RSDB
a flash-sympathetic persistent lock-free B+ tree, pagecache, and log
progress
- lock-free log with reservable slots
- lock-free pagecache with cache-friendly partial updates
- lock-free b-link tree
- recovery
- zstd compression
- LRU cache
- pagetable snapshotting for faster recovery
- epoch-based gc
- higher-level interface with multi-key transaction support
- merge operator support
- formal verification of lock-free algorithms via symbolic execution
Goals
- don't use so much electricity. our data structures should play to modern hardware's strengths.
- don't surprise users with performance traps.
- bring reliability techniques from academia into real-world practice.
Architecture
Lock-free trees on a lock-free pagecache on a lock-free log. The pagecache scatters partial page fragments across the log, rather than rewriting entire pages at a time as B+ trees for spinning disks historically have. On page reads, we concurrently scatter-gather reads across the log to materialize the page from its fragments. We are friendly to cache by minimizing the
The system is largely inspired by the Deuteronomy architecture, and aims to implement the best features from RocksDB as well.