skyshelve 0.1.0

Backend-agnostic persistent dictionary for Python with BadgerDB and SlateDB support

Skyshelve: Backend-Agnostic Persistent Python Mapping

Skyshelve exposes a minimal dictionary-shaped interface to embedded key-value stores such as BadgerDB and SlateDB. The core is a Go library compiled in c-shared mode which provides a handful of exported functions that manage the selected backend and offer CRUD primitives. A small ctypes shim (src/skyshelve/__init__.py) loads the shared object and presents a Python-friendly API.

Layout

• skyshelve.go — Go implementation of the shared library exports.
• src/skyshelve/__init__.py — Python package exposing the SkyShelve class.
• src/skyshelve/libskyshelve.* — Platform-specific shared library produced by the Go compiler.
• PersistentObject base class (in src/skyshelve/__init__.py) offers an inheritable ORM-style helper that uses file locks so multiple processes can safely read and mutate shared records.
• examples/demo.py — Minimal usage example.
• examples/scan_example.py — Demonstrates scanning keys and persistent objects.
• examples/indexed_profiles.py — Pydantic-backed parent/child models with secondary indexes.
• examples/simple_counter.py — Uses PersistentObject to track run counts.
• examples/slatedb_backend.py — Demonstrates opting into the SlateDB backend.

Prerequisites

• Go 1.20 or newer (Go 1.25 used while developing the library).
• Python 3.9 or newer.
• A C toolchain for building cgo shared objects (e.g. build-essential on Debian/Ubuntu, Xcode Command Line Tools on macOS, or MSYS2 on Windows).

Installation

The package is published on PyPI; install it with:

pip install skyshelve

You will still need a platform-appropriate Go build of the shared library if you are building from source or developing locally (see below).

Getting backend dependencies

The Go module requires github.com/dgraph-io/badger/v4 (always) and optionally slatedb.io/slatedb-go when you intend to use SlateDB. With network access, run:

go mod tidy

This will download both dependencies and produce an up-to-date go.sum.

Building the shared library

# Linux / macOS (run from the repository root)
go build -buildmode=c-shared -o src/skyshelve/libskyshelve.so

# Windows (PowerShell)
go build -buildmode=c-shared -o src/skyshelve/libskyshelve.dll

The command produces two files inside src/skyshelve/: the shared library (.so/.dll) and a matching C header (libskyshelve.h). Keep the header if you plan to integrate through other FFI layers.

Using from Python

Place the compiled shared library next to src/skyshelve/__init__.py, then interact with the store:

from skyshelve import SkyShelve

with SkyShelve("data") as store:
    store["username"] = "alice"          # stored as UTF-8 text
    store["profile"] = {"plan": "pro"}  # auto-pickled
    store["avatar"] = b"\x89PNG"        # raw bytes stay bytes

    print(store["username"])           # 'alice'
    print(store["profile"])            # {'plan': 'pro'}
    print(store["avatar"])             # b'\x89PNG'
    store.sync()                        # flush to disk

By default the wrapper expects the shared library to be named libskyshelve.so/.dylib/.dll in the same directory as the skyshelve package. If you relocate it, pass lib_path="..." when constructing SkyShelve.

Compatibility note: the class is also exported as BadgerDict for projects that previously depended on the old package name.

Values that are bytes-like or str are stored as-is; everything else is serialized with pickle.dumps by default. Disable that behaviour with SkyShelve(..., auto_pickle=False) if you need stricter type enforcement.

For richer models, inherit from PersistentObject and call YourModel.configure_storage(...) once per process, then use save(), load(), and update() to modify state atomically across processes.

Alternatively, set private configuration attributes on your model. When no path is provided, a default of ./data/<model-name-lowercase> is used. The helper PersistentBaseModel combines Pydantic with PersistentObject, automatically serializing models (and stdlib dataclasses) while keeping secondary indexes in sync:

from skyshelve import PersistentBaseModel


class User(PersistentBaseModel):
    __persistent_key_field__ = "username"
    __persistent_path__ = "data/users"
    __persistent_secondary_indexes__ = {"email": lambda u: [u.email]}

    username: str
    email: str


User(username="alice", email="[e-mail]").save()
print(User.scan_index("email", "[e-mail]"))  # -> [User(...)]
print(User.children("email", "[e-mail]"))     # same as scan_index

Using the SlateDB backend

The same Python API can target SlateDB by passing a slatedb: URI-like path to SkyShelve or any PersistentObject configuration:

from skyshelve import SkyShelve

with SkyShelve("slatedb://") as store:  # defaults to ./data/slatedb
    store["answer"] = 42

The path segment after slatedb:// points to the SlateDB data directory. Use a JSON payload for advanced configuration—e.g. "slatedb:{\"path\":\"/srv/slate\",\"store\":{\"provider\":\"s3\"}}"—which is forwarded to the SlateDB Go client. When no path is supplied, the library stores data under ./data/slatedb by default. Badger-backed stores still expect an explicit path unless you use PersistentObject, which defaults to ./data/<model-name>.

SlateDB support relies on the upstream Go bindings. Build the native library in the SlateDB repository with cargo build -p slatedb-go --release, then ensure the resulting shared object (often libslatedb_go.so/.dylib/.dll) is discoverable at run time—typically by adding it to your system library path or placing it next to libskyshelve before running go build.

To use an in-memory Badger store without touching disk, call SkyShelve(None, in_memory=True).

Quick demo & throughput glimpse

python examples/demo.py

The demo populates a store in ./data, reads a few values, and runs a small threaded benchmark, reporting elapsed time and effective operations-per-second before flushing the data to disk.

Running the concurrent stress tests

pytest tests/test_concurrency.py

The suite spins up multiple worker threads that hammer a single SkyShelve instance with random read/write/delete workloads, then verifies durability by reopening the store. Building the shared library is attempted automatically; if the Go dependencies have not been downloaded yet you will see a skip message reminding you to run go mod tidy.

Building wheels / sdists

This project uses a pyproject.toml with a setuptools backend and src/ layout. After building the Go shared library for your platform, create distributable artifacts with:

python -m pip install build
python -m build

The resulting wheel embeds the shared object located in src/skyshelve/.

Automated releases

A GitHub Actions workflow (.github/workflows/publish.yml) builds the shared library, runs the test suite, and publishes wheels/sdists to PyPI whenever a release is published (or the workflow is triggered manually). Configure the repository secret PYPI_API_TOKEN with an API token generated from your PyPI account before running the workflow.

Cleanup & caveats

• Always call close() (or use the context manager) to release the underlying handle; the backend flushes outstanding writes on close.
• Empty string keys are not supported by the wrapper.
• If you need advanced backend features (TTL, transactions, iteration), extend skyshelve.go with additional exported functions and surface them through src/skyshelve/__init__.py.