v26.43.0 release notes

[bosconi]

Starts the v26.43.0 release-note train. Dates in RC snapshots are provisional until the release ships. One commit is added per snapshot; the final snapshot sets the real dates.

This commit (rc.1) adds the ## v26.43.0 section — 3 Improvements, 4 Bug Fixes — and the v26.43 row in the self-managed operator compatibility table. No Features section: nothing in this snapshot's 109 in-scope PRs rises to one. No Agent Skills section either — the agent-skills window contains 0 PRs.

⚠️ Reviewer check on the dates. Two things to weigh:

• The provisional dates collide with the v26.42.0 train. Cloud 2026-09-24 / Self-Managed 2026-09-25 is the first Thursday strictly after the v26.43.0-rc.1 tag (2026-09-18T02:20:07Z), plus one day. But v26.42.0 has not shipped — no v26.42.0 tag, release#579 still open on rc.4 — and v26.42.0 release notes (rc.4) #38925 carries the same provisional pair. Two consecutive trains cannot ship on one day; either v26.42.0 ships this week and v26.43.0 slips a week, or v26.42.0 has slipped and this train follows it.
• Expect the Thursday estimate to be one day high. Every train since v26.37.0 has shipped on the Wednesday before the predicted Thursday, which would put this release at 2026-09-23 Cloud / 2026-09-24 Self-Managed. The cadence rule is followed as written rather than overridden here.

Three of the seven entries refine hydration history, which shipped in v26.42.0 (#38654, published in that train's rc.3). None of them should be read as introducing the capability — they add per-process rows and peaks, and fix object_count.

Docs follow-up, not a release-note matter. doc/user/content/sql/system-catalog/mz_internal.md still documents object_count as "Includes the replica's system introspection dataflows, so it exceeds the number of indexes and materialized views you created." #38729 makes that sentence stale; it is deliberately not repeated in these notes, but the page itself wants a separate fix.

Five PRs in this snapshot's window are excluded as already published by earlier trains (#38436, #38809, #38654, #38863 in v26.42.0; #38664 in v26.41.1). They merged to main in this window but were cherry-picked onto an earlier release branch and published there — restating them here would put the same entry in two consecutive releases. Accounted for in omitted.md under "Already published in an earlier release".

Snapshot drafts (with PR links) in mz-skills:

• rc.1: release_notes.md / omitted.md

Borderline PRs omitted: rc.1: 8 — review borderline calls. The closest call is #38718 (columnar accumulators for sum/count/any/all): a measured win — one arrangement 1.21 GB → 436 MB, replica RSS during hydration down 21% — held only because enable_columnar_accumulable_diff reads false at this tag. If the release owner confirms the flag is on for customers in v26.43.0, publish it as an Improvement. Four further borderline items are gated the same way (enable_metric_sink, enable_unified_cluster).