performance: tuning rocksdb bloom filter

[quake]

What problem does this PR solve?

We previously used a global lru cache as rocksdb cache (default 128M) and setting the cache_index_and_filter_blocks parameter to true, the rocksdb's IndexBlock is placed in the cache also (note, we did not use FilterBlock in current code). With the increasing of stored data, DataBlock and IndexBlock will compete for cache expiration, there will be more cache misses, resulting in read performance degradation.

rocksdb statistics log

Block cache LRUCache@0x7f41c443d3d0#3747917 capacity: 128.00 MB seed: 886014176 usage: 117.24 MB table_size: 32768 occupancy: 5795 collections: 155 last_copies: 19 last_secs: 0.001642 secs_since: 0
Block cache entry stats(count,size,portion): DataBlock(5543,23.49 MB,18.3507%) IndexBlock(251,93.22 MB,72.8312%) Misc(1,0.00 KB,0%)

What is changed and how it works?

This PR tries to adjust the cache configuration, and also uses a bloom filter to improve the performance of prefix seek.

through the mainnet synchronization test, the overall performance has been improved by about 6%.

and memory usage has also decreased

Check List

Tests

• Unit test
• Integration test

Release note

Title Only: Include only the PR title in the release note.

[zhangsoledad]

nervosnetwork/rust-rocksdb#52