[zvonand]

Changelog category (leave one):

• Critical Bug Fix (crash, data loss, RBAC)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fix a possible crash when querying the view over Distributed with enable_analyzer=0 (ClickHouse#123431 by @mkmkme)

Documentation entry for user-facing changes

...

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Performance tests
• Aarch64 tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• CAS (content-addressed storage; Antalya only)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• OAuth (5m)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[zvonand]

Backporter: cherry-pick conflicts resolved by AI

Conflicted files:

• src/Interpreters/InterpreterSelectQuery.cpp (both modified)

Commits:

• 5c77c46: cherry-pick with the conflicts committed as-is

• c2e47fe: resolution

• src/Interpreters/InterpreterSelectQuery.cpp: the conflict was in the namespace Setting declarations. The source PR's new line sat next to allow_calculating_subcolumns_sizes_for_merge_tree_reading, which this file doesn't declare in antalya-26.3.

• I kept only the extern const SettingsBool allow_experimental_analyzer; declaration, which the fix needs.

• The initSettings() change applied cleanly: after the stored SELECT settings are applied, the analyzer is switched back off. The 26.3 IInterpreterUnionOrSelectQuery uses the same pattern.

• The new stateless test 05315_view_analyzer_setting_over_distributed came in unchanged.

[github-actions]

Workflow [PR], commit [c2e47fe]

[mkmkme]

@blau-ai

[blau-ai]

CI triage

Verdict: 1 failing check — not caused by this PR (infra/timeout). Everything else is green.

The only red check is SQLLogic test; it hit the job's 3-hour wall-clock cap and was killed. All 128 other non-skipped checks passed (Fast test, every Build, all Stateless/Integration/Unit/Stress, AST fuzzer, BuzzHouse). The PR line is just the aggregate gate reflecting that one failure.

SQLLogic test — timeout (not PR-related)

Decisive log lines (job.log):

[2026-10-24 08:29:48] Run statements-test
[2026-10-24 10:13:36] Run complete-test          # statements-test took ~1h44m
[2026-10-24 11:27:52] WARNING: Timeout exceeded [10800] for [4519]
[2026-10-24 11:28:22] WARNING: Job timed out: [SQLLogic test], timeout [10800], exit code [137]

The job ran out of its 10800s (3h) budget mid complete-test and was SIGKILLed (exit 137). No individual SQL logic test was reported as failed — the suite simply didn't finish in time.

Why this is not the PR's fault:

• On recent base-branch (antalya-26.3) PRs the same job passes but already runs right up against the cap: Antalya 26.3 Backport of #123436 - 26.8: Fix ALIAS row policies on legacy distributed reads #2654 → 2h27m, Antalya 26.3 Backport of #107339 - Reuse AES key schedules and compose plain ECB/CBC on a streaming ECB context in encrypt/decrypt #2629 → 2h24m, Antalya 26.3 Backport of #109051 - Respect use_client_time_zone for string datetime literals parsed on the server #2628 → 2h25m, Antalya 26.3 Backport of #107059 - Fix NOT_FOUND_COLUMN_IN_BLOCK in the native Parquet V3 reader prewhere handling #2624 → 2h23m, all against the same 3h limit. There's only ~30 min of head-room, so a slower shared self-hosted runner tips it over — a classic timeout flake.
• This PR's only code change is a narrow old-interpreter path in InterpreterSelectQuery::initSettings(): it forces allow_experimental_analyzer=0 only when a query carries inline SETTINGS enable_analyzer=1 and is run through the old interpreter. SQLLogic's generated plain-SQL statements don't use ClickHouse SETTINGS clauses, so this code essentially never fires there and can't account for a ~30-min slowdown.

Next step: re-run the SQLLogic job (safe to re-run). If it keeps crossing the cap, that's a branch-wide budget problem (suite sits at ~2h25m vs a 3h limit) worth raising separately with CI — not something to fix in this backport.

Sanity checks on the PR's own change

• The new test 0_stateless/05315_view_analyzer_setting_over_distributed passed everywhere (it runs in the distributed plan stateless jobs, all green).
• The one Failed: 1 inside the otherwise-green Stateless tests (amd_asan, db disk, distributed plan, sequential) job was 04306_delta_lake_merge_missing_table_state ("result differs with reference") — an unrelated flaky Delta Lake test; the job passed within tolerance.

Bottom line: no PR-related failures. The backport looks healthy — the C++ fix and its regression test pass across ASAN/UBSAN/debug stateless, integration, and unit suites; only the perennially-near-the-limit SQLLogic job timed out and just needs a re-run.