Backport of ClickHouse/ClickHouse#104123 — fix DataLake getAllTableNames heavyweight path

[il9ue]

Summary

Antalya users on antalya-25.8 are exposed to the DataLake catalog OOM described in ClickHouse#92494 — SHOW TABLES against a Glue catalog with thousands of tables triggers per-table S3 metadata fetches and can exhaust memory. Upstream addressed this via ClickHouse#94467/ClickHouse#97062; those landed on antalya-26.1 (via Altinity PR #1552) and antalya-26.3 (inherited from upstream history) but were not ported to antalya-25.8.

Additionally, all three branches have a related residual hole that upstream ClickHouse#94467/ClickHouse#97062 did not close: the typo-hint path through getAllTableNames() still triggers the same heavyweight per-table fetch on any query referencing a non-existent table in a DataLake database. I filed ClickHouse#104123 upstream for this; PR ClickHouse#104124 is pending merge.

This issue tracks porting the typo-hint fix (ClickHouse#104123) to all three maintained Antalya branches, and asks for direction on whether to also backport the larger ClickHouse#97062 fix to 25.8.

State across maintained Antalya branches (verified)

Branch	ClickHouse#94467 / ClickHouse#97062 (SHOW TABLES fix)	getAllTableNames overridden?
antalya-25.8 (tip e1a7eb14)	No — getLightweightTablesIterator is still the original heavyweight implementation (50-thread pool + per-table tryGetTableImpl)	No — falls through to IDatabase::getAllTableNames
antalya-26.1 (tip 889ea2b0)	Yes — via Altinity PR #1552 (merged 2026-03-26)	No — same fallthrough
antalya-26.3 (tip 12361002)	Yes — picked up natively from upstream history	No — same fallthrough

No prior issue, PR, or commit message indicates ClickHouse#97062 was deliberately deferred for 25.8. Flagging this before assuming.

Proposed: ports for all three branches

I plan to make the typo-hint fix available on all three Antalya branches — 25.8, 26.1, and 26.3 — once ClickHouse#104124 merges upstream.

Easy ports: antalya-26.1 and antalya-26.3

Mechanically a cherry-pick of the upstream fix. The override is a five-line getAllTableNames() mirroring the existing getLightweightTablesIterator() override that PR #1552 brought into 26.1 (and that 26.3 inherits from upstream). Plus regression tests under tests/integration/test_database_glue/.

I'll open one PR per branch — separately rather than bundled — to match the pattern set by PR #1552.

antalya-25.8 needs more work — please advise on scope

The typo-hint fix itself is straightforward to port to 25.8. The override returns Strings, and getCatalog()->getTables() exists on 25.8 already (it's used inside the heavyweight getLightweightTablesIterator). So the mechanical fix is a five-line override, no API dependencies.

However, porting only my fix to 25.8 leaves the much larger SHOW TABLES OOM unaddressed. The original ClickHouse#92494 failure mode (53 GB RSS at 10k+ tables) remains fully reproducible on 25.8 because ClickHouse#97062 was never backported there.

Backporting ClickHouse#97062 to 25.8 is not a mechanical cherry-pick. It involves an API reshape:

• The getLightweightTablesIterator virtual on IDatabase changes return type from DatabaseTablesIteratorPtr to
  std::vector<LightWeightTableDetails>.
• A new LightWeightTableDetails struct is introduced.
• Caller sites in StorageSystemTables and StorageSystemCompletions are updated to consume the new return type.
• PR Antalya 26.1 Backport of #97062 - Improve catalog show tables query #1552's diff on 26.1 touched ~10 files for this reshape.

Three paths for 25.8 — I'd like guidance:

1. Port my typo-hint fix only. Closes the typo-hint hole. Honest disclosure in the PR description that the SHOW TABLES heavyweight path remains until Revert "Revert "Improve catalog show tables query"" ClickHouse/ClickHouse#97062 is backported separately. Smallest change, ships fast, leaves the bigger bug alive.

2. Backport Revert "Revert "Improve catalog show tables query"" ClickHouse/ClickHouse#97062 first, then my fix on top. Closes both holes. Substantially larger change to 25.8 (10+ files, virtual signature changes). Requires careful review against any in-flight 25.8 work I can't see.

3. Skip 25.8 entirely. If 25.8 is end-of-life soon, or if there's in-flight work that would conflict with Revert "Revert "Improve catalog show tables query"" ClickHouse/ClickHouse#97062's API reshape, this is the right call. I'd hold off and ports go to 26.1 / 26.3 only.

I'm comfortable doing any of the three; what's the preferred shape for 25.8 given your maintenance plans?

Timeline

I'll wait for ClickHouse#104124 to merge upstream before opening Altinity PRs, so the ports cherry-pick from finalized code. The 26.1 and 26.3 ports can move as soon as you confirm direction; the 25.8 port shape depends on your answer above.