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
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766 | class PGVectorStore(BasePydanticVectorStore):
"""Postgres向量存储。
示例:
`pip install llama-index-vector-stores-postgres`
```python
from llama_index.vector_stores.postgres import PGVectorStore
# 创建PGVectorStore实例
vector_store = PGVectorStore.from_params(
database="vector_db",
host="localhost",
password="password",
port=5432,
user="postgres",
table_name="paul_graham_essay",
embed_dim=1536 # openai嵌入维度
)
```"""
from sqlalchemy.sql.selectable import Select
stores_text = True
flat_metadata = False
connection_string: Union[str, sqlalchemy.URL]
async_connection_string: Union[str, sqlalchemy.URL]
table_name: str
schema_name: str
embed_dim: int
hybrid_search: bool
text_search_config: str
cache_ok: bool
perform_setup: bool
debug: bool
use_jsonb: bool
_base: Any = PrivateAttr()
_table_class: Any = PrivateAttr()
_engine: Any = PrivateAttr()
_session: Any = PrivateAttr()
_async_engine: Any = PrivateAttr()
_async_session: Any = PrivateAttr()
_is_initialized: bool = PrivateAttr(default=False)
def __init__(
self,
connection_string: Union[str, sqlalchemy.URL],
async_connection_string: Union[str, sqlalchemy.URL],
table_name: str,
schema_name: str,
hybrid_search: bool = False,
text_search_config: str = "english",
embed_dim: int = 1536,
cache_ok: bool = False,
perform_setup: bool = True,
debug: bool = False,
use_jsonb: bool = False,
) -> None:
table_name = table_name.lower()
schema_name = schema_name.lower()
if hybrid_search and text_search_config is None:
raise ValueError(
"Sparse vector index creation requires "
"a text search configuration specification."
)
from sqlalchemy.orm import declarative_base
# sqlalchemy model
self._base = declarative_base()
self._table_class = get_data_model(
self._base,
table_name,
schema_name,
hybrid_search,
text_search_config,
cache_ok,
embed_dim=embed_dim,
use_jsonb=use_jsonb,
)
super().__init__(
connection_string=connection_string,
async_connection_string=async_connection_string,
table_name=table_name,
schema_name=schema_name,
hybrid_search=hybrid_search,
text_search_config=text_search_config,
embed_dim=embed_dim,
cache_ok=cache_ok,
perform_setup=perform_setup,
debug=debug,
use_jsonb=use_jsonb,
)
async def close(self) -> None:
if not self._is_initialized:
return
self._session.close_all()
self._engine.dispose()
await self._async_engine.dispose()
@classmethod
def class_name(cls) -> str:
return "PGVectorStore"
@classmethod
def from_params(
cls,
host: Optional[str] = None,
port: Optional[str] = None,
database: Optional[str] = None,
user: Optional[str] = None,
password: Optional[str] = None,
table_name: str = "llamaindex",
schema_name: str = "public",
connection_string: Optional[Union[str, sqlalchemy.URL]] = None,
async_connection_string: Optional[Union[str, sqlalchemy.URL]] = None,
hybrid_search: bool = False,
text_search_config: str = "english",
embed_dim: int = 1536,
cache_ok: bool = False,
perform_setup: bool = True,
debug: bool = False,
use_jsonb: bool = False,
) -> "PGVectorStore":
"""从数据库参数返回连接字符串。"""
conn_str = (
connection_string
or f"postgresql+psycopg2://{user}:{password}@{host}:{port}/{database}"
)
async_conn_str = async_connection_string or (
f"postgresql+asyncpg://{user}:{password}@{host}:{port}/{database}"
)
return cls(
connection_string=conn_str,
async_connection_string=async_conn_str,
table_name=table_name,
schema_name=schema_name,
hybrid_search=hybrid_search,
text_search_config=text_search_config,
embed_dim=embed_dim,
cache_ok=cache_ok,
perform_setup=perform_setup,
debug=debug,
use_jsonb=use_jsonb,
)
@property
def client(self) -> Any:
if not self._is_initialized:
return None
return self._engine
def _connect(self) -> Any:
from sqlalchemy import create_engine
from sqlalchemy.ext.asyncio import AsyncSession, create_async_engine
from sqlalchemy.orm import sessionmaker
self._engine = create_engine(self.connection_string, echo=self.debug)
self._session = sessionmaker(self._engine)
self._async_engine = create_async_engine(self.async_connection_string)
self._async_session = sessionmaker(self._async_engine, class_=AsyncSession) # type: ignore
def _create_schema_if_not_exists(self) -> None:
with self._session() as session, session.begin():
from sqlalchemy import text
# Check if the specified schema exists with "CREATE" statement
check_schema_statement = text(
f"SELECT schema_name FROM information_schema.schemata WHERE schema_name = :schema_name"
)
result = session.execute(
check_schema_statement, {"schema_name": self.schema_name}
).fetchone()
# If the schema does not exist, then create it
if not result:
create_schema_statement = text(
f"CREATE SCHEMA IF NOT EXISTS :schema_name"
)
session.execute(
create_schema_statement, {"schema_name": self.schema_name}
)
session.commit()
def _create_tables_if_not_exists(self) -> None:
with self._session() as session, session.begin():
self._base.metadata.create_all(session.connection())
def _create_extension(self) -> None:
import sqlalchemy
with self._session() as session, session.begin():
statement = sqlalchemy.text("CREATE EXTENSION IF NOT EXISTS vector")
session.execute(statement)
session.commit()
def _initialize(self) -> None:
if not self._is_initialized:
self._connect()
if self.perform_setup:
self._create_extension()
self._create_schema_if_not_exists()
self._create_tables_if_not_exists()
self._is_initialized = True
def _node_to_table_row(self, node: BaseNode) -> Any:
return self._table_class(
node_id=node.node_id,
embedding=node.get_embedding(),
text=node.get_content(metadata_mode=MetadataMode.NONE),
metadata_=node_to_metadata_dict(
node,
remove_text=True,
flat_metadata=self.flat_metadata,
),
)
def add(self, nodes: List[BaseNode], **add_kwargs: Any) -> List[str]:
self._initialize()
ids = []
with self._session() as session, session.begin():
for node in nodes:
ids.append(node.node_id)
item = self._node_to_table_row(node)
session.add(item)
session.commit()
return ids
async def async_add(self, nodes: List[BaseNode], **kwargs: Any) -> List[str]:
self._initialize()
ids = []
async with self._async_session() as session, session.begin():
for node in nodes:
ids.append(node.node_id)
item = self._node_to_table_row(node)
session.add(item)
await session.commit()
return ids
def _to_postgres_operator(self, operator: FilterOperator) -> str:
if operator == FilterOperator.EQ:
return "="
elif operator == FilterOperator.GT:
return ">"
elif operator == FilterOperator.LT:
return "<"
elif operator == FilterOperator.NE:
return "!="
elif operator == FilterOperator.GTE:
return ">="
elif operator == FilterOperator.LTE:
return "<="
elif operator == FilterOperator.IN:
return "IN"
elif operator == FilterOperator.NIN:
return "NOT IN"
elif operator == FilterOperator.CONTAINS:
return "@>"
else:
_logger.warning(f"Unknown operator: {operator}, fallback to '='")
return "="
def _build_filter_clause(self, filter_: MetadataFilter) -> Any:
from sqlalchemy import text
if filter_.operator in [FilterOperator.IN, FilterOperator.NIN]:
# Expects a single value in the metadata, and a list to compare
return text(
f"metadata_->>'{filter_.key}' {self._to_postgres_operator(filter_.operator)} :values"
).bindparams(values=tuple(filter_.value))
elif filter_.operator == FilterOperator.CONTAINS:
# Expects a list stored in the metadata, and a single value to compare
return text(
f"metadata_::jsonb->'{filter_.key}' "
f"{self._to_postgres_operator(filter_.operator)} "
f"'[\"{filter_.value}\"]'"
)
else:
# Check if value is a number. If so, cast the metadata value to a float
# This is necessary because the metadata is stored as a string
try:
return text(
f"(metadata_->>'{filter_.key}')::float "
f"{self._to_postgres_operator(filter_.operator)} "
f"{float(filter_.value)}"
)
except ValueError:
# If not a number, then treat it as a string
return text(
f"metadata_->>'{filter_.key}' "
f"{self._to_postgres_operator(filter_.operator)} "
f"'{filter_.value}'"
)
def _recursively_apply_filters(self, filters: List[MetadataFilters]) -> Any:
"""
返回一个SQLAlchemy的where子句。
"""
import sqlalchemy
sqlalchemy_conditions = {
"or": sqlalchemy.sql.or_,
"and": sqlalchemy.sql.and_,
}
if filters.condition not in sqlalchemy_conditions:
raise ValueError(
f"Invalid condition: {filters.condition}. "
f"Must be one of {list(sqlalchemy_conditions.keys())}"
)
return sqlalchemy_conditions[filters.condition](
*(
(
self._build_filter_clause(filter_)
if not isinstance(filter_, MetadataFilters)
else self._recursively_apply_filters(filter_)
)
for filter_ in filters.filters
)
)
def _apply_filters_and_limit(
self,
stmt: Select,
limit: int,
metadata_filters: Optional[MetadataFilters] = None,
) -> Any:
if metadata_filters:
stmt = stmt.where( # type: ignore
self._recursively_apply_filters(metadata_filters)
)
return stmt.limit(limit) # type: ignore
def _build_query(
self,
embedding: Optional[List[float]],
limit: int = 10,
metadata_filters: Optional[MetadataFilters] = None,
) -> Any:
from sqlalchemy import select, text
stmt = select( # type: ignore
self._table_class.id,
self._table_class.node_id,
self._table_class.text,
self._table_class.metadata_,
self._table_class.embedding.cosine_distance(embedding).label("distance"),
).order_by(text("distance asc"))
return self._apply_filters_and_limit(stmt, limit, metadata_filters)
def _query_with_score(
self,
embedding: Optional[List[float]],
limit: int = 10,
metadata_filters: Optional[MetadataFilters] = None,
**kwargs: Any,
) -> List[DBEmbeddingRow]:
stmt = self._build_query(embedding, limit, metadata_filters)
with self._session() as session, session.begin():
from sqlalchemy import text
if kwargs.get("ivfflat_probes"):
ivfflat_probes = kwargs.get("ivfflat_probes")
session.execute(
text(f"SET ivfflat.probes = :ivfflat_probes"),
{"ivfflat_probes": ivfflat_probes},
)
if kwargs.get("hnsw_ef_search"):
hnsw_ef_search = kwargs.get("hnsw_ef_search")
session.execute(
text(f"SET hnsw.ef_search = :hnsw_ef_search"),
{"hnsw_ef_search": hnsw_ef_search},
)
res = session.execute(
stmt,
)
return [
DBEmbeddingRow(
node_id=item.node_id,
text=item.text,
metadata=item.metadata_,
similarity=(1 - item.distance) if item.distance is not None else 0,
)
for item in res.all()
]
async def _aquery_with_score(
self,
embedding: Optional[List[float]],
limit: int = 10,
metadata_filters: Optional[MetadataFilters] = None,
**kwargs: Any,
) -> List[DBEmbeddingRow]:
stmt = self._build_query(embedding, limit, metadata_filters)
async with self._async_session() as async_session, async_session.begin():
from sqlalchemy import text
if kwargs.get("hnsw_ef_search"):
hnsw_ef_search = kwargs.get("hnsw_ef_search")
await async_session.execute(
text(f"SET hnsw.ef_search = :hnsw_ef_search"),
{"hnsw_ef_search": hnsw_ef_search},
)
if kwargs.get("ivfflat_probes"):
ivfflat_probes = kwargs.get("ivfflat_probes")
await async_session.execute(
text(f"SET ivfflat.probes = :ivfflat_probes"),
{"ivfflat_probes": ivfflat_probes},
)
res = await async_session.execute(stmt)
return [
DBEmbeddingRow(
node_id=item.node_id,
text=item.text,
metadata=item.metadata_,
similarity=(1 - item.distance) if item.distance is not None else 0,
)
for item in res.all()
]
def _build_sparse_query(
self,
query_str: Optional[str],
limit: int,
metadata_filters: Optional[MetadataFilters] = None,
) -> Any:
from sqlalchemy import select, type_coerce
from sqlalchemy.sql import func, text
from sqlalchemy.types import UserDefinedType
class REGCONFIG(UserDefinedType):
# The TypeDecorator.cache_ok class-level flag indicates if this custom TypeDecorator is safe to be used as part of a cache key.
# If the TypeDecorator is not guaranteed to produce the same bind/result behavior and SQL generation every time,
# this flag should be set to False; otherwise if the class produces the same behavior each time, it may be set to True.
cache_ok = True
def get_col_spec(self, **kw: Any) -> str:
return "regconfig"
if query_str is None:
raise ValueError("query_str must be specified for a sparse vector query.")
# Replace '&' with '|' to perform an OR search for higher recall
ts_query = func.to_tsquery(
func.replace(
func.text(
func.plainto_tsquery(
type_coerce(self.text_search_config, REGCONFIG), query_str
)
),
"&",
"|",
)
)
stmt = (
select( # type: ignore
self._table_class.id,
self._table_class.node_id,
self._table_class.text,
self._table_class.metadata_,
func.ts_rank(self._table_class.text_search_tsv, ts_query).label("rank"),
)
.where(self._table_class.text_search_tsv.op("@@")(ts_query))
.order_by(text("rank desc"))
)
# type: ignore
return self._apply_filters_and_limit(stmt, limit, metadata_filters)
async def _async_sparse_query_with_rank(
self,
query_str: Optional[str] = None,
limit: int = 10,
metadata_filters: Optional[MetadataFilters] = None,
) -> List[DBEmbeddingRow]:
stmt = self._build_sparse_query(query_str, limit, metadata_filters)
async with self._async_session() as async_session, async_session.begin():
res = await async_session.execute(stmt)
return [
DBEmbeddingRow(
node_id=item.node_id,
text=item.text,
metadata=item.metadata_,
similarity=item.rank,
)
for item in res.all()
]
def _sparse_query_with_rank(
self,
query_str: Optional[str] = None,
limit: int = 10,
metadata_filters: Optional[MetadataFilters] = None,
) -> List[DBEmbeddingRow]:
stmt = self._build_sparse_query(query_str, limit, metadata_filters)
with self._session() as session, session.begin():
res = session.execute(stmt)
return [
DBEmbeddingRow(
node_id=item.node_id,
text=item.text,
metadata=item.metadata_,
similarity=item.rank,
)
for item in res.all()
]
async def _async_hybrid_query(
self, query: VectorStoreQuery, **kwargs: Any
) -> List[DBEmbeddingRow]:
import asyncio
if query.alpha is not None:
_logger.warning("postgres hybrid search does not support alpha parameter.")
sparse_top_k = query.sparse_top_k or query.similarity_top_k
results = await asyncio.gather(
self._aquery_with_score(
query.query_embedding,
query.similarity_top_k,
query.filters,
**kwargs,
),
self._async_sparse_query_with_rank(
query.query_str, sparse_top_k, query.filters
),
)
dense_results, sparse_results = results
all_results = dense_results + sparse_results
return _dedup_results(all_results)
def _hybrid_query(
self, query: VectorStoreQuery, **kwargs: Any
) -> List[DBEmbeddingRow]:
if query.alpha is not None:
_logger.warning("postgres hybrid search does not support alpha parameter.")
sparse_top_k = query.sparse_top_k or query.similarity_top_k
dense_results = self._query_with_score(
query.query_embedding,
query.similarity_top_k,
query.filters,
**kwargs,
)
sparse_results = self._sparse_query_with_rank(
query.query_str, sparse_top_k, query.filters
)
all_results = dense_results + sparse_results
return _dedup_results(all_results)
def _db_rows_to_query_result(
self, rows: List[DBEmbeddingRow]
) -> VectorStoreQueryResult:
nodes = []
similarities = []
ids = []
for db_embedding_row in rows:
try:
node = metadata_dict_to_node(db_embedding_row.metadata)
node.set_content(str(db_embedding_row.text))
except Exception:
# NOTE: deprecated legacy logic for backward compatibility
node = TextNode(
id_=db_embedding_row.node_id,
text=db_embedding_row.text,
metadata=db_embedding_row.metadata,
)
similarities.append(db_embedding_row.similarity)
ids.append(db_embedding_row.node_id)
nodes.append(node)
return VectorStoreQueryResult(
nodes=nodes,
similarities=similarities,
ids=ids,
)
async def aquery(
self, query: VectorStoreQuery, **kwargs: Any
) -> VectorStoreQueryResult:
self._initialize()
if query.mode == VectorStoreQueryMode.HYBRID:
results = await self._async_hybrid_query(query, **kwargs)
elif query.mode in [
VectorStoreQueryMode.SPARSE,
VectorStoreQueryMode.TEXT_SEARCH,
]:
sparse_top_k = query.sparse_top_k or query.similarity_top_k
results = await self._async_sparse_query_with_rank(
query.query_str, sparse_top_k, query.filters
)
elif query.mode == VectorStoreQueryMode.DEFAULT:
results = await self._aquery_with_score(
query.query_embedding,
query.similarity_top_k,
query.filters,
**kwargs,
)
else:
raise ValueError(f"Invalid query mode: {query.mode}")
return self._db_rows_to_query_result(results)
def query(self, query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult:
self._initialize()
if query.mode == VectorStoreQueryMode.HYBRID:
results = self._hybrid_query(query, **kwargs)
elif query.mode in [
VectorStoreQueryMode.SPARSE,
VectorStoreQueryMode.TEXT_SEARCH,
]:
sparse_top_k = query.sparse_top_k or query.similarity_top_k
results = self._sparse_query_with_rank(
query.query_str, sparse_top_k, query.filters
)
elif query.mode == VectorStoreQueryMode.DEFAULT:
results = self._query_with_score(
query.query_embedding,
query.similarity_top_k,
query.filters,
**kwargs,
)
else:
raise ValueError(f"Invalid query mode: {query.mode}")
return self._db_rows_to_query_result(results)
def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
from sqlalchemy import delete
self._initialize()
with self._session() as session, session.begin():
stmt = delete(self._table_class).where(
self._table_class.metadata_["doc_id"].astext == ref_doc_id
)
session.execute(stmt)
session.commit()
|