2.3.1

Blog post: https://fjall-rs.github.io/post/fjall-2-3/

• [perf] Increase default journal preallocated size to 32 MB
• [perf] Skip journal flush to OS buffers if they are not dirty
• [doc] Update Readme
• [doc] Fix some docs
• [misc] Relaxed write stall thresholds to more-or-less match RocksDB's

2.3.0

coming soon

2.2.0

Blog post: https://fjall-rs.github.io/post/announcing-fjall-22

• [feat] Added ssi_tx feature flag for serializable multi-writer transactions #39 #79 by @jeromegn
• [api] Bind lifetime of Write Tx iterators to the Write Tx
• [example] Added SSI transaction examples

2.1.2

• [fix] make sure non-transactional workloads can reclaim space #85
• [misc] give spawned threads names
• [feat] add '$' to partition name character set #89
• [test] the above

2.1.1

• [fix] update lsm-tree to 2.1.1 - fix disjoint tree logic
• [fix] correctly check for fatal I/O error on write batch commit
• [fix] benchmarks by @jonahlund
• [perf] skip insert op if fetch_update or update_fetch return the same value as read

2.1.0

Blog post: https://fjall-rs.github.io/post/fjall-2-1/

• [docs] Update TxKeyspace::persist doc

From lsm-tree

• [perf] Improve performance of open ranges fjall-rs/lsm-tree#56
• [perf] Improve performance of range culling in large trees fjall-rs/lsm-tree#55
• [perf] Use rustc_hash inside BlockCache
• [perf] Do less Arc clones inside Segment construction
• [refactor] Internal refactors

2.0.3

• [fix] Fix config builder blob compression handling when not using default
• [test] the above

2.0.2

• [perf] Fix snapshot point read performance

2.0.1

• [breaking fix] Compression option not being applied to blobs

This is unfortunately a breaking bug fix IF key-value separation was enabled in 2.0.0.

2.0.0

Blog: https://fjall-rs.github.io/post/announcing-fjall-2/

NOTE: This version is not compatible with 1.x.x databases

Features

• Key-value separation & GC API
• Added miniz compression (feature flag miniz)
• Added keys() & values() iterators
• Allow values bigger than 64K
• Added durability setting for write batches and write transactions (default = Flush buffers to OS)

Other changes

• Changed partition config to also store memtable size & compaction strategy
• Simplified journal implementation at the cost of some multithread performance if large values are written
• Improved journal eviction performance
• Made write stalling less aggressive
• Updated lsm-tree to 2.0.0
• Update docs
• Internal refactors
• Changed all Arc<[u8]> output types to Slice
• Added more examples