pg-dlock

PostgreSQL advisory-lock based distributed lock, sync + async.

from pg_dlock import Locker

with Locker("postgres://...") as locker:
    with locker.lock("my-key"):
        ...  # critical section

Async:

from pg_dlock import AsyncLocker

async with AsyncLocker("postgres://...") as locker:
    async with locker.lock("my-key"):
        ...

API

• Locker(conninfo, *, pool_min_size=0, pool_max_size=10)
• locker.lock(key, *, scope="session"|"transaction", shared=False) -> Lock
• lock.acquire(blocking=True, timeout=-1) -> bool
  • blocking=True, timeout=-1: block forever
  • blocking=False: non-blocking
  • blocking=True, timeout>=0 (seconds): wait up to N seconds (uses statement_timeout)
• lock.release()
• Context manager support (with lock: / async with lock:).

Scopes

• session: each lock opens its own dedicated connection (autocommit). Released via pg_advisory_unlock.
• transaction: connection is checked out from an internal pool and a transaction is begun; lock is released by the commit/rollback on exit.

Keys

• str / bytes: hashed with xxhash.xxh3_64 into a signed int64.
• int: used directly (must fit in signed int64).
• tuple[int, int]: each element must fit in signed int32; packed into a single int64.