Antalya 25.8 Backport of #90059: Unblock ttl part drops for cold volumes

[mkmkme]

25.8.16 Stable Backport of ClickHouse#90059: Unblock ttl part drops for cold volumes

Same as #1363, but for Antalya 25.8.

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

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

Split part ranges by volume characteristics to enable TTL drop merges for cold volumes. After this patch, parts with a max TTL < now will be removed from cold storage. The algorithm will schedule only single part drops. (ClickHouse#90059 by @Michicosun)

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)
• 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 [dba57d4]

[mkmkme]

@blau-ai

[blau-ai]

CI triage

Verdict: 7 red checks, but only 3 distinct failures — all in the Iceberg / Swarms regression suites. 0 are caused by this PR; all are pre-existing flaky / test-framework issues in the external clickhouse-regression suite. The FinishCIReport red is just the aggregate gate reflecting those three.

Everything that actually exercises this change is green: all builds (amd/arm), Fast test, every Stateless combo, Stateful, Stress, AST fuzzer, BuzzHouse, and — most importantly — Integration tests (fail: 0 on all 5 shards), which include this PR's own new test test_no_merges_volume_ttl.

Why the PR is not responsible

This is a MergeTree TTL / cold-volume compaction backport. The diff only touches:

• src/Storages/MergeTree/Compaction/** (PartProperties, TTLMergeSelector, merge predicates, parts collectors)
• src/Storages/MergeTree/MergeTreeDataMergerMutator.cpp
• tests/integration/test_no_merges_volume_ttl/** (new)

None of that is on any Iceberg, object-storage, REST/Glue-catalog, swarm, or JOIN code path. There is no plausible mechanism by which it affects these suites.

Per-failure breakdown

1. Regression x86 iceberg_1 / RegressionTestsRelease / Iceberg (1) — pre-existing / test-framework

✘ [ Fail ]  /iceberg/iceberg engine/rest catalog/position delete reads/read position deletes  — AssertionError
✘ [ Error ] /iceberg/iceberg engine/rest catalog/sort key timezone  — AttributeError: feature_supported

The sort key timezone case is an AttributeError: feature_supported — a Python bug inside the regression framework itself, not a ClickHouse defect. position delete reads is a data-assertion against a live REST catalog + MinIO.

2. Regression x86 iceberg_2 / RegressionTestsRelease / Iceberg (2) — pre-existing / flaky

✘ [ Fail ] /iceberg/iceberg engine/glue catalog/position delete reads/read position deletes  — AssertionError

Same "position delete reads" assertion, this time against the Glue catalog path.

3. Regression x86 swarms / RegressionTestsRelease / Swarms — pre-existing / flaky

✘ [ Fail ] /swarms/feature/node failure/check restart swarm node  — AssertionError (r.exitcode == exitcode)
✘ [ Fail ] /swarms/feature/swarm joins/join clause/join 775 of 816480: iceberg(...) with s3(...) LEFT ANY JOIN  — AssertionError

A node-restart resilience assertion and an iceberg×s3 JOIN case — both depend on external MinIO/catalog services and node-restart timing, the classic flaky surface of these suites.

4. FinishCIReport — aggregate gate only; red solely because of the three suites above. No independent failure.

Suggested next step

No code change is needed on this PR. Safe to re-run the Iceberg (1/2) and Swarms regression jobs — these Altinity regression suites hit real MinIO/Glue/REST-catalog services and node-restart timing and are known to be intermittent on the antalya branches; the AttributeError: feature_supported in particular is a framework bug that lives in Altinity/clickhouse-regression, not in this binary. If any of them reproduces deterministically across reruns, it should be tracked against the regression suite rather than blocking this backport.