Python client (`datap-rs-client`)¶

A standalone Python client for a running DataPress server, backed by the native datapress-client Rust crate via PyO3. Requests are plain Python dicts; responses come back as dicts. Structured queries can optionally be decoded into a pyarrow.Table.

Two Python clients

This is the standalone package (datap-rs-client), independent of the server. If you already use the embedded datap-rs wheel — which can also run a server — its bundled stdlib client is documented under Python → Client. Use datap-rs-client when you only need to talk to a server and want the fast Rust transport.

Install¶

uv pip install datap-rs-client          # core
uv pip install datap-rs-client[arrow]   # + pyarrow for query_arrow()

Wheels are published for Linux (x86_64/aarch64), macOS (arm64), and Windows (x86_64) against CPython 3.9+ (abi3).

Usage¶

from datap_rs_client import DataPressClient

client = DataPressClient("http://127.0.0.1:8000")

client.datasets()
# ['accidents']

client.count("accidents", predicates=[{"col": "Severity", "op": "gte", "val": 3}])
# 123456

rows = client.query(
    "accidents",
    columns=["State", "Severity"],
    predicates=[{"col": "Severity", "op": "gte", "val": 3}],
    page_size=1000,
)
rows["page"], len(rows["data"])
# (1, 1000)

# Arrow (requires the [arrow] extra)
table = client.query_arrow("accidents", columns=["State", "Severity"], page_size=100_000)
table.num_rows

Authentication¶

client = DataPressClient(
    "http://127.0.0.1:8000",
    bearer_token="…",     # servers with auth enabled
    admin_token="…",      # required by reload()
)

SQL¶

client.sql("SELECT State, COUNT(*) AS n FROM accidents GROUP BY State", max_rows=100)

DataFrames¶

query_arrow(...) returns a pyarrow.Table (install the [arrow] extra). Arrow is the zero-copy interchange format for every popular dataframe library, so a single query feeds them all:

table = client.query_arrow(
    "accidents",
    columns=["State", "Severity"],
    predicates=[{"col": "Severity", "op": "gte", "val": 3}],
    page_size=1_000_000,
)

PolarspandasDuckDBPySparkDataFusion

import polars as pl

# Zero-copy from the Arrow table.
df = pl.from_arrow(table)
df.group_by("State").len().sort("len", descending=True)

import pandas as pd  # noqa: F401  (pyarrow drives the conversion)

# Arrow-backed dtypes (recommended) …
df = table.to_pandas(types_mapper=pd.ArrowDtype)
# … or classic NumPy-backed dtypes:
df = table.to_pandas()
df.groupby("State")["Severity"].mean()

import duckdb

# DuckDB queries the Arrow table in place — no copy, no temp files.
duckdb.sql("SELECT State, COUNT(*) AS n FROM table GROUP BY State ORDER BY n DESC")

from pyspark.sql import SparkSession

spark = SparkSession.builder.getOrCreate()
# Spark has no direct Arrow-table constructor; go via pandas (Arrow-accelerated).
sdf = spark.createDataFrame(table.to_pandas())
sdf.groupBy("State").count().orderBy("count", ascending=False).show()

from datafusion import SessionContext

ctx = SessionContext()
df = ctx.from_arrow(table)
df.aggregate([df["State"]], [df["Severity"].mean()])

Python client (datap-rs-client)¶