[subkanthi]

…d::vector + binary search to improve performance

Changelog category (leave one):

• Improvement

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

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 [84e16b2]

[subkanthi]

@blau-ai

[blau-ai]

CI triage for 84e16b2

Verdict: 7 red checks. None caused by this PR. 3 integration shards fail the same way on other antalya-26.8 PRs, and all 4 regression suites also fail on the base branch's latest MasterCI. Nothing to fix in this PR.

Check	Failures	Classification
Integration tests (amd_asan_ubsan, db disk, old analyzer, 5/8)	22/1532	Pre-existing (21) + infra (1)
Integration tests (amd_asan_ubsan, db disk, old analyzer, 6/8)	17/957	Pre-existing
Integration tests (amd_asan_ubsan, db disk, old analyzer, 7/8)	2/889	Pre-existing
Regression release oauth	1 feature	Pre-existing on base
Regression release swarms	1 module	Pre-existing on base
Regression release settings	1 module	Pre-existing on base
Regression release s3_azure_1	1 module	Pre-existing on base

The PR changes Parquet dictionary-filter pruning (Parquet/Reader.cpp, parquetBloomFilterHash.cpp, KeyCondition). None of the failing tests read Parquet through that path. The suites that would catch a problem in it passed: Fast test (0 failed / 11444), Stateless tests (amd_debug, parallel) (0 failed / 13638), and the asan/tsan "selected tests" jobs (0 failed).

Integration shards 5/8, 6/8, 7/8: pre-existing

PR #2730 (Fix early manifest deletion in CAS GC, also on antalya-26.8, no Parquet changes) has the same failure set on the same shards: 21 / 17 / 2. #2737 fails the same test_export_partition_iceberg tests on its arm_binary shards. These shards passed on base MasterCI run 39839856866 (2026-10-22), so something that landed on antalya-26.8 after that run probably broke them. It isn't this PR.

• 5/8: test_storage_iceberg_with_spark/test_export_partition_iceberg.py::* (all 21 tests). CREATE TABLE ... ENGINE=IcebergS3(...) fails with (ACCESS_DENIED) during setup, before any Parquet read happens. Identical on Fix early manifest deletion in CAS GC #2730 and Antalya 26.8: Antalya-26:6 - iceberg v3: rewriteFiles - Added logic to merge small files into bins #2737.
• 5/8: test_cas_gcs_relink_liveness/test.py::test_both_queues_drain_under_slow_checkpoints. This is the one test not in Fix early manifest deletion in CAS GC #2730's list. It's an infra/environment failure: the fakegcs container never started:

  Timed out while waiting for instance `fakegcs' ... to start.
  bash: line 1: /usr/bin/clickhouse: cannot execute: required file not found
  

  No server code ran. Safe to re-run.
• 6/8: test_database_backup/test.py::* (17 tests). The server aborts on a LOGICAL_ERROR assertion in S3 copy during BACKUP DATABASE ... TO Disk('backup_disk_s3_plain_rustfs', ...). The later tests in the module then get Connection refused:

  Received signal Aborted (6)  (query: BACKUP DATABASE test_database TO Disk('backup_disk_s3_plain_rustfs', 'test_database_backup');)
  12. src/IO/S3/copyS3File.cpp:274:23: UploadHelper::performMultipartUpload
  13. src/IO/S3/copyS3File.cpp:815:31: CopyFileHelper::performMultipartUploadCopy
  

  Same 17 failures on Fix early manifest deletion in CAS GC #2730. This path (S3 multipart copy) has nothing to do with this PR.
• 7/8: test_backward_compatibility/test_custom_s3_disk_remote_url_allow_hosts.py::{test_table,test_database}_on_custom_s3_disk_loads_after_upgrade. The node instance never comes up in cluster.start(). Same 2 failures on Fix early manifest deletion in CAS GC #2730.

Regression release oauth / swarms / settings / s3_azure_1: pre-existing on base

All four failed on the last antalya-26.8 MasterCI run (39839856866: RegressionTestsRelease and RegressionTestsAarch64 both list oauth, swarms, settings, s3_azure_1 as failure). oauth and swarms also fail on #2729 and #2715.

• settings: every /settings/default values/<name> scenario fails with a bare AssertionError, about 900 settings starting from adaptive_aggregator_freeze_threshold. That pattern means the suite's expected-defaults snapshot doesn't match this branch, not a single bad setting. The PR doesn't touch any settings.
• s3_azure_1: /s3/azure/part 1/disk/{remote host filter,syntax} fail with URL "http://azurite:10000" is not allowed in configuration file, see <remote_url_allow_hosts>. (UNACCEPTABLE_URL). This looks like the same remote_url_allow_hosts behaviour as integration 7/8, and unrelated to Parquet.

Not run (coverage note, not failures)

Praktika marked several jobs DROPPED (GitHub shows them as skipped): Stateless amd_debug sequential, distributed plan s3, the cas/cas-s3 storage variants, Stress (amd_debug, amd_asan_ubsan), AST fuzzer / BuzzHouse (amd_debug), SQLLogic, SQLStorm, Install packages, Compatibility check. They produced no results, so I can't say anything about them.

Suggested next steps

• Nothing to change in this PR.
• Optionally re-run integration 5/8 to clear the fakegcs startup flake. The other 21 failures in that shard will stay red until antalya-26.8 is fixed.
• The base-branch breakages (Iceberg export ACCESS_DENIED, copyS3File LOGICAL_ERROR on rustfs backup, remote_url_allow_hosts startup, regression settings snapshot) should be tracked separately against antalya-26.8.

_{Evidence: result_pr.json for this SHA and for #2730 / #2737 / #2740 from the praktika S3 bucket, and GitHub job logs for the regression suites. Analysis only; I made no commits.}