bloomfilter-py

Overview

Yet another Bloomfilter implementation in Python, compatible with Java's Guava library.

Compatibility targets Guava's compact BloomFilter.writeTo / readFrom format. Java ObjectOutputStream / ObjectInputStream serialization is a different format and is not supported. The compact format does not include the funnel; callers must provide the matching encoder when loading a filter.

I was looking for a Python library which is capable of reading what Bloomfilter of Java's Guava library serializes and is also able to output byte array which is recognizable by Java. But unfortunately failed. Hence I developed this library by borrowing how Guava implements Bloomfilter serialization/deserialization a lot to deal with Bloomfilters on both Python and Java sides.

As for Bloomfilter usage in Java world, please refer to this post.

Here's a brief introduction to Bloomfilter.

Requirements

• Python 3.8+

Install

pip install bloomfilter-py

Usage Examples

Basic Usage

>>> from bloomfilter import BloomFilter
>>> bloom_filter = BloomFilter(expected_insertions=500, err_rate=0.01)
>>> for i in range(100):
...     bloom_filter.put(i)
...
>>> 1 in bloom_filter
True
>>> 100 in bloom_filter
False
>>>

Guava Funnel Compatibility

Guava's compact BloomFilter format stores the hash strategy and bit array, but does not store its Funnel. To query a filter produced by Guava, select the same funnel that was used in Java:

from bloomfilter import BloomFilter, INTEGER_FUNNEL

with open("guava-filter.out", "rb") as f:
    bloom_filter = BloomFilter.loads(f.read(), funnel=INTEGER_FUNNEL)

The built-in mappings are:

StringFunnel supports Java's six standard charsets: US-ASCII, ISO-8859-1, UTF-8, UTF-16, UTF-16BE, and UTF-16LE. Custom Guava funnels are not encoded in the serialized filter; use a matching custom Python Funnel implementation when reading them.

Use the same funnel when creating a filter that Java will read:

from bloomfilter import BloomFilter, LONG_FUNNEL

bloom_filter = BloomFilter(500, 0.01, funnel=LONG_FUNNEL)
bloom_filter.put(1)  # Encoded as a Java long even though 1 fits in an int.

Calls that omit funnel retain the historical bloomfilter-py behavior, which chooses integer width from each value and therefore is not suitable for all Guava interoperability scenarios.

Serialize Bloomfilter

You can easily serialize BloomFilter instance to a byte array

>>> dumps = bloom_filter.dumps()
>>> with open("dumps.out", "wb") as f:
...     f.write(dumps)
...
>>>

or to a hex string

>>> hex_str = bloom_filter.dumps_to_hex()

or to a base64 encoded bytes

base64_bytes = bloom_filter.dumps_to_base64()

Deserialize Bloomfilter

And you can easily initialize a BloomFilter instance from a byte array

>>> with open("dumps.out", "rb") as f:
...     bf = BloomFilter.loads(f.read())
...
>>> 1 in bf
True
>>> 100 in bf
False
>>>

or from a hex string

>>> bf = BloomFilter.loads_from_hex(hex_str)
>>> 1 in bf
True
>>> 100 in bf
False

or from a base64 encoded bytes

>>> bf = BloomFilter.loads_from_base64(base64_bytes)
>>> 100 in bf
False
>>> 200 in bf
False
>>> 1 in bf
True
>>> 99 in bf
True

Deserialization limits the backing bit array to 2³⁰ bits by default. For a trusted larger Guava filter, pass an explicit limit to any loading method:

bf = BloomFilter.loads(dumps, funnel=INTEGER_FUNNEL, max_num_bits=2**31)

Stream I/O

Read and write compact Guava data directly with binary streams:

with open("guava-filter.out", "rb") as stream:
    bf = BloomFilter.read_from(stream, funnel=INTEGER_FUNNEL)

with open("python-filter.out", "wb") as stream:
    bf.write_to(stream)

byte_count = bf.serialized_size()

read_from reads exactly one filter, leaving any following stream data unread. It accepts the same max_num_bits limit as loads, and validates the header before reading the payload. loads requires exactly one filter with no trailing bytes. Neither stream method closes the caller's stream. Both handle partial reads or writes on blocking binary streams.

Copies, Merging, and Statistics

snapshot = bf.copy()
other = BloomFilter(500, 0.01, funnel=INTEGER_FUNNEL)
other.put(123)

if bf.is_compatible(other):
    bf.put_all(other)

false_positive_probability = bf.expected_fpp()
estimated_distinct_elements = bf.approximate_element_count()

copy duplicates the backing bit array and preserves the loading limit. It shares the funnel and stateless strategy, as Guava does; custom funnels should be immutable. put_all performs an in-place union without modifying the source. Compatibility requires distinct instances with the same bit-array size, number of hashes, strategy, and equal funnels. Custom funnels must define equality when separate instances represent the same encoding.

Statistics use the current bit occupancy, so they also work on loaded and merged filters. A fully saturated filter has an expected FPP of 1.0; its element count cannot be estimated and raises OverflowError, corresponding to Guava's ArithmeticException. Filter equality compares configuration, funnel, and bit contents. Mutable filters are unhashable in Python.

Unlike Guava's atomic implementation, this library does not guarantee safe concurrent mutations. Use an external lock around mutations, queries, copies, statistics, and serialization when sharing a filter between threads. Java's deprecated predicate aliases and stream collectors are language-specific APIs; use might_contain (or in) and a loop calling put in Python.

Development and Testing

From the repository root, create an isolated virtual environment and install the project in editable mode with pytest:

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e . pytest

Run the complete test suite:

python -m pytest -v

The bidirectional Java integration test requires a JDK and a Guava JAR. Point GUAVA_JAR at the JAR to compile the included Java helper and verify both Guava-to-Python and Python-to-Guava serialization:

GUAVA_JAR=/path/to/guava-33.7.2-jre.jar \
  python -m pytest -v tests/test_guava_interop.py

The test is skipped when those prerequisites are unavailable. CI downloads the pinned Guava release, verifies its SHA-256 checksum, and runs the test as a required separate job.

To run the same formatting, type, and security checks used during development, install the additional tools and execute:

python -m pip install black mypy bandit
python -m black --check bloomfilter tests
python -m mypy bloomfilter --strict
python -m bandit -r bloomfilter

On Windows PowerShell, activate the virtual environment with .venv\Scripts\Activate.ps1 instead of source .venv/bin/activate.