9 unstable releases (3 breaking)

0.6.3 Feb 27, 2024
0.6.2 Feb 13, 2024
0.6.1 Jan 25, 2024
0.5.0 Jan 18, 2024
0.1.0 Dec 21, 2023

#26 in Database implementations

Download history 75/week @ 2024-01-07 97/week @ 2024-01-14 47/week @ 2024-01-21 22/week @ 2024-01-28 59/week @ 2024-02-04 83/week @ 2024-02-11 58/week @ 2024-02-18 184/week @ 2024-02-25 56/week @ 2024-03-03 48/week @ 2024-03-10 1/week @ 2024-03-17 34/week @ 2024-03-31

87 downloads per month

MIT/Apache

150KB
3K SLoC

CI docs.rs Crates.io MSRV

Fjall is an LSM-based embedded key-value storage engine written in Rust. It features:

  • Thread-safe BTreeMap-like API
  • 100% safe & stable Rust
  • Range & prefix searching with forward and reverse iteration
  • Cross-partition snapshots (MVCC)
  • Automatic background maintenance

Each Keyspace is a single logical database and is split into partitions (a.k.a. column families) - you should probably only use a single keyspace for your application. Each partition is physically a single LSM-tree and its own logical collection; however, write operations across partitions are atomic as they are persisted in a single database-level journal, which will be recovered on restart.

It is not:

  • a standalone server
  • a relational database
  • a wide-column database: it has no notion of columns

Keys are limited to 65536 bytes, values are limited to 2^32 bytes. As is normal with any kind of storage engine, larger keys and values have a bigger performance impact.

For the underlying LSM-tree implementation, see: https://crates.io/crates/lsm-tree.

Basic usage

cargo add fjall
use fjall::{Config, Keyspace, PartitionCreateOptions};

let keyspace = Config::new(folder).open()?;

// Each partition is its own physical LSM-tree
let items = keyspace.open_partition("my_items", PartitionCreateOptions::default())?;

// Write some data
items.insert("a", "hello")?;

// And retrieve it
let bytes = items.get("a")?;

// Or remove it again
items.remove("a")?;

// Search by prefix
for item in &items.prefix("prefix") {
  // ...
}

// Search by range
for item in &items.range("a"..="z") {
  // ...
}

// Iterators implement DoubleEndedIterator, so you can search backwards, too!
for item in items.prefix("prefix").into_iter().rev() {
  // ...
}

// Atomic write batches (multiple partitions can be used in a single batch)
let mut batch = keyspace.batch();
batch.insert(&items, "1", "abc");
batch.insert(&items, "3", "abc");
batch.insert(&items, "5", "abc");
batch.commit()?;

// Sync the journal to disk to make sure data is definitely durable
// When the keyspace is dropped, it will try to persist
// Also, by default every second the keyspace will be persisted asynchronously
keyspace.persist()?;

// Destroy the partition, removing all data in it.
// This may be useful when using temporary tables or indexes,
// as it is essentially an O(1) operation.
keyspace.delete_partition(items)?;

Details

  • Partitions (a.k.a. column families) with cross-partition atomic semantics (atomic write batches)
  • Sharded journal for concurrent writes
  • Cross-partition snapshots (MVCC)
  • anything else implemented in lsm-tree

Stable disk format

The disk format will be stable from 1.0.0 (oh, the dreaded 1.0.0...) onwards. Any breaking change after that will result in a major bump.

Examples

See here for practical examples.

And checkout Smoltable, a standalone Bigtable-inspired mini wide-column database using fjall as its storage engine.

Contributing

How can you help?

License

All source code is licensed under MIT OR Apache-2.0.

All contributions are to be licensed as MIT OR Apache-2.0.

Dependencies

~5–17MB
~209K SLoC