- 基于redis和memory
- 低时间复杂度
from pyfilters import MemoryBloomFilter
bf = MemoryBloomFilter(10000, 0.00001)
for i in range(1000):
bf.add(i)
for i in range(1000):
assert i in bf
assert 1001 not in bf
- 计数形布隆过滤器,可以删除数据
from pyfilters import CountMemoryBloomFilter
cbf = CountMemoryBloomFilter(10000, 0.00001)
for i in range(1000):
cbf.add(i)
for i in range(1000):
assert i in cbf
cbf.remove(1)
assert 1 not in cbf
- redis分块布隆过滤器,避免单key过大
from redis import Redis
from pyfilters import ChunkedRedisBloomFilter
bf = ChunkedRedisBloomFilter(Redis(), "test_bloomfilter", 10000, 0.00001)
for i in range(1000):
bf.add(i)
for i in range(1000):
assert i in bf
assert 1001 not in bf
- 分块计数形redis布隆过滤器,可以删除数据
from redis import Redis
from pyfilters import CountRedisBloomFilter
rcbf = CountRedisBloomFilter(Redis(), "test_countbloomfilter", 10000, 0.00001)
for i in range(1000):
rcbf.add(i)
for i in range(1000):
assert i in rcbf
rcbf.remove(1)
assert 1 not in rcbf
在pyfilters.asyncio包
import asyncio
from aioredis import Redis
from pyfilters.asyncio import CountRedisBloomFilter
async def main():
rcbf = CountRedisBloomFilter(Redis(), "test_countbloomfilter", 10000, 0.00001)
for i in range(1000):
await rcbf.add(i)
for i in range(1000):
assert await rcbf.contains(i)
await rcbf.remove(1)
assert not await rcbf.contains(1)
asyncio.run(main())