Spaces:
Running
Running
File size: 18,057 Bytes
a55ea57 5583f46 a55ea57 6663ad4 a55ea57 6663ad4 a55ea57 6663ad4 a55ea57 6663ad4 a55ea57 5583f46 a55ea57 5583f46 a55ea57 5583f46 a55ea57 5583f46 a55ea57 5583f46 a55ea57 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 | """
File: /vectors.py
Project: postgresql
Created Date: Sunday May 24th 2026
Author: Christian Nonis <alch.infoemail@gmail.com>
"""
from __future__ import annotations
from contextlib import contextmanager
import hashlib
import json
import logging
import re
import threading
from typing import Any, Iterator, Optional
import psycopg2
import psycopg2.extras
from pgvector.psycopg2 import register_vector
from src.adapters.interfaces.embeddings import VectorStoreClient
from src.config import config
from src.constants.embeddings import EMBEDDING_STORES_SIZES, Vector
from src.lib.embeddings.profile import EmbeddingProfileError, index_profile, require_matching_profile
from ._provisioning import (
borrow,
clear_brain_database,
require_registered_brain,
get_brain_pool,
)
logger = logging.getLogger(__name__)
_STORE_NAME_RE = re.compile(r"^[a-zA-Z_][a-zA-Z0-9_]*$")
def string_to_int64(s: str) -> int:
sha256 = hashlib.sha256(s.encode()).digest()
return int.from_bytes(sha256[:8], byteorder="big") % (2**63)
def _to_int_id(value: Any) -> int:
if isinstance(value, int):
return value
if isinstance(value, str):
try:
return int(value)
except ValueError:
return string_to_int64(value)
return string_to_int64(str(value))
def _safe_store(store: str) -> str:
if not _STORE_NAME_RE.match(store):
raise ValueError(f"Invalid store name: {store}")
return store
def _table_name(store: str) -> str:
return f"vectors_{_safe_store(store)}"
def _vector_index_ddl(
table: str, dimension: int, *, search_enabled: bool | None = None
) -> str:
dim = int(dimension)
if search_enabled is None:
search_enabled = bool(config.search_enabled)
if dim <= 2000:
return f"""
CREATE INDEX IF NOT EXISTS idx_{table}_embeddings
ON {table} USING hnsw (embeddings vector_cosine_ops);
"""
if not search_enabled:
return ""
return f"""
CREATE INDEX IF NOT EXISTS idx_{table}_embeddings_halfvec
ON {table} USING hnsw
((embeddings::halfvec({dim})) halfvec_cosine_ops);
"""
def _uses_halfvec_ann(dimension: int, *, search_enabled: bool | None = None) -> bool:
if search_enabled is None:
search_enabled = bool(config.search_enabled)
try:
dim = int(dimension)
except (TypeError, ValueError):
return False
return search_enabled and dim > 2000
def _table_vector_dimension(cur: psycopg2.extensions.cursor, table: str) -> Optional[int]:
cur.execute(
"""
SELECT a.atttypmod
FROM pg_attribute a
JOIN pg_class c ON a.attrelid = c.oid
JOIN pg_namespace n ON c.relnamespace = n.oid
WHERE c.relname = %s
AND a.attname = 'embeddings'
AND NOT a.attisdropped
AND n.nspname = current_schema()
""",
(table,),
)
row = cur.fetchone()
if not row or row[0] is None or row[0] < 1:
return None
return int(row[0])
class PostgreSQLVectorStoreClient(VectorStoreClient):
"""
Vector store backed by PostgreSQL with the pgvector extension.
Each brain owns its own database (`brain_<sanitized_id>`); per-store tables
live inside that database with no `brain_id` column. The `vector` extension
is enabled lazily, once per brain database.
"""
def __init__(self):
self._lock = threading.RLock()
self._brain_extensions_ready: set[str] = set()
self._initialized_stores: set[str] = set()
@contextmanager
def _connection(self, brain_id: str) -> Iterator[psycopg2.extensions.connection]:
self._ensure_extension(brain_id)
with borrow(get_brain_pool(brain_id)) as conn:
register_vector(conn)
yield conn
def _ensure_extension(self, brain_id: str) -> None:
require_registered_brain(brain_id)
if brain_id in self._brain_extensions_ready:
return
with self._lock:
if brain_id in self._brain_extensions_ready:
return
with borrow(get_brain_pool(brain_id)) as conn:
with conn.cursor() as cur:
cur.execute("CREATE EXTENSION IF NOT EXISTS vector")
conn.commit()
self._brain_extensions_ready.add(brain_id)
def clear_brain_data(self, brain_id: str) -> None:
clear_brain_database(brain_id, table_prefixes=("vectors_",))
self._brain_extensions_ready.discard(brain_id)
self._initialized_stores = {
key for key in self._initialized_stores if not key.startswith(f"{brain_id}:")
}
def _ensure_store(self, store: str, brain_id: str) -> None:
require_registered_brain(brain_id)
if store not in EMBEDDING_STORES_SIZES:
raise ValueError(f"Store {store} not available")
dimension = EMBEDDING_STORES_SIZES[store]
profile = index_profile(config, dimension)
key = f"{brain_id}:{store}:{json.dumps(profile, sort_keys=True)}"
if key in self._initialized_stores:
return
with self._lock:
if key in self._initialized_stores:
return
table = _table_name(store)
index_ddl = _vector_index_ddl(table, dimension)
ddl = f"""
CREATE TABLE IF NOT EXISTS {table} (
id BIGINT PRIMARY KEY,
uuid TEXT NOT NULL,
embeddings vector({dimension}),
metadata JSONB NOT NULL DEFAULT '{{}}'::jsonb
);
CREATE INDEX IF NOT EXISTS idx_{table}_uuid
ON {table} (uuid);
{index_ddl}
"""
with self._connection(brain_id) as conn:
# Roll back on any mismatch, including pooled connections.
with conn, conn.cursor() as cur:
cur.execute("SELECT pg_advisory_xact_lock(hashtext(%s))", ("brainapi-embedding-profiles",))
existing_dim = _table_vector_dimension(cur, table)
if existing_dim is not None and existing_dim != dimension:
raise EmbeddingProfileError(
f"Embedding index/profile mismatch for {table}: existing dimension "
f"{existing_dim}, configured {dimension}. Use a new empty brain/index "
"or explicitly re-index; existing vectors will not be deleted."
)
cur.execute("""
CREATE TABLE IF NOT EXISTS embedding_profiles (
store TEXT PRIMARY KEY,
profile JSONB NOT NULL
)
""")
cur.execute("SELECT profile FROM embedding_profiles WHERE store = %s", (store,))
row = cur.fetchone()
require_matching_profile(row[0] if row else None, profile, legacy_index=existing_dim is not None)
try:
cur.execute(ddl)
except Exception as exc:
if _uses_halfvec_ann(dimension):
raise RuntimeError(
"SEARCH_ENABLED=true requires a halfvec HNSW index "
f"on {table} (dimension {dimension} > 2000). "
"The vector backend could not create it: "
f"{exc}"
) from exc
raise
cur.execute(
"INSERT INTO embedding_profiles (store, profile) VALUES (%s, %s::jsonb) ON CONFLICT (store) DO NOTHING",
(store, json.dumps(profile)),
)
conn.commit()
self._initialized_stores.add(key)
def add_vectors(
self, vectors: list[Vector], store: str, brain_id: str
) -> list[str]:
if not vectors:
return []
self._ensure_store(store, brain_id)
table = _table_name(store)
rows = []
ids: list[int] = []
for vector in vectors:
int_id = string_to_int64(vector.id)
ids.append(int_id)
metadata = dict(vector.metadata or {})
metadata.pop("id", None)
metadata.pop("embeddings", None)
metadata.pop("distance", None)
rows.append(
(
int_id,
vector.id,
vector.embeddings,
json.dumps(metadata, default=str),
)
)
with self._connection(brain_id) as conn:
with conn.cursor() as cur:
psycopg2.extras.execute_values(
cur,
f"""
INSERT INTO {table}
(id, uuid, embeddings, metadata)
VALUES %s
ON CONFLICT (id) DO UPDATE SET
uuid = EXCLUDED.uuid,
embeddings = EXCLUDED.embeddings,
metadata = EXCLUDED.metadata
""",
rows,
template="(%s, %s, %s, %s::jsonb)",
)
conn.commit()
return ids
def search_vectors(
self, data_vector: list[float], brain_id: str, store: str, k: int = 10
) -> list[Vector]:
from src.utils.vector_search import ann_overfetch_k, stable_top_k_vectors
self._ensure_store(store, brain_id)
if k <= 0:
return []
table = _table_name(store)
fetch_k = ann_overfetch_k(k)
dimension = EMBEDDING_STORES_SIZES[store]
use_halfvec = _uses_halfvec_ann(dimension)
if use_halfvec:
order_sql = (
f"embeddings::halfvec({dimension}) "
f"<=> (%s::vector)::halfvec({dimension})"
)
else:
order_sql = "embeddings <=> %s::vector"
with self._connection(brain_id) as conn:
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
cur.execute(
"SELECT set_config('hnsw.ef_search', %s, true)",
(str(max(100, fetch_k)),),
)
cur.execute(
f"""
SELECT id, uuid, metadata,
(embeddings <=> %s::vector) AS distance
FROM {table}
ORDER BY {order_sql}, uuid ASC, id ASC
LIMIT %s
""",
(data_vector, data_vector, fetch_k),
)
rows = cur.fetchall()
results: list[Vector] = []
for row in rows:
metadata = dict(row["metadata"] or {})
row_uuid = row.get("uuid")
if row_uuid and not metadata.get("uuid"):
metadata["uuid"] = str(row_uuid)
results.append(
Vector(
id=str(row_uuid or row["id"]),
metadata=metadata,
distance=float(row["distance"]),
)
)
return stable_top_k_vectors(results, k)
def get_by_ids(self, ids: list[str], store: str, brain_id: str) -> list[Vector]:
if not ids:
return []
self._ensure_store(store, brain_id)
table = _table_name(store)
int_ids = [_to_int_id(i) for i in ids]
with self._connection(brain_id) as conn:
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
cur.execute(
f"""
SELECT id, uuid, embeddings, metadata
FROM {table}
WHERE id = ANY(%s)
""",
(int_ids,),
)
rows = cur.fetchall()
results: list[Vector] = []
for row in rows:
embeddings = row["embeddings"]
if embeddings is not None and not isinstance(embeddings, list):
embeddings = list(embeddings)
results.append(
Vector(
id=row["uuid"] or str(row["id"]),
embeddings=embeddings,
metadata=dict(row["metadata"] or {}),
)
)
return results
def search_similar_by_ids(
self,
vector_ids: list[str],
brain_id: str,
store: str,
min_similarity: float,
limit: int = 10,
) -> dict[str, list[Vector]]:
if not vector_ids:
return {}
self._ensure_store(store, brain_id)
table = _table_name(store)
int_ids = [_to_int_id(i) for i in vector_ids]
with self._connection(brain_id) as conn:
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
cur.execute(
f"""
SELECT id, uuid, embeddings
FROM {table}
WHERE id = ANY(%s)
""",
(int_ids,),
)
origin_rows = cur.fetchall()
embeddings_by_id: dict[int, list[float]] = {}
for row in origin_rows:
embedding = row["embeddings"]
if embedding is None:
continue
if not isinstance(embedding, list):
embedding = list(embedding)
embeddings_by_id[row["id"]] = embedding
out: dict[str, list[Vector]] = {str(v_id): [] for v_id in vector_ids}
if not embeddings_by_id:
return out
fetch_limit = max(limit * 5, limit) + 1
with self._connection(brain_id) as conn:
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
for origin_id, embedding in embeddings_by_id.items():
cur.execute(
f"""
SELECT id, uuid, metadata,
(embeddings <=> %s::vector) AS distance
FROM {table}
ORDER BY embeddings <=> %s::vector
LIMIT %s
""",
(embedding, embedding, fetch_limit),
)
rows = cur.fetchall()
collected: list[Vector] = []
for row in rows:
if row["id"] == origin_id:
continue
distance = float(row["distance"])
similarity = 1.0 - distance
if similarity < min_similarity:
continue
collected.append(
Vector(
id=str(row["id"]),
metadata=dict(row["metadata"] or {}),
distance=distance,
)
)
if len(collected) >= limit:
break
out[str(origin_id)] = collected
return out
def remove_vectors(self, ids: list[str], store: str, brain_id: str) -> None:
if not ids:
return
self._ensure_store(store, brain_id)
table = _table_name(store)
int_ids = [_to_int_id(i) for i in ids]
try:
with self._connection(brain_id) as conn:
with conn.cursor() as cur:
cur.execute(
f"""
DELETE FROM {table}
WHERE id = ANY(%s)
""",
(int_ids,),
)
conn.commit()
except Exception as exc:
print(f"[PostgreSQLVectorStore] Failed to remove vectors from {store}: {exc}")
def list_vectors(
self,
store: str,
brain_id: str,
limit: int = 10,
skip: int = 0,
include_embeddings: bool = False,
) -> tuple[list[Vector], int]:
self._ensure_store(store, brain_id)
table = _table_name(store)
embedding_col = ", embeddings" if include_embeddings else ""
with self._connection(brain_id) as conn:
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
cur.execute(f"SELECT COUNT(*) AS total FROM {table}")
total = int(cur.fetchone()["total"])
cur.execute(
f"""
SELECT id, uuid, metadata{embedding_col}
FROM {table}
ORDER BY id
LIMIT %s OFFSET %s
""",
(limit, skip),
)
rows = cur.fetchall()
results: list[Vector] = []
for row in rows:
embeddings = None
if include_embeddings:
embeddings = row.get("embeddings")
if embeddings is not None and not isinstance(embeddings, list):
embeddings = list(embeddings)
results.append(
Vector(
id=row["uuid"] or str(row["id"]),
embeddings=embeddings,
metadata=dict(row["metadata"] or {}),
)
)
return results, total
_postgresql_vector_store_client: Optional[PostgreSQLVectorStoreClient] = None
def get_postgresql_vector_store_client() -> PostgreSQLVectorStoreClient:
global _postgresql_vector_store_client
if _postgresql_vector_store_client is None:
_postgresql_vector_store_client = PostgreSQLVectorStoreClient()
return _postgresql_vector_store_client
|