Releases: chdb-io/chdb
Releases · chdb-io/chdb
Release list
v4.3.0
v4.2.1
chdb 4.2.1
Requires chdb-core>=26.5.0.
New
- clickhouse-connect backend (#593) — use chDB as a drop-in execution backend for clickhouse-connect:
clickhouse_connect.get_client("chdb://memory")(in-memory) orget_client("chdb:///path/to/data.db")(on-disk) returns a client backed by the embedded engine instead of an HTTP server. Ships aschdb.cc_backendplus aclickhouse_connect.backendsentry point; install withpip install "chdb[clickhouse-connect]". End-to-end usage requires a clickhouse-connect release that includes the pluggable-backend mechanism (ClickHouse/clickhouse-connect#811). chdb.agentscontract 0.2.0 (#602) — descriptors as the single source of truth,capabilities(),acall(), hardened descriptor and dispatch edges, external-session probing, allowlist scanner fixes.
Docs
CI
- Pin
pyarrow<25in the Linux arm64 DataStore test environment to work around a pyarrow 25.0.0 parquet read corruption on linux-aarch64; the pin will be removed once a fixed pyarrow release is out (#605).
v4.2.0
chdb 4.2.0
Requires chdb-core>=26.5.0.
New
chdb.agents.ChDBTool(beta/experimental) — the canonical Python agent tool for chDB, and the reference implementation of a cross-language contract (chdb/agents/CONTRACT.md). Query + introspection with engine-enforced read-only (readonly=2, keepsfile()/s3()/url()usable), server-side parameter binding, result caps with an explicittruncatedflag, error-to-model dispatch, optional query timeout,attach_file, an optional file allowlist, and a Python-onlydataframe_queryoverPython(). A shared conformance fixture keeps other-language bindings in lock-step. The API is beta and may change in a minor release — pin a version if you depend on it.- DataStore → ClickHouse writeback —
to_clickhousewrites an in-memory DataFrame to a ClickHouse server (same- and cross-server). - DataStore pandas 3.x —
pd.col()expressions; lazy metadata (.columns/.dtypes/.shape/.indexresolved without materializing). - DataStore ergonomics —
output_format="DataStore"dispatch and mirrored UDF surfaces from chdb-core.
Notes
All additions are backward-compatible. See the merged PRs since v4.1.9 (#561, #568, #570, #584, #591, #592, #594, #597).
v4.1.9
What's Changed
- feat(datastore): mirror pandas iteration on LazyGroupBy / SeriesGroupBy by @wudidapaopao in #582
- Push unbounded ORDER BY to SQL for local from_df source by @wudidapaopao in #586
- test(datastore): require chdb-core>=26.5.0; un-xfail Array-in-Nullable cases it fixes by @ShawnChen-Sirius in #588
Full Changelog: v4.1.8...v4.1.9
v4.1.8
What's Changed
- Rewrite chdb-datastore and chdb-sql skill descriptions for trigger accuracy by @ShawnChen-Sirius in #573
- Fix pre-load _chdb.abi3.so with RTLD_DEEPBIND to avoid host libstdc++ conflicts by @wudidapaopao in #572
- Fix explain() mislabeling pushed-down GroupBy/apply as Pandas by @auxten in #576
- Unify SQL builder behind a single _build_layer dispatcher by @auxten in #577
- Resolve 3 SQL builder bugs found by user-path tests by @auxten in #578
Full Changelog: v4.1.7...v4.1.8
v4.1.7
What's Changed
- Upgrade macos arm64 runner to macos-15 by @wudidapaopao in #548
- Disable default memory limits by @wudidapaopao in #551
- Increase the timeout for macOS x86 CI to 120 minutes by @wudidapaopao in #554
- Add chdb vs pandas peak-memory benchmark by @wudidapaopao in #557
- Skip flaky groupby_agg_sum timing test on macOS x86_64 by @wudidapaopao in #559
- Fixes chdb-ds: UrlTableFunction.to_sql() emits HEADERS outside url() call by @nklmish in #563
- Optimize memory usage of DataStore by @wudidapaopao in #553
- Fix reading Parquet files from S3 by @mneedham in #562
- Migrate the PR-CI lint gate from flake8 to ruff by @ShawnChen-Sirius in #567
New Contributors
- @nklmish made their first contribution in #563
- @mneedham made their first contribution in #562
- @ShawnChen-Sirius made their first contribution in #567
Full Changelog: v4.1.6...v4.1.7