Cas separate phases during GC round

[k-morozov]

Changelog category (leave one):

• Not for changelog (changelog entry is not required)

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

GC round: split the phases into separate functions.

Documentation entry for user-facing changes

...

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 [683c7ca]

[ilejn]

LGTM

[ilejn]

@k-morozov , do we have state diagram that illustrates GC states?

[DimensionWieldr]

CI Failures Analysis

HEAD 683c7cac, run 34585156117. The seven failing jobs were rerun on 21 Sep (attempt 2). None of the failures are caused by this PR. The diff is CasGc.cpp / CasGc.h only, and every failure that stayed red is a bug already filed on other CAS work, or an asan flake that changed tests between attempts.

Related to this PR

None.

Cleared on rerun

• swarms — passed in 26m. Attempt 1 was one race in /swarms/feature/node failure/check restart swarm node (query returned 0, test required exit 138).
• cas_lightweight_delete_4 — passed in 2h 38m. Attempt 1 left count() = 0 where the test expected 2 after disjoint concurrent deletes. That miss did not repeat.
• cas_alter_attach_1 — passed, 99.9% OK. Attempt 1 was a Code 210 mount lease not held storm on CREATE TABLE.

Pre-existing flaky (still red)

• cas_selects — failed again, 21m, 99% OK. Same Code 210 mount lease not held on /selects/final/force/concurrent (895 hits). This is #2332, which already records this suite failing about 15 of 47 runs.
• cas_alter_attach_3 — failed again, 2h 4m, 59.9% OK. Writes die with Code 210 CAS write could not be committed (persistent CAS contention on _ckpt, ref catalog gave up at the policy deadline). Code 341 is that error inside a mutation. This is #2343, filed from PR CAS improvements #2300, where every CAS attach part-3 shard failed, and the same load already failed on the CAS baseline. The low OK rate is that one catalog timeout spreading through the part-3 pool.

Infrastructure (still red)

• Stateless tests amd_asan_ubsan, cas s3 storage, parallel 1/2 and 2/2. Attempt 2 failed a different set of tests from attempt 1 (no shared names). 03229_json_structure_comparison and 02474_unhex_in_fix_string failed once under randomized settings and then passed every diagnosis rerun. 02435_rollback_cancelled_queries (556s) and 02421_truncate_isolation_no_merges (359s) are timeouts. 02122_join_group_by_timeout missed its reference. The release CAS stateless jobs on this SHA are green.

Issue references

• #2332 — cas_selects mount lease
• #2343 — cas_alter_attach_3 ref catalog / _ckpt deadline
• #2243 — earlier ephemeral-port form of the same lease error