vkmem for Python

vkmem starts the Go vkmem-server binary as a child process and returns a normal Redis/Valkey DSN. Existing clients such as redis-py, valkey-py, redis-om, and SQL-free test helpers can use it without an adapter.

pip install vkmem redis

import redis
import vkmem

with vkmem.start() as server:
    client = redis.Redis.from_url(server.dsn)
    client.set("answer", 42)
    assert client.get("answer") == b"42"
    client.close()

Prepared snapshots

Prepare a template once, then fork an independent Valkey server for each test. snapshot() writes a synchronous RDB into vkmem's in-memory filesystem; fork() starts a new guest over a private copy of that filesystem. Connections and server runtime state are not copied.

import redis
import vkmem

with vkmem.start() as process:
    prepare(redis.Redis.from_url(process.template.dsn))
    snapshot = process.template.snapshot(max_forks=4)
    try:
        with snapshot.fork() as fork:
            client = redis.Redis.from_url(fork.dsn)
            # Starts with the prepared keyspace; writes stay in this fork.
            client.set("test-only", "value")
            client.close()
    finally:
        snapshot.close()

The package also registers pgmem-style pytest fixtures when pytest is installed. Override vkmem_snapshot to prepare data, then use vkmem_fork or vkmem_dsn in each test. vkmem_class_fork and vkmem_class_dsn share one fork for a test class.

# conftest.py
import pytest
import redis

@pytest.fixture(scope="session")
def vkmem_snapshot(vkmem_server):
    client = redis.Redis.from_url(vkmem_server.dsn)
    prepare(client)
    client.close()
    return vkmem_server.snapshot(max_forks=4)

# test_orders.py
def test_write_isolated(vkmem_dsn):
    client = redis.Redis.from_url(vkmem_dsn)
    client.set("order:1", "ready")
    client.close()

The server exposes these useful attributes:

• server.dsn / server.url: redis://127.0.0.1:<port>
• server.host, server.port, and server.addr
• server.unix_socket: the Unix socket path, when enabled
• server.pid, server.version, and server.valkey_version

Pass Valkey server arguments after args, choose a port, or disable the Unix socket when a test only needs TCP:

with vkmem.start(
    args=["--maxmemory", "32mb"],
    unix_socket=False,
    timeout=30,
) as server:
    client = redis.Redis.from_url(server.dsn)

Binary lookup uses VKMEM_SERVER_BIN, then the binary bundled in the platform wheel, then vkmem-server on PATH. vkmem.find_binary() exposes that lookup directly. A local checkout can use binary="/path/to/vkmem-server".

The wheel has no runtime Python dependencies. It bundles a static binary for Linux x86_64/aarch64, macOS arm64, and Windows x86_64/arm64. The Go binary also exits when the Python process closes its stdin pipe or disappears, so a test failure does not leave a server running.