nitrite_vector 0.4.2

Vector / HNSW ANN indexing and RAG store for the Nitrite database
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

//! The HNSW index must survive a database close/reopen (durability), honor the
//! configured stored-vector precision, and heal itself from corrupt or stale
//! index storage by rebuilding from the collection.

mod common;

use common::{doc_with_vector, names, open_db, open_plain_db};
use nitrite::common::Value;
use nitrite_vector::{
    vector_field, vector_index_options, IndexBackend, Metric, Precision, VectorIndexConfig,
};

#[test]
fn index_survives_close_and_reopen() {
    let dir = tempfile::tempdir().unwrap();
    let path = dir.path().to_str().unwrap().to_string();
    let config = VectorIndexConfig::new(3, Metric::Cosine);

    // Phase 1: build the index, then close the database.
    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        collection
            .create_index(vec!["embedding"], &vector_index_options())
            .unwrap();
        collection.insert(doc_with_vector("a", &[1.0, 0.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("b", &[0.0, 1.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("c", &[0.0, 0.0, 1.0])).unwrap();
        collection.insert(doc_with_vector("d", &[0.9, 0.1, 0.0])).unwrap();
        db.close().unwrap();
    }

    // Phase 2: reopen at the same path and query — the graph must be intact.
    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        let filter = vector_field("embedding").nearest(vec![1.0, 0.0, 0.0], 2).build();
        let got = names(&collection, filter);
        assert_eq!(got.len(), 2);
        assert_eq!(got[0], "a");
        assert_eq!(got[1], "d");
        db.close().unwrap();
    }
}

#[test]
fn hnsw_honors_stored_vector_precision_across_reopen() {
    let dir = tempfile::tempdir().unwrap();
    let path = dir.path().to_str().unwrap().to_string();
    // F16 stored vectors: search must still resolve correctly after a reopen
    // forces vectors to round-trip through the half-precision codec.
    let config = VectorIndexConfig::new(3, Metric::Cosine).precision(Precision::F16);

    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        collection.create_index(vec!["embedding"], &vector_index_options()).unwrap();
        collection.insert(doc_with_vector("a", &[1.0, 0.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("b", &[0.0, 1.0, 0.0])).unwrap();
        db.close().unwrap();
    }
    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        let filter = vector_field("embedding").nearest(vec![1.0, 0.0, 0.0], 1).build();
        assert_eq!(names(&collection, filter), vec!["a"]);
        db.close().unwrap();
    }
}

#[test]
fn corrupt_hnsw_header_triggers_automatic_rebuild() {
    let dir = tempfile::tempdir().unwrap();
    let path = dir.path().to_str().unwrap().to_string();
    let config = VectorIndexConfig::new(3, Metric::Cosine);

    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        collection.create_index(vec!["embedding"], &vector_index_options()).unwrap();
        collection.insert(doc_with_vector("a", &[1.0, 0.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("b", &[0.0, 1.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("c", &[0.9, 0.1, 0.0])).unwrap();
        db.close().unwrap();
    }
    // Corrupt the index header through a plain (no vector module) session.
    {
        let db = open_plain_db(&path);
        let store = db.config().nitrite_store().unwrap();
        let map = store.open_map("docs_embedding_vector_idx").unwrap();
        map.put(
            Value::String("__hnsw_meta__".to_string()),
            Value::Bytes(vec![0xDE, 0xAD, 0xBE, 0xEF]),
        )
        .unwrap();
        db.close().unwrap();
    }
    // Reopen with the module: the damaged index must be rebuilt and correct.
    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        let filter = vector_field("embedding").nearest(vec![1.0, 0.0, 0.0], 2).build();
        let got = names(&collection, filter);
        assert_eq!(got, vec!["a", "c"], "rebuilt index must rank correctly");
        db.close().unwrap();
    }
}

#[test]
fn lost_diskann_sidecar_triggers_automatic_rebuild() {
    let dir = tempfile::tempdir().unwrap();
    let path = dir.path().to_str().unwrap().to_string();
    let config = VectorIndexConfig::new(3, Metric::Cosine)
        .backend(IndexBackend::DiskAnn)
        .pq_subvectors(0);

    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        collection.create_index(vec!["embedding"], &vector_index_options()).unwrap();
        collection.insert(doc_with_vector("a", &[1.0, 0.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("b", &[0.0, 1.0, 0.0])).unwrap();
        collection.insert(doc_with_vector("c", &[0.9, 0.1, 0.0])).unwrap();
        db.close().unwrap();
    }
    // Simulate a crash artifact: delete the sidecar, keep the data file.
    let meta = std::fs::read_dir(dir.path())
        .unwrap()
        .filter_map(|e| e.ok().map(|e| e.path()))
        .find(|p| p.to_string_lossy().ends_with(".dann.meta"))
        .expect("sidecar exists");
    std::fs::remove_file(meta).unwrap();

    {
        let db = open_db(&path, config);
        let collection = db.collection("docs").unwrap();
        let filter = vector_field("embedding").nearest(vec![1.0, 0.0, 0.0], 2).build();
        let got = names(&collection, filter);
        assert_eq!(got, vec!["a", "c"], "rebuilt DiskANN index must rank correctly");
        db.close().unwrap();
    }
}

#[test]
fn per_index_configs_allow_mixed_dimensions() {
    use nitrite::nitrite::Nitrite;
    use nitrite_fjall_adapter::FjallModule;
    use nitrite_vector::VectorModule;

    let dir = tempfile::tempdir().unwrap();
    let path = dir.path().to_str().unwrap().to_string();
    let db = Nitrite::builder()
        .load_module(FjallModule::with_config().db_path(&path).low_memory_preset().build())
        .load_module(
            VectorModule::builder(3, Metric::Cosine)
                .index_config("wide", "embedding", VectorIndexConfig::new(5, Metric::Euclidean))
                .build(),
        )
        .open_or_create(None, None)
        .unwrap();

    let narrow = db.collection("docs").unwrap();
    narrow.create_index(vec!["embedding"], &vector_index_options()).unwrap();
    narrow.insert(doc_with_vector("n", &[1.0, 0.0, 0.0])).unwrap();

    let wide = db.collection("wide").unwrap();
    wide.create_index(vec!["embedding"], &vector_index_options()).unwrap();
    wide.insert(doc_with_vector("w", &[1.0, 0.0, 0.0, 0.0, 0.0])).unwrap();

    let got = names(&narrow, vector_field("embedding").nearest(vec![1.0, 0.0, 0.0], 1).build());
    assert_eq!(got, vec!["n"]);
    let got = names(&wide, vector_field("embedding").nearest(vec![1.0, 0.0, 0.0, 0.0, 0.0], 1).build());
    assert_eq!(got, vec!["w"]);
    db.close().unwrap();
}

#[test]
fn knn_query_without_index_is_an_error() {
    let (_dir, db) = common::temp_db(VectorIndexConfig::new(3, Metric::Cosine));
    let collection = db.collection("noindex").unwrap();
    collection.insert(doc_with_vector("a", &[1.0, 0.0, 0.0])).unwrap();

    let filter = vector_field("embedding").nearest(vec![1.0, 0.0, 0.0], 1).build();
    let result = collection.find(filter).and_then(|c| c.map(|r| r.map(|_| ())).collect::<Result<Vec<_>, _>>());
    assert!(result.is_err(), "kNN without a vector index must error, not scan");
}