Skip to content

symfonic.memory.layers.procedural.review_lock

review_lock

One in-process ordering point for review decisions and generic graph writes.

The lock is shared by wrappers over the same backend. Cross-process ordering still requires a backend compare-and-set operation; this is not that claim.

durable_backend_for 

durable_backend_for(candidate: Any) -> Any

Resolve graph/store/journal/fence wrappers to the same backend identity.

Source code in src/symfonic/memory/layers/procedural/review_lock.py 
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
def durable_backend_for(candidate: Any) -> Any:
    """Resolve graph/store/journal/fence wrappers to the same backend identity."""
    seen: list[Any] = []
    for _ in range(8):
        if any(candidate is item for item in seen):
            break
        seen.append(candidate)
        for attribute in ("unfenced", "durable", "backend", "graph"):
            wrapped = getattr(candidate, attribute, None)
            if wrapped is not None and wrapped is not candidate:
                candidate = wrapped
                break
        else:
            break
    return candidate

review_lock_for 

review_lock_for(backend: Any) -> asyncio.Lock

Return the task-reentrant review/write lock for one backend identity.

Source code in src/symfonic/memory/layers/procedural/review_lock.py 
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
def review_lock_for(backend: Any) -> asyncio.Lock:
    """Return the task-reentrant review/write lock for one backend identity."""
    backend = durable_backend_for(backend)
    try:
        lock = _LOCKS.get(backend)
        if lock is None:
            lock = _TaskOwnedLock()
            _LOCKS[backend] = lock
        return lock
    except TypeError:
        held = _STRONG_LOCKS.get(id(backend))
        if held is None:
            held = (backend, _TaskOwnedLock())
            _STRONG_LOCKS[id(backend)] = held
        return held[1]