Skip to main content

Python bindings for Alopex DB

Project description

Alopex Python バインディング

Python から AlopexDB を操作するためのバインディングです。
Database/Transaction の基本機能に加え、ベクトル検索(numpy)と Unity Catalog 互換の Catalog API(polars)を提供します。

インストール

pip install alopex

Catalog API (polars) を使う場合:

pip install alopex[polars]

開発中は maturin を利用できます。

maturin develop -m crates/alopex-py/pyproject.toml

オプション依存:

  • numpy を使う場合: pip install alopex[numpy]
  • polars を使う場合: pip install alopex[polars]

対応バージョン

依存関係 対応バージョン
Python 3.8+
Polars 0.20+ (Catalog API)
NumPy 1.20+ (Vector API)

基本的な使い方

Database / Transaction

from alopex import Database, TxnMode

db = Database.new()

with db.begin(TxnMode.READ_WRITE) as txn:
    txn.put(b"user:1", b"alice")
    txn.commit()

with db.begin(TxnMode.READ_ONLY) as txn:
    value = txn.get(b"user:1")
    print(value)

db.close()

v0.8 embedded-local stream / DataFrame API

v0.8 の stream API は embedded-local database 専用です。Client、endpoint、remote session、remote DataFrame execution は提供しません。Database は既定で thread_mode="multi" であり、 thread_mode="single" を選ぶと database、transaction、stream、DataFrame、LazyFrame は作成した Python thread だけで使用できます。

同期 SQL / scan stream

execute_sql_stream() は、単一 table の local SELECT、row-local WHERE、projection、LIMIT / OFFSET の documented subset を一行ずつ返します。query_stream()LocalScan.tablecsvparquetcolumnar_segmentlazyframe の五つだけを受け取ります。非対応 SQL、callback、remote source は stream を開く前に AlopexError となります。

from alopex import Database, LocalScan

db = Database.new()
db.execute_sql("CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT)")
db.execute_sql("INSERT INTO users (id, name) VALUES (1, 'one'), (2, 'two')")

with db.execute_sql_stream(
    "SELECT id, name FROM users WHERE id >= 1",
    resource_limit_bytes=64 * 1024 * 1024,
    timeout=5.0,
) as stream:
    for row in stream:
        print(row)
    print(stream.status)
    # {"terminal": "exhausted", "rows_delivered": 2,
    #  "resource_limit_bytes": ..., "resource_scope": "sql_row",
    #  "transaction_effect": "none"}

DataFrameStream.status は同じ構造で、rows_delivered の代わりに batches_deliveredresource_scope="dataframe_batch" を返します。normal exhaustion は以後 end-of-stream、close/cancel/timeout/failure は以後も同じ識別可能な terminal error を返します。

transaction 内の stream は暗黙 commit しません。正常に exhaustion した read stream は commit 可能であり、 early close、cancel、failure は transaction.status["stream_effect"] と stream の transaction_effect で確認した後に rollback してください。active stream 中の commit は拒否されます。

Python DataFrame streaming / expressions

LazyFrame.collect(streaming=True) は Phase 3 と同じ source、row order、schema、NULL、resource contract で有限 DataFrame batch を返します。concatconcat_strselectfilterwith_columns は同じ expression semantics を使用します。

from alopex import DataFrame, col, concat_str, lit

plan = (
    DataFrame({"id": [1, 2, 3], "left": ["a", "b", "c"], "right": ["x", "y", "z"]})
    .lazy()
    .filter(col("id").gt(lit(1)))
    .select([
        col("id").add(lit(10)).alias("next_id"),
        concat_str([col("left"), col("right")], "-").alias("label"),
    ])
)

with plan.collect(streaming=True, batch_rows=1) as batches:
    for batch in batches:
        print(batch.to_dict())

asyncio

alopex.asyncio は Python 3.8 以上の標準 asyncio loop をサポートし、caller に Rust/Tokio runtime を 要求しません。prefetch_batches は read ahead の上限、max_buffered_batches は ready result buffer の上限です。両者は 0 <= prefetch_batches <= max_buffered_batches、かつ max_buffered_batches >= 1 でなければなりません。consumer_idle_timeouttimeout の単位は秒です。 native worker は bounded Rust payload だけを保持し、Python の row / DataFrame 変換は asyncio consumer 側で 行われます。Python の producer queue や callback は使用しません。

import asyncio
from alopex.asyncio import AsyncDatabase

async def main() -> None:
    async with await AsyncDatabase.new() as db:
        await db.execute_sql("CREATE TABLE events (id INTEGER PRIMARY KEY)")
        await db.execute_sql("INSERT INTO events (id) VALUES (1), (2)")
        stream = await db.execute_sql_stream(
            "SELECT id FROM events",
            prefetch_batches=1,
            max_buffered_batches=1,
            consumer_idle_timeout=5.0,
        )
        async with stream:
            async for row in stream:
                print(row)

asyncio.run(main())

同じ stream で同時に二つの anext() を実行すると stream_busy です。task cancellation、aclose()cancel()、idle timeout は stream の native source を終端させ、後続の独立した database operation を継続できます。

ベクトル検索(numpy 必須)

import numpy as np
from alopex import Database, Metric, TxnMode

db = Database.new()
with db.begin(TxnMode.READ_WRITE) as txn:
    vec = np.array([1.0, 0.0, 0.0], dtype=np.float32)
    txn.upsert_vector(b"k1", None, vec, Metric.COSINE)
    results = txn.search_similar(vec, Metric.COSINE, 1, return_vectors=True)
    print(results[0].key, results[0].score)
    if results[0].vector is not None:
        print(results[0].vector.dtype, results[0].vector.shape)

NumPy 入出力とゼロコピー条件(v0.3.5)

入力(Python → Rust):

  • dtype: float32 が優先。float64float32 に変換して処理します。
  • layout: C-contiguous が優先。非連続(strided/Fortran order 等)は C-contiguous に変換して処理します。
  • ゼロコピー入力: float32 かつ C-contiguous の場合は Rust 側でコピーなしに参照します。

出力(Rust → Python):

  • Transaction.search_similar(..., return_vectors=True) の場合、SearchResult.vectornumpy.ndarray[float32] を含められます。
  • Transaction.search_similar(..., zero_copy_return=True) / Transaction.get_vector(..., zero_copy_return=True) の場合、可能なら所有権移譲によるゼロコピー返却を行います(False の場合はコピー)。

GIL:

  • upsert_vector / search_similar / search_hnsw は重い処理中に GIL を解放します。

HNSW インデックス(numpy 必須)

import numpy as np
from alopex import Database, HnswConfig, TxnMode

db = Database.new()
db.create_hnsw_index("idx", HnswConfig(2))

with db.begin(TxnMode.READ_WRITE) as txn:
    vec = np.array([1.0, 0.0], dtype=np.float32)
    txn.upsert_to_hnsw("idx", b"k1", vec, None)
    txn.commit()

results, stats = db.search_hnsw("idx", np.array([1.0, 0.0], dtype=np.float32), 1)
print(stats.node_count)

Catalog API(polars 必須)

import polars as pl
from alopex import Catalog, ColumnInfo

Catalog.create_catalog("main")
Catalog.create_namespace("main", "default")

columns = [ColumnInfo("id", "int", 0, False)]
Catalog.create_table("main", "default", "users", columns, "/tmp/users.parquet")

df = pl.DataFrame({"id": [1, 2], "name": ["a", "b"]})
Catalog.write_table(
    df,
    "main",
    "default",
    "users",
    delta_mode="overwrite",
    storage_location="/tmp/users.parquet",
)

lazy_frame = Catalog.scan_table("main", "default", "users")
print(lazy_frame.collect())

注意事項

  • numpy / polars が未インストールの場合、対応 API は AlopexError を返します。
  • Phase 1 では Parquet のみ対応しています。

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

alopex-0.8.0.tar.gz (1.2 MB view details)

Uploaded Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

alopex-0.8.0-cp38-abi3-win_amd64.whl (6.6 MB view details)

Uploaded CPython 3.8+Windows x86-64

alopex-0.8.0-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (6.7 MB view details)

Uploaded CPython 3.8+manylinux: glibc 2.17+ x86-64

alopex-0.8.0-cp38-abi3-macosx_11_0_arm64.whl (5.8 MB view details)

Uploaded CPython 3.8+macOS 11.0+ ARM64

alopex-0.8.0-cp38-abi3-macosx_10_12_x86_64.whl (6.3 MB view details)

Uploaded CPython 3.8+macOS 10.12+ x86-64

File details

Details for the file alopex-0.8.0.tar.gz.

File metadata

  • Download URL: alopex-0.8.0.tar.gz
  • Upload date:
  • Size: 1.2 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for alopex-0.8.0.tar.gz
Algorithm Hash digest
SHA256 89b6c8d00f1976dd9e117601abffeac2d6d82eb9bc340aee1043b4718ad1c11f
MD5 796a823b916bcdc0e44bdf041092e219
BLAKE2b-256 fc63eb952251fb11163559e29324f8f0a853eb4c25272be37038ad29dca8f091

See more details on using hashes here.

Provenance

The following attestation bundles were made for alopex-0.8.0.tar.gz:

Publisher: alopex-py-release.yml on alopex-db/alopex

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file alopex-0.8.0-cp38-abi3-win_amd64.whl.

File metadata

  • Download URL: alopex-0.8.0-cp38-abi3-win_amd64.whl
  • Upload date:
  • Size: 6.6 MB
  • Tags: CPython 3.8+, Windows x86-64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for alopex-0.8.0-cp38-abi3-win_amd64.whl
Algorithm Hash digest
SHA256 fca1b3b5af79e76f175aef153db6bd22c5c17505ba9fce84782f382f435dc559
MD5 4abec4955733de5a61fe157a44f3e6a9
BLAKE2b-256 911305cb76a13fe6e9ff6418b792f07660a879e6313637a81718992692898294

See more details on using hashes here.

Provenance

The following attestation bundles were made for alopex-0.8.0-cp38-abi3-win_amd64.whl:

Publisher: alopex-py-release.yml on alopex-db/alopex

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file alopex-0.8.0-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.

File metadata

File hashes

Hashes for alopex-0.8.0-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
Algorithm Hash digest
SHA256 721c1f6263ee51c5440c2384364e6bcf3e9ef56d531b433bf400f7b82673d875
MD5 7854341970b14f230e8db4d1e4c3b3ab
BLAKE2b-256 5ea4ea6bef2af48c7b00682d8f7eda4bb17a925c89c79ff8743289e69a1305ad

See more details on using hashes here.

Provenance

The following attestation bundles were made for alopex-0.8.0-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl:

Publisher: alopex-py-release.yml on alopex-db/alopex

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file alopex-0.8.0-cp38-abi3-macosx_11_0_arm64.whl.

File metadata

File hashes

Hashes for alopex-0.8.0-cp38-abi3-macosx_11_0_arm64.whl
Algorithm Hash digest
SHA256 fd93cb212b3fb797a23c8e85f1058fc181fe6ec94b6bf4a6c86462d85056e9a4
MD5 5fc2d82d9a509f0a3c13b3f5cef56bcb
BLAKE2b-256 b77c1d007db26d07b39fa97a4c77bf13b134dc5837249ca2b4ad4cbae950f586

See more details on using hashes here.

Provenance

The following attestation bundles were made for alopex-0.8.0-cp38-abi3-macosx_11_0_arm64.whl:

Publisher: alopex-py-release.yml on alopex-db/alopex

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file alopex-0.8.0-cp38-abi3-macosx_10_12_x86_64.whl.

File metadata

File hashes

Hashes for alopex-0.8.0-cp38-abi3-macosx_10_12_x86_64.whl
Algorithm Hash digest
SHA256 a54b3bc4481cbe5030da5c9b3fe85929f707888d7fe7555e7a44b5e08e3ab136
MD5 bae893c31495128585e4260b0c514196
BLAKE2b-256 b86f1fdfa6134dec3c6e16d38ffb2e3345fd44340e6322fc37d8c39f26b7f1d8

See more details on using hashes here.

Provenance

The following attestation bundles were made for alopex-0.8.0-cp38-abi3-macosx_10_12_x86_64.whl:

Publisher: alopex-py-release.yml on alopex-db/alopex

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page