BloomFilter(s) in Ruby: Native counting filter + Redis counting/non-counting filters
Branch: master
Clone or download
Latest commit a092bc1 Apr 26, 2017
Type Name Latest commit message Commit time
Failed to load latest commit information.
benchmark for redis lookups, first query the first bit, if not set then missing, Mar 31, 2011
ext/cbloomfilter fix bit padding bug in bucket_set/unset Aug 15, 2013
lib bump version (bugfix) Jan 5, 2015
spec disable deletes on non-counting redis filter Jan 5, 2015
.gitignore changed marshal #dump #load format to be size of in-memory bitmap Dec 5, 2012
Gemfile bundlerize + rspec2 cleanup Jan 5, 2011 Update Readme Apr 26, 2017
Rakefile Reorder requires and add spec prerequisites Mar 6, 2012
bloomfilter-rb.gemspec Update to rspec 3 and modern "expect" syntax. Aug 1, 2014

BloomFilter(s) in Ruby

  • Native (MRI/C) counting bloom filter
  • Redis-backed getbit/setbit non-counting bloom filter
  • Redis-backed set-based counting (+TTL) bloom filter

Bloom filter is a space-efficient probabilistic data structure that is used to test whether an element is a member of a set. False positives are possible, but false negatives are not. For more detail, check the wikipedia article. Instead of using k different hash functions, this implementation seeds the CRC32 hash with k different initial values (0, 1, ..., k-1). This may or may not give you a good distribution, it all depends on the data.

Performance of the Bloom filter depends on a number of variables:

  • size of the bit array
  • size of the counter bucket
  • number of hash functions


MRI/C API Example

MRI/C implementation which creates an in-memory filter which can be saved and reloaded from disk.

require 'bloomfilter-rb'

bf = => 100, :hashes => 2, :seed => 1, :bucket => 3, :raise => false)
bf.include?("test")     # => true
bf.include?("blah")     # => false

bf.include?("test")     # => false

# Hash with a bloom filter!
bf["test2"] = "bar"
bf["test2"]             # => true
bf["test3"]             # => false

# => Number of filter bits (m): 10
# => Number of filter elements (n): 2
# => Number of filter hashes (k) : 2
# => Predicted false positive rate = 10.87%

Redis-backed setbit/getbit bloom filter

Uses getbit/setbit on Redis strings - efficient, fast, can be shared by multiple/concurrent processes.

bf =

bf.include?('test')     # => true
bf.include?('blah')     # => false

bf.include?('test')     # => false

Memory footprint

  • 1.0% error rate for 1M items, 10 bits/item: 2.5 mb
  • 1.0% error rate for 150M items, 10 bits per item: 358.52 mb
  • 0.1% error rate for 150M items, 15 bits per item: 537.33 mb

Redis-backed counting bloom filter with TTLs

Uses regular Redis get/set counters to implement a counting filter with optional TTL expiry. Because each "bit" requires its own key in Redis, you do incur a much larger memory overhead.

bf = => 2)

bf.include?('test')     # => true

bf.include?('test')     # => false


Tatsuya Mori (Original C implementation:


MIT License - Copyright (c) 2011 Ilya Grigorik