Antalya 26.8: add UniqApacheHLL

[zvonand]

Changelog category (leave one):

• New Feature

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

Aggregate function, which states are compatible with apache data sketches HLL implementation (#2398 by @UnamedRus).

CI/CD Options

Exclude tests:

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

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

Cherry-picked from #2398.

[github-actions]

Workflow [PR], commit [22cdd1a]

[zvonand]

@blau-ai

[blau-ai]

CI triage — all failures are pre-existing / infra / flaky (0 PR-caused)

Verdict: None of the red checks are caused by this PR. This PR only adds the uniqApacheHLL aggregate function (+ the mergeImpl combinator plumbing and its own stateless tests); it touches nothing in S3 export, the swarm feature, the parser-memory baseline, or container packaging. Its own tests pass. The failing checks are all environmental and also fail on unrelated PRs on the same antalya-26.8 base.

Key cross-check: sibling PR #2515 (an unrelated change on the same base) shows the identical set — Parser memory check, Regression swarms (aarch64 + release), and Grype scans all failing there too.

---

1. Regression aarch64/release swarms — ❌ not PR-related (pre-existing/flaky)

Same 3 scenarios fail on both architectures:

• /swarms/feature/cluster discovery
• /swarms/feature/cluster discovery … with wrong cluster name
• /swarms/feature/node failure/cpu overload (+ … all swarm nodes)

All AssertionError in dynamic cluster-membership / CPU-overload timing tests. Unrelated to aggregate functions, and the same swarm failures are present on #2515. Next step: safe to ignore for this PR; these are branch-level flaky swarm tests.

2. Parser memory check — ❌ not PR-related (infra)

ERROR  Resolve master binary — No master `clickhouse-examples` artifact was found

The check can't find the master/baseline clickhouse-examples binary to diff memory against — an artifact-availability problem, not a code issue. Also failing on #2515. Next step: infra; nothing to fix here.

3. Grype Scan (keeper / server / server-alpine) — ❌ not PR-related (infra)

Container-image CVE scans. These depend on the Grype vulnerability DB at scan time, not on source changes — note the non-alpine server image reports "0 high/critical" yet still gates, and on #2515 a different subset of the three images is red. Next step: infra/security-DB driven; not actionable in this PR.

4. Stateless tests (amd_debug, distributed plan, s3 storage, parallel) — ❌ not PR-related (flaky, randomized settings)

Single failure: 03572_export_merge_tree_part_limits_and_table_functions — not a test added or modified by this PR (it's a pre-existing S3 hive-partitioned Parquet export test). CI's own auto-diagnosis pinned it to randomization:

Step 2: Re-running without randomized settings … Passes without randomization.
Confirmed: the failure is caused by randomized settings.
Step 3: Culprit MergeTree setting: --index_granularity 52492

Symptom is a file-count off-by-one (should be 5 … got 4) in partitioned export under that granularity. Independent of aggregate-function merge logic. Next step: safe to re-run; if it persists it's a 03572 flakiness issue to track separately, unowned by this PR.

---

PR health check

The PR itself looks healthy. All required builds are green (amd/arm debug, binary, release). The uniqApacheHLL feature is validated: the new stateless tests (04327_uniq_apache_hll, 05055_uniq_theta_compressed_state, 05136–05139_uniq_apache_hll_*) all pass — Stateless tests (amd_debug, parallel/sequential) and (arm_binary, parallel/sequential) are all OK, and the praktika report lists zero failures among those tests. Stress, AST fuzzer, BuzzHouse, compatibility, and install checks are green too. I'd consider this PR effectively green on its own merits; the red marks are branch-wide noise.

No code change proposed — nothing in the PR needs fixing.

_{🤖 Triage by @blau-ai from CI evidence (praktika result_pr.json, swarms report, and cross-check vs #2515). I can't build/run ClickHouse in-container, so verdicts are evidence-based from CI artifacts.}