A pure Python, zero dependency bloom filter.
Project description
VSTG
VSTG Skips Table Gets
A pure Python, zero dependency bloom filter, built so a check that would otherwise hit your database on every request almost never has to.
The name is also short for vestige, a trace left behind by something that passed through, which is what the underlying bit array actually is.
The problem this solves
Before:
def check_username(request):
username = request.GET["username"]
taken = User.objects.filter(username=username).exists()
return JsonResponse({"available": not taken})
After:
import vstg
def check_username(request):
username = request.GET["username"]
if not vstg.might_contain("usernames", username.encode("utf-8")):
return JsonResponse({"available": True})
taken = User.objects.filter(username=username).exists()
return JsonResponse({"available": not taken})
Every username that was never registered returns without the database being touched. The rare "maybe" still runs the exact same query as before, so correctness never changes, only how often the expensive path is needed.
Install
Not yet published to PyPI. Until it is:
pip install git+https://github.com/vszlx4/vstg.git
Or, cloned and editable, for contributing:
git clone https://github.com/vszlx4/vstg.git
cd vstg
pip install -e '.[dev]'
Once published, this becomes pip install vstg. No dependencies are
ever installed alongside it either way.
Usage
from pathlib import Path
import vstg
vstg.init(Path("bloom_state"))
vstg.register("usernames", capacity=10_000_000, error_rate=0.001)
vstg.might_contain("usernames", username.encode("utf-8")) # check
vstg.add("usernames", username.encode("utf-8")) # record
init and register run once, at startup. might_contain and add
run everywhere else, for the life of the process.
Customization
vstg.register(
"usernames",
capacity=10_000_000,
error_rate=0.001,
policy=vstg.PersistPolicy(
checkpoint_interval_seconds=300,
checkpoint_insert_threshold=50_000,
checkpoint_on_shutdown=True,
),
)
vstg.register(
"email_verification_tokens",
capacity=500_000,
error_rate=0.01,
shard_count=8,
)
Every named filter is independent, its own capacity, error rate, and
policy. shard_count, left unset, gives a single checkpoint file. Set
it and a checkpoint only rewrites the shards that actually changed.
Capacity, error rate, and policy are fixed the moment register is
called, for the rest of the process.
The math
A filter is a bit array of length $m$, all bits starting at $0$. Inserting a member sets $k$ of those bits, each position derived by hashing the member. Checking a member reads the same $k$ positions: any bit still $0$ proves the member was never inserted, since insertion only ever sets bits. All $k$ bits set means probably inserted, since an unrelated combination of other members could have set the same positions by coincidence.
Sizing, given an expected item count $n$ (capacity) and a target
false positive probability $p$ (error_rate):
$$ m = -\frac{n \ln p}{(\ln 2)^2} \qquad\qquad k = \frac{m}{n} \ln 2 $$
optimal_size and optimal_hash_count in core.py compute exactly
these two values.
Bit positions, rather than running $k$ separate hash functions, two hashes are computed once and every position derived from them (Kirsch-Mitzenmacher):
$$ i_j = (h_1 + j \cdot h_2) \bmod m, \qquad j = 0, \dots, k-1 $$
Resulting false positive rate, after $n$ insertions into an $m$-bit array with $k$ hash rounds:
$$ p \approx \left(1 - e^{-kn/m}\right)^k $$
The sizing formulas above are exactly this expression solved for the
$m$ and $k$ that minimize it at the requested $n$ and $p$. Exceeding
the configured capacity degrades the real rate past what was
requested, since $n$ in this formula grows while $m$ and $k$ stay
fixed.
Full API
BloomFilter— the bit array,add,might_contain, in memory only.ManagedBloomFilter— aBloomFilterplus automatic checkpointing.ShardedBloomFilter— checkpointing partitioned across shard files.BloomFilterRegistry— the named collectionvstg.registeruses internally.PersistPolicy— checkpoint trigger configuration for all three above.
Contributing
Two space indentation, mypy --strict, no runtime dependencies, a
docstring on every module, class, and function.
git clone https://github.com/vszlx4/vstg.git
cd vstg
pip install -e '.[dev]'
python -m unittest discover -s tests -v
License
MIT, see LICENSE.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file vstg-0.1.0.tar.gz.
File metadata
- Download URL: vstg-0.1.0.tar.gz
- Upload date:
- Size: 33.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.10.8
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
d1c534ed48b56141253c927262e290575e5c805b51a041dc1d2193599cf0980a
|
|
| MD5 |
aa07e40bd502543c967008a879b89a94
|
|
| BLAKE2b-256 |
f899c4a7d04de3e1b0471c55e0dbe6a6b837d95fd54085d4e7f08c82b7902929
|
File details
Details for the file vstg-0.1.0-py3-none-any.whl.
File metadata
- Download URL: vstg-0.1.0-py3-none-any.whl
- Upload date:
- Size: 24.3 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.10.8
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
214e67e6e8bab9b101e068c331044ba51f3074c683f21e944f73caa0ae52bd87
|
|
| MD5 |
9ee78c6ce975ff1288fdec4fcd818b0c
|
|
| BLAKE2b-256 |
966a56a77e9651a5674ba2d95880048dffa2cdcdecb1cacb22fbcfa232ed3ded
|