Pythomnic - platform for building reliable network services in Python

Like any other concurrent database BerkeleyDB also suffers from locking problems, and like many it resolves deadlocks by aborting one of the conflicting transactions. This is one other thing your application code should be aware of. Unless you lock the safe exclusively (more on this below), there is a chance that upon a concurrent write access even the simplest

pmnc_safe["foo"] = "bar"

will throw a DBLockDeadlockError which needs to be handled in the following manner:

while True:
    try:
        pmnc_safe["foo"] = "bar"
    except db.DBLockDeadlockError:
        continue
    else:
        break

To resolve both of the mentioned problems, pmnc_safe supports advisory interlocking across different threads. A safe could be locked exclusively:

pmnc_safe.acquire()
try:

    # this thread has exclusive access to the safe,
    # no other thread could interfere or deadlock

    pmnc_safe["foo"] = "bar"
    pmnc_safe["biz"] = "baz"

finally:
    pmnc_safe.release()

or in shared mode:

pmnc_safe.acquire_shared()
try:

    # other threads could access the safe, but none 
    # of them could have acquired exclusive lock

    # as soon as this access is read-only, you don't
    # have to worry about the deadlocks

    assert "foo" in pmnc_safe and "biz" in pmnc_safe

finally:
    pmnc_safe.release_shared()

This approach looks like a sufficient replacement to serializable transactions whenever the module enforces interlocking policy as necessary.

« prev

Features
Download
Documentation
Tutorial