Rocksdb vs boltdb
- 1. 1 TELENAV CONFIDENTIAL RocksDB vs. BoltDB Xun Liu & Jay
- 2. 2 TELENAV CONFIDENTIAL Agenda •RocksDB • What & Why & How • Internal • Use cases •BoltDB • History & Keywords & Project Status • RocksDB vs. BoltDB • Practice in OSRM
- 5. 5 TELENAV CONFIDENTIAL What is RocksDB 1. Key-value persistent store 2. Embedded 4. Optimized for flash 3. Log Structure base tree 5. Fork of LevelDB More info: https://github.com/CodeBear801/tech_summary/blob/master/t ech-summary/tools/rocksdb/rocksdb_index.md
- 6. 6 TELENAV CONFIDENTIAL Why RocksDB Write amplification Traditional DB use B+ tree Optimize for heavy write
- 14. 14 TELENAV CONFIDENTIAL RocksDB architecture Topics we will focus on - Avoid random write - Single write + multiple read - Snapshot + multiple version concurrency control
- 16. 16 TELENAV CONFIDENTIAL Internal(Based on LevelDB) 1. Put 2. Compaction 3. Get 4. Concurrent read/write
- 17. 17 TELENAV CONFIDENTIAL RocksDB Example Write operation is extremely fast, just one disk update(WAL) + one memory write(skiplist)
- 19. 19 TELENAV CONFIDENTIAL Write Ahead Log key =bob value =3
- 20. 20 TELENAV CONFIDENTIAL Write Ahead Log key =bob value =3
- 21. 21 TELENAV CONFIDENTIAL Write Ahead Log key =bob value =3
- 23. 23 TELENAV CONFIDENTIAL Skiplist – a balance of space and performance, CAS, lock free
- 24. 24 TELENAV CONFIDENTIAL Internal(Based on LevelDB) 1. Put 2. Compaction 3. Get 4. Concurrent read/write
- 25. 25 TELENAV CONFIDENTIAL When to trigger - Size - Seek - Manual compaction
- 26. 26 TELENAV CONFIDENTIAL RocksDB put() -> compaction
- 27. 27 TELENAV CONFIDENTIAL RocksDB put() -> compaction
- 28. 28 TELENAV CONFIDENTIAL RocksDB put() -> compaction
- 29. 29 TELENAV CONFIDENTIAL Internal(Based on LevelDB) 1. Put 2. Compaction 3. Get 4. Concurrent read/write
- 31. 31 TELENAV CONFIDENTIAL RocksDB get() example
- 32. 32 TELENAV CONFIDENTIAL bloom filter If bloom filter detect key is not in the set, then it must not: never false negative If bloom filter detect the key inside the set, then there is high possibility its in, but also possible not exist, which is false positive For rocksDB, 99.9% of get() finds value in single sstable load
- 34. 34 TELENAV CONFIDENTIAL Internal(Based on LevelDB) 1. Put 2. Compaction 3. Get 4. Concurrent read/write
- 35. 35 TELENAV CONFIDENTIAL Concurrent Read/Write Challenges - Performance - What if compaction happens during query? - What if a key is updated during reading? - Consistent result(no stale data, timeout)
- 36. 36 TELENAV CONFIDENTIAL Put write tasks into queue
- 41. 41 TELENAV CONFIDENTIAL Challenges - Performance - What if compaction happens during query? - What if a key is updated during reading? - Consistent result(no stale data, timeout) Solutions: - Isolate action of compaction and query - Sequence ID + Version
- 46. 46 TELENAV CONFIDENTIAL When to delete old sstable? reference count by version
- 49. 49 TELENAV CONFIDENTIAL #1 Chrome – storage engine
- 50. 50 TELENAV CONFIDENTIAL #1 Chrome – storage engine
- 51. 51 TELENAV CONFIDENTIAL #2 Calculation Engine – record state
- 52. 52 TELENAV CONFIDENTIAL #3 Realtime newsfeed processing More Writes More Data More Query
- 53. 53 TELENAV CONFIDENTIAL #4. Map Making/Map data aggregator
- 54. 54 TELENAV CONFIDENTIAL RocksDB Resources Main pages • https://github.com/facebook/rocksdb • https://github.com/facebook/rocksdb/wiki • https://github.com/google/leveldb/tree/master/doc Posts • RocksDB: A High Performance Embedded Key-Value Store for Flash Storage • Code reading notes about LevelDB Practices • RocksDB usage at LinkedIn • Linked FollowFeed - (ALT - Aggregator Leaf Tailer)
- 56. 56 TELENAV CONFIDENTIAL BoltDB History • 2011, Howard Chu introduced MDB, a memory-mapped database backend for OpenLDAP, later renamed to LMDB (Lightning Memory-Mapped Database). • 2013, Bolt initially started by Ben Johnson as a port of LMDB to Go, but then the two projects diverged. • The author of Bolt decided to focus on simplicity and providing the easy-to-use Go API. • 2018, Blot stable and widely used. Etcd-io forked for further improvement.
- 57. 57 TELENAV CONFIDENTIAL BoltDB Keywords • Key/value • Embedded • Simplicity and easy-to-use API • Fast • B+ tree • Pure Go • Memory-mapped
- 58. 58 TELENAV CONFIDENTIAL BoltDB vs RocksDB • B+ tree vs. LSM tree • Optimizes for • read-heavy/range scans vs. random writes • Go vs. C++ https://github.com/etcd-io/bbolt#comparison-with-other-databases
- 59. 59 TELENAV CONFIDENTIAL BoltDB – Stable & Widely used Project Status • Bolt is stable, the API is fixed, and the file format is fixed. • Full unit test coverage and randomized black box testing are used to ensure database consistency and thread safety. • Bolt is currently used in high-load production environments serving databases as large as 1TB. • Many companies such as Shopify and Heroku use Bolt- backed services every day. ÞBackend store of etcd Þhttps://github.com/boltdb/bolt#other-projects-using-bolt
- 60. 60 TELENAV CONFIDENTIAL BoltDB Practice in OSRM Generate DB • Stores fromNodeID,toNodeID->wayID mapping, each key costs 16bytes, each value costs 8 bytes. • DB file is about 28 GB(about 9.2GB after snappy compressed), ~5 billions keys. • Generating process is about 4 hours. Query • Query tens of nodes only takes 1ms! • Query 226390 wayIDs from 226394 nodeIDs, takes 1.399967 seconds. (shm) References § https://github.com/Telenav/osrm-backend/issues/257 § https://github.com/Telenav/osrm-backend/issues/272
- 61. 61 TELENAV CONFIDENTIAL BoltDB Resources Main pages • https://github.com/boltdb/bolt • https://github.com/etcd-io/bbolt • https://github.com/bmatsuo/lmdb-go • https://github.com/dgraph-io/badger Posts • Intro to BoltDB: Painless Performant Persistence by Nate Finch. • Bolt -- an embedded key/value database for Go by Progville • https://tech.townsourced.com/post/boltdb-vs-badger/ • https://dgraph.io/blog/post/badger-lmdb-boltdb/ Practices • https://github.com/Telenav/osrm-backend/issues/257 • https://github.com/Telenav/osrm-backend/issues/272