vecstore 1.0.0

The perfect vector database - 100/100 score, embeddable, high-performance, production-ready with RAG toolkit
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
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301

use tempfile::TempDir;
use vecstore::{HNSWSearchParams, Metadata, PrefetchQuery, Query, QueryStage, VecStore};

#[test]
fn test_hnsw_parameter_tuning() {
    let temp_dir = TempDir::new().unwrap();
    let mut store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // Insert test data
    for i in 0..100 {
        let vector = vec![i as f32 / 100.0, (100 - i) as f32 / 100.0, 0.5];
        let mut meta = Metadata {
            fields: std::collections::HashMap::new(),
        };
        meta.fields.insert("id".into(), serde_json::json!(i));
        store.upsert(format!("doc{}", i), vector, meta).unwrap();
    }

    let query_vec = vec![0.5, 0.5, 0.5];

    // Test different HNSW search parameters
    let fast_results = store
        .query_with_params(
            Query::new(query_vec.clone()).with_limit(10),
            HNSWSearchParams::fast(),
        )
        .unwrap();

    let balanced_results = store
        .query_with_params(
            Query::new(query_vec.clone()).with_limit(10),
            HNSWSearchParams::balanced(),
        )
        .unwrap();

    let high_recall_results = store
        .query_with_params(
            Query::new(query_vec.clone()).with_limit(10),
            HNSWSearchParams::high_recall(),
        )
        .unwrap();

    let max_recall_results = store
        .query_with_params(
            Query::new(query_vec.clone()).with_limit(10),
            HNSWSearchParams::max_recall(),
        )
        .unwrap();

    // All should return results
    assert_eq!(fast_results.len(), 10);
    assert_eq!(balanced_results.len(), 10);
    assert_eq!(high_recall_results.len(), 10);
    assert_eq!(max_recall_results.len(), 10);

    // Higher recall settings may find better results (higher scores)
    // But for small datasets like this, they should be similar
    assert!(balanced_results[0].score > 0.0);
    assert!(high_recall_results[0].score > 0.0);
}

#[test]
fn test_query_planning() {
    let temp_dir = TempDir::new().unwrap();
    let mut store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // Insert test data
    for i in 0..1000 {
        let vector = vec![i as f32 / 1000.0, (1000 - i) as f32 / 1000.0, 0.5];
        let mut meta = Metadata {
            fields: std::collections::HashMap::new(),
        };
        meta.fields
            .insert("category".into(), serde_json::json!("tech"));
        meta.fields
            .insert("score".into(), serde_json::json!(i as f64 / 1000.0));
        store.upsert(format!("doc{}", i), vector, meta).unwrap();
    }

    // Test query planning without filter
    let query1 = Query::new(vec![0.5, 0.5, 0.5]).with_limit(10);
    let plan1 = store.explain_query(query1).unwrap();

    assert_eq!(plan1.query_type, "Vector Search");
    assert!(plan1.steps.len() >= 2); // HNSW + Top-K selection
    assert!(plan1.estimated_cost > 0.0);
    assert!(plan1.estimated_duration_ms > 0.0);

    // Test query planning with filter
    let query2 = Query::new(vec![0.5, 0.5, 0.5])
        .with_limit(10)
        .with_filter("score > 0.5");
    let plan2 = store.explain_query(query2).unwrap();

    assert_eq!(plan2.query_type, "Filtered Vector Search");
    assert!(plan2.steps.len() >= 3); // HNSW + Filter + Top-K
    assert!(plan2.estimated_cost > plan1.estimated_cost); // Filtering adds cost

    // Check for recommendations on large k
    let query3 = Query::new(vec![0.5, 0.5, 0.5]).with_limit(150);
    let plan3 = store.explain_query(query3).unwrap();

    assert!(plan3.recommendations.iter().any(|r| r.contains("Large k")));
}

#[test]
fn test_prefetch_query_vector_search() {
    let temp_dir = TempDir::new().unwrap();
    let mut store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // Insert test data
    for i in 0..50 {
        let vector = vec![i as f32 / 50.0, (50 - i) as f32 / 50.0, 0.5];
        let mut meta = Metadata {
            fields: std::collections::HashMap::new(),
        };
        meta.fields.insert("id".into(), serde_json::json!(i));
        store.upsert(format!("doc{}", i), vector, meta).unwrap();
    }

    // Single-stage prefetch query (vector search)
    let prefetch_query = PrefetchQuery {
        stages: vec![QueryStage::VectorSearch {
            vector: vec![0.5, 0.5, 0.5],
            k: 10,
            filter: None,
        }],
    };

    let results = store.prefetch_query(prefetch_query).unwrap();
    assert_eq!(results.len(), 10);
}

#[test]
fn test_prefetch_query_with_mmr() {
    let temp_dir = TempDir::new().unwrap();
    let mut store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // Insert test data with similar vectors
    for i in 0..50 {
        let vector = vec![0.5 + (i as f32 * 0.01), 0.5 - (i as f32 * 0.01), 0.5];
        let mut meta = Metadata {
            fields: std::collections::HashMap::new(),
        };
        meta.fields.insert("id".into(), serde_json::json!(i));
        store.upsert(format!("doc{}", i), vector, meta).unwrap();
    }

    // Two-stage prefetch: broad search + MMR for diversity
    let prefetch_query = PrefetchQuery {
        stages: vec![
            QueryStage::VectorSearch {
                vector: vec![0.5, 0.5, 0.5],
                k: 20, // Fetch 20 candidates
                filter: None,
            },
            QueryStage::MMR {
                k: 5,        // Select 5 diverse results
                lambda: 0.7, // 70% relevance, 30% diversity
            },
        ],
    };

    let results = store.prefetch_query(prefetch_query).unwrap();
    assert_eq!(results.len(), 5);

    // Results should exist and have metadata
    for result in &results {
        assert!(result.metadata.fields.contains_key("id"));
        assert!(result.score > 0.0);
    }

    // MMR should have successfully selected 5 results (diversity not strictly testable with small dataset)
    assert_eq!(results.len(), 5);
}

#[test]
fn test_prefetch_query_with_filter() {
    let temp_dir = TempDir::new().unwrap();
    let mut store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // Insert test data
    for i in 0..50 {
        let vector = vec![i as f32 / 50.0, (50 - i) as f32 / 50.0, 0.5];
        let mut meta = Metadata {
            fields: std::collections::HashMap::new(),
        };
        meta.fields.insert(
            "category".into(),
            serde_json::json!(if i % 2 == 0 { "even" } else { "odd" }),
        );
        meta.fields.insert("id".into(), serde_json::json!(i));
        store.upsert(format!("doc{}", i), vector, meta).unwrap();
    }

    // Multi-stage prefetch with filter
    let prefetch_query = PrefetchQuery {
        stages: vec![
            QueryStage::VectorSearch {
                vector: vec![0.5, 0.5, 0.5],
                k: 20,
                filter: None,
            },
            QueryStage::Filter {
                expr: vecstore::FilterExpr::Cmp {
                    field: "category".into(),
                    op: vecstore::FilterOp::Eq,
                    value: serde_json::json!("even"),
                },
            },
        ],
    };

    let results = store.prefetch_query(prefetch_query).unwrap();

    // All results should have category = "even"
    for result in results {
        assert_eq!(
            result.metadata.fields.get("category").unwrap(),
            &serde_json::json!("even")
        );
    }
}

#[test]
fn test_query_builder_api() {
    let temp_dir = TempDir::new().unwrap();
    let mut store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // Insert test data
    let mut meta = Metadata {
        fields: std::collections::HashMap::new(),
    };
    meta.fields
        .insert("category".into(), serde_json::json!("tech"));
    store
        .upsert("doc1".into(), vec![0.1, 0.2, 0.3], meta)
        .unwrap();

    // Test builder API
    let query = Query::new(vec![0.1, 0.2, 0.3])
        .with_limit(5)
        .with_filter("category = 'tech'");

    assert_eq!(query.k, 5);
    assert!(query.filter.is_some());

    let results = store.query(query).unwrap();
    assert_eq!(results.len(), 1);
}

#[test]
fn test_prefetch_empty_stages() {
    let temp_dir = TempDir::new().unwrap();
    let store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    let prefetch_query = PrefetchQuery { stages: vec![] };

    let result = store.prefetch_query(prefetch_query);
    assert!(result.is_err());
    assert!(result
        .unwrap_err()
        .to_string()
        .contains("at least one stage"));
}

#[test]
fn test_prefetch_mmr_without_previous_stage() {
    let temp_dir = TempDir::new().unwrap();
    let store = VecStore::open(temp_dir.path().join("test.db")).unwrap();

    // MMR stage without a previous search stage should fail
    let prefetch_query = PrefetchQuery {
        stages: vec![QueryStage::MMR { k: 5, lambda: 0.7 }],
    };

    let result = store.prefetch_query(prefetch_query);
    assert!(result.is_err());
    assert!(result
        .unwrap_err()
        .to_string()
        .contains("requires previous stage"));
}

#[test]
fn test_hnsw_params_values() {
    let fast = HNSWSearchParams::fast();
    assert_eq!(fast.ef_search, 20);

    let balanced = HNSWSearchParams::balanced();
    assert_eq!(balanced.ef_search, 50);

    let high_recall = HNSWSearchParams::high_recall();
    assert_eq!(high_recall.ef_search, 100);

    let max_recall = HNSWSearchParams::max_recall();
    assert_eq!(max_recall.ef_search, 200);

    let default = HNSWSearchParams::default();
    assert_eq!(default.ef_search, 50);
}