polnor 1.1.0

Polnor Python SDK — query Iceberg lakehouses, manage notebooks/jobs/models, MLflow-compatible tracking.

Polnor Python SDK

The official Python client for the Polnor lakehouse — query Iceberg tables with SQL, manage notebooks / jobs / models / endpoints, read & write pandas DataFrames, and use the PEP 249 driver for dbt, pandas, sqlalchemy, and any tool that speaks standard Python DB API.

Install

pip install polnor

Quick start

import polnor

# `polnor.sql()` runs against your default warehouse. Inside a Polnor
# notebook or job, credentials are injected automatically. On a laptop,
# set env vars first (see "Auth" below).
result = polnor.sql("SELECT * FROM demo.orders LIMIT 5")
print(result.rows)            # list of lists
print(result.to_pandas())     # pandas DataFrame

Auth

The SDK resolves credentials in this order (first match wins):

1. polnor.set_token("...", api_url="...", warehouse_id="...") — explicit call
2. Environment variables: POLNOR_API_URL, POLNOR_TOKEN, POLNOR_WAREHOUSE_ID
3. ~/.polnor/config.toml — TOML config file written by the polnor CLI

Inside a Polnor notebook or job, all three POLNOR_ env vars are injected by the runtime* — your cell-1 polnor.sql(...) works with no setup.

Local example (TOML config):

# ~/.polnor/config.toml
host = "https://api.polnor.net"
token = "<personal access token>"
workspace_id = "<your workspace>"
warehouse_id = "<your default warehouse>"

What's in the box

Module	What it does
polnor.sql(query)	Run SQL on a SQL warehouse, return rows / pandas / dicts.
polnor.warehouses	List / get / start / stop / default SQL warehouses.
polnor.compute	Manage compute VMs (start, stop, install_library, logs).
polnor.notebooks	Create / start / stop Jupyter notebooks on a compute.
polnor.jobs	Create + run DAG jobs, fetch logs, cancel, wait.
polnor.models	Model registry: versions, artifact URIs, framework tags.
polnor.endpoints	Serve a model image: deploy, predict, scale, stop.
polnor.tables	Browse databases (Iceberg namespaces) and tables.
polnor.read_pandas	Read a table (with where=, limit=) into a DataFrame.
polnor.write_pandas	Bulk-insert a DataFrame (append / overwrite / create).
polnor.dbapi	PEP 249 driver: connect() / cursor() for dbt / pandas.
polnor.mlflow	MLflow-compatible tracking: log_param, log_metric, etc.

Examples

SQL → pandas DataFrame

df = polnor.sql("SELECT region, COUNT(*) c FROM demo.orders GROUP BY region").to_pandas()
df.plot(kind="bar")

Push a DataFrame to a table

import pandas as pd
df = pd.DataFrame({"id": [1, 2, 3], "name": ["alice", "bob", "carol"]})
polnor.write_pandas(df, "polnor.demo.default.users", mode="create")

PEP 249 (works with dbt, pandas.read_sql, sqlalchemy)

import pandas as pd
import polnor.dbapi as dbapi

conn = dbapi.connect()
df = pd.read_sql("SELECT * FROM demo.orders", conn)

Notebooks, jobs, models, endpoints

# Start a notebook on a CPU compute
polnor.compute.start("my-cpu", wait=True)
nb = polnor.notebooks.create(name="explore", compute_id="my-cpu")
polnor.notebooks.start("explore", wait=True)

# Run a DAG job
job = polnor.jobs.create(name="etl", tasks=[
    {"key": "extract",   "command": "python extract.py",   "image": "python:3.11"},
    {"key": "transform", "command": "python transform.py", "image": "python:3.11",
     "depends_on": ["extract"]},
])
run_id = polnor.jobs.run("etl")
polnor.jobs.wait(run_id)
print(polnor.jobs.logs(run_id))

# Serve a model
polnor.endpoints.create(name="fraud-v1", image="my-registry/fraud:1.0",
                        environment={"CONTAINER_PORT": "8080"})
polnor.endpoints.deploy("fraud-v1", wait=True)
pred = polnor.endpoints.predict("fraud-v1", inputs=[[0.1, 0.2, 0.3]])

MLflow-compatible tracking

from polnor import mlflow
with mlflow.start_run() as run:
    mlflow.log_param("lr", 0.01)
    mlflow.log_metric("accuracy", 0.93)
    mlflow.log_artifact("./model.pkl")

Errors

The SDK raises these typed exceptions:

• polnor.PolnorAPIError — non-2xx HTTP from the control plane.
• polnor.SQLError — statement failed, cancelled, or timed out.
• polnor.PolnorConfigError — no credentials could be resolved.
• polnor.dbapi.ProgrammingError / OperationalError / etc. — PEP 249 hierarchy.

Compatibility

• Python ≥ 3.9 (tomli falls in for 3.9 / 3.10; 3.11+ has tomllib built in)
• pandas integration: optional, lazily imported
• dbt: works via polnor.dbapi (a dedicated dbt-polnor adapter is in V2)

Development

pip install -e ".[test]"
pytest

Tests use responses to mock HTTP — no live API needed. The tests/test_e2e_prod.py suite (opt-in via env) hits api.polnor.net and requires a running warehouse + valid PAT.

Links

• Homepage: https://polnor.net
• API docs: https://docs.polnor.net
• Source / issues: https://github.com/polnor/polnor

License

Apache-2.0.