Libraries: Guava, Redis, RocksDB, and Others

In one line

Use proven implementations: Guava's BloomFilter (Java, thread-safe, serializable, mergeable), RedisBloom's BF.* commands (shared, scalable), storage-engine filters in RocksDB, Cassandra and HBase, and libraries like bits-and-blooms (Go) or pybloom-live (Python). Know their parameters and limits.

0/4 · 0%

Think of it like this

Buying a certified fire extinguisher instead of building one. You still need to know how it works and where to mount it.

Key ideas

01
Guava: BloomFilter.create(Funnels.stringFunnel(UTF_8), expectedInsertions, fpp), put, mightContain, expectedFpp(), approximateElementCount(), putAll(other) for merging compatible filters, writeTo/readFrom for persistence. Thread-safe since Guava 23.
02
Redis (Redis 8 core, RedisBloom, or valkey-bloom on Valkey): BF.RESERVE key error_rate capacity [EXPANSION n] [NONSCALING], BF.ADD, BF.MADD, BF.EXISTS, BF.MEXISTS, BF.INSERT, BF.INFO, BF.CARD, BF.SCANDUMP/BF.LOADCHUNK for backup. Scaling by default (adds sub-filters when full).
03
Storage engines: RocksDB (bloom_bits_per_key, often 10, with cache-local Bloom and Ribbon filters), LevelDB (NewBloomFilterPolicy(10)), Cassandra (bloom_filter_fp_chance per table), HBase (row or row+column blooms per HFile), Parquet (split-block Bloom filters per column chunk).
04
Other languages: Go (bits-and-blooms/bloom), Python (pybloom-live, rbloom), Rust (fastbloom, xorf for Xor filters).
05
Library checks before adopting: hash algorithm and stability across versions, serialization compatibility, thread safety, maximum size, merge support, and whether scalable growth is automatic.

Code & diagrams

GuavaExample.javajava

BloomFilter<CharSequence> seen = BloomFilter.create(Funnels.stringFunnel(UTF_8), 100_000_000, 0.01);
seen.put("https://example.com/a");
boolean maybe = seen.mightContain("https://example.com/a");          // true
double fpp = seen.expectedFpp();                                      // current estimate
long approx = seen.approximateElementCount();

BloomFilter<CharSequence> other = BloomFilter.create(Funnels.stringFunnel(UTF_8), 100_000_000, 0.01);
if (seen.isCompatible(other)) seen.putAll(other);                     // bitwise OR merge

redis-bloom.redisredis

127.0.0.1:6379> BF.RESERVE bf:products 0.01 100000000 EXPANSION 2
OK
127.0.0.1:6379> BF.MADD bf:products p:1001 p:1002 p:1003
1) (integer) 1
2) (integer) 1
3) (integer) 1
127.0.0.1:6379> BF.MEXISTS bf:products p:1002 p:999999
1) (integer) 1
2) (integer) 0
127.0.0.1:6379> BF.INFO bf:products
 1) Capacity                 2) (integer) 100000000
 3) Size                     4) (integer) 119816312
 5) Number of filters        6) (integer) 1
 7) Number of items inserted 8) (integer) 3
 9) Expansion rate          10) (integer) 2

Explain it without notes

What should you verify before adopting a Bloom filter library in production?

Practice

Create the same 10M-item, 1% filter with Guava and RedisBloom and compare memory, lookup latency and measured FPR.

Trade-offs

↔
Local libraries are fastest but per-instance; Redis is shared and scalable but adds network latency and a dependency.

Done when you can

I know the main Bloom filter libraries and engine settings and what to check before using one.