Releases: fjall-rs/lsm-tree
Releases · fjall-rs/lsm-tree
2.5.0
- [api] (Possibly breaking?) Use Into<...> for insert operations #85 by @carlsverre
- [perf] L0 and L1 segments now use the full block index #51
- [perf] Fix snapshot point reads in memtable for update-heavy workloads 8db1f38
- [perf] Lazily evaluate range bounds #62
- [perf] Allow parallel compactions for Leveled compaction strategy #83
- [perf] Decreased bloom filter FPR for L0 and L1 segments
- [perf] Group disjoint runs in compaction worker to reduce key comparisons #84
- [perf] Simplify segment snapshot point reads, removing a heap allocation 9113cb4 ab28434
- [misc] Internal refactors
- [misc] Better handle I/O errors in compaction worker
- [fix] make sure KV versions cannot span segments
2.5.0-pre.0
revert crossbeam-skiplist for now
2.4.0
- [feat] Added Tokio Bytes feature flag, by @carlsverre
- [feat] Added Tree::size_of
2.3.3
- [refactor] Internal refactor
- [misc] Change size tiered segment base size to 64 MB
2.3.2
- [fix] Don't build bloom filters if bits_per_key == -1
2.3.0
Blog post: https://fjall-rs.github.io/post/fjall-2-3/
- [perf] Improved performance of Leveled compaction for random writes
- [perf] Use lz4_flex unsafe mode
- [perf] Fix performance degradation of point reads in certain workloads
- [perf] Rewrite segments to reach the segment target size and decrease bloom filter memory pressure, even in monotonic workloads
- [fix] Make sure blob GC cannot scan blob files that are referenced by LSM-tree segments that are still in-flight
- [fix] Make sure blob GC cannot drop blob files that could still be referenced by a snapshot
- [fix] Don't try reading bloom filter from a database that was built without the
bloom
feature - [dep] Update value-log
- [refactors] Internal refactors
- [api] Moved blob GC stuff into the
gc
module
2.1.1
- [fix] Disjoint tree checking logic
- [refactor] Internal newtype refactors
2.1.0
2.0.2
- [perf] Fix snapshot point read performance
2.0.0
recreate v2 test fixture with bloom filter