postgresql-lock 0.2.2

postgresql-lock

Lock mechanism implemented with PostgreSQL advisory locks.

Easily implement distributed database locking.

Install

pip install postgresql-lock

Supported database interfaces

• asyncpg
  • asynchronous
• psycopg2
  • synchronous
• psycopg3
  • asynchronous
  • synchronous
• sqlalchemy (supports version 1 & 2; can use any underlying database interface)
  • asynchronous
  • synchronous

Why would I use this?

• PostgreSQL table locks aren't sufficient for your use-case
• PostgreSQL row locks don't work on INSERT
• You want to prevent race conditions between INSERT and UPDATE on the same primary key
• None of the aforementioned details fit your use-case, but you have PostgreSQL installed and need to prevent race conditions in a distributed system

Default operation

By default postgresql-lock will use session lock scope in blocking mode with rollback_on_error enabled. The session lock scope means only a single database connection can acquire the lock at a time.

Usage

All work revolves around the Lock class.

The easiest way to use Lock is with with or async with statements. The lock will be released automatically. If rollback_on_error is enabled (default), rollbacks are automatically handled prior to release.

Using with and async with implies blocking mode.

from postgresql_lock import Lock

# setup connection
conn = ...

# create and use lock
with Lock(conn, "shared-identifier"):
    print("Acquired lock!")

    # do something here

Now compare the above example to the equivalent try/finally example below:

from postgresql_lock import Lock

# setup connection
conn = ...

# create lock
lock = Lock(conn, "shared-identifier")

try:
    # acquire lock
    lock.acquire()

    print("Acquired lock!")

    try:
        # do something here
        pass

    except Exception as exc:
        # handle_error() will rollback the transaction by default
        lock.handle_error(exc)

        raise exc
finally:
    # release lock (this is safe to run even if the lock has not been acquired)
    lock.release()

Asynchronous usage (without async with)

from postgresql_lock import Lock

# setup connection
conn = ...

# create lock
lock = Lock(conn, "shared-identifier")

try:
    # acquire lock
    await lock.acquire_async()

    print("Acquired lock!")

    try:
        # do something here
        pass

    except Exception as exc:
        # handle_error_async() will rollback the transaction by default
        await lock.handle_error_async(exc)

        raise exc
finally:
    # release lock (this is safe to run even if the lock has not been acquired)
    await lock.release_async()

Non-blocking mode (supports async as well)

from postgresql_lock import Lock

# setup connection
conn = ...

# create lock
lock = Lock(conn, "shared-identifier")

# acquire lock
if lock.acquire(block=False):
    # do something here
    pass

else:
    # could not acquire lock
    pass

# release lock (this is safe to run even if the lock has not been acquired)
lock.release()

Specify the database interface manually

from postgresql_lock import Lock

# setup connection
conn = ...

# create and use lock
lock = Lock(conn, "shared-identifier", interface="asyncpg")

# do things with the lock

Handle rollbacks manually

from postgresql_lock import Lock

# setup connection
conn = ...

# create and use lock
lock = Lock(conn, "shared-identifier", rollback_on_error=False)

# do things with the lock

Changelog

• 0.2.2
  • Fix: error raised when releasing transaction level locks
• 0.2.1
  • Moved public Lock fields to properties
• 0.1.9
  • Fix: release_async() bug for sqlalchemy connections
• 0.1.8
  • Add logger() function
  • Use "postgresql_lock" logger name
• 0.1.7
  • Add logging statements
• 0.1.6
  • Use int.from_bytes() to convert lock key into integer
  • Fix: psycopg3 close() not being awaited bug
• 0.1.5
  • Rename package from postgres-lock to postgresql-lock
• 0.1.4
  • Add py.typed for mypy
• 0.1.3
  • Key can be any object
• 0.1.2
  • Add Lock.rollback_on_error (default true)
  • Add Lock.handle_error() & Lock.handle_error_async()
• 0.1.1
  • Key can be str or int