v26.42.0 release notes (rc.4)

[bosconi]

Completes the v26.42.0 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.

Why this is a new PR. The long-lived PR for this train, #38761, carried rc.1–rc.3 plus the docs team's edits and was merged on 2026-09-17, which deleted its branch. This PR recreated docs/v26.42.0-release-notes off main and resumed adding one commit per remaining snapshot — rc.4, then final. A third PR was deliberately not opened.

The final commit changes dates only. The v26.42.0 tag landed 2026-09-18T17:38:13Z, under twenty hours after the rc.4 tag, and the only thing that reached the release branch in between was the materialize-bot version bump 859daf038, which has no associated PR. The final snapshot therefore has 0 PRs and publishes no entries; the release's content is entirely in the rc.1–rc.3 increments already on this branch, all of which the docs team has since edited. The commit does not regenerate the assembled section.

It replaces rc.4's provisional 2026-09-25 with Self-Managed 2026-09-19 in both _index.md and self_managed_operator_compatibility.yml, superseding the stale one-week roll-forward rc.4 applied.

⚠️ Two things for the reviewer to confirm before this merges.

1. The dates are inferred, not supplied. No dates were given, so these come from the v26.42.0 tag's commit date — Cloud 2026-09-18, Self-Managed 2026-09-19 (tag date + 1). Across the last three trains the final tag date equalled the published Cloud date for v26.39.0 and v26.40.0 and was one day early for v26.41.0, so the tag is a reliable lower bound and usually exact — but please confirm with the release owner.
2. The ## v26.42.0 section has no Cloud date line, and this commit did not restore it. d07211371 deleted *Released to Materialize Cloud: ...* from the section. Every other .0 release in _index.md carries both lines, so the deletion is left in place as a deliberate editorial change rather than reinstated — restoring it is a reviewer decision. If it is restored, 2026-09-18 is the best available Cloud date. Note also that the editor's remaining line read 2026-09-18, which under the house Self-Managed = Cloud + 1 convention would imply a Cloud date one day before the final tag; the likelier reading is that the then-current provisional value was kept, so that value should not be treated as a confirmed Self-Managed date.

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

• rc.1: release_notes.md / omitted.md
• rc.2: release_notes.md / omitted.md
• rc.3: release_notes.md / omitted.md
• rc.4: release_notes.md / omitted.md
• final: release_notes.md / omitted.md

Borderline PRs omitted: rc.1: 7 — review borderline calls, rc.2: 0 — none, rc.3: 1 — review borderline call, rc.4: 2 — review borderline calls, final: 0 — none

---

Notes carried forward

Two editorial cuts worth a second look before the release ships. Flagged, not overturned — the final commit reinstates neither.

• rustls security fix, RUSTSEC-2026-0285 (#38863, drafted in rc.3) — cut in 6f71a2f45. This is the train's only security-advisory item. The fix still ships in v26.42.0, and customers who track RUSTSEC advisories are the ones most likely to look for it here.
• WAIT options on ALTER CLUSTER are no longer in private preview (#38627, drafted in rc.1) — cut in d07211371. This announced an availability change: users previously told WITH (WAIT FOR ...) and WITH (WAIT UNTIL READY ...) were in private preview have no other signal that the flag is no longer needed.

Two further entries were cut in d07211371 and are also left cut: rc.2's Connection limit in balancerd (#38436) and rc.2's Tighter limits on pre-authentication messages (#38809). Because the latter is now cut, rc.4's note that #38907's user-visible half was "already published in rc.2" is stale — nothing to fix, but the cross-reference no longer holds.

Resolved: hydration history was promoted to a Feature. rc.3 published #38654 as the Improvement "Hydration history available by default" while noting that two releases of deferred announcement arguably add up to a Feature heading. The docs team wrote exactly that — ### Size clusters using hydration history {#v26.42-hydration-history}, from the mz_internal docs rather than the snapshot. No action needed; recorded so the resolution is on file next time a deferred announcement lands.

### Guides {#v26.42-guides} is the docs team's. Added in d329f3eeb; the section is house practice but is not generated, and the final commit preserves it untouched.

Carry-forward note from rc.4. #38839 documents a production behavior while excluding it from a test: introspection arrangement compaction depends on the lower bound across replicas, so a non-running replica can hold it back. That is user-observable on a cluster with an unreachable replica and is not fixed in v26.42.0 — CPU-255 tracks it. Correctly absent from these notes (nothing changed), but the same PR removes the coverage that would have caught it, so it will not resurface from CI.

🤖 Generated with Claude Code