PersistDict 0.1.0

Looks like a dict and acts like a dict but is persistent via an sqlite3 db, like sqldict

PersistDict

Just a DIY version sqldict: looks like a dict and acts like a dict but is persistent via an sqlite3 db.

Why?

I ran into issue with langchain's caches when developping wdoc (my RAG lib, optimized for my use) and after months of waiting I decided to fix it myself. And instead of trusting sqldict's implementation with langchain's concurrency I made my own. This makes it very easy to add persistent cache to anything. Also it was easy to do thanks to my BrownieCutter.

Features:

• threadsafe: if several threads try to access the same db it won't be a problem. Even if multiple other threads use also another db. Thanks to a singleton class. And if several python scripts run at the same time and try to access the same db, sqlite3 should make them wait appropriately.
• atime and ctime: each entry includes a creation time and a last access time.
• expiration: won't grow too large because old keys are automatically removed.
• cached: an actual python dict is used to cache the access to the db. This cache is shared among instances, and dropped if another scripts uses the same db.
• compression: using the builtin sqlite3 compression.
• customizable serializer for the value: by default pickle is used, but could be numpy.npz, joblib.dumps, dill.dumps etc
• encryption: unsing the UNMAINTAINED library pysqlcipher3, because it was very easy to add. In the future will use an up to date library and encrypt value in place directly.
• no dependencies needed If you have beartype installed it will be used, same with loguru. Encryption comes from the UNMAINTAINED pysqlcipher3 lib. For now as I plan to move on to a simple in place encryption instead.

Differences with python dict:

• keys have to be str, that's what the sqlite db table is expecting.
• an object stored at self.missing_value is used to designate a MISSING value, so you can't pickle this object. By default it's dataclasses.MISSING.
• .clear() will throw a NotImplementedError to avoid erasing the db. If you just want to clear the cache use self.clear_cache()
• add 3 methods to 'slice' the dict with multiple key/values:
  • .getitems
  • .setitems
  • .delitems
  • Note that calling getitems with some keys missing will not return a KeyError but a self.missing_value for those keys, which by default is dataclasses.MISSING.

Usage:

• git clone https://github.com/thiswillbeyourgithub/PersistDict
• cd PersistDict
• pip install -e .
• To test that the code woks: cd PersistDict ; python PersistDict.py

# create the object
d = PersistDict(
    database_path=a_path,
    compression=True,
    password="J4mesB0nd",
    verbose=True,
)
# then treat it like a dict:
d["a"] = 1

# You can even create it via __call__, like a dict:
# d = d(a=1, b="b", c=str)  # this actually calls __call__ but is only
# allowed once per SqlDict, just like regular dict

# it's a child from dict
assert isinstance(d, dict)

# prints like a dict
print(d)
# {'a': 1, 'b': 'b', 'c': str}

# Supports the same methodas dict
assert sorted(list(d.keys())) == ["a", "b", "c"], d
assert "b" in d
del d["b"]
assert list(d.keys()) == ["a", "c"], d
assert len(d) == 2, d
assert d.__repr__() == {"a": 1, "c": str}.__repr__()
assert d.__str__() == {"a": 1, "c": str}.__str__()

# supports all the same types as value as pickle (or more if you change
# the serializer)
d["d"] = None

# new method to get and set multiple elements at the same time
assert d.__getitems__(["c", "d", "a"]) == [str, None, 1]

d.__setitems__(( ("a", 1), ("b", 2), ("c", 3), ('d', 4)))
assert d.__getitems__(["c", "d", "a", "b"]) == [3, 4, 1, 2], d.__getitems__(["c", "d", "a", "b"])

d.__delitems__(["c", "a"])
assert d.__getitems__(["b", "d"]) == [2, 4], d
assert len(d) == 2, d

# If you create another object pointing at the same db, they will share the
# same cache and won't corrupt the db:
d2 = SQLiteDict(
database_path=dbp,
compression=compr,
password=pw,
verbose=True,
)
list(d.keys()) == list(d2.keys()), d2