infinitree 0.3.0

Embedded, encrypted database with tiered cache
Documentation

Infinitree is a versioned, embedded database that uses uniform, encrypted blobs to store data.

Multiple writers can use the same storage, but not the same tree safely at the same time.

Calling Infinitree a database may be generous, as all persistence-related operations are explicit.

Features

  • Thread-safe by default
  • Transparently handle hot/warm/cold storage tiers; currently S3-compatible backends is supported
  • Versioned data structures that can be queried using the Iterator trait without loading in full
  • Encrypt all on-disk data, and only decrypt it on use
  • Focus on performance and control over memory use
  • Extensible for custom data types, storage backends, and serialization

Example use

use infinitree::{
Infinitree,
Index,
Key,
anyhow,
backends::Directory,
fields::{VersionedMap},
};
use serde::{Serialize, Deserialize};

fn main() -> anyhow::Result<()> {
let mut tree = Infinitree::<VersionedMap<String, usize>>::empty(
Directory::new("test_data")?,
Key::from_credentials("username", "password")?
);

tree.index().insert("sample_size".into(), 1234);

tree.commit("first measurement! yay!");
Ok(())
}

Core concepts

[Infinitree] is a versioned data store interface that is the first point of contact with the library. It provides convenience functions to work on different versions of the database index, and access random access data using pointers.

There are 2 types of interactions with an infinitree: one that's happening through an index, and one that's directly exposing random access data.

Any data stored outside of an index will receive a ChunkPointer, which must be stored somewhere to retrieve the data. Hence the need for an index.

Indexes can be any struct that implement the [Index] trait. There's also a helpful derive macro that helps you do this. An index will consist of various fields, which act like regular old Rust types, but need to implement a few traits to help serialization.

Infinitree

[Infinitree] provides high-level versioning, querying, and key and memory management operations for working with the different [fields] in the [Index].

An Infinitree instance is mainly acting as a context for all operations on the tree, and will be your first entry point when working with trees and persisting them.

Here you can select different versions for the index to interact with, and create new commits.

Index

You can think about your Index as a schema. Or really just the central struct definition for your data.

In a more abstract sense, the [Index] trait and corresponding derive macro represent a view into a single version of your database. Using an [Infinitree] you can swap between the various versions and mix-and-match data from various versions into a single Index instance.

Interaction with Index member fields is straightforward. The derive macro will generate functions that produce an Intent for any operation that touches the persistence layer, such as Store and Load.

Fields

An Index contains serializable fields. These are thread-safe data structures with internal mutation, which support some kind of serialization Strategy.

You can use any type that implements [serde::Serialize] as a field through the fields::Serialized wrapper type, but there are incremental hash map and list-like types available for you to use to track and only save changes between versions of your data.

Persisting and loading fields is done using an Intent wrapper. If you use the [Index][derive@Index] macro, this will automatically create accessor functions for each field in an index, that return an Intent wrapped strategy.

Intents elide the specific types of the field and allow doing batch operations, e.g. when calling [Infinitree::commit] using a different strategy for each field in an Index.

Strategy

To tell Infinitree how to serialize a field, you can use different strategies. A strategy has full control over the field and the serializers/loader transactions for it, which means you can control the placement of pieces of data.

Every strategy receives an Index transaction, and a [object::Reader] or [object::Writer]. It is the responsibility of the strategy to store references so you can load back the data once persisted.

There are 2 strategies in the base library:

  • LocalField: Store all of the data in the index. This is the default.
  • SparseField: Store values in a Map outside of the index. Best suited for large structs as values.

Deciding which strategy is best for your use case may mean you have to run some experiments and benchmarks. A SparseField is generally useful for indexing large structs that you want to query rather than load all at once.

See the documentation for the [Index][derive@Index] macro to see how to use strategies.