lance 0.8.17

A columnar data format that is 100x faster than Parquet for random access.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251

// Copyright 2023 Lance Developers.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

use std::sync::Arc;

use lance_core::{format::Index as IndexMetadata, Error, Result};
use lance_index::scalar::lance_format::LanceIndexStore;
use lance_index::IndexType;
use log::info;
use roaring::RoaringBitmap;
use snafu::{location, Location};
use uuid::Uuid;

use crate::dataset::index::unindexed_fragments;
use crate::dataset::scanner::ColumnOrdering;
use crate::dataset::Dataset;
use crate::index::vector::ivf::IVFIndex;

use super::DatasetIndexInternalExt;

/// Append new data to the index, without re-train.
///
/// Returns the UUID of the new index along with a vector of newly indexed fragment ids
pub async fn append_index(
    dataset: Arc<Dataset>,
    old_index: &IndexMetadata,
) -> Result<Option<(Uuid, Option<RoaringBitmap>)>> {
    let unindexed = unindexed_fragments(old_index, dataset.as_ref()).await?;
    if unindexed.is_empty() {
        return Ok(None);
    };

    let frag_bitmap = old_index.fragment_bitmap.as_ref().map(|bitmap| {
        let mut bitmap = bitmap.clone();
        bitmap.extend(unindexed.iter().map(|frag| frag.id as u32));
        bitmap
    });

    let column = dataset
        .schema()
        .field_by_id(old_index.fields[0])
        .ok_or(Error::Index {
            message: format!(
                "Append index: column {} does not exist",
                old_index.fields[0]
            ),
            location: location!(),
        })?;

    let index = dataset
        .open_generic_index(&column.name, &old_index.uuid.to_string())
        .await?;

    match index.index_type() {
        IndexType::Scalar => {
            let index = dataset
                .open_scalar_index(&column.name, &old_index.uuid.to_string())
                .await?;

            let mut scanner = dataset.scan();
            scanner
                .with_fragments(unindexed)
                .with_row_id()
                .order_by(Some(vec![ColumnOrdering::asc_nulls_first(
                    column.name.clone(),
                )]))?
                .project(&[&column.name])?;
            let new_data_stream = scanner.try_into_stream().await?;

            let new_uuid = Uuid::new_v4();

            let index_dir = dataset.indices_dir().child(new_uuid.to_string());
            let new_store = LanceIndexStore::new((*dataset.object_store).clone(), index_dir);

            index.update(new_data_stream.into(), &new_store).await?;

            Ok(Some((new_uuid, frag_bitmap)))
        }
        IndexType::Vector => {
            let mut scanner = dataset.scan();
            scanner.with_fragments(unindexed);
            scanner.with_row_id();
            scanner.project(&[&column.name])?;
            let new_data_stream = scanner.try_into_stream().await?;

            let index = dataset
                .open_vector_index(&column.name, old_index.uuid.to_string().as_str())
                .await?;

            let Some(ivf_idx) = index.as_any().downcast_ref::<IVFIndex>() else {
                info!("Index type: {:?} does not support append", index);
                return Ok(None);
            };

            let new_index = ivf_idx
                .append(dataset.as_ref(), new_data_stream, old_index, &column.name)
                .await?;

            Ok(Some((new_index, frag_bitmap)))
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    use arrow_array::cast::AsArray;
    use arrow_array::{FixedSizeListArray, RecordBatch, RecordBatchIterator};
    use arrow_schema::{DataType, Field, Schema};
    use futures::{stream, StreamExt, TryStreamExt};
    use lance_arrow::FixedSizeListArrayExt;
    use lance_index::{
        vector::{ivf::IvfBuildParams, pq::PQBuildParams},
        IndexType,
    };
    use lance_linalg::distance::MetricType;
    use lance_testing::datagen::generate_random_array;
    use tempfile::tempdir;

    use crate::index::vector::{pq::PQIndex, VectorIndexParams};
    use crate::index::DatasetIndexExt;

    #[tokio::test]
    async fn test_append_index() {
        const DIM: usize = 64;
        const IVF_PARTITIONS: usize = 2;

        let test_dir = tempdir().unwrap();
        let test_uri = test_dir.path().to_str().unwrap();

        let vectors = generate_random_array(1000 * DIM);

        let schema = Arc::new(Schema::new(vec![Field::new(
            "vector",
            DataType::FixedSizeList(
                Arc::new(Field::new("item", DataType::Float32, true)),
                DIM as i32,
            ),
            true,
        )]));
        let array = Arc::new(FixedSizeListArray::try_new_from_values(vectors, DIM as i32).unwrap());
        let batch = RecordBatch::try_new(schema.clone(), vec![array.clone()]).unwrap();

        let batches = RecordBatchIterator::new(vec![batch].into_iter().map(Ok), schema.clone());
        let mut dataset = Dataset::write(batches, test_uri, None).await.unwrap();

        let ivf_params = IvfBuildParams::new(IVF_PARTITIONS);
        let pq_params = PQBuildParams {
            num_sub_vectors: 2,
            ..Default::default()
        };
        let params = VectorIndexParams::with_ivf_pq_params(MetricType::L2, ivf_params, pq_params);

        dataset
            .create_index(&["vector"], IndexType::Vector, None, &params, true)
            .await
            .unwrap();

        let vectors = generate_random_array(1000 * DIM);
        let array = Arc::new(FixedSizeListArray::try_new_from_values(vectors, DIM as i32).unwrap());
        let batch = RecordBatch::try_new(schema.clone(), vec![array.clone()]).unwrap();

        let batches = RecordBatchIterator::new(vec![batch].into_iter().map(Ok), schema.clone());
        dataset.append(batches, None).await.unwrap();

        let index = &dataset.load_indices().await.unwrap()[0];
        assert!(!unindexed_fragments(index, &dataset)
            .await
            .unwrap()
            .is_empty());

        let q = array.value(5);
        let mut scanner = dataset.scan();
        scanner.nearest("vector", q.as_primitive(), 10).unwrap();
        let results = scanner
            .try_into_stream()
            .await
            .unwrap()
            .try_collect::<Vec<_>>()
            .await
            .unwrap();
        assert_eq!(results[0].num_rows(), 10); // Flat search.

        dataset.optimize_indices().await.unwrap();
        let index = &dataset.load_indices().await.unwrap()[0];
        assert!(unindexed_fragments(index, &dataset)
            .await
            .unwrap()
            .is_empty());

        // There should be two indices directories existed.
        let object_store = dataset.object_store();
        let index_dirs = object_store
            .read_dir_all(&dataset.indices_dir(), None)
            .await
            .unwrap()
            .try_collect::<Vec<_>>()
            .await
            .unwrap();
        assert_eq!(index_dirs.len(), 2);

        let mut scanner = dataset.scan();
        scanner.nearest("vector", q.as_primitive(), 10).unwrap();
        let results = scanner
            .try_into_stream()
            .await
            .unwrap()
            .try_collect::<Vec<_>>()
            .await
            .unwrap();
        let vectors = &results[0]["vector"];
        // Second batch of vectors should be in the index.
        let contained = vectors.as_fixed_size_list().iter().any(|v| {
            let vec = v.as_ref().unwrap();
            array.iter().any(|a| a.as_ref().unwrap() == vec)
        });
        assert!(contained);

        // Check that the index has all 2000 rows.
        let binding = dataset
            .open_vector_index("vector", index.uuid.to_string().as_str())
            .await
            .unwrap();
        let ivf_index = binding.as_any().downcast_ref::<IVFIndex>().unwrap();
        let row_in_index = stream::iter(0..IVF_PARTITIONS)
            .map(|part_id| async move {
                let part = ivf_index.load_partition(part_id, true).await.unwrap();
                let pq_idx = part.as_any().downcast_ref::<PQIndex>().unwrap();
                pq_idx.row_ids.as_ref().unwrap().len()
            })
            .buffered(2)
            .collect::<Vec<usize>>()
            .await
            .iter()
            .sum::<usize>();
        assert_eq!(row_in_index, 2000);
        assert_eq!(dataset.index_cache_entry_count(), 6)
    }
}