Files
lancedb/python/src/index.rs
Shengan Zhang 650f173236 feat(python): add IVF_HNSW_FLAT vector index support (#3366)
## Summary

Wire up `IVF_HNSW_FLAT` in the Rust core and Python SDK. The index was
documented at https://docs.lancedb.com/indexing/vector-index but
`lancedb.Table.create_index(index_type="IVF_HNSW_FLAT")` raised
`ValueError: Unknown index type IVF_HNSW_FLAT` — the underlying
`pylance` already accepted it, only the LanceDB wrapper was missing the
wiring.

**Rust core (`rust/lancedb`):**
- Add `Index::IvfHnswFlat` / `IndexType::IvfHnswFlat` variants and the
`IvfHnswFlatIndexBuilder` (modelled on `IvfHnswSqIndexBuilder`).
- Build Lance params via the existing `VectorIndexParams::ivf_hnsw(...)`
helper, keeping symmetry with the other `IVF_HNSW_*` variants.
- Forward the variant in `RemoteTable::create_index` and add two
parametrised tests (default + customised config) for the JSON
serialisation.
- New `NativeTable` integration test
(`test_create_index_ivf_hnsw_flat`).

**Python binding (`python/`):**
- New `HnswFlat` dataclass + backwards-compat `IvfHnswFlat` alias.
- PyO3 `extract_index_params` recognises the `HnswFlat` config.
- `LanceTable.create_index(index_type="IVF_HNSW_FLAT", …)` and the sync
`RemoteTable.create_index` both dispatch to the new config.
- `IndexStatistics.index_type` `Literal` and `_lancedb.pyi` stubs cover
the new type so `pyright`/`make check` stays clean.
- Async integration tests (`HnswFlat` + `IvfHnswFlat` alias) and a sync
dispatcher test, mirroring the existing `IVF_HNSW_SQ` coverage.
- Existing `test_index_statistics_index_type_lists_all_supported_values`
updated to include `IVF_HNSW_FLAT`.

A matching Node.js / TypeScript binding is in a follow-up PR.

Closes #3331

## Test plan

- [ ] \`cargo check --quiet --features remote --tests --examples\`
- [ ] \`cargo test --quiet --features remote -p lancedb\` (covers the
new \`test_create_index_ivf_hnsw_flat\` and the two new parametrised
\`RemoteTable::create_index\` cases)
- [ ] \`cargo fmt --all\` / \`cargo clippy --quiet --features remote
--tests --examples\`
- [ ] \`cd python && make develop && make check && make test\` (covers
the two new async tests, the alias test, the dispatcher test, and the
updated \`test_index_statistics_index_type_lists_all_supported_values\`
assertion)
2026-05-11 15:08:32 -07:00

329 lines
13 KiB
Rust

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright The LanceDB Authors
use lancedb::index::vector::{
IvfFlatIndexBuilder, IvfHnswFlatIndexBuilder, IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder,
IvfPqIndexBuilder, IvfRqIndexBuilder, IvfSqIndexBuilder,
};
use lancedb::index::{
Index as LanceDbIndex,
scalar::{BTreeIndexBuilder, FtsIndexBuilder},
};
use pyo3::IntoPyObject;
use pyo3::types::PyStringMethods;
use pyo3::{
Bound, FromPyObject, PyAny, PyResult, Python,
exceptions::{PyKeyError, PyValueError},
intern, pyclass, pymethods,
types::{PyAnyMethods, PyString},
};
use crate::util::parse_distance_type;
pub fn class_name(ob: &'_ Bound<'_, PyAny>) -> PyResult<String> {
let full_name = ob
.getattr(intern!(ob.py(), "__class__"))?
.getattr(intern!(ob.py(), "__name__"))?;
let full_name = full_name.cast::<PyString>()?.to_string_lossy();
match full_name.rsplit_once('.') {
Some((_, name)) => Ok(name.to_string()),
None => Ok(full_name.to_string()),
}
}
pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<LanceDbIndex> {
if let Some(source) = source {
match class_name(source)?.as_str() {
"BTree" => Ok(LanceDbIndex::BTree(BTreeIndexBuilder::default())),
"Bitmap" => Ok(LanceDbIndex::Bitmap(Default::default())),
"LabelList" => Ok(LanceDbIndex::LabelList(Default::default())),
"FTS" => {
let params = source.extract::<FtsParams>()?;
let inner_opts = FtsIndexBuilder::default()
.base_tokenizer(params.base_tokenizer)
.language(&params.language)
.map_err(|_| {
PyValueError::new_err(format!(
"LanceDB does not support the requested language: '{}'",
params.language
))
})?
.with_position(params.with_position)
.lower_case(params.lower_case)
.max_token_length(params.max_token_length)
.remove_stop_words(params.remove_stop_words)
.stem(params.stem)
.ascii_folding(params.ascii_folding)
.ngram_min_length(params.ngram_min_length)
.ngram_max_length(params.ngram_max_length)
.ngram_prefix_only(params.prefix_only);
Ok(LanceDbIndex::FTS(inner_opts))
}
"IvfFlat" => {
let params = source.extract::<IvfFlatParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut ivf_flat_builder = IvfFlatIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate);
if let Some(num_partitions) = params.num_partitions {
ivf_flat_builder = ivf_flat_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
ivf_flat_builder =
ivf_flat_builder.target_partition_size(target_partition_size);
}
Ok(LanceDbIndex::IvfFlat(ivf_flat_builder))
}
"IvfPq" => {
let params = source.extract::<IvfPqParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut ivf_pq_builder = IvfPqIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate)
.num_bits(params.num_bits);
if let Some(num_partitions) = params.num_partitions {
ivf_pq_builder = ivf_pq_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
ivf_pq_builder = ivf_pq_builder.target_partition_size(target_partition_size);
}
if let Some(num_sub_vectors) = params.num_sub_vectors {
ivf_pq_builder = ivf_pq_builder.num_sub_vectors(num_sub_vectors);
}
Ok(LanceDbIndex::IvfPq(ivf_pq_builder))
}
"IvfSq" => {
let params = source.extract::<IvfSqParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut ivf_sq_builder = IvfSqIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate);
if let Some(num_partitions) = params.num_partitions {
ivf_sq_builder = ivf_sq_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
ivf_sq_builder = ivf_sq_builder.target_partition_size(target_partition_size);
}
Ok(LanceDbIndex::IvfSq(ivf_sq_builder))
}
"IvfRq" => {
let params = source.extract::<IvfRqParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut ivf_rq_builder = IvfRqIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate)
.num_bits(params.num_bits);
if let Some(num_partitions) = params.num_partitions {
ivf_rq_builder = ivf_rq_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
ivf_rq_builder = ivf_rq_builder.target_partition_size(target_partition_size);
}
Ok(LanceDbIndex::IvfRq(ivf_rq_builder))
}
"HnswPq" => {
let params = source.extract::<IvfHnswPqParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut hnsw_pq_builder = IvfHnswPqIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate)
.num_edges(params.m)
.ef_construction(params.ef_construction)
.num_bits(params.num_bits);
if let Some(num_partitions) = params.num_partitions {
hnsw_pq_builder = hnsw_pq_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
hnsw_pq_builder = hnsw_pq_builder.target_partition_size(target_partition_size);
}
if let Some(num_sub_vectors) = params.num_sub_vectors {
hnsw_pq_builder = hnsw_pq_builder.num_sub_vectors(num_sub_vectors);
}
Ok(LanceDbIndex::IvfHnswPq(hnsw_pq_builder))
}
"HnswSq" => {
let params = source.extract::<IvfHnswSqParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut hnsw_sq_builder = IvfHnswSqIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate)
.num_edges(params.m)
.ef_construction(params.ef_construction);
if let Some(num_partitions) = params.num_partitions {
hnsw_sq_builder = hnsw_sq_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
hnsw_sq_builder = hnsw_sq_builder.target_partition_size(target_partition_size);
}
Ok(LanceDbIndex::IvfHnswSq(hnsw_sq_builder))
}
"HnswFlat" => {
let params = source.extract::<IvfHnswFlatParams>()?;
let distance_type = parse_distance_type(params.distance_type)?;
let mut hnsw_flat_builder = IvfHnswFlatIndexBuilder::default()
.distance_type(distance_type)
.max_iterations(params.max_iterations)
.sample_rate(params.sample_rate)
.num_edges(params.m)
.ef_construction(params.ef_construction);
if let Some(num_partitions) = params.num_partitions {
hnsw_flat_builder = hnsw_flat_builder.num_partitions(num_partitions);
}
if let Some(target_partition_size) = params.target_partition_size {
hnsw_flat_builder =
hnsw_flat_builder.target_partition_size(target_partition_size);
}
Ok(LanceDbIndex::IvfHnswFlat(hnsw_flat_builder))
}
not_supported => Err(PyValueError::new_err(format!(
"Invalid index type '{}'. Must be one of BTree, Bitmap, LabelList, FTS, IvfPq, IvfSq, IvfHnswPq, IvfHnswSq, or IvfHnswFlat",
not_supported
))),
}
} else {
Ok(LanceDbIndex::Auto)
}
}
#[derive(FromPyObject)]
struct FtsParams {
with_position: bool,
base_tokenizer: String,
language: String,
max_token_length: Option<usize>,
lower_case: bool,
stem: bool,
remove_stop_words: bool,
ascii_folding: bool,
ngram_min_length: u32,
ngram_max_length: u32,
prefix_only: bool,
}
#[derive(FromPyObject)]
struct IvfFlatParams {
distance_type: String,
num_partitions: Option<u32>,
max_iterations: u32,
sample_rate: u32,
target_partition_size: Option<u32>,
}
#[derive(FromPyObject)]
struct IvfPqParams {
distance_type: String,
num_partitions: Option<u32>,
num_sub_vectors: Option<u32>,
num_bits: u32,
max_iterations: u32,
sample_rate: u32,
target_partition_size: Option<u32>,
}
#[derive(FromPyObject)]
struct IvfSqParams {
distance_type: String,
num_partitions: Option<u32>,
max_iterations: u32,
sample_rate: u32,
target_partition_size: Option<u32>,
}
#[derive(FromPyObject)]
struct IvfRqParams {
distance_type: String,
num_partitions: Option<u32>,
num_bits: u32,
max_iterations: u32,
sample_rate: u32,
target_partition_size: Option<u32>,
}
#[derive(FromPyObject)]
struct IvfHnswPqParams {
distance_type: String,
num_partitions: Option<u32>,
num_sub_vectors: Option<u32>,
num_bits: u32,
max_iterations: u32,
sample_rate: u32,
m: u32,
ef_construction: u32,
target_partition_size: Option<u32>,
}
#[derive(FromPyObject)]
struct IvfHnswSqParams {
distance_type: String,
num_partitions: Option<u32>,
max_iterations: u32,
sample_rate: u32,
m: u32,
ef_construction: u32,
target_partition_size: Option<u32>,
}
#[derive(FromPyObject)]
struct IvfHnswFlatParams {
distance_type: String,
num_partitions: Option<u32>,
max_iterations: u32,
sample_rate: u32,
m: u32,
ef_construction: u32,
target_partition_size: Option<u32>,
}
#[pyclass(get_all)]
/// A description of an index currently configured on a column
pub struct IndexConfig {
/// The type of the index
pub index_type: String,
/// The columns in the index
///
/// Currently this is always a list of size 1. In the future there may
/// be more columns to represent composite indices.
pub columns: Vec<String>,
/// Name of the index.
pub name: String,
}
#[pymethods]
impl IndexConfig {
pub fn __repr__(&self) -> String {
format!(
"Index({}, columns={:?}, name=\"{}\")",
self.index_type, self.columns, self.name
)
}
// For backwards-compatibility with the old sync SDK, we also support getting
// attributes via __getitem__.
pub fn __getitem__<'a>(&self, key: String, py: Python<'a>) -> PyResult<Bound<'a, PyAny>> {
match key.as_str() {
"index_type" => Ok(self.index_type.clone().into_pyobject(py)?.into_any()),
"columns" => Ok(self.columns.clone().into_pyobject(py)?.into_any()),
"name" | "index_name" => Ok(self.name.clone().into_pyobject(py)?.into_any()),
_ => Err(PyKeyError::new_err(format!("Invalid key: {}", key))),
}
}
}
impl From<lancedb::index::IndexConfig> for IndexConfig {
fn from(value: lancedb::index::IndexConfig) -> Self {
let index_type = format!("{:?}", value.index_type);
Self {
index_type,
columns: value.columns,
name: value.name,
}
}
}