed in the FLIPs section below" would be cleaner.

2. Factual Errors and Inconsistencies
   FLIP-584 link is inconsistent. In the Kubernetes Operator section, FLIP-584 is linked to a Jira ticket (https://issues.apache.org/jira/browse/FLINK-39911) while every other FLIP in the post links to Confluence. This looks like a copy/paste error — either the link is wrong, or the FLIP number is wrong.

"The announcement blog covers all fifteen FLIPs" — but the New Releases section says the release "implements the full or core functionality of fifteen FLIPs". These are consistent with each other, which is good. Just flagging they should stay in sync if the final count changes.

StateFun vote status — "The formal vote... opened in late May and remains open at the time of writing." If this post is publishing on July 30, 2026, and the vote opened in late May, a two-month-open vote is unusual and may have concluded by publication. This should be verified and updated before publishing.

3. Readability
   Generally solid. The writing is clear and appropriately technical without being dry. The quarterly format works well — the thematic intro followed by detailed sections is a good structure.

One paragraph that loses its thread:

"The release also brings cross-language actions (Java actions inside Python agents and vice versa) and a durable reconciler for recovering in-flight external side effects after failure recovery, Observability also improved, agent event logs now appear in the Web UI by default."

This runs two separate thoughts together with a comma where there should be a sentence break. The second sentence also starts with "Observability" but the sentence before it already mentions observability indirectly. Tighten it.

4. Logic and Arguments
   Weak: The Agents 0.3.0 section says:

"The pattern across this release is a project adopting the AI ecosystem's emerging standards rather than rolling its own."

This framing is stated as a conclusion but only loosely supported. Agent Skills and Mem0 support the claim. The YAML config API and the durable reconciler don't obviously fit "adopting emerging standards". The claim is slightly too broad for what the evidence supports — either narrow it or give a second example that clearly fits.

Hand-wavy: Under "Before you upgrade":

"If you use mini-batch aggregation with ONE_PHASE aggregation, it would be worth reviewing the 2.3.0 upgrade."

This is vague to the point of being useless. What specifically should they look for? What changed? Either add the detail or link directly to the relevant upgrade notes section. As written, a reader affected by this has learned nothing actionable.

Good discipline: The ON CONFLICT clause section correctly notes that a query that compiled on 2.2 will now fail at planning time on 2.3 and frames it as a feature, not a regression. That's honest and useful framing.

5. Focus and Structure
   This post has a clearly defined, single purpose: quarterly community roundup. It is not trying to be a tutorial, opinion piece, or think piece simultaneously. The structure is declared upfront and followed. No scope creep issues.

The "Also in 2.3.0" subsection is slightly weaker than the rest — it feels like a clearinghouse for things that didn't quite make the cut for full treatment. That's fine for a community update, but the FLIP-547 entry ("provides relief for large jobs recovering with unaligned checkpoints") is thin. Even one sentence on why this matters would help it feel less like a bullet dump.

6. Images
   No images in this post. Nothing to check.

7. AsciiDoc Markup
   This post is Markdown (frontmatter uses YAML, links are text style). The AsciiDoc checks don't apply.

8. Image Format and Size
   No images. Not applicable.

9. Category Taxonomy
   No categories are defined in the frontmatter provided. If this blog system uses categories, they are missing entirely from the header shown — worth checking whether they need to be added before publication.

Summary
Blockers before publishing:

Verify the StateFun vote has not already concluded — update the text accordingly.
Fix the FLIP-584 link (Jira vs Confluence mismatch).
Fix the comma splice / double-space in the Kafka FLIP-572 sentence.
Should fix:
4. The trailing comma sentence in the Agents section.
5. "Congratulations to all" needs a full stop.
6. The ONE_PHASE mini-batch upgrade note is too vague — add substance or a direct link to the specific upgrade note.
7. The "more info found in FLIPs being discussed" phrasing is awkward.

Consider:
8. The "emerging standards" framing in the Agents intro is slightly overreaching given the evidence in the section — either tighten the claim or add a clearer second example.
9. Add a sentence of context to the FLIP-547 entry.

[davidradl]

@sofie204 I have sent an email to the dev list and contacted the author. As you rightly highlight - FLIP-587 does not look to have enough binding votes to be accepted.

[snuyanzin]

I wonder how you picked the topics, tend to think based on ML only

At the same time there are major changes which are completely missed

Calcite upgrade from 1.37.0 to 1.41.0 which unblocks implementation of variety of features like lambda support, recursive WITH, brings a number of improvements in correlated queries and etc.
https://issues.apache.org/jira/browse/FLINK-35856
https://issues.apache.org/jira/browse/FLINK-36602
https://issues.apache.org/jira/browse/FLINK-39817
https://issues.apache.org/jira/browse/FLINK-39828
https://issues.apache.org/jira/browse/FLINK-39859

Introduction of common subexpression elimination
https://issues.apache.org/jira/browse/FLINK-39268
and reuse result of JSON parse in case of multiple JSON functions in SQL https://issues.apache.org/jira/browse/FLINK-39638 which should bring perf boost for such cases.

Lots of improvements towards jdk25
https://issues.apache.org/jira/browse/FLINK-37719

I'm pretty sure there are more,
based on current state I don't feel confident that all major features are covered here

[sofie204]

@davidradl Thanks for helping resolve this issue David

@snuyanzin:

On how topics were picked, the selection of topics, FLIPs, tickets, etc was a manual process that came from a sweep of the dev@ archive (all the [VOTE]/[RESULT]/[ANNOUNCEMENT] threads from april–june), the 2.3.0 release announcement (because of the impact to users), and a cross-reference against the most recent community update which would be April, I included all these items into a sort of inventory spreadsheet with the links to the post, FLIPs and relevant Jira pages. The idea was to rank them based on the impact of the change and the relevance as well (I read through the flips in order to determine the relevance).

the reason why the items you listed got missed was that my focus was skewed towards the vote threads, announcements and major FLIPs and as a result, the work that lives only at the Jira level like the ones you linked above the calcite upgrade, json parse reuse, did not enter into the list. Especially when they are not in the 2.3.0 release version.

The community announcements scope is "the changes we think matter most across April - June" so my aim was not full coverage but rather things that are most impactful and the way i determined that is what I stated above.

I will take a look at what you linked above. Based on my process, it's clear that the individual Jira tickets with 2.4.0 labels are a part that I missed. That said, with three months of activity across a project this size, some judgment calls are going to be made as this update can't cover everything. What I will commit to is applying the same criteria to everything that gets flagged like you have just done. So where some items may not meet the bar for user impact then the release notes and Jira include the complete record.