-
Notifications
You must be signed in to change notification settings - Fork 1.6k
/
pgvector2.py
389 lines (338 loc) · 15.1 KB
/
pgvector2.py
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
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
from typing import Optional, List, Union, Dict, Any
from hashlib import md5
try:
from sqlalchemy.dialects import postgresql
from sqlalchemy.engine import create_engine, Engine
from sqlalchemy.inspection import inspect
from sqlalchemy.orm import Session, sessionmaker
from sqlalchemy.schema import MetaData, Table, Column
from sqlalchemy.sql.expression import text, func, select
from sqlalchemy.types import DateTime, String
except ImportError:
raise ImportError("`sqlalchemy` not installed")
try:
from pgvector.sqlalchemy import Vector
except ImportError:
raise ImportError("`pgvector` not installed")
from phi.document import Document
from phi.embedder import Embedder
from phi.vectordb.base import VectorDb
from phi.vectordb.distance import Distance
from phi.vectordb.pgvector.index import Ivfflat, HNSW
from phi.utils.log import logger
class PgVector2(VectorDb):
def __init__(
self,
collection: str,
schema: Optional[str] = "ai",
db_url: Optional[str] = None,
db_engine: Optional[Engine] = None,
embedder: Optional[Embedder] = None,
distance: Distance = Distance.cosine,
index: Optional[Union[Ivfflat, HNSW]] = HNSW(),
):
_engine: Optional[Engine] = db_engine
if _engine is None and db_url is not None:
_engine = create_engine(db_url)
if _engine is None:
raise ValueError("Must provide either db_url or db_engine")
# Collection attributes
self.collection: str = collection
self.schema: Optional[str] = schema
# Database attributes
self.db_url: Optional[str] = db_url
self.db_engine: Engine = _engine
self.metadata: MetaData = MetaData(schema=self.schema)
# Embedder for embedding the document contents
_embedder = embedder
if _embedder is None:
from phi.embedder.openai import OpenAIEmbedder
_embedder = OpenAIEmbedder()
self.embedder: Embedder = _embedder
self.dimensions: Optional[int] = self.embedder.dimensions
# Distance metric
self.distance: Distance = distance
# Index for the collection
self.index: Optional[Union[Ivfflat, HNSW]] = index
# Database session
self.Session: sessionmaker[Session] = sessionmaker(bind=self.db_engine)
# Database table for the collection
self.table: Table = self.get_table()
def get_table(self) -> Table:
return Table(
self.collection,
self.metadata,
Column("id", String, primary_key=True),
Column("name", String),
Column("meta_data", postgresql.JSONB, server_default=text("'{}'::jsonb")),
Column("content", postgresql.TEXT),
Column("embedding", Vector(self.dimensions)),
Column("usage", postgresql.JSONB),
Column("created_at", DateTime(timezone=True), server_default=text("now()")),
Column("updated_at", DateTime(timezone=True), onupdate=text("now()")),
Column("content_hash", String),
extend_existing=True,
)
def table_exists(self) -> bool:
logger.debug(f"Checking if table exists: {self.table.name}")
try:
return inspect(self.db_engine).has_table(self.table.name, schema=self.schema)
except Exception as e:
logger.error(e)
return False
def create(self) -> None:
if not self.table_exists():
with self.Session() as sess:
with sess.begin():
logger.debug("Creating extension: vector")
sess.execute(text("create extension if not exists vector;"))
if self.schema is not None:
logger.debug(f"Creating schema: {self.schema}")
sess.execute(text(f"create schema if not exists {self.schema};"))
logger.debug(f"Creating table: {self.collection}")
self.table.create(self.db_engine)
def doc_exists(self, document: Document) -> bool:
"""
Validating if the document exists or not
Args:
document (Document): Document to validate
"""
columns = [self.table.c.name, self.table.c.content_hash]
with self.Session() as sess:
with sess.begin():
cleaned_content = document.content.replace("\x00", "\ufffd")
stmt = select(*columns).where(self.table.c.content_hash == md5(cleaned_content.encode()).hexdigest())
result = sess.execute(stmt).first()
return result is not None
def name_exists(self, name: str) -> bool:
"""
Validate if a row with this name exists or not
Args:
name (str): Name to check
"""
with self.Session() as sess:
with sess.begin():
stmt = select(self.table.c.name).where(self.table.c.name == name)
result = sess.execute(stmt).first()
return result is not None
def id_exists(self, id: str) -> bool:
"""
Validate if a row with this id exists or not
Args:
id (str): Id to check
"""
with self.Session() as sess:
with sess.begin():
stmt = select(self.table.c.id).where(self.table.c.id == id)
result = sess.execute(stmt).first()
return result is not None
def insert(self, documents: List[Document], batch_size: int = 10) -> None:
with self.Session() as sess:
counter = 0
for document in documents:
document.embed(embedder=self.embedder)
cleaned_content = document.content.replace("\x00", "\ufffd")
content_hash = md5(cleaned_content.encode()).hexdigest()
_id = document.id or content_hash
stmt = postgresql.insert(self.table).values(
id=_id,
name=document.name,
meta_data=document.meta_data,
content=cleaned_content,
embedding=document.embedding,
usage=document.usage,
content_hash=content_hash,
)
sess.execute(stmt)
counter += 1
logger.debug(f"Inserted document: {document.name} ({document.meta_data})")
# Commit every `batch_size` documents
if counter >= batch_size:
sess.commit()
logger.info(f"Committed {counter} documents")
counter = 0
# Commit any remaining documents
if counter > 0:
sess.commit()
logger.info(f"Committed {counter} documents")
def upsert_available(self) -> bool:
return True
def upsert(self, documents: List[Document], batch_size: int = 20) -> None:
"""
Upsert documents into the database.
Args:
documents (List[Document]): List of documents to upsert
batch_size (int): Batch size for upserting documents
"""
with self.Session() as sess:
counter = 0
for document in documents:
document.embed(embedder=self.embedder)
cleaned_content = document.content.replace("\x00", "\ufffd")
content_hash = md5(cleaned_content.encode()).hexdigest()
_id = document.id or content_hash
stmt = postgresql.insert(self.table).values(
id=_id,
name=document.name,
meta_data=document.meta_data,
content=cleaned_content,
embedding=document.embedding,
usage=document.usage,
content_hash=content_hash,
)
# Update row when id matches but 'content_hash' is different
stmt = stmt.on_conflict_do_update(
index_elements=["id"],
set_=dict(
name=stmt.excluded.name,
meta_data=stmt.excluded.meta_data,
content=stmt.excluded.content,
embedding=stmt.excluded.embedding,
usage=stmt.excluded.usage,
content_hash=stmt.excluded.content_hash,
),
)
sess.execute(stmt)
counter += 1
logger.debug(f"Upserted document: {document.id} | {document.name} | {document.meta_data}")
# Commit every `batch_size` documents
if counter >= batch_size:
sess.commit()
logger.info(f"Committed {counter} documents")
counter = 0
# Commit any remaining documents
if counter > 0:
sess.commit()
logger.info(f"Committed {counter} documents")
def search(self, query: str, limit: int = 5, filters: Optional[Dict[str, Any]] = None) -> List[Document]:
query_embedding = self.embedder.get_embedding(query)
if query_embedding is None:
logger.error(f"Error getting embedding for Query: {query}")
return []
columns = [
self.table.c.name,
self.table.c.meta_data,
self.table.c.content,
self.table.c.embedding,
self.table.c.usage,
]
stmt = select(*columns)
if filters is not None:
for key, value in filters.items():
if hasattr(self.table.c, key):
stmt = stmt.where(getattr(self.table.c, key) == value)
if self.distance == Distance.l2:
stmt = stmt.order_by(self.table.c.embedding.max_inner_product(query_embedding))
if self.distance == Distance.cosine:
stmt = stmt.order_by(self.table.c.embedding.cosine_distance(query_embedding))
if self.distance == Distance.max_inner_product:
stmt = stmt.order_by(self.table.c.embedding.max_inner_product(query_embedding))
stmt = stmt.limit(limit=limit)
logger.debug(f"Query: {stmt}")
# Get neighbors
try:
with self.Session() as sess:
with sess.begin():
if self.index is not None:
if isinstance(self.index, Ivfflat):
sess.execute(text(f"SET LOCAL ivfflat.probes = {self.index.probes}"))
elif isinstance(self.index, HNSW):
sess.execute(text(f"SET LOCAL hnsw.ef_search = {self.index.ef_search}"))
neighbors = sess.execute(stmt).fetchall() or []
except Exception as e:
logger.error(f"Error searching for documents: {e}")
logger.error("Table might not exist, creating for future use")
self.create()
return []
# Build search results
search_results: List[Document] = []
for neighbor in neighbors:
search_results.append(
Document(
name=neighbor.name,
meta_data=neighbor.meta_data,
content=neighbor.content,
embedder=self.embedder,
embedding=neighbor.embedding,
usage=neighbor.usage,
)
)
return search_results
def delete(self) -> None:
if self.table_exists():
logger.debug(f"Deleting table: {self.collection}")
self.table.drop(self.db_engine)
def exists(self) -> bool:
return self.table_exists()
def get_count(self) -> int:
with self.Session() as sess:
with sess.begin():
stmt = select(func.count(self.table.c.name)).select_from(self.table)
result = sess.execute(stmt).scalar()
if result is not None:
return int(result)
return 0
def optimize(self) -> None:
from math import sqrt
logger.debug("==== Optimizing Vector DB ====")
if self.index is None:
return
if self.index.name is None:
_type = "ivfflat" if isinstance(self.index, Ivfflat) else "hnsw"
self.index.name = f"{self.collection}_{_type}_index"
index_distance = "vector_cosine_ops"
if self.distance == Distance.l2:
index_distance = "vector_l2_ops"
if self.distance == Distance.max_inner_product:
index_distance = "vector_ip_ops"
if isinstance(self.index, Ivfflat):
num_lists = self.index.lists
if self.index.dynamic_lists:
total_records = self.get_count()
logger.debug(f"Number of records: {total_records}")
if total_records < 1000000:
num_lists = int(total_records / 1000)
elif total_records > 1000000:
num_lists = int(sqrt(total_records))
with self.Session() as sess:
with sess.begin():
logger.debug(f"Setting configuration: {self.index.configuration}")
for key, value in self.index.configuration.items():
sess.execute(text(f"SET {key} = '{value}';"))
logger.debug(
f"Creating Ivfflat index with lists: {num_lists}, probes: {self.index.probes} "
f"and distance metric: {index_distance}"
)
sess.execute(text(f"SET ivfflat.probes = {self.index.probes};"))
sess.execute(
text(
f"CREATE INDEX IF NOT EXISTS {self.index.name} ON {self.table} "
f"USING ivfflat (embedding {index_distance}) "
f"WITH (lists = {num_lists});"
)
)
elif isinstance(self.index, HNSW):
with self.Session() as sess:
with sess.begin():
logger.debug(f"Setting configuration: {self.index.configuration}")
for key, value in self.index.configuration.items():
sess.execute(text(f"SET {key} = '{value}';"))
logger.debug(
f"Creating HNSW index with m: {self.index.m}, ef_construction: {self.index.ef_construction} "
f"and distance metric: {index_distance}"
)
sess.execute(
text(
f"CREATE INDEX IF NOT EXISTS {self.index.name} ON {self.table} "
f"USING hnsw (embedding {index_distance}) "
f"WITH (m = {self.index.m}, ef_construction = {self.index.ef_construction});"
)
)
logger.debug("==== Optimized Vector DB ====")
def clear(self) -> bool:
from sqlalchemy import delete
with self.Session() as sess:
with sess.begin():
stmt = delete(self.table)
sess.execute(stmt)
return True