thadam

Python SDK for the Thadam edge capture runtime, built with PyO3 and maturin.

• thadam.validate(path) — validate a pipeline configuration, returning every error and warning with positions and hints.
• thadam.config_hash(path) — the stable configuration hash the runtime records in thadam.json.
• thadam.read_segments(sink_dir, dataset=None) — read a Parquet sink's ready segments into a pyarrow.Table, zero-copy over the Arrow PyCapsule interface.
• thadam.Pipeline(path) — build and run a pipeline in-process on a Rust-owned Tokio runtime, with the GIL released and graceful shutdown from another thread via stop().

Install

pip install thadam

To pin the current pre-release explicitly:

pip install thadam==0.1.0b2

Requires Python >= 3.9; pyarrow >= 14 is installed as a dependency. Prebuilt abi3 wheels (CPython 3.9+) are published for Linux glibc (manylinux_2_28) and musl (musllinux_1_2) on x86_64 and aarch64, and for macOS arm64 (11.0+) and x86_64. Other platforms (such as Windows) build from the source distribution and need a Rust toolchain (1.89 or newer).

This package is the Python SDK; the thadam command-line tool installs with cargo install thadam --version 0.1.0-beta.2 --locked.

Examples

Validate a configuration

import thadam

for d in thadam.validate("pipeline.yaml"):
    where = f":{d.line}:{d.column}" if d.line else ""
    print(f"{d.severity}[{d.code}] {d.path}{where}: {d.message}")
    if d.hint:
        print(f"  hint: {d.hint}")

The configuration hash

assert thadam.config_hash("pipeline.yaml") == "1734a596e8a251683f1e829377f1e2c5"

Read a Parquet sink

table = thadam.read_segments("/var/lib/thadam/sinks/local", dataset="readings")
print(table.schema)
print(table.num_rows)
df = table.to_pandas()

read_segments lists files through the sink's catalog.log when one exists (every record that still has a file on disk, oldest WAL range first) or walks ready/ directly when there is no catalog (fixtures, copied-out data). All files must share a schema — they do when one sink wrote them.

Run a pipeline in-process

import threading

import thadam

pipeline = thadam.Pipeline("pipeline.yaml")
t = threading.Thread(target=lambda: print("exit", pipeline.run(timeout=300)), daemon=True)
t.start()
# ... other work ...
pipeline.stop()   # graceful shutdown, like SIGINT to the CLI
t.join()
exit_code = pipeline.run(timeout=300)  # raises: a Pipeline runs once

run(timeout=seconds) triggers graceful shutdown when the timeout elapses (the shutdown deadline then bounds the drain); it is not a hard kill. Exit codes mirror the CLI: 0 clean, 1 runtime failure, 2 invalid configuration, 3 data directory, 130 interrupted.

Catching errors

try:
    h = thadam.config_hash("missing.yaml")
except thadam.ThadamError as e:
    print(e.code, e.domain)  # config.invalid config
    print(e.message)

Building from a checkout

For contributors working on the SDK from a checkout of the Thadam repository, building from source requires the workspace Rust toolchain (rust-toolchain.toml) and a CPython >= 3.9 build on PATH (PYO3_PYTHON to pick a specific one):

cd sdk/python
python3 -m venv .venv
. .venv/bin/activate
pip install maturin pytest pyarrow
maturin develop
pytest tests

maturin develop builds the thadam._native cdylib with pyo3/extension-module (no libpython linked) and installs the package in the virtualenv. Cargo's plain cargo test -p thadam-python builds without that feature (rlib + libpython) so the Rust unit tests run too.

Arrow versions

The workspace writes segments with arrow/parquet 60, while pyo3-arrow 0.19 (the newest release) pairs with arrow 59 — no release tracks arrow 60 yet. read_segments therefore opens the Parquet files itself with parquet 59 and hands the batches to pyo3-arrow; the decode types never cross between majors, and the Python-side transfer stays zero-copy through the Arrow C stream capsule. When pyo3-arrow publishes an arrow-60 release this constraint goes away. One wheel per platform covers all supported Pythons via PyO3's abi3 (abi3-py39); pyo3-arrow's capsule and buffer-protocol APIs are stable-ABI-safe.

Layout

sdk/python/
  Cargo.toml            Rust crate `thadam-python`, cdylib `thadam`
  pyproject.toml        maturin build, Python >= 3.9, abi3 wheel
  src/                  PyO3 bindings and Rust unit tests
  python/thadam/         pure-Python package (re-exports, .pyi stubs, py.typed)
  tests/                pytest suite (run inside a venv, see above)