Antalya 26.6: add UniqApacheHLL

[UnamedRus]

Changelog category (leave one):

• New Feature

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

Aggregate function, which states are compatible with apache data sketches HLL implementation

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Unit 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)

[github-actions]

Workflow [PR], commit [a6477a1]

[zvonand]

@blau-ai

[zvonand]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-30T08:50:26.470323Z	9c9c7fb	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[blau-ai]

CI triage — @zvonand

Verdict: Only one class of failure is caused by this PR, and it's the expected one — the regression suite noticing the new function isn't covered yet. Everything else is pre-existing / flaky / infra. Nothing points to a bug in the uniqApacheHLL implementation: the PR's own tests (04327, 05026, 05055, 05136) all passed, and every functional/build/unit/integration job for the core code is green.

Head SHA analyzed: 9c9c751 (9c9c7fbb49f510a5ecd72c41c14ab5952d1d4d58).

---

🔴 PR-caused (1 root cause — action needed, but not in this repo)

Regression … aggregate_functions_1 — fails on every variant (release + aarch64, plain / cas / cas_s3_cache). All share one message:

Aggregate function 'uniqApacheHLL' was added, but not tested.
/aggregate functions/part 1/function_list/untested function uniqApacheHLL → Fail

This is the Altinity clickhouse-regression function_list guard: it asserts that every aggregate function reported by system.functions is present in the suite's known/tested list. Adding uniqApacheHLL (registered at AggregateFunctionUniqApacheHLL.cpp:128) trips it by design.

Fix: add uniqApacheHLL to the aggregate-function list (and, ideally, a small coverage test) in the clickhouse-regression repo — not in this PR. This is a separate repository I don't have write access to, so a human needs to open that PR. Once the function is listed there, all aggregate_functions_1 variants go green. No change is required to the ClickHouse code in this PR.

---

🟡 Not PR-caused — pre-existing / flaky / infra (safe to re-run or ignore)

Stateless tests — all failures are unrelated to aggregate functions and are confirmed transient by CI's own auto-rerun:

Suite	Test	CI rerun verdict
amd_msan WasmEdge 2/4	03572_export_merge_tree_part_limits_and_table_functions	All reruns passed — not reproducible
amd_debug distributed s3	03572_export_merge_tree_part_limits_and_table_functions	All reruns passed — not reproducible
amd_tsan cas s3 1/2	03913_…statistics, 03164_materialize_skip_index_on_merge, 03100_lwu_23_apply_patches, 04049_tuple_inside_nullable_…, 02346_text_index_partially_materialized	All reruns passed — not reproducible
amd_tsan cas s3 1/2	04661_refreshable_mv_cancel_during_planning	flaky (1/5 reruns failed)
arm_asan_ubsan azure	02354_vector_search_rescoring_distance_in_select_list	fails 3/3 — small BFloat16 distance rounding diff (0.203125 vs 0.1875); vector-search FP precision on ARM, no overlap with this PR

Stress test (amd_asan_ubsan, cas s3 storage) — Cannot start clickhouse-server / Check failed. Server-startup/infra failure in the cas-s3 stress variant; all other 15 stress variants (incl. amd_asan_ubsan plain) are green → infra/flaky, not code.

Integration tests (arm_binary, distributed plan, 3/4) — ERROR: Failed to pre-pull Docker images needed by the test batch. Pure infra; the other 3/4 shards passed. Safe to re-run.

Grype Scan (keeper + server-alpine) — CVE-2026-85091 (High) flagged in the base image. Decisive tell: the keeper image contains none of this PR's aggregate-function code yet fails identically → it's a base-image CVE, pre-existing across the branch, not introduced here.

Other regression suites — cas_alter_attach_3, cas_lightweight_delete_4, cas_s3_cache_lightweight_delete_4, cas_selects, s3_azure_1, settings, swarms. These cover SELECT FINAL concurrency, lightweight delete, ALTER ATTACH, S3/Azure, an iceberg setting default (iceberg_manifest_min_count_to_compact AssertionError), swarms, etc. — feature areas with zero code overlap with a single new aggregate function. Spot-checked reports show generic AssertionErrors / timeouts / errored features typical of the branch's regression baseline. Pre-existing noise on antalya-26.6, not caused by this PR.

---

TL;DR

• PR-related: aggregate_functions_1 regression only — expected; fix by registering uniqApacheHLL in the clickhouse-regression suite (separate repo, needs a human).
• Everything else (other regression suites, stateless, stress, integration pre-pull, Grype): pre-existing / flaky / infra — re-run the flaky ones; the Grype CVE and the regression-baseline failures are branch-wide, not from this PR.
• The uniqApacheHLL code itself looks healthy: its dedicated tests and all core build/functional/unit jobs are green.

_{🤖 Automated CI triage by @blau-ai. Evidence: praktika result_pr.json + testflows reports for 9c9c751. I can't build/run ClickHouse locally, so these are read from CI artifacts.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9c9c7fbb49

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[DimensionWieldr]

AI audit note: This review comment was generated by AI.

Audit update for PR #2398 (uniqApacheHLL, Apache DataSketches HLL aggregate):

Confirmed defects:

Medium: Fast test and any ENABLE_LIBRARIES=0 build fail to link

• Impact: clickhouse does not link when DataSketches is disabled. Fast test configures with -DENABLE_LIBRARIES=0, which turns off ENABLE_DATASKETCHES / USE_DATASKETCHES.
• Anchor: src/AggregateFunctions/registerAggregateFunctions.cpp / registerAggregateFunctions; src/AggregateFunctions/AggregateFunctionUniqApacheHLL.cpp / registerAggregateFunctionUniqApacheHLL
• Trigger: Configure with -DENABLE_LIBRARIES=0 and link the server.
• Why defect: The call is unconditional, but the definition is compiled only under #if USE_DATASKETCHES. uniqTheta avoids this by registering inside registerAggregateFunctionUniq, which always exists. There is no #else stub here.
• Fix direction (short): Provide an empty registerAggregateFunctionUniqApacheHLL when USE_DATASKETCHES is off.
• Regression test direction (short): Build with -DENABLE_LIBRARIES=0 (the Fast test CMake line in ci/jobs/fast_test.py).

Medium: Empty String values are counted only when the column's character buffer is already allocated

• Impact: Distinct count for '' changes with block layout. A block of only empty strings contributes nothing. A block that also contains a non-empty string counts '' as a distinct value. Apache DataSketches update(std::string) never records an empty string, so exchanged states disagree.
• Anchor: AggregateFunctionUniqApacheHLL::add → HllSketchData::insertData → hll_sketch::update(const void *, size_t)
• Trigger: Compare uniqApacheHLL on a String column of only '' with one that also contains a non-empty value (small enough to stay in coupon mode, so the count is exact).
• Why defect: insertData forwards a zero length. update(const void *) returns only when the pointer is null, and still hashes a non-null zero-length pointer into a coupon (value is at least 1, so the coupon is not empty). ColumnString::getDataAt yields a null pointer only while chars has never been allocated (c_start == nullptr). Any non-empty value in that column allocates the buffer, so the empty row's pointer becomes non-null.
• Fix direction (short): Skip size == 0 in insertData so an empty string is never recorded, matching hll_sketch::update(const std::string &).
• Regression test direction (short): SELECT uniqApacheHLL(s) for [''] and for ['', 'a'] must return the same treatment of '', including when those rows arrive in separate blocks.

Coverage summary:

• Scope reviewed: factory registration and parameter checks, type dispatch, HllSketchData insert/merge/serialize/deserialize, empty-state and lg_k < 7 fixes, DataSketches 5.2.0 HLL union backport (23bd9b07), and the four stateless tests.
• Categories failed: optional-library link, empty-string hashing.
• Categories passed: parameter bounds, rejected types, corrupt-state translation to CORRUPTED_DATA, empty-sketch serialization, merge resolution in [4, 21], HLL union KxQ rebuild (submodule patch), per-state ownership under normal aggregation (no shared mutable sketch).
• Assumptions/limits: static review only; integer hashing was checked against the documented 8-byte zero-extension contract, not against update(uint32_t) sign-extension; std::llround above 2^63-1 and big-endian UUID byte order were not treated as confirmed defects.

[CarlosFelipeOR]

I added uniqApacheHLL coverage to the clickhouse-regression aggregate_functions suite (57cff3f, 25030e1). This also clears the function_list failure in the regression job. One check fails on this PR's build: uniqApacheHLL throws on an all-NULL argument instead of returning 0.

SELECT uniq(NULL), uniqTheta(NULL);  -- 0  0
SELECT uniqApacheHLL(NULL);          -- Code: 43. Illegal type Nullable(Nothing) of argument for aggregate function uniqApacheHLL

The failing checks are /aggregate functions/part 1/uniqApacheHLL/NULL for all rows and /aggregate functions/part 3/state/uniqApacheHLLState/NULL for all rows.

AI analysis: for an all-NULL argument, the factory still calls the creator with Nullable(Nothing) before the Null combinator replaces it with nothing (AggregateFunctionNull.cpp:113). uniq and uniqTheta accept any type at that step. createAggregateFunctionUniqApacheHLL throws instead. Returning an instance for argument_type.onlyNull() before the final throw should fix it.