vectordb-bench 0.0.12__py3-none-any.whl → 0.0.14__py3-none-any.whl

vectordb_bench/backend/clients/pgvecto_rs/pgvecto_rs.py

@@ -1,73 +1,138 @@
 """Wrapper around the Pgvecto.rs vector database over VectorDB"""
 
-import io
 import logging
+import pprint
 from contextlib import contextmanager
-from typing import Any
-import pandas as pd
-import psycopg2
-import psycopg2.extras
+from typing import Any, Generator, Optional, Tuple
 
-from ..api import VectorDB, DBCaseConfig
+import numpy as np
+import psycopg
+from psycopg import Connection, Cursor, sql
+from pgvecto_rs.psycopg import register_vector
+
+from ..api import VectorDB
+from .config import PgVectoRSConfig, PgVectoRSIndexConfig
 
 log = logging.getLogger(__name__)
 
+
 class PgVectoRS(VectorDB):
-    """Use SQLAlchemy instructions"""
+    """Use psycopg instructions"""
+
+    conn: psycopg.Connection[Any] | None = None
+    cursor: psycopg.Cursor[Any] | None = None
+    _unfiltered_search: sql.Composed
+    _filtered_search: sql.Composed
 
     def __init__(
         self,
         dim: int,
-        db_config: dict,
-        db_case_config: DBCaseConfig,
-        collection_name: str = "PgVectorCollection",
+        db_config: PgVectoRSConfig,
+        db_case_config: PgVectoRSIndexConfig,
+        collection_name: str = "PgVectoRSCollection",
         drop_old: bool = False,
         **kwargs,
     ):
+
+        self.name = "PgVectorRS"
         self.db_config = db_config
         self.case_config = db_case_config
         self.table_name = collection_name
         self.dim = dim
 
-        self._index_name = "pqvector_index"
+        self._index_name = "pgvectors_index"
         self._primary_field = "id"
         self._vector_field = "embedding"
 
         # construct basic units
-        self.conn = psycopg2.connect(**self.db_config)
-        self.conn.autocommit = False
-        self.cursor = self.conn.cursor()
+        self.conn, self.cursor = self._create_connection(**self.db_config)
 
-        # create vector extension
-        self.cursor.execute("CREATE EXTENSION IF NOT EXISTS vectors")
-        self.conn.commit()
+        log.info(f"{self.name} config values: {self.db_config}\n{self.case_config}")
+        if not any(
+            (
+                self.case_config.create_index_before_load,
+                self.case_config.create_index_after_load,
+            )
+        ):
+            err = f"{self.name} config must create an index using create_index_before_load or create_index_after_load"
+            log.error(err)
+            raise RuntimeError(
+                f"{err}\n{pprint.pformat(self.db_config)}\n{pprint.pformat(self.case_config)}"
+            )
 
         if drop_old:
             log.info(f"Pgvecto.rs client drop table : {self.table_name}")
             self._drop_index()
             self._drop_table()
             self._create_table(dim)
-            self._create_index()
+            if self.case_config.create_index_before_load:
+                self._create_index()
 
         self.cursor.close()
         self.conn.close()
         self.cursor = None
         self.conn = None
 
+    @staticmethod
+    def _create_connection(**kwargs) -> Tuple[Connection, Cursor]:
+        conn = psycopg.connect(**kwargs)
+
+        # create vector extension
+        conn.execute("CREATE EXTENSION IF NOT EXISTS vectors")
+        conn.commit()
+        register_vector(conn)
+
+        conn.autocommit = False
+        cursor = conn.cursor()
+
+        assert conn is not None, "Connection is not initialized"
+        assert cursor is not None, "Cursor is not initialized"
+
+        return conn, cursor
+
     @contextmanager
-    def init(self) -> None:
+    def init(self) -> Generator[None, None, None]:
         """
         Examples:
             >>> with self.init():
             >>>     self.insert_embeddings()
             >>>     self.search_embedding()
         """
-        self.conn = psycopg2.connect(**self.db_config)
-        self.conn.autocommit = False
-        self.cursor = self.conn.cursor()
-        self.cursor.execute('SET search_path = "$user", public, vectors')
+
+        self.conn, self.cursor = self._create_connection(**self.db_config)
+
+        # index configuration may have commands defined that we should set during each client session
+        session_options = self.case_config.session_param()
+
+        for key, val in session_options.items():
+            command = sql.SQL("SET {setting_name} " + "= {val};").format(
+                setting_name=sql.Identifier(key),
+                val=val,
+            )
+            log.debug(command.as_string(self.cursor))
+            self.cursor.execute(command)
         self.conn.commit()
 
+        self._filtered_search = sql.Composed(
+            [
+                sql.SQL(
+                    "SELECT id FROM public.{table_name} WHERE id >= %s ORDER BY embedding "
+                ).format(table_name=sql.Identifier(self.table_name)),
+                sql.SQL(self.case_config.search_param()["metric_fun_op"]),
+                sql.SQL(" %s::vector LIMIT %s::int"),
+            ]
+        )
+
+        self._unfiltered_search = sql.Composed(
+            [
+                sql.SQL(
+                    "SELECT id FROM public.{table_name} ORDER BY embedding "
+                ).format(table_name=sql.Identifier(self.table_name)),
+                sql.SQL(self.case_config.search_param()["metric_fun_op"]),
+                sql.SQL(" %s::vector LIMIT %s::int"),
+            ]
+        )
+
         try:
             yield
         finally:
@@ -79,42 +144,65 @@ class PgVectoRS(VectorDB):
     def _drop_table(self):
         assert self.conn is not None, "Connection is not initialized"
         assert self.cursor is not None, "Cursor is not initialized"
+        log.info(f"{self.name} client drop table : {self.table_name}")
 
-        self.cursor.execute(f'DROP TABLE IF EXISTS public."{self.table_name}"')
+        self.cursor.execute(
+            sql.SQL("DROP TABLE IF EXISTS public.{table_name}").format(
+                table_name=sql.Identifier(self.table_name)
+            )
+        )
         self.conn.commit()
 
     def ready_to_load(self):
         pass
 
     def optimize(self):
-        pass
+        self._post_insert()
 
-    def ready_to_search(self):
-        pass
+    def _post_insert(self):
+        log.info(f"{self.name} post insert before optimize")
+        if self.case_config.create_index_after_load:
+            self._drop_index()
+            self._create_index()
 
     def _drop_index(self):
         assert self.conn is not None, "Connection is not initialized"
         assert self.cursor is not None, "Cursor is not initialized"
+        log.info(f"{self.name} client drop index : {self._index_name}")
 
-        self.cursor.execute(f'DROP INDEX IF EXISTS "{self._index_name}"')
+        drop_index_sql = sql.SQL("DROP INDEX IF EXISTS {index_name}").format(
+            index_name=sql.Identifier(self._index_name)
+        )
+        log.debug(drop_index_sql.as_string(self.cursor))
+        self.cursor.execute(drop_index_sql)
         self.conn.commit()
 
     def _create_index(self):
         assert self.conn is not None, "Connection is not initialized"
         assert self.cursor is not None, "Cursor is not initialized"
+        log.info(f"{self.name} client create index : {self._index_name}")
 
         index_param = self.case_config.index_param()
 
+        index_create_sql = sql.SQL(
+            """
+            CREATE INDEX IF NOT EXISTS {index_name} ON public.{table_name} 
+            USING vectors (embedding {embedding_metric}) WITH (options = {index_options})
+            """
+        ).format(
+            index_name=sql.Identifier(self._index_name),
+            table_name=sql.Identifier(self.table_name),
+            embedding_metric=sql.Identifier(index_param["metric"]),
+            index_options=index_param["options"],
+        )
         try:
-            # create table
-            self.cursor.execute(
-                f'CREATE INDEX IF NOT EXISTS {self._index_name} ON public."{self.table_name}" \
-                    USING vectors (embedding {index_param["metric"]}) WITH (options = $${index_param["options"]}$$);'
-            )
+            log.debug(index_create_sql.as_string(self.cursor))
+            self.cursor.execute(index_create_sql)
             self.conn.commit()
         except Exception as e:
             log.warning(
-                f"Failed to create pgvecto.rs table: {self.table_name} error: {e}"
+                f"Failed to create pgvecto.rs index {self._index_name} \
+                    at table {self.table_name} error: {e}"
             )
             raise e from None
 
@@ -122,12 +210,18 @@ class PgVectoRS(VectorDB):
         assert self.conn is not None, "Connection is not initialized"
         assert self.cursor is not None, "Cursor is not initialized"
 
+        table_create_sql = sql.SQL(
+            """
+            CREATE TABLE IF NOT EXISTS public.{table_name}
+            (id BIGINT PRIMARY KEY, embedding vector({dim}))
+            """
+        ).format(
+            table_name=sql.Identifier(self.table_name),
+            dim=dim,
+        )
         try:
             # create table
-            self.cursor.execute(
-                f'CREATE TABLE IF NOT EXISTS public."{self.table_name}" \
-                    (id Integer PRIMARY KEY, embedding vector({dim}));'
-            )
+            self.cursor.execute(table_create_sql)
             self.conn.commit()
         except Exception as e:
             log.warning(
@@ -140,7 +234,7 @@ class PgVectoRS(VectorDB):
         embeddings: list[list[float]],
         metadata: list[int],
         **kwargs: Any,
-    ) -> (int, Exception):
+    ) -> Tuple[int, Optional[Exception]]:
         assert self.conn is not None, "Connection is not initialized"
         assert self.cursor is not None, "Cursor is not initialized"
 
@@ -148,19 +242,27 @@ class PgVectoRS(VectorDB):
         assert self.cursor is not None, "Cursor is not initialized"
 
         try:
-            items = {
-                "id": metadata,
-                "embedding": embeddings
-            }
-            df = pd.DataFrame(items)
-            csv_buffer = io.StringIO()
-            df.to_csv(csv_buffer, index=False, header=False)
-            csv_buffer.seek(0)
-            self.cursor.copy_expert(f"COPY public.\"{self.table_name}\" FROM STDIN WITH (FORMAT CSV)", csv_buffer)
+            metadata_arr = np.array(metadata)
+            embeddings_arr = np.array(embeddings)
+
+            with self.cursor.copy(
+                sql.SQL("COPY public.{table_name} FROM STDIN (FORMAT BINARY)").format(
+                    table_name=sql.Identifier(self.table_name)
+                )
+            ) as copy:
+                copy.set_types(["bigint", "vector"])
+                for i, row in enumerate(metadata_arr):
+                    copy.write_row((row, embeddings_arr[i]))
             self.conn.commit()
+
+            if kwargs.get("last_batch"):
+                self._post_insert()
+
             return len(metadata), None
         except Exception as e:
-            log.warning(f"Failed to insert data into pgvecto.rs table ({self.table_name}), error: {e}")
+            log.warning(
+                f"Failed to insert data into pgvecto.rs table ({self.table_name}), error: {e}"
+            )
             return 0, e
 
     def search_embedding(
@@ -173,20 +275,18 @@ class PgVectoRS(VectorDB):
         assert self.conn is not None, "Connection is not initialized"
         assert self.cursor is not None, "Cursor is not initialized"
 
-        search_param = self.case_config.search_param()
+        q = np.asarray(query)
 
         if filters:
+            log.debug(self._filtered_search.as_string(self.cursor))
             gt = filters.get("id")
-            self.cursor.execute(
-                f"SELECT id FROM (SELECT * FROM public.\"{self.table_name}\" ORDER BY embedding \
-                    {search_param['metrics_op']} '{query}' LIMIT {k}) AS X WHERE id > {gt} ;"
+            result = self.cursor.execute(
+                self._filtered_search, (gt, q, k), prepare=True, binary=True
             )
         else:
-            self.cursor.execute(
-                f"SELECT id FROM public.\"{self.table_name}\" ORDER BY embedding \
-                    {search_param['metrics_op']} '{query}' LIMIT {k};"
+            log.debug(self._unfiltered_search.as_string(self.cursor))
+            result = self.cursor.execute(
+                self._unfiltered_search, (q, k), prepare=True, binary=True
             )
-        self.conn.commit()
-        result = self.cursor.fetchall()
 
-        return [int(i[0]) for i in result]
+        return [int(i[0]) for i in result.fetchall()]

vectordb_bench/backend/clients/pgvector/cli.py

@@ -10,6 +10,7 @@ from ....cli.cli import (
     IVFFlatTypedDict,
     cli,
     click_parameter_decorators_from_typed_dict,
+    get_custom_case_config,
     run,
 )
 from vectordb_bench.backend.clients import DB
@@ -56,7 +57,15 @@ class PgVectorTypedDict(CommonTypedDict):
             required=False,
         ),
     ]
-
+    quantization_type: Annotated[
+        Optional[str],
+        click.option(
+            "--quantization-type",
+            type=click.Choice(["none", "halfvec"]),
+            help="quantization type for vectors",
+            required=False,
+        ),
+    ]
 
 class PgVectorIVFFlatTypedDict(PgVectorTypedDict, IVFFlatTypedDict):
     ...
@@ -69,6 +78,7 @@ def PgVectorIVFFlat(
 ):
     from .config import PgVectorConfig, PgVectorIVFFlatConfig
 
+    parameters["custom_case"] = get_custom_case_config(parameters)
     run(
         db=DB.PgVector,
         db_config=PgVectorConfig(
@@ -79,7 +89,10 @@ def PgVectorIVFFlat(
             db_name=parameters["db_name"],
         ),
         db_case_config=PgVectorIVFFlatConfig(
-            metric_type=None, lists=parameters["lists"], probes=parameters["probes"]
+            metric_type=None,
+            lists=parameters["lists"],
+            probes=parameters["probes"],
+            quantization_type=parameters["quantization_type"],
         ),
         **parameters,
     )
@@ -96,6 +109,7 @@ def PgVectorHNSW(
 ):
     from .config import PgVectorConfig, PgVectorHNSWConfig
 
+    parameters["custom_case"] = get_custom_case_config(parameters)
     run(
         db=DB.PgVector,
         db_config=PgVectorConfig(
@@ -111,6 +125,7 @@ def PgVectorHNSW(
             ef_search=parameters["ef_search"],
             maintenance_work_mem=parameters["maintenance_work_mem"],
             max_parallel_workers=parameters["max_parallel_workers"],
+            quantization_type=parameters["quantization_type"],
         ),
         **parameters,
     )

vectordb_bench/backend/clients/pgvector/config.py

@@ -59,11 +59,18 @@ class PgVectorIndexConfig(BaseModel, DBCaseConfig):
     create_index_after_load: bool = True
 
     def parse_metric(self) -> str:
-        if self.metric_type == MetricType.L2:
-            return "vector_l2_ops"
-        elif self.metric_type == MetricType.IP:
-            return "vector_ip_ops"
-        return "vector_cosine_ops"
+        if self.quantization_type == "halfvec":
+            if self.metric_type == MetricType.L2:
+                return "halfvec_l2_ops"
+            elif self.metric_type == MetricType.IP:
+                return "halfvec_ip_ops"
+            return "halfvec_cosine_ops"
+        else:
+            if self.metric_type == MetricType.L2:
+                return "vector_l2_ops"
+            elif self.metric_type == MetricType.IP:
+                return "vector_ip_ops"
+            return "vector_cosine_ops"
 
     def parse_metric_fun_op(self) -> LiteralString:
         if self.metric_type == MetricType.L2:
@@ -143,9 +150,12 @@ class PgVectorIVFFlatConfig(PgVectorIndexConfig):
     index: IndexType = IndexType.ES_IVFFlat
     maintenance_work_mem: Optional[str] = None
     max_parallel_workers: Optional[int] = None
+    quantization_type: Optional[str] = None
 
     def index_param(self) -> PgVectorIndexParam:
         index_parameters = {"lists": self.lists}
+        if self.quantization_type == "none":
+            self.quantization_type = None
         return {
             "metric": self.parse_metric(),
             "index_type": self.index.value,
@@ -154,6 +164,7 @@ class PgVectorIVFFlatConfig(PgVectorIndexConfig):
             ),
             "maintenance_work_mem": self.maintenance_work_mem,
             "max_parallel_workers": self.max_parallel_workers,
+            "quantization_type": self.quantization_type,
         }
 
     def search_param(self) -> PgVectorSearchParam:
@@ -183,9 +194,12 @@ class PgVectorHNSWConfig(PgVectorIndexConfig):
     index: IndexType = IndexType.ES_HNSW
     maintenance_work_mem: Optional[str] = None
     max_parallel_workers: Optional[int] = None
+    quantization_type: Optional[str] = None
 
     def index_param(self) -> PgVectorIndexParam:
         index_parameters = {"m": self.m, "ef_construction": self.ef_construction}
+        if self.quantization_type == "none":
+            self.quantization_type = None
         return {
             "metric": self.parse_metric(),
             "index_type": self.index.value,
@@ -194,6 +208,7 @@ class PgVectorHNSWConfig(PgVectorIndexConfig):
             ),
             "maintenance_work_mem": self.maintenance_work_mem,
             "max_parallel_workers": self.max_parallel_workers,
+            "quantization_type": self.quantization_type,
         }
 
     def search_param(self) -> PgVectorSearchParam:

vectordb_bench/backend/clients/pgvector/pgvector.py

@@ -22,7 +22,7 @@ class PgVector(VectorDB):
     conn: psycopg.Connection[Any] | None = None
     cursor: psycopg.Cursor[Any] | None = None
 
-    # TODO add filters support
+    _filtered_search: sql.Composed
     _unfiltered_search: sql.Composed
 
     def __init__(
@@ -112,15 +112,63 @@ class PgVector(VectorDB):
                 self.cursor.execute(command)
             self.conn.commit()
 
-        self._unfiltered_search = sql.Composed(
-            [
-                sql.SQL("SELECT id FROM public.{} ORDER BY embedding ").format(
-                    sql.Identifier(self.table_name)
-                ),
-                sql.SQL(self.case_config.search_param()["metric_fun_op"]),
-                sql.SQL(" %s::vector LIMIT %s::int"),
-            ]
-        )
+        index_param = self.case_config.index_param()
+        # The following sections assume that the quantization_type value matches the quantization function name
+        if index_param["quantization_type"] != None:
+            self._filtered_search = sql.Composed(
+                [
+                    sql.SQL(
+                        "SELECT id FROM public.{table_name} WHERE id >= %s ORDER BY embedding::{quantization_type}({dim}) "
+                    ).format(
+                        table_name=sql.Identifier(self.table_name),
+                        quantization_type=sql.SQL(index_param["quantization_type"]),
+                        dim=sql.Literal(self.dim),
+                    ),
+                    sql.SQL(self.case_config.search_param()["metric_fun_op"]),
+                    sql.SQL(" %s::{quantization_type}({dim}) LIMIT %s::int").format(
+                        quantization_type=sql.SQL(index_param["quantization_type"]),
+                        dim=sql.Literal(self.dim),
+                    ),
+                ]
+            )
+        else:
+            self._filtered_search = sql.Composed(
+                [
+                    sql.SQL(
+                        "SELECT id FROM public.{table_name} WHERE id >= %s ORDER BY embedding "
+                        ).format(table_name=sql.Identifier(self.table_name)),
+                    sql.SQL(self.case_config.search_param()["metric_fun_op"]),
+                    sql.SQL(" %s::vector LIMIT %s::int"),
+                ]
+            )
+
+        if index_param["quantization_type"] != None:
+            self._unfiltered_search = sql.Composed(
+                [
+                    sql.SQL(
+                        "SELECT id FROM public.{table_name} ORDER BY embedding::{quantization_type}({dim}) "
+                    ).format(
+                        table_name=sql.Identifier(self.table_name),
+                        quantization_type=sql.SQL(index_param["quantization_type"]),
+                        dim=sql.Literal(self.dim),
+                    ),
+                    sql.SQL(self.case_config.search_param()["metric_fun_op"]),
+                    sql.SQL(" %s::{quantization_type}({dim}) LIMIT %s::int").format(
+                        quantization_type=sql.SQL(index_param["quantization_type"]),
+                        dim=sql.Literal(self.dim),
+                    ),
+                ]
+            )
+        else:
+            self._unfiltered_search = sql.Composed(
+                [
+                    sql.SQL("SELECT id FROM public.{} ORDER BY embedding ").format(
+                        sql.Identifier(self.table_name)
+                    ),
+                    sql.SQL(self.case_config.search_param()["metric_fun_op"]),
+                    sql.SQL(" %s::vector LIMIT %s::int"),
+                ]
+            )
 
         try:
             yield
@@ -255,17 +303,34 @@ class PgVector(VectorDB):
         else:
             with_clause = sql.Composed(())
 
-        index_create_sql = sql.SQL(
-            """
-            CREATE INDEX IF NOT EXISTS {index_name} ON public.{table_name} 
-            USING {index_type} (embedding {embedding_metric})
-            """
-        ).format(
-            index_name=sql.Identifier(self._index_name),
-            table_name=sql.Identifier(self.table_name),
-            index_type=sql.Identifier(index_param["index_type"]),
-            embedding_metric=sql.Identifier(index_param["metric"]),
-        )
+        if index_param["quantization_type"] != None:
+            index_create_sql = sql.SQL(
+                """
+                CREATE INDEX IF NOT EXISTS {index_name} ON public.{table_name} 
+                USING {index_type} ((embedding::{quantization_type}({dim})) {embedding_metric})
+                """
+            ).format(
+                index_name=sql.Identifier(self._index_name),
+                table_name=sql.Identifier(self.table_name),
+                index_type=sql.Identifier(index_param["index_type"]),
+                # This assumes that the quantization_type value matches the quantization function name
+                quantization_type=sql.SQL(index_param["quantization_type"]),
+                dim=self.dim,
+                embedding_metric=sql.Identifier(index_param["metric"]),
+            )
+        else:
+            index_create_sql = sql.SQL(
+                """
+                CREATE INDEX IF NOT EXISTS {index_name} ON public.{table_name} 
+                USING {index_type} (embedding {embedding_metric})
+                """
+            ).format(
+                index_name=sql.Identifier(self._index_name),
+                table_name=sql.Identifier(self.table_name),
+                index_type=sql.Identifier(index_param["index_type"]),
+                embedding_metric=sql.Identifier(index_param["metric"]),
+            )
+
         index_create_sql_with_with_clause = (
             index_create_sql + with_clause
         ).join(" ")
@@ -342,9 +407,14 @@ class PgVector(VectorDB):
         assert self.cursor is not None, "Cursor is not initialized"
 
         q = np.asarray(query)
-        # TODO add filters support
-        result = self.cursor.execute(
-            self._unfiltered_search, (q, k), prepare=True, binary=True
-        )
+        if filters:
+            gt = filters.get("id")
+            result = self.cursor.execute(
+                    self._filtered_search, (gt, q, k), prepare=True, binary=True
+                    )
+        else:
+            result = self.cursor.execute(
+                    self._unfiltered_search, (q, k), prepare=True, binary=True
+                    )
 
         return [int(i[0]) for i in result.fetchall()]