Fjall is an LSM-based embeddable 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.

Like any typical key-value store, keys are stored in lexicographic order. If you are storing integer keys (e.g. timeseries data), you should use the big endian form to adhere to locality.

Basic usage

cargo add fjall

use fjall::{Config, FlushMode, 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").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(FlushMode::SyncAll)?;

// 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

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

Durability

To support different kinds of workloads, Fjall is agnostic about the type of durability your application needs. After writing data (be it, insert, remove or committing a write batch), you can choose to call Keyspace::persist which takes a FlushMode parameter. By default every 1000ms data is fsynced asynchronously.

Multithreading, Async and Multiprocess

Fjall is internally synchronized for multi-threaded access, so you can clone around the Keyspace and Partitions as needed, without needing to lock yourself. Common operations like inserting and reading are generally lock free.

For an async example, see the tokio.

A single keyspace may not be loaded in parallel from separate processes however.

Feature flags

bloom

Uses bloom filters to reduce disk I/O for non-existing keys. Improves point read performance, but increases memory usage.

Disabled by default.

Stable disk format

The disk format is stable as of 1.0.0. Future breaking changes will result in a major version bump and a migration path.

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?

• Ask a question
• Post benchmarks and things you created
• Open an issue (bug report, weirdness)
• Open a PR

License

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

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