merge: ARCH-003 OceanBase vector adapter

This commit is contained in:
d-robotics 2026-09-02 11:10:05 +08:00
commit 92d098cc25
17 changed files with 1021 additions and 9 deletions

View File

@ -31,6 +31,14 @@ MOCHOW_TABLE=knowledge_chunks
MOCHOW_BATCH_SIZE=100
MOCHOW_SEARCH_EF=200
MOCHOW_TIMEOUT_SECONDS=30
OCEANBASE_VECTOR_URI=
OCEANBASE_VECTOR_USER=
OCEANBASE_VECTOR_PASSWORD=
OCEANBASE_VECTOR_DATABASE=
OCEANBASE_VECTOR_TABLE=knowledge_chunks_v1
OCEANBASE_VECTOR_INDEX=vidx_knowledge_chunks_embedding
OCEANBASE_VECTOR_BATCH_SIZE=100
OCEANBASE_VECTOR_TIMEOUT_SECONDS=30
EMBEDDING_PROVIDER=qianfan
EMBEDDING_MODEL=bge-large-zh

View File

@ -97,6 +97,9 @@ demo providers without changing PostgreSQL canonical ownership. Until the runtim
cutover task is merged, the paragraphs above describe the current implementation.
ARCH-002 adds the selectable BOS adapter and safe runtime configuration documented
in `docs/development/BOS_SETUP.md`; local storage remains the default until ARCH-005.
ARCH-003 adds the selectable OceanBase Vector adapter and pre-provisioned cosine
HNSW schema documented in `docs/development/OCEANBASE_VECTOR_SETUP.md`; Mochow
remains the runtime default until ARCH-005.
DEMO-007 adds query embedding and profile-scoped Dense Top-N retrieval. The
default is Top 20 with a configurable Mochow HNSW `ef`; the in-memory fake uses

View File

@ -23,7 +23,11 @@ from ekp_object_storage import ObjectStorage, create_bos_object_storage
from ekp_object_storage.local import LocalObjectStorage
from ekp_parser import ParserRouter
from ekp_reranker import QianfanRerankerProvider, RerankerProvider
from ekp_vector_store import MochowVectorStore, VectorStore, create_mochow_vector_store
from ekp_vector_store import (
VectorStore,
create_mochow_vector_store,
create_oceanbase_vector_store,
)
BGE_LARGE_ZH_DIMENSIONS = 1024
DEMO_CHUNK_STRATEGY_VERSION = "structure_v1"
@ -85,8 +89,29 @@ def create_object_storage(settings: Settings) -> ObjectStorage:
)
def create_vector_store(settings: Settings) -> MochowVectorStore:
"""Create the Mochow adapter for an explicitly configured database/table."""
def create_vector_store(settings: Settings) -> VectorStore:
"""Create the selected vector adapter from explicit, secret-safe settings."""
if settings.vector_store_provider == "oceanbase":
if settings.oceanbase_vector_uri is None:
raise ProviderConfigurationError("OCEANBASE_VECTOR_URI is required")
if settings.oceanbase_vector_user is None:
raise ProviderConfigurationError("OCEANBASE_VECTOR_USER is required")
if settings.oceanbase_vector_password is None:
raise ProviderConfigurationError("OCEANBASE_VECTOR_PASSWORD is required")
if settings.oceanbase_vector_database is None:
raise ProviderConfigurationError("OCEANBASE_VECTOR_DATABASE is required")
return create_oceanbase_vector_store(
uri=settings.oceanbase_vector_uri,
user=settings.oceanbase_vector_user,
password=settings.oceanbase_vector_password.get_secret_value(),
database_name=settings.oceanbase_vector_database,
table_name=settings.oceanbase_vector_table,
index_name=settings.oceanbase_vector_index,
vector_dimension=BGE_LARGE_ZH_DIMENSIONS,
batch_size=settings.oceanbase_vector_batch_size,
timeout_seconds=settings.oceanbase_vector_timeout_seconds,
)
if settings.baidu_mochow_api_key is None:
raise ProviderConfigurationError("BAIDU_MOCHOW_API_KEY is required")
if not settings.baidu_mochow_endpoint:

View File

@ -0,0 +1,30 @@
-- ARCH-003 rebuildable OceanBase Vector index for bge-large-zh (1024 dimensions).
-- PostgreSQL remains canonical; this table may be dropped and rebuilt from canonical chunks.
-- Requires OceanBase 4.6.0 or newer for explicit ANN scalar pre-filtering.
CREATE TABLE IF NOT EXISTS knowledge_chunks_v1 (
index_record_id VARCHAR(36) NOT NULL,
chunk_id VARCHAR(36) NOT NULL,
document_id VARCHAR(36) NOT NULL,
filename VARCHAR(255) NOT NULL,
page INT NULL,
page_end INT NULL,
section VARCHAR(1024) NULL,
chunk_index INT NOT NULL,
content MEDIUMTEXT NOT NULL,
embedding VECTOR(1024) NOT NULL,
embedding_model VARCHAR(128) NOT NULL,
index_profile_id VARCHAR(128) NOT NULL,
PRIMARY KEY (index_record_id),
KEY idx_knowledge_chunks_document_profile (document_id, index_profile_id),
KEY idx_knowledge_chunks_profile (index_profile_id),
VECTOR KEY vidx_knowledge_chunks_embedding (embedding)
WITH (
DISTANCE = COSINE,
TYPE = HNSW,
LIB = VSAG,
M = 16,
EF_CONSTRUCTION = 256,
EF_SEARCH = 200
) BLOCK_SIZE 16384
) ORGANIZATION = HEAP;

View File

@ -364,7 +364,8 @@ DEMO-002 Document 数据模型 + 本地文件存储
DEMO-003 PDF/DOCX/TXT/MD 上传
DEMO-004 Parser + Cleaner
DEMO-005 Chunk Pipeline
DEMO-006 bge-large-zh Embedding + 初始 Mochow 入库(将由 ARCH-003 迁移)
DEMO-006 bge-large-zh Embedding + 初始 Mochow 入库(ARCH-003 已提供 OceanBase Adapter,
ARCH-005 切换运行时)
DEMO-007 Dense Retrieval
DEMO-008 Reranker + Source Result
DEMO-009 Document/Search REST API

View File

@ -0,0 +1,72 @@
# ARCH-003 OceanBase Vector 配置与索引契约
ARCH-003 在既有 `VectorStore` 契约后新增 OceanBase Vector Adapter。PostgreSQL 仍保存
canonical Document、Chunk、Index Profile 和 Chunk Index State;OceanBase 表只保存可由这些
数据重建的检索行。运行时默认切换和 Mochow 配置清理留在 ARCH-005。
## 数据库前提
- OceanBase MySQL 租户版本至少为 4.6.0:1024 维 Cosine HNSW 本身可由更早版本支持,
但本项目要求使用 `VECTOR_INDEX(... pre_filter)` 显式保证 Profile 在 ANN 前过滤;
- 目标数据库和最小权限账号由受控部署流程预创建;
- Vector 内存、资源规格和 HNSW 参数由 DBA 在受控环境确认;
- 应用启动和请求路径不会创建、修改或删除表结构。
使用 [001_knowledge_chunks_v1.sql](../../deploy/oceanbase/001_knowledge_chunks_v1.sql)
在专用数据库中创建索引表。脚本固定:
```text
VECTOR(1024)
DISTANCE = COSINE
TYPE = HNSW
EF_CONSTRUCTION = 256
EF_SEARCH = 200
```
表的 `index_record_id` 是确定性的 Chunk/Profile 主键;`document_id + index_profile_id` 和
`index_profile_id` 是清理与 Profile 过滤的标量索引。表名或向量维度变化必须建立新表或
Shadow Profile,验证后再激活,禁止在生产索引上原地修改。
## 运行配置
配置只能由部署环境或 Secret Manager 注入,禁止把真实地址、用户名和密码写入 `.env`、
镜像、日志、测试快照或提交记录。
```text
VECTOR_STORE_PROVIDER=oceanbase
OCEANBASE_VECTOR_URI=<host:port>
OCEANBASE_VECTOR_USER=<运行时注入>
OCEANBASE_VECTOR_PASSWORD=<运行时注入>
OCEANBASE_VECTOR_DATABASE=<预创建数据库>
OCEANBASE_VECTOR_TABLE=knowledge_chunks_v1
OCEANBASE_VECTOR_INDEX=vidx_knowledge_chunks_embedding
OCEANBASE_VECTOR_BATCH_SIZE=100
OCEANBASE_VECTOR_TIMEOUT_SECONDS=30
```
- URI 只接受 `host:port`,不得包含 Scheme、用户信息、Path、Query 或 Fragment;
- Database/Table/Index 只接受安全标识符;
- Password 使用 `SecretStr`,Provider 异常只返回收敛后的错误,不回显 SQL、连接信息、
Chunk 内容或 SDK Response;
- `pyobvector` 只存在于 Adapter/Factory 内,业务服务不依赖其 Row、Engine 或距离函数类型。
## 写入、删除和检索语义
- Upsert 使用 `index_record_id` 主键,按可配置批次写入;
- 删除使用参数化的 `document_id + index_profile_id` 条件,不拼接用户输入;
- Dense Search 在同一 ANN 查询中使用 `VECTOR_INDEX(... pre_filter)`,保证
`index_profile_id` 在候选近邻搜索前过滤,并返回 Dense Top20;禁止依赖普通 `WHERE`
的自适应过滤语义,也禁止全局 TopK 后再过滤;
- OceanBase `cosine_distance` 是低值优先距离,Adapter 统一转换为
`score = 1 - cosine_distance`,保持 `VectorStore` 的高分优先契约;
- 返回 UUID、Profile、字段数、距离范围和排序不满足契约时失败关闭,不生成 Source Result。
默认测试使用 Fake Client,不连接 OceanBase。真实版本、DDL、权限、Cosine HNSW 召回、过滤
行为与 30~50 条脱敏 Golden Query 统一在 ARCH-006 的受控资源中验证,并且只允许使用轮换
后的凭据。
官方参考:
- [pyobvector 官方仓库](https://github.com/oceanbase/pyobvector)
- [OceanBase 创建向量索引](https://en.oceanbase.com/docs/common-oceanbase-database-10000000003075357)
- [OceanBase 向量索引查询 Hint](https://en.oceanbase.com/docs/common-oceanbase-database-10000000004311163)

View File

@ -2,11 +2,11 @@
当前阶段:Provider 架构迁移
已完成:DEMO-001~DEMO-009、ARCH-001、ARCH-002
已完成:DEMO-001~DEMO-009、ARCH-001、ARCH-002、ARCH-003
进行中:无
阻塞:无
下一任务:ARCH-003
下一任务:ARCH-004
Provider 架构迁移按 `ARCH-001 → ARCH-002 → ARCH-003 → ARCH-004 → ARCH-005 → ARCH-006`
执行;每个 Task 验证后独立推送并合并 `main`。全部迁移完成后再继续 DEMO-010。
@ -154,3 +154,17 @@ Provider 架构迁移按 `ARCH-001 → ARCH-002 → ARCH-003 → ARCH-004 → AR
Prefix 和超时均由环境注入。Local 仍是测试/开发及 ARCH-005 前的默认实现。
- 新增完全离线的 BOS Adapter、配置、Provider wiring 测试和中文配置说明;未使用内部配置
清单中的真实地址或凭据,真实 BOS 联调留在 ARCH-006。
### ARCH-003
- 新增官方 `pyobvector` 驱动的 `OceanBaseVectorStore`,覆盖批量 Upsert、按 Document/Profile
参数化删除和 Profile 条件下推的 Cosine ANN TopN;SDK 类型不进入业务服务。检索固定使用
OceanBase 4.6 的 `VECTOR_INDEX(... pre_filter)`,不依赖普通 `WHERE` 的自适应过滤,也不做
全局 TopK 后过滤。
- 将低值优先的 `cosine_distance` 转为 `score = 1 - distance`,并对向量维度、有限值、
非零范数、返回字段、UUID、Profile、距离范围和排序执行失败关闭校验。
- 新增 `VECTOR(1024)`、Cosine HNSW、Profile/Document 标量索引的预置 DDL;应用不自动执行
外部索引 DDL,PostgreSQL 继续保存 canonical 数据,OceanBase 表可重建。
- 新增 `mochow|oceanbase` 类型化配置、SecretStr 密码和运行时工厂;Mochow 保持 ARCH-005
前的默认实现。全部 Adapter、配置、Schema 和 wiring 测试均离线运行,真实数据库验收留在
ARCH-006。

View File

@ -4,7 +4,7 @@
- [x] ARCH-001 Architecture decision, PRD, Scope and execution baseline
- [x] ARCH-002 BOS ObjectStorage adapter and runtime configuration
- [ ] ARCH-003 OceanBase VectorStore adapter and index schema
- [x] ARCH-003 OceanBase VectorStore adapter and index schema
- [ ] ARCH-004 Qianfan `bce-reranker-base` alignment
- [ ] ARCH-005 Runtime cutover, shadow rebuild workflow and legacy configuration cleanup
- [ ] ARCH-006 Controlled live integration and RAG acceptance

View File

@ -42,7 +42,7 @@ class Settings(BaseSettings):
embedding_endpoint: str = "https://qianfan.baidubce.com/v2/embeddings"
baidu_qianfan_api_key: SecretStr | None = None
embedding_timeout_seconds: float = Field(default=30.0, gt=0)
vector_store_provider: Literal["mochow"] = "mochow"
vector_store_provider: Literal["mochow", "oceanbase"] = "mochow"
baidu_mochow_account: str = "root"
baidu_mochow_api_key: SecretStr | None = None
baidu_mochow_endpoint: str | None = None
@ -51,6 +51,14 @@ class Settings(BaseSettings):
mochow_batch_size: int = Field(default=100, ge=1, le=1000)
mochow_search_ef: int = Field(default=200, ge=1, le=10000)
mochow_timeout_seconds: float = Field(default=30.0, gt=0)
oceanbase_vector_uri: str | None = None
oceanbase_vector_user: str | None = None
oceanbase_vector_password: SecretStr | None = None
oceanbase_vector_database: str | None = None
oceanbase_vector_table: str = "knowledge_chunks_v1"
oceanbase_vector_index: str = "vidx_knowledge_chunks_embedding"
oceanbase_vector_batch_size: int = Field(default=100, ge=1, le=1000)
oceanbase_vector_timeout_seconds: float = Field(default=30.0, gt=0)
index_profile_id: str = "demo-bge-large-zh-structure-v1"
dense_top_n: int = Field(default=20, ge=1, le=64)
reranker_provider: Literal["qianfan"] = "qianfan"

View File

@ -68,7 +68,7 @@ class DocumentIndexingService:
self._index_profile = index_profile
def process(self, document_id: UUID) -> DocumentIndexingResult:
"""Index all chunks for one profile, compensating any partial Mochow write."""
"""Index all chunks for one profile, compensating any partial provider write."""
chunks = self._start_indexing(document_id)
upsert_attempted = False
try:

View File

@ -12,6 +12,12 @@ from ekp_vector_store.mochow import (
MochowVectorStoreError,
create_mochow_vector_store,
)
from ekp_vector_store.oceanbase import (
OceanBaseConfigurationError,
OceanBaseVectorStore,
OceanBaseVectorStoreError,
create_oceanbase_vector_store,
)
__all__ = [
"FakeVectorStore",
@ -19,7 +25,11 @@ __all__ = [
"DenseSearchRequest",
"MochowVectorStore",
"MochowVectorStoreError",
"OceanBaseConfigurationError",
"OceanBaseVectorStore",
"OceanBaseVectorStoreError",
"VectorRecord",
"VectorStore",
"create_mochow_vector_store",
"create_oceanbase_vector_store",
]

View File

@ -0,0 +1,403 @@
"""OceanBase Vector adapter isolated behind the provider-neutral VectorStore contract."""
import re
from collections.abc import Iterable
from math import isfinite
from numbers import Real
from typing import Any, Protocol
from urllib.parse import urlsplit
from uuid import UUID
from sqlalchemy import and_, column, text
from ekp_vector_store.base import DenseSearchCandidate, DenseSearchRequest, VectorRecord
OCEANBASE_IDENTIFIER = re.compile(r"^[A-Za-z][A-Za-z0-9_]{0,63}$")
PROFILE_ID = re.compile(r"^[A-Za-z0-9][A-Za-z0-9_.:-]{0,127}$")
SEARCH_OUTPUT_FIELDS = [
"index_record_id",
"chunk_id",
"document_id",
"filename",
"page",
"page_end",
"section",
"chunk_index",
"content",
"index_profile_id",
]
class OceanBaseConfigurationError(ValueError):
"""Raised when OceanBase connection or index settings are invalid."""
class OceanBaseVectorStoreError(RuntimeError):
"""Sanitized OceanBase failure without SQL, endpoints, content, or credentials."""
class _OceanBaseVectorClient(Protocol):
"""Narrow pyobvector client surface used by the EKP adapter."""
def upsert(self, table_name: str, data: list[dict[str, object]]) -> object:
"""Insert or replace a row batch by primary key."""
def delete(
self,
table_name: str,
*,
where_clause: list[object],
) -> object:
"""Delete rows matching bound SQLAlchemy predicates."""
def dense_search(
self,
table_name: str,
*,
index_name: str,
query_vector: tuple[float, ...],
index_profile_id: str,
top_n: int,
output_column_names: list[str],
) -> Iterable[object]:
"""Return explicitly pre-filtered ANN rows followed by their distance."""
class _PyObVectorClient:
"""Keep SDK operations and the required pre-filter SQL inside the adapter."""
def __init__(self, sdk_client: Any) -> None:
self._sdk_client = sdk_client
def upsert(self, table_name: str, data: list[dict[str, object]]) -> object:
"""Delegate bounded row writes to the official client."""
result: object = self._sdk_client.upsert(table_name, data)
return result
def delete(
self,
table_name: str,
*,
where_clause: list[object],
) -> object:
"""Delegate parameterized cleanup to the official client."""
result: object = self._sdk_client.delete(table_name, where_clause=where_clause)
return result
def dense_search(
self,
table_name: str,
*,
index_name: str,
query_vector: tuple[float, ...],
index_profile_id: str,
top_n: int,
output_column_names: list[str],
) -> Iterable[object]:
"""Execute cosine ANN with an explicit OceanBase pre-filter hint."""
statement = text(
_build_dense_search_sql(
table_name=table_name,
index_name=index_name,
output_column_names=output_column_names,
)
)
parameters = {
"query_vector": _serialize_vector(query_vector),
"index_profile_id": index_profile_id,
"top_n": top_n,
}
# Materialize rows before closing the connection; no SDK Result leaks upward.
with self._sdk_client.engine.connect() as connection:
rows: list[object] = list(connection.execute(statement, parameters))
return rows
class OceanBaseVectorStore:
"""Translate neutral vector records into one rebuildable OceanBase index table."""
def __init__(
self,
client: _OceanBaseVectorClient,
table_name: str,
*,
index_name: str = "vidx_knowledge_chunks_embedding",
vector_dimension: int = 1024,
batch_size: int = 100,
) -> None:
if not OCEANBASE_IDENTIFIER.fullmatch(table_name):
raise OceanBaseConfigurationError("invalid OceanBase vector table name")
if not OCEANBASE_IDENTIFIER.fullmatch(index_name):
raise OceanBaseConfigurationError("invalid OceanBase vector index name")
if vector_dimension <= 0:
raise OceanBaseConfigurationError("OceanBase vector dimension must be positive")
if not 1 <= batch_size <= 1000:
raise OceanBaseConfigurationError(
"OceanBase vector batch_size must be between 1 and 1000"
)
self._client = client
self._table_name = table_name
self._index_name = index_name
self._vector_dimension = vector_dimension
self._batch_size = batch_size
def upsert(self, records: tuple[VectorRecord, ...]) -> None:
"""Validate and upsert records in bounded pyobvector batches."""
rows = [self._to_row(record) for record in records]
for start in range(0, len(rows), self._batch_size):
try:
self._client.upsert(
self._table_name,
rows[start : start + self._batch_size],
)
except Exception as exc:
raise OceanBaseVectorStoreError("OceanBase vector upsert failed") from exc
def delete_document(self, document_id: UUID, index_profile_id: str) -> None:
"""Delete only one document/profile using bound scalar predicates."""
_validate_profile_id(index_profile_id)
predicate = and_(
column("document_id") == str(document_id),
column("index_profile_id") == index_profile_id,
)
try:
self._client.delete(
self._table_name,
where_clause=[predicate],
)
except Exception as exc:
raise OceanBaseVectorStoreError("OceanBase vector cleanup failed") from exc
def dense_search(self, request: DenseSearchRequest) -> tuple[DenseSearchCandidate, ...]:
"""Run profile-scoped cosine ANN and convert distance to high-is-better score."""
_validate_profile_id(request.index_profile_id)
self._validate_vector(request.query_vector)
try:
rows = self._client.dense_search(
self._table_name,
index_name=self._index_name,
query_vector=request.query_vector,
index_profile_id=request.index_profile_id,
top_n=request.top_n,
output_column_names=list(SEARCH_OUTPUT_FIELDS),
)
candidates = tuple(
_to_candidate(row, expected_profile=request.index_profile_id) for row in rows
)
if len(candidates) > request.top_n:
raise ValueError("OceanBase returned more rows than requested")
if any(
earlier.score < later.score
for earlier, later in zip(candidates, candidates[1:], strict=False)
):
raise ValueError("OceanBase ANN rows are not ordered by ascending distance")
return candidates
except Exception as exc:
raise OceanBaseVectorStoreError("OceanBase dense search failed") from exc
def _to_row(self, record: VectorRecord) -> dict[str, object]:
"""Map one rebuildable row to the pre-provisioned OceanBase schema."""
_validate_profile_id(record.index_profile_id)
self._validate_vector(record.embedding)
return {
"index_record_id": str(record.index_record_id),
"chunk_id": str(record.chunk_id),
"document_id": str(record.document_id),
"filename": record.filename,
"page": record.page,
"page_end": record.page_end,
"section": record.section,
"chunk_index": record.chunk_index,
"content": record.content,
"embedding": list(record.embedding),
"embedding_model": record.embedding_model,
"index_profile_id": record.index_profile_id,
}
def _validate_vector(self, vector: tuple[float, ...]) -> None:
"""Reject dimensions and values that cannot satisfy the cosine index contract."""
if len(vector) != self._vector_dimension:
raise OceanBaseConfigurationError("OceanBase vector dimension mismatch")
if any(not isfinite(value) for value in vector):
raise OceanBaseConfigurationError("OceanBase vector values must be finite")
if not any(value != 0 for value in vector):
raise OceanBaseConfigurationError("OceanBase cosine vector norm must be positive")
def create_oceanbase_vector_store(
*,
uri: str,
user: str,
password: str,
database_name: str,
table_name: str,
index_name: str = "vidx_knowledge_chunks_embedding",
vector_dimension: int = 1024,
batch_size: int = 100,
timeout_seconds: float = 30.0,
) -> OceanBaseVectorStore:
"""Create the official pyobvector client from runtime-injected settings."""
_validate_uri(uri)
if not user or not password:
raise OceanBaseConfigurationError("OceanBase vector user and password are required")
if not OCEANBASE_IDENTIFIER.fullmatch(database_name):
raise OceanBaseConfigurationError("invalid OceanBase vector database name")
if timeout_seconds <= 0:
raise OceanBaseConfigurationError("OceanBase timeout_seconds must be positive")
try:
from pyobvector import ObVecClient # type: ignore[import-untyped]
timeout = max(1, int(timeout_seconds))
sdk_client = ObVecClient(
uri=uri,
user=user,
password=password,
db_name=database_name,
pool_pre_ping=True,
connect_args={
"connect_timeout": timeout,
"read_timeout": timeout,
"write_timeout": timeout,
},
)
except Exception as exc:
raise OceanBaseVectorStoreError("OceanBase vector client initialization failed") from exc
_validate_oceanbase_version(sdk_client.ob_version.version_nums)
return OceanBaseVectorStore(
_PyObVectorClient(sdk_client),
table_name,
index_name=index_name,
vector_dimension=vector_dimension,
batch_size=batch_size,
)
def _to_candidate(row: object, *, expected_profile: str) -> DenseSearchCandidate:
"""Validate one pyobvector result tuple and normalize cosine distance to score."""
if isinstance(row, (str, bytes, dict)) or not isinstance(row, Iterable):
raise TypeError("OceanBase ANN row must be tuple-like")
values = tuple(row)
if len(values) != len(SEARCH_OUTPUT_FIELDS) + 1:
raise ValueError("OceanBase ANN row has an unexpected field count")
profile = _required_str(values[9], "index_profile_id")
if profile != expected_profile:
raise ValueError("OceanBase returned a row from another index profile")
distance = values[10]
if isinstance(distance, bool) or not isinstance(distance, Real):
raise ValueError("OceanBase cosine distance must be numeric")
numeric_distance = float(distance)
if not isfinite(numeric_distance) or not -1e-6 <= numeric_distance <= 2.000001:
raise ValueError("OceanBase cosine distance is outside the supported range")
score = max(-1.0, min(1.0, 1.0 - numeric_distance))
return DenseSearchCandidate(
index_record_id=UUID(_required_str(values[0], "index_record_id")),
chunk_id=UUID(_required_str(values[1], "chunk_id")),
document_id=UUID(_required_str(values[2], "document_id")),
filename=_required_str(values[3], "filename"),
page=_optional_int(values[4]),
page_end=_optional_int(values[5]),
section=_optional_str(values[6]),
chunk_index=_required_int(values[7], "chunk_index"),
content=_required_str(values[8], "content"),
score=score,
index_profile_id=profile,
)
def _validate_profile_id(profile_id: str) -> None:
"""Apply the canonical IndexProfile identifier subset at the provider edge."""
if not PROFILE_ID.fullmatch(profile_id):
raise OceanBaseConfigurationError("invalid OceanBase index profile ID")
def _validate_uri(uri: str) -> None:
"""Accept only a host:port target without credentials, paths, or query data."""
if "://" in uri:
raise OceanBaseConfigurationError("invalid OceanBase vector URI")
try:
parsed = urlsplit(f"//{uri}")
port = parsed.port
except ValueError as exc:
raise OceanBaseConfigurationError("invalid OceanBase vector URI") from exc
if (
not parsed.hostname
or port is None
or parsed.username is not None
or parsed.password is not None
or parsed.path
or parsed.query
or parsed.fragment
):
raise OceanBaseConfigurationError("invalid OceanBase vector URI")
def _validate_oceanbase_version(version_nums: object) -> None:
"""Require the first release with an explicit vector pre-filter hint."""
if (
not isinstance(version_nums, (list, tuple))
or len(version_nums) != 4
or any(isinstance(part, bool) or not isinstance(part, int) for part in version_nums)
):
raise OceanBaseVectorStoreError("OceanBase version validation failed")
if tuple(version_nums) < (4, 6, 0, 0):
raise OceanBaseConfigurationError(
"OceanBase 4.6.0 or newer is required for vector pre-filtering"
)
def _build_dense_search_sql(
*,
table_name: str,
index_name: str,
output_column_names: list[str],
) -> str:
"""Build a fixed-identifier query whose values remain bind parameters."""
identifiers = [table_name, index_name, *output_column_names]
if any(not OCEANBASE_IDENTIFIER.fullmatch(value) for value in identifiers):
raise OceanBaseConfigurationError("invalid OceanBase dense-search identifier")
fields = ", ".join(output_column_names)
return (
f"SELECT /*+ VECTOR_INDEX({table_name} {index_name} pre_filter) */ "
f"{fields}, cosine_distance(embedding, :query_vector) AS distance "
f"FROM {table_name} "
"WHERE index_profile_id = :index_profile_id "
"ORDER BY cosine_distance(embedding, :query_vector) APPROXIMATE "
"LIMIT :top_n"
)
def _serialize_vector(vector: tuple[float, ...]) -> str:
"""Serialize already-validated finite floats for OceanBase's VECTOR parameter."""
return "[" + ",".join(format(value, ".17g") for value in vector) + "]"
def _required_str(value: object, name: str) -> str:
if not isinstance(value, str) or not value:
raise ValueError(f"OceanBase field {name} must be non-empty text")
return value
def _required_int(value: object, name: str) -> int:
if isinstance(value, bool) or not isinstance(value, int):
raise ValueError(f"OceanBase field {name} must be an integer")
return value
def _optional_int(value: object) -> int | None:
if value is None:
return None
if isinstance(value, bool) or not isinstance(value, int):
raise ValueError("OceanBase nullable integer metadata must be an integer")
return value
def _optional_str(value: object) -> str | None:
if value is None:
return None
if not isinstance(value, str):
raise ValueError("OceanBase nullable text metadata must be text")
return value

View File

@ -20,6 +20,7 @@ dependencies = [
"python-docx>=1.2,<2",
"sqlalchemy>=2.0,<3",
"uvicorn[standard]>=0.34,<1",
"pyobvector>=0.2.28,<0.3",
]
[dependency-groups]

View File

@ -36,6 +36,14 @@ def test_settings_have_demo_defaults(isolated_settings_environment: None) -> Non
assert settings.mochow_table == "knowledge_chunks"
assert settings.mochow_search_ef == 200
assert settings.mochow_timeout_seconds == 30
assert settings.oceanbase_vector_uri is None
assert settings.oceanbase_vector_user is None
assert settings.oceanbase_vector_password is None
assert settings.oceanbase_vector_database is None
assert settings.oceanbase_vector_table == "knowledge_chunks_v1"
assert settings.oceanbase_vector_index == "vidx_knowledge_chunks_embedding"
assert settings.oceanbase_vector_batch_size == 100
assert settings.oceanbase_vector_timeout_seconds == 30
assert settings.index_profile_id == "demo-bge-large-zh-structure-v1"
assert settings.dense_top_n == 20
assert settings.reranker_provider == "qianfan"
@ -70,6 +78,7 @@ def test_settings_redact_provider_secrets(isolated_settings_environment: None) -
baidu_mochow_api_key=SecretStr("mochow-secret"),
bos_access_key_id=SecretStr("bos-access-key"),
bos_secret_access_key=SecretStr("bos-secret-key"),
oceanbase_vector_password=SecretStr("oceanbase-secret"),
reranker_api_key=SecretStr("reranker-secret"),
)
@ -78,4 +87,5 @@ def test_settings_redact_provider_secrets(isolated_settings_environment: None) -
assert "mochow-secret" not in rendered
assert "bos-access-key" not in rendered
assert "bos-secret-key" not in rendered
assert "oceanbase-secret" not in rendered
assert "reranker-secret" not in rendered

View File

@ -0,0 +1,310 @@
"""Offline tests for neutral-record translation into pyobvector operations."""
from dataclasses import replace
from pathlib import Path
from typing import Any, cast
from uuid import UUID, uuid4
import pytest
from ekp_vector_store import DenseSearchRequest, OceanBaseVectorStore, VectorRecord
from ekp_vector_store.oceanbase import (
OceanBaseConfigurationError,
OceanBaseVectorStoreError,
_build_dense_search_sql,
_PyObVectorClient,
_serialize_vector,
_validate_oceanbase_version,
)
class FakeOceanBaseClient:
"""Capture ObVecClient-shaped calls without a database connection."""
def __init__(self) -> None:
self.upsert_calls: list[tuple[str, list[dict[str, object]]]] = []
self.delete_calls: list[tuple[str, list[object]]] = []
self.search_calls: list[dict[str, object]] = []
self.search_rows: list[tuple[object, ...]] = []
self.failure: Exception | None = None
def upsert(self, table_name: str, data: list[dict[str, object]]) -> object:
self._raise_failure()
self.upsert_calls.append((table_name, data))
return object()
def delete(
self,
table_name: str,
*,
where_clause: list[object],
) -> object:
self._raise_failure()
self.delete_calls.append((table_name, where_clause))
return object()
def dense_search(
self,
table_name: str,
*,
index_name: str,
query_vector: tuple[float, ...],
index_profile_id: str,
top_n: int,
output_column_names: list[str],
) -> list[tuple[object, ...]]:
self._raise_failure()
self.search_calls.append(
{
"table_name": table_name,
"index_name": index_name,
"query_vector": query_vector,
"index_profile_id": index_profile_id,
"top_n": top_n,
"output_column_names": output_column_names,
}
)
return self.search_rows
def _raise_failure(self) -> None:
if self.failure is not None:
raise self.failure
class FakeConnection:
"""Capture SQLAlchemy text and bound values without executing vendor SQL."""
def __init__(self) -> None:
self.executions: list[tuple[str, dict[str, object]]] = []
def __enter__(self) -> "FakeConnection":
return self
def __exit__(self, *args: object) -> None:
return None
def execute(self, statement: object, parameters: dict[str, object]) -> list[object]:
self.executions.append((str(statement), parameters))
return []
class FakeEngine:
"""Return one inspectable fake connection."""
def __init__(self) -> None:
self.connection = FakeConnection()
def connect(self) -> FakeConnection:
return self.connection
class FakeSdkClient:
"""Provide the official client's engine shape for the SQL boundary test."""
def __init__(self) -> None:
self.engine = FakeEngine()
def _record(document_id: UUID, chunk_index: int) -> VectorRecord:
return VectorRecord(
index_record_id=uuid4(),
chunk_id=uuid4(),
document_id=document_id,
filename="policy.md",
page=1,
page_end=2,
section="产品 / 退款",
chunk_index=chunk_index,
content="退款周期为七天。",
embedding=(0.1, 0.2, 0.3),
embedding_model="fake",
index_profile_id="profile-v1",
)
def _search_row(record: VectorRecord, distance: float) -> tuple[object, ...]:
return (
str(record.index_record_id),
str(record.chunk_id),
str(record.document_id),
record.filename,
record.page,
record.page_end,
record.section,
record.chunk_index,
record.content,
record.index_profile_id,
distance,
)
def test_oceanbase_adapter_maps_and_batches_upserts() -> None:
client = FakeOceanBaseClient()
store = OceanBaseVectorStore(client, "knowledge_chunks_v1", vector_dimension=3, batch_size=2)
document_id = uuid4()
records = tuple(_record(document_id, index) for index in range(3))
store.upsert(records)
assert [len(data) for _, data in client.upsert_calls] == [2, 1]
first = client.upsert_calls[0][1][0]
assert first["index_record_id"] == str(records[0].index_record_id)
assert first["document_id"] == str(document_id)
assert first["embedding"] == [0.1, 0.2, 0.3]
assert first["index_profile_id"] == "profile-v1"
def test_oceanbase_delete_uses_bound_document_and_profile_predicate() -> None:
client = FakeOceanBaseClient()
store = OceanBaseVectorStore(client, "knowledge_chunks_v1", vector_dimension=3)
document_id = uuid4()
store.delete_document(document_id, "profile-v1")
table_name, clauses = client.delete_calls[0]
compiled = cast(Any, clauses[0]).compile()
assert table_name == "knowledge_chunks_v1"
assert (
str(compiled) == "document_id = :document_id_1 AND index_profile_id = :index_profile_id_1"
)
assert set(compiled.params.values()) == {str(document_id), "profile-v1"}
def test_oceanbase_dense_search_filters_profile_and_normalizes_cosine_distance() -> None:
client = FakeOceanBaseClient()
first = _record(uuid4(), 0)
second = _record(uuid4(), 1)
client.search_rows = [_search_row(first, 0.1), _search_row(second, 0.4)]
store = OceanBaseVectorStore(client, "knowledge_chunks_v1", vector_dimension=3)
candidates = store.dense_search(
DenseSearchRequest((0.3, 0.2, 0.1), index_profile_id="profile-v1", top_n=20)
)
call = client.search_calls[0]
assert call["table_name"] == "knowledge_chunks_v1"
assert call["index_name"] == "vidx_knowledge_chunks_embedding"
assert call["query_vector"] == (0.3, 0.2, 0.1)
assert call["index_profile_id"] == "profile-v1"
assert call["top_n"] == 20
assert [candidate.score for candidate in candidates] == pytest.approx([0.9, 0.6])
assert candidates[0].chunk_id == first.chunk_id
def test_oceanbase_rejects_wrong_dimension_and_zero_cosine_vector() -> None:
store = OceanBaseVectorStore(
FakeOceanBaseClient(),
"knowledge_chunks_v1",
vector_dimension=3,
)
with pytest.raises(OceanBaseConfigurationError, match="dimension mismatch"):
store.dense_search(DenseSearchRequest((0.1, 0.2), "profile-v1"))
with pytest.raises(OceanBaseConfigurationError, match="norm must be positive"):
store.upsert((replace(_record(uuid4(), 0), embedding=(0.0, 0.0, 0.0)),))
def test_oceanbase_sanitizes_malformed_or_failed_search() -> None:
client = FakeOceanBaseClient()
client.search_rows = [("malformed",)]
store = OceanBaseVectorStore(client, "knowledge_chunks_v1", vector_dimension=3)
with pytest.raises(OceanBaseVectorStoreError, match="dense search failed"):
store.dense_search(DenseSearchRequest((0.1, 0.2, 0.3), "profile-v1"))
client.failure = RuntimeError("private-host password=secret-value")
with pytest.raises(OceanBaseVectorStoreError, match="cleanup failed") as caught:
store.delete_document(uuid4(), "profile-v1")
assert "private-host" not in str(caught.value)
assert "secret-value" not in str(caught.value)
@pytest.mark.parametrize(
"uri",
["", "https://db.example:2881", "user:password@db.example:2881", "db.example"],
)
def test_oceanbase_factory_rejects_unsafe_uri_without_connecting(uri: str) -> None:
from ekp_vector_store.oceanbase import create_oceanbase_vector_store
with pytest.raises(OceanBaseConfigurationError, match="URI"):
create_oceanbase_vector_store(
uri=uri,
user="test-user",
password="test-password",
database_name="test_database",
table_name="knowledge_chunks_v1",
)
def test_oceanbase_dense_sql_requires_profile_pre_filter_and_bound_values() -> None:
sql = _build_dense_search_sql(
table_name="knowledge_chunks_v1",
index_name="vidx_knowledge_chunks_embedding",
output_column_names=["chunk_id", "index_profile_id"],
)
assert "VECTOR_INDEX(knowledge_chunks_v1 vidx_knowledge_chunks_embedding pre_filter)" in sql
assert "WHERE index_profile_id = :index_profile_id" in sql
assert "cosine_distance(embedding, :query_vector)" in sql
assert "APPROXIMATE LIMIT :top_n" in sql
assert "profile-v1" not in sql
assert (
_serialize_vector((0.1, -0.2, 0.3))
== "[0.10000000000000001,-0.20000000000000001,0.29999999999999999]"
)
def test_oceanbase_rejects_unsafe_index_identifier() -> None:
with pytest.raises(OceanBaseConfigurationError, match="index name"):
OceanBaseVectorStore(
FakeOceanBaseClient(),
"knowledge_chunks_v1",
index_name="vector-index;drop",
vector_dimension=3,
)
def test_oceanbase_requires_version_with_explicit_pre_filter() -> None:
with pytest.raises(OceanBaseConfigurationError, match="4.6.0"):
_validate_oceanbase_version([4, 3, 5, 1])
_validate_oceanbase_version([4, 6, 0, 0])
with pytest.raises(OceanBaseVectorStoreError, match="version validation"):
_validate_oceanbase_version("4.6.0.0")
def test_pyobvector_boundary_binds_profile_vector_and_limit() -> None:
sdk_client = FakeSdkClient()
client = _PyObVectorClient(sdk_client)
rows = client.dense_search(
"knowledge_chunks_v1",
index_name="vidx_knowledge_chunks_embedding",
query_vector=(0.1, 0.2, 0.3),
index_profile_id="profile-v1",
top_n=20,
output_column_names=["chunk_id", "index_profile_id"],
)
assert list(rows) == []
sql, parameters = sdk_client.engine.connection.executions[0]
assert "pre_filter" in sql
assert parameters == {
"query_vector": "[0.10000000000000001,0.20000000000000001,0.29999999999999999]",
"index_profile_id": "profile-v1",
"top_n": 20,
}
assert "profile-v1" not in sql
def test_oceanbase_schema_declares_rebuildable_cosine_index() -> None:
schema_path = Path(__file__).parents[2] / "deploy" / "oceanbase" / "001_knowledge_chunks_v1.sql"
schema = schema_path.read_text(encoding="utf-8")
assert "embedding VECTOR(1024) NOT NULL" in schema
assert "index_profile_id VARCHAR(128) NOT NULL" in schema
assert "DISTANCE = COSINE" in schema
assert "TYPE = HNSW" in schema
assert "document_id, index_profile_id" in schema

View File

@ -42,6 +42,44 @@ def test_provider_wiring_requires_explicit_mochow_target(
create_vector_store(settings)
def test_provider_wiring_requires_complete_oceanbase_target(
isolated_settings_environment: None,
) -> None:
settings = Settings(vector_store_provider="oceanbase")
with pytest.raises(ProviderConfigurationError, match="OCEANBASE_VECTOR_URI"):
create_vector_store(settings)
def test_provider_wiring_builds_oceanbase_from_secret_safe_settings(
monkeypatch: pytest.MonkeyPatch,
isolated_settings_environment: None,
) -> None:
captured: dict[str, object] = {}
expected = FakeVectorStore()
def fake_factory(**kwargs: object) -> FakeVectorStore:
captured.update(kwargs)
return expected
monkeypatch.setattr("ekp_api.providers.create_oceanbase_vector_store", fake_factory)
settings = Settings(
vector_store_provider="oceanbase",
oceanbase_vector_uri="db.example:2881",
oceanbase_vector_user="test-user@test-tenant",
oceanbase_vector_password=SecretStr("test-password"),
oceanbase_vector_database="test_database",
)
vector_store = create_vector_store(settings)
assert vector_store is expected
assert captured["uri"] == "db.example:2881"
assert captured["password"] == "test-password"
assert captured["index_name"] == "vidx_knowledge_chunks_embedding"
assert captured["vector_dimension"] == 1024
def test_provider_wiring_selects_local_storage_by_default(
isolated_settings_environment: None,
) -> None:

79
uv.lock generated
View File

@ -2,6 +2,18 @@ version = 1
revision = 3
requires-python = ">=3.12, <3.14"
[[package]]
name = "aiomysql"
version = "0.3.2"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "pymysql" },
]
sdist = { url = "https://files.pythonhosted.org/packages/29/e0/302aeffe8d90853556f47f3106b89c16cc2ec2a4d269bdfd82e3f4ae12cc/aiomysql-0.3.2.tar.gz", hash = "sha256:72d15ef5cfc34c03468eb41e1b90adb9fd9347b0b589114bd23ead569a02ac1a", size = 108311, upload-time = "2025-10-22T00:15:21.278Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/4c/af/aae0153c3e28712adaf462328f6c7a3c196a1c1c27b491de4377dd3e6b52/aiomysql-0.3.2-py3-none-any.whl", hash = "sha256:c82c5ba04137d7afd5c693a258bea8ead2aad77101668044143a991e04632eb2", size = 71834, upload-time = "2025-10-22T00:15:15.905Z" },
]
[[package]]
name = "alembic"
version = "1.19.1"
@ -230,6 +242,7 @@ dependencies = [
{ name = "pydantic-settings" },
{ name = "pymochow" },
{ name = "pymupdf" },
{ name = "pyobvector" },
{ name = "python-docx" },
{ name = "python-multipart" },
{ name = "sqlalchemy" },
@ -254,6 +267,7 @@ requires-dist = [
{ name = "pydantic-settings", specifier = ">=2.7,<3" },
{ name = "pymochow", specifier = ">=2.4,<3" },
{ name = "pymupdf", specifier = ">=1.26,<2" },
{ name = "pyobvector", specifier = ">=0.2.28,<0.3" },
{ name = "python-docx", specifier = ">=1.2,<2" },
{ name = "python-multipart", specifier = ">=0.0.20,<1" },
{ name = "sqlalchemy", specifier = ">=2.0,<3" },
@ -580,6 +594,36 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/79/7b/2c79738432f5c924bef5071f933bcc9efd0473bac3b4aa584a6f7c1c8df8/mypy_extensions-1.1.0-py3-none-any.whl", hash = "sha256:1be4cccdb0f2482337c4743e60421de3a356cd97508abadd57d47403e94f5505", size = 4963, upload-time = "2025-04-22T14:54:22.983Z" },
]
[[package]]
name = "numpy"
version = "2.5.2"
source = { registry = "https://pypi.org/simple" }
sdist = { url = "https://files.pythonhosted.org/packages/9a/80/db0b4559e57ec36362bedbb05530a87fafbcb6067708c946967a41d449e7/numpy-2.5.2.tar.gz", hash = "sha256:d482d171c406ae88c5b19cad3b6a1c4c5209f886ab74bc44c2c865c23f52d860", size = 20773161, upload-time = "2026-08-09T13:48:27.962Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/69/72/dccb0aaf40972777283303919f613964227266d0c13adebb79ac124f1c3e/numpy-2.5.2-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:14e373cfc6387177e8409dac3c7159be8eb05cd77096cd7c950268b86f62831c", size = 16891693, upload-time = "2026-08-09T13:44:51.702Z" },
{ url = "https://files.pythonhosted.org/packages/60/2e/b5aee50a1f74ac815cf8331812cb8251e29024025de462e0c047641c614c/numpy-2.5.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:4bbd96c833ecc8cc069ce518078fc8c60cb9cbfb0fea5b7a803ad65035596d03", size = 11903109, upload-time = "2026-08-09T13:44:55.501Z" },
{ url = "https://files.pythonhosted.org/packages/f3/f4/29e78102a80601cf034d4e9767022cffeca2c3b4c926e1754572ca95593d/numpy-2.5.2-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:6e8172ddfcf5cf74b811d372b570b83c60bd2de87a6fbfbebdadb4a9bd9c6cbb", size = 5350202, upload-time = "2026-08-09T13:44:58.401Z" },
{ url = "https://files.pythonhosted.org/packages/11/4b/dcd3b7eadaf4035d2c7a4289d232523a6964f602598ef7674e4bd7291f93/numpy-2.5.2-cp312-cp312-macosx_14_0_x86_64.whl", hash = "sha256:65f188481f1669e26f62b701e8205d19e460fa4a9b52a1414ba382330e4a3414", size = 6687736, upload-time = "2026-08-09T13:45:00.813Z" },
{ url = "https://files.pythonhosted.org/packages/e5/21/4947e0e9d6c9fc2e2ff15b8949049ee44f63adb9cacc729ab8793f97e712/numpy-2.5.2-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:8ee9c4eeb8454b3660a8b53493563c3e121c2fc94fbd72b848ef814ed7b676a9", size = 15612696, upload-time = "2026-08-09T13:45:04.151Z" },
{ url = "https://files.pythonhosted.org/packages/3a/5f/62d28cf019460c7f1394105b4d49d9911a9c444cb77ab0bd95a204c5a6de/numpy-2.5.2-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:3cdec01fa790a186d430433fdd4d4ffb70eed6f0eeb4bf05c8dbe2dce0a9bcb8", size = 16722264, upload-time = "2026-08-09T13:45:07.714Z" },
{ url = "https://files.pythonhosted.org/packages/14/25/3f0be4c1b9fdf5dd5e708a6806978564d7c46a055c000496309ff2a2f8af/numpy-2.5.2-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:7999d4ddb0c4025018373fd787510d46e04c769467af22869707b3c1cfd459ab", size = 16974396, upload-time = "2026-08-09T13:45:11.316Z" },
{ url = "https://files.pythonhosted.org/packages/22/72/6262cbdeeb45da9d971e40715f579d791603ba8ec0b5e2db1ac55454421d/numpy-2.5.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:c1f017dc0875c9209d219f97feceb7d54c2661bb243deb4114478e1295808af7", size = 18476044, upload-time = "2026-08-09T13:45:14.869Z" },
{ url = "https://files.pythonhosted.org/packages/36/33/29208b8b075bde62d26a81d14b358c42b0f69b6cabd98d4ff97f37f22b05/numpy-2.5.2-cp312-cp312-win32.whl", hash = "sha256:d6a48072864e3324e194a8fbb3c657bcc5b5c869dbc64c9537b1d5c862572c0a", size = 6072817, upload-time = "2026-08-09T13:45:17.867Z" },
{ url = "https://files.pythonhosted.org/packages/7f/b9/87fea2769fe1c47c1b5b01d8310772c9d1a85d485de7cf386ef7a3332b02/numpy-2.5.2-cp312-cp312-win_amd64.whl", hash = "sha256:28ac63476ec7651484215ee7fa15a1f78b57c14621f01e392afe17b9a1390ce4", size = 12464674, upload-time = "2026-08-09T13:45:20.734Z" },
{ url = "https://files.pythonhosted.org/packages/14/52/032b97e00461ab0809bbe4c588b035620e5a14b8cdee47ecddefc7b17d33/numpy-2.5.2-cp312-cp312-win_arm64.whl", hash = "sha256:27650bb0e7140fa3d37b9923b4803645e0b125d190f326eecfd3f4dad8e8ade1", size = 10397131, upload-time = "2026-08-09T13:45:23.73Z" },
{ url = "https://files.pythonhosted.org/packages/f5/d2/6b24738a0ef4557d189b150046cd07823c50e4273e8aebd651222e24306f/numpy-2.5.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:8e4cb9a754c8a0c62eaa88273a5fba3391f4a610d1dee893c0755da31c083f15", size = 16886595, upload-time = "2026-08-09T13:45:27.323Z" },
{ url = "https://files.pythonhosted.org/packages/65/60/f2d208d366f263f39c6e69ed309290717aab41078b6d04c9be2a84fa2a07/numpy-2.5.2-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:52c808f96484f5571a5cc863775ce50247c17dfb3b0361f8ed6b4b0456f80080", size = 11896845, upload-time = "2026-08-09T13:45:31.638Z" },
{ url = "https://files.pythonhosted.org/packages/3c/79/81e0bf24f4d020a2b1d5cd297a9f60c3f24eeb116f9bba5870443f7b6a4a/numpy-2.5.2-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:29d81e97f668489cba8ebfd796b9bdd453525d35dd9e162e2daec94bf3fc7740", size = 5343880, upload-time = "2026-08-09T13:45:34.373Z" },
{ url = "https://files.pythonhosted.org/packages/ba/cc/e3141cf06d1a8a2c7e107543fe1269c1d1af760d4d683c0794a4ee1127c2/numpy-2.5.2-cp313-cp313-macosx_14_0_x86_64.whl", hash = "sha256:afb3f0632d6b2e3ba04dbce8d1e48d321b369138b73830b5ca371a0e8d479d56", size = 6682264, upload-time = "2026-08-09T13:45:36.7Z" },
{ url = "https://files.pythonhosted.org/packages/29/f1/2a64a307d92c5d98f5255a4014eb43bb6103ee477087b61ecae44a3aa9b9/numpy-2.5.2-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0aadf13b60048d501e05fa699efaf7734e2494f3498a4c2a5521d822640324f3", size = 15609566, upload-time = "2026-08-09T13:45:39.518Z" },
{ url = "https://files.pythonhosted.org/packages/7b/44/59a1eb68e773c4098d107ef34a0dbdeca501d72ffcfbff9a7707343921ce/numpy-2.5.2-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:29b86ff8a6cc556b47ec6b64b194815cc80e6bf5eedcc6cddfd65318cb0b4eee", size = 16709995, upload-time = "2026-08-09T13:45:43.661Z" },
{ url = "https://files.pythonhosted.org/packages/8a/4c/3e54d4ddbc359a1295f8b633e8106bcd4d7d4a206e82df051bdfb3058755/numpy-2.5.2-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:6950c4b7dd562453090548ba7f5da7e59f57f85663f15d5dcc60e249192f7e59", size = 16972511, upload-time = "2026-08-09T13:45:47.094Z" },
{ url = "https://files.pythonhosted.org/packages/f2/9f/02e371638ebf19b66d46231e4be52999e87f32d1961b113bc45656608b22/numpy-2.5.2-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:b9727f472d2f3888053b8a75ab0cb94745a9de224bb5846dbadc0092101bc71d", size = 18465609, upload-time = "2026-08-09T13:45:50.808Z" },
{ url = "https://files.pythonhosted.org/packages/eb/ae/ad6645abc7a3510fe48e8ea1ab4598166f500057ef4ebf38bfad4f1577de/numpy-2.5.2-cp313-cp313-win32.whl", hash = "sha256:4f9744f9fbdcea0bc552e8f19e1f141f811a3f9bc2be2cc6e86d982cab23e3f4", size = 6070204, upload-time = "2026-08-09T13:45:54.111Z" },
{ url = "https://files.pythonhosted.org/packages/15/20/f3489f86d81ea460b2bcdceaed094142ca6579f6be0ec527b781d39afe68/numpy-2.5.2-cp313-cp313-win_amd64.whl", hash = "sha256:85aaccb24182c25df891ad0ec333585967e115269d5f1b17f2c9ae005bc96657", size = 12460532, upload-time = "2026-08-09T13:45:57.167Z" },
{ url = "https://files.pythonhosted.org/packages/d5/21/35b31dde1b283b79de828b80f876afd8c94e28fe1e9c375f89e261cc4c0d/numpy-2.5.2-cp313-cp313-win_arm64.whl", hash = "sha256:bd68ece1553d2023c09a4226d9e41c586ad2d20594d1a456186c33513d2cb3f2", size = 10396725, upload-time = "2026-08-09T13:46:00.478Z" },
]
[[package]]
name = "orjson"
version = "3.12.0"
@ -828,6 +872,32 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/58/8c/d897dcd32a25b58186c968b15ce4324ca029e9d96460de12325314e390be/pymupdf-1.28.2-cp313-abi3-pyemscripten_2025_0_wasm32.whl", hash = "sha256:2e1b574c0fd2cb238021033fd3c0f9c4388816638df064e4bfb56d9d81736dc8", size = 18399403, upload-time = "2026-08-06T21:39:25.008Z" },
]
[[package]]
name = "pymysql"
version = "1.2.0"
source = { registry = "https://pypi.org/simple" }
sdist = { url = "https://files.pythonhosted.org/packages/c9/bc/1c6a92f385940f727daeecf3bacaf186e03875dff57197801046c583bcf0/pymysql-1.2.0.tar.gz", hash = "sha256:6c7b17ca686988104d7426c27895b455cdeea3e9d3ceb1270f0c3704fead8c33", size = 49021, upload-time = "2026-05-19T08:26:22.302Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/c4/bd/2534e130295c8cfd4f0a2e31623baab7502278f1e97bcfe61db75656a77f/pymysql-1.2.0-py3-none-any.whl", hash = "sha256:62169ce6d5510f08e140c5e7990ee884a9764024e4a9a27b2cc11f1099322ae0", size = 45716, upload-time = "2026-05-19T08:26:20.974Z" },
]
[[package]]
name = "pyobvector"
version = "0.2.29"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "aiomysql" },
{ name = "numpy" },
{ name = "pydantic" },
{ name = "pymysql" },
{ name = "sqlalchemy" },
{ name = "sqlglot" },
]
sdist = { url = "https://files.pythonhosted.org/packages/32/47/80f5f62572b00348d6864eb14d495cf26e52d7b9dba0ad7365cdbac86851/pyobvector-0.2.29.tar.gz", hash = "sha256:3f0feac0ac964bb56c2e3ed661b5de84ef5bc97253ae8e9935a65fea367e3736", size = 85589, upload-time = "2026-07-29T02:53:22.676Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/82/66/1aa4ca29ee2d1780aa4761d58ccac82771747a75f9146db0d7285a245bdb/pyobvector-0.2.29-py3-none-any.whl", hash = "sha256:f7cfdf1ca0c5c6ccbc10fd096aef944cd3970970f1382441a77dff1be78744cf", size = 70210, upload-time = "2026-07-29T02:53:21.307Z" },
]
[[package]]
name = "pytest"
version = "9.1.1"
@ -993,6 +1063,15 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/b3/3f/3582293d1e185e71d19d7c731c3e2ee20ba21981c4a1115c0806c1f62120/sqlalchemy-2.0.52-py3-none-any.whl", hash = "sha256:3b81b8363a919ce53453591cdb93702e6bd54ade6c4fa2f468fc053baee5ed89", size = 1950700, upload-time = "2026-08-11T20:47:21.603Z" },
]
[[package]]
name = "sqlglot"
version = "30.17.0"
source = { registry = "https://pypi.org/simple" }
sdist = { url = "https://files.pythonhosted.org/packages/56/d4/da49abcc81beebbb25f29ddf87f2980c63c949569d7f2da40c06d95fa415/sqlglot-30.17.0.tar.gz", hash = "sha256:2d6b8def93304fa300f4d20f48e3909e7f436fda56ca1fafd8975f6c561ef62c", size = 5999019, upload-time = "2026-08-12T19:36:50.587Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/f3/bc/2a07cef49046e6cf5d1a8b1de553aaef8491b4170dbfe254c8687c764408/sqlglot-30.17.0-py3-none-any.whl", hash = "sha256:84435ac283a60173da31b5fd7d11a725037a1c3fd6ed1e21fb065de74ddb579f", size = 741795, upload-time = "2026-08-12T19:36:48.699Z" },
]
[[package]]
name = "starlette"
version = "1.6.0"