velesdb-core 7.0.0

High-performance vector database engine written in Rust
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
252
253
254
255
256

//! Tests for batch and multi-query search methods.

#![cfg(all(test, feature = "persistence"))]

use crate::collection::types::Collection;
use crate::filter::{Condition, Filter};
use crate::test_fixtures::fixtures::{make_point_with_payload, setup_collection};

/// Helper: create a collection with 6 test points spanning the unit hypercube.
fn setup_batch_collection() -> (tempfile::TempDir, Collection) {
    let points = vec![
        make_point_with_payload(
            1,
            vec![1.0, 0.0, 0.0, 0.0],
            serde_json::json!({ "category": "tech" }),
        ),
        make_point_with_payload(
            2,
            vec![0.0, 1.0, 0.0, 0.0],
            serde_json::json!({ "category": "sports" }),
        ),
        make_point_with_payload(
            3,
            vec![0.7, 0.7, 0.0, 0.0],
            serde_json::json!({ "category": "tech" }),
        ),
        make_point_with_payload(
            4,
            vec![0.0, 0.0, 1.0, 0.0],
            serde_json::json!({ "category": "music" }),
        ),
        make_point_with_payload(
            5,
            vec![0.5, 0.5, 0.5, 0.5],
            serde_json::json!({ "category": "tech" }),
        ),
        make_point_with_payload(
            6,
            vec![0.9, 0.1, 0.0, 0.0],
            serde_json::json!({ "category": "sports" }),
        ),
    ];
    let (dir, col) = setup_collection(4);
    col.upsert(points).expect("test: upsert");
    (dir, col)
}

// -----------------------------------------------------------------------
// search_batch_parallel
// -----------------------------------------------------------------------

#[test]
fn test_batch_parallel_two_queries_returns_per_query_results() {
    let (_dir, col) = setup_batch_collection();

    let q1: Vec<f32> = vec![1.0, 0.0, 0.0, 0.0];
    let q2: Vec<f32> = vec![0.0, 1.0, 0.0, 0.0];
    let queries: Vec<&[f32]> = vec![&q1, &q2];

    let results = col
        .search_batch_parallel(&queries, 3)
        .expect("batch search");

    assert_eq!(results.len(), 2, "one result set per query");
    assert!(!results[0].is_empty(), "first query should have results");
    assert!(!results[1].is_empty(), "second query should have results");

    // First query is close to point 1 ([1,0,0,0]).
    assert_eq!(results[0][0].point.id, 1, "q1 best match should be id=1");
    // Second query is close to point 2 ([0,1,0,0]).
    assert_eq!(results[1][0].point.id, 2, "q2 best match should be id=2");
}

// -----------------------------------------------------------------------
// search_batch_with_filters — per-query filters
// -----------------------------------------------------------------------

#[test]
fn test_batch_with_filters_applies_distinct_filter_per_query() {
    let (_dir, col) = setup_batch_collection();

    let q1: Vec<f32> = vec![1.0, 0.0, 0.0, 0.0];
    let q2: Vec<f32> = vec![0.0, 1.0, 0.0, 0.0];
    let queries: Vec<&[f32]> = vec![&q1, &q2];

    let filter_tech = Some(Filter::new(Condition::eq("category", "tech")));
    let filter_sports = Some(Filter::new(Condition::eq("category", "sports")));
    let filters = vec![filter_tech, filter_sports];

    let results = col
        .search_batch_with_filters(&queries, 5, &filters)
        .expect("batch with filters");

    assert_eq!(results.len(), 2);

    // All q1 results should be category=tech.
    for r in &results[0] {
        let cat = r.point.payload.as_ref().and_then(|p| p.get("category"));
        assert_eq!(cat.and_then(|v| v.as_str()), Some("tech"));
    }
    // All q2 results should be category=sports.
    for r in &results[1] {
        let cat = r.point.payload.as_ref().and_then(|p| p.get("category"));
        assert_eq!(cat.and_then(|v| v.as_str()), Some("sports"));
    }
}

// -----------------------------------------------------------------------
// search_batch_with_filters — mismatched lengths
// -----------------------------------------------------------------------

#[test]
fn test_batch_with_filters_rejects_mismatched_lengths() {
    let (_dir, col) = setup_batch_collection();

    let q1: Vec<f32> = vec![1.0, 0.0, 0.0, 0.0];
    let queries: Vec<&[f32]> = vec![&q1];
    let filters = vec![None, None]; // 2 filters for 1 query

    let result = col.search_batch_with_filters(&queries, 5, &filters);
    assert!(result.is_err(), "should reject mismatched lengths");
}

// -----------------------------------------------------------------------
// Batch search on empty collection
// -----------------------------------------------------------------------

#[test]
fn test_batch_parallel_empty_collection_returns_empty_vecs() {
    let (_dir, col) = setup_collection(4);

    let q: Vec<f32> = vec![1.0, 0.0, 0.0, 0.0];
    let queries: Vec<&[f32]> = vec![&q];

    let results = col
        .search_batch_parallel(&queries, 5)
        .expect("batch search on empty");

    assert_eq!(results.len(), 1);
    assert!(
        results[0].is_empty(),
        "empty collection should return no hits"
    );
}

// -----------------------------------------------------------------------
// Dimension mismatch
// -----------------------------------------------------------------------

#[test]
fn test_batch_parallel_rejects_wrong_dimension() {
    let (_dir, col) = setup_batch_collection();

    let bad_q: Vec<f32> = vec![1.0, 0.0]; // dim=2, expected=4
    let queries: Vec<&[f32]> = vec![&bad_q];

    let result = col.search_batch_parallel(&queries, 3);
    assert!(result.is_err(), "should reject wrong dimension");
}

// -----------------------------------------------------------------------
// Per-query top_k: separate calls with different k values (#419)
// -----------------------------------------------------------------------

#[test]
fn test_batch_with_filters_respects_per_call_k() {
    let (_dir, col) = setup_batch_collection();

    let q1: Vec<f32> = vec![1.0, 0.0, 0.0, 0.0];
    let q2: Vec<f32> = vec![0.0, 1.0, 0.0, 0.0];

    // Call with k=2: both queries truncated to 2 results.
    let results_k2 = col
        .search_batch_with_filters(&[&q1, &q2], 2, &[None, None])
        .expect("batch k=2");
    assert!(results_k2[0].len() <= 2, "q1 should have at most 2 results");
    assert!(results_k2[1].len() <= 2, "q2 should have at most 2 results");

    // Call with k=5: both queries can return up to 5 (collection has 6 points).
    let results_k5 = col
        .search_batch_with_filters(&[&q1, &q2], 5, &[None, None])
        .expect("batch k=5");
    assert!(
        results_k5[0].len() > results_k2[0].len(),
        "k=5 should return more results than k=2"
    );
}

// -----------------------------------------------------------------------
// Regression (#2102): multi-query fusion on a distance metric must rank
// the NEAREST documents first. Pre-fix, score-level strategies fused raw
// Euclidean distances as if higher were better and returned the farthest.
// -----------------------------------------------------------------------

fn setup_euclidean_collection() -> (tempfile::TempDir, Collection) {
    use crate::distance::DistanceMetric;
    let dir = tempfile::tempdir().expect("test: tempdir");
    let col = Collection::create(
        std::path::PathBuf::from(dir.path()),
        2,
        DistanceMetric::Euclidean,
    )
    .expect("test: collection");
    col.upsert(vec![
        make_point_with_payload(1, vec![0.1, 0.0], serde_json::json!({})),
        make_point_with_payload(2, vec![5.0, 0.0], serde_json::json!({})),
        make_point_with_payload(3, vec![9.0, 0.0], serde_json::json!({})),
    ])
    .expect("test: upsert");
    (dir, col)
}

#[test]
fn test_multi_query_euclidean_maximum_ranks_nearest_first() {
    let (_dir, col) = setup_euclidean_collection();
    let q1: Vec<f32> = vec![0.0, 0.0];
    let q2: Vec<f32> = vec![0.2, 0.0];
    let queries: Vec<&[f32]> = vec![&q1, &q2];

    let results = col
        .multi_query_search(&queries, 3, crate::fusion::FusionStrategy::Maximum, None)
        .expect("test: multi query");

    let ids: Vec<u64> = results.iter().map(|r| r.point.id).collect();
    assert_eq!(
        ids,
        vec![1, 2, 3],
        "nearest-first order for Euclidean fusion, got scores {:?}",
        results.iter().map(|r| r.score).collect::<Vec<_>>()
    );
    // "Best" per doc across queries is its SMALLEST distance, in metric units.
    assert!(
        (results[0].score - 0.1).abs() < 1e-5,
        "doc1 best distance is 0.1 from both queries, got {}",
        results[0].score
    );
}

#[test]
fn test_multi_query_euclidean_average_ranks_nearest_first() {
    let (_dir, col) = setup_euclidean_collection();
    let q1: Vec<f32> = vec![0.0, 0.0];
    let queries: Vec<&[f32]> = vec![&q1];

    let results = col
        .multi_query_search(&queries, 3, crate::fusion::FusionStrategy::Average, None)
        .expect("test: multi query");

    let ids: Vec<u64> = results.iter().map(|r| r.point.id).collect();
    assert_eq!(ids, vec![1, 2, 3], "ascending average distance");
    assert!(
        results.windows(2).all(|w| w[0].score <= w[1].score),
        "scores must ascend in metric units: {:?}",
        results.iter().map(|r| r.score).collect::<Vec<_>>()
    );
}