Alexey MilovidovandClaude Opus 4.8 c7cfd22d30 Key hive_files_cache by the table-specific pruning schema and settings ...

The `hive_files_cache` lives on the per-remote-table `HiveTableMetadata`
and is shared by every `StorageHive` table over the same remote Hive
table, but a cached `IHiveFile` bakes in table-specific state: the
partition values, the minmax-index column layout (`index_names_and_types`),
the file format, and the `enable_orc_*` / `enable_parquet_rowgroup_minmax_index`
settings. `Hive` allows different ClickHouse schemas/settings over one
remote table, so two such tables previously shared one cached file (keyed
only by the remote `file_info.path`); the second table then pruned and
interpreted the cached minmax index against the first table's layout and
could silently keep or skip the wrong data.

Key the cache by the remote path plus a `SipHash` signature of that
table-specific state, so mismatched tables get independent cache entries
while an identical table still reuses the cache. Add
`test_hive_files_cache_is_table_local`, which reads the same remote ORC
table through two ClickHouse `Hive` tables that differ in column order and
minmax settings and asserts each returns its own correct rows.

Addresses review: https://github.com/ClickHouse/ClickHouse/pull/108094#discussion_r3507709555

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

2026-07-05 06:11:45 +00:00

..

configs

…

data

test_hive_query: probe metastore readiness before loading data

2026-06-30 16:53:48 +00:00

__init__.py

…

test.py

Key hive_files_cache by the table-specific pruning schema and settings

2026-07-05 06:11:45 +00:00