Changelog category (leave one):

• CI Fix or Improvement (changelog entry is not required)

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

...

Description

Fixes flaky test_keeper_disks/test.py::test_snapshots_with_disks. Two distinct races, same area (cross-disk snapshot moves); both are test-only issues, no server behavior is changed.

1) Local snapshot count race (assert len(local_snapshot_files) == 1 -> assert 2 == 1).
When a separate latest_snapshot_storage_disk is configured, creating a snapshot relocates the previous (non-latest) snapshot off the local "latest" disk. In KeeperStateMachine::create_snapshot the Created persistent snapshot {idx} log line is emitted in the under-lock publish step (publishWrittenSnapshot), while the actual move runs afterwards in the deferred, outside-the-lock maintenance step (runSnapshotMaintenance -> moveSnapshotCandidate). The test waited on that log line and checked the local count once, racing the still-pending move (selectSnapshotsToMove also defers a move while the old snapshot is transiently pinned, use_count > 1). The move is eventually consistent, so the fix polls for the count to settle to 1 (up to 60s), mirroring the assert_single_local_log helper already used by test_logs_with_disks.

2) Transient tmp_ marker counted as a snapshot (assert set(local_snapshot_files) == set(previous_snapshot_files)).
This is now the dominant variant and also hits master. moveFileBetweenDisks (KeeperCommon.cpp) writes an empty tmp_<name> marker on the target disk during a move and removes it once the copy completes. The test's list_s3_objects / get_local_files helpers listed everything, so a listing taken while a move was in flight captured the transient tmp_snapshot_<idx> marker into previous_snapshot_files. After the next restart only the real snapshot_<idx> remained, so the set comparison failed with a lone tmp_-prefixed entry on the expected side. The server itself never treats tmp_-prefixed entries as snapshots/logs (it skips them when scanning disks), so the helpers now exclude them too; the disk-cleanup path still wipes them (exclude_tmp=False).

Validation: for (1), injecting an 8s delay before the move made the bare assertion fail 3/3 and the polling version pass 3/3. For (2), injecting a transient tmp_ marker at the listing point made the raw listing fail with the exact CI signature ({snapshot_34} vs {snapshot_34, tmp_snapshot_34}) and the tmp_-excluding listing pass. The full test passed 6/6 (both tests, random order) on a clean build.

CI failure (variant 2, master): https://s3.amazonaws.com/clickhouse-test-reports/json.html?REF=master&sha=b3461b9116ed78b84a6c37fa846134523a46e21f&name_0=MasterCI&name_1=Integration%20tests%20%28arm_binary%2C%20distributed%20plan%2C%202%2F4%29
CI failure (variant 1): https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=108458&sha=c79ebbf375b04bea319369ed0efff2d275aae491&name_0=PR&name_1=Integration%20tests%20%28arm_binary%2C%20distributed%20plan%2C%202%2F4%29

Version info

• Merged into: 26.7.1.175