llama-index-vector-stores-postgres 0.7.0__tar.gz → 0.7.2__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: llama-index-vector-stores-postgres
3
- Version: 0.7.0
3
+ Version: 0.7.2
4
4
  Summary: llama-index vector_stores postgres integration
5
5
  Author-email: Your Name <you@example.com>
6
6
  License-Expression: MIT
@@ -49,6 +49,8 @@ PGType = Literal[
49
49
  "date",
50
50
  "timestamp",
51
51
  "uuid",
52
+ # Array type for GIN indexing
53
+ "text[]",
52
54
  ]
53
55
 
54
56
 
@@ -81,6 +83,7 @@ def get_data_model(
81
83
  from pgvector.sqlalchemy import Vector
82
84
  from sqlalchemy import Column, Computed
83
85
  from sqlalchemy.dialects.postgresql import (
86
+ ARRAY,
84
87
  BIGINT,
85
88
  JSON,
86
89
  JSONB,
@@ -105,6 +108,8 @@ def get_data_model(
105
108
  "date": Date,
106
109
  "timestamp": DateTime,
107
110
  "uuid": UUID,
111
+ # Array type for GIN indexing
112
+ "text[]": ARRAY(String),
108
113
  }
109
114
 
110
115
  indexed_metadata_keys = indexed_metadata_keys or set()
@@ -131,15 +136,33 @@ def get_data_model(
131
136
  else:
132
137
  embedding_col = Column(Vector(embed_dim)) # type: ignore
133
138
 
134
- metadata_indices = [
139
+ # BTREE indices for scalar types (existing behavior)
140
+ btree_indices = [
135
141
  Index(
136
142
  f"{indexname}_{key}_{pg_type.replace(' ', '_')}",
137
143
  cast(column("metadata_").op("->>")(key), pg_type_map[pg_type]),
138
144
  postgresql_using="btree",
139
145
  )
140
146
  for key, pg_type in indexed_metadata_keys
147
+ if pg_type != "text[]"
141
148
  ]
142
149
 
150
+ # GIN indices for text arrays (enables fast array operations with ?|, ?&, @> operators)
151
+ gin_indices = [
152
+ Index(
153
+ f"{indexname}_{key}_gin",
154
+ cast(
155
+ column("metadata_").op("->")(key), JSONB
156
+ ), # Cast to JSONB for GIN index compatibility
157
+ postgresql_using="gin",
158
+ )
159
+ for key, pg_type in indexed_metadata_keys
160
+ if pg_type == "text[]"
161
+ ]
162
+
163
+ # Combine both types of indices
164
+ metadata_indices = btree_indices + gin_indices
165
+
143
166
  if hybrid_search:
144
167
 
145
168
  class HybridAbstractData(base): # type: ignore
@@ -373,7 +396,6 @@ class PGVectorStore(BasePydanticVectorStore):
373
396
  if not self._is_initialized:
374
397
  return
375
398
 
376
- self._session.close_all()
377
399
  if self._engine:
378
400
  self._engine.dispose()
379
401
  if self._async_engine:
@@ -677,9 +699,9 @@ class PGVectorStore(BasePydanticVectorStore):
677
699
  f"({filter_value})"
678
700
  )
679
701
  elif filter_.operator in [FilterOperator.ANY, FilterOperator.ALL]:
680
- # Expects a list stored in the metadata, and a single value to compare
681
-
682
- # We apply same logic as above, but as an array
702
+ # Expects a text array stored in the metadata, and a list of values to compare
703
+ # Works with text[] arrays using PostgreSQL ?| (ANY) and ?& (ALL) operators
704
+ # Example: metadata_::jsonb->'tags' ?| array['AI', 'ML']
683
705
  filter_value = ", ".join(f"'{e}'" for e in filter_.value)
684
706
 
685
707
  return text(
@@ -27,7 +27,7 @@ dev = [
27
27
 
28
28
  [project]
29
29
  name = "llama-index-vector-stores-postgres"
30
- version = "0.7.0"
30
+ version = "0.7.2"
31
31
  description = "llama-index vector_stores postgres integration"
32
32
  authors = [{name = "Your Name", email = "you@example.com"}]
33
33
  requires-python = ">=3.9,<4.0"