Crate airone

Airone

Airone is a library inspired from Aral Balkan’s JSDB principles applied to a Rust library.

The name has nothing to do with “air” or “one”, it simply comes from the Italian word “Airone”, which means “Heron”. Hence, it has to be read [aiˈrone].

This library persists a list of structs from memory to disk. Every change in the data in memory is saved to disk automatically incrementally, avoiding the full dump of the whole list.

As saving a long list of objects to a file each time a change happens may be cumbersome, airone writes each incremental change to an append-only file. This incremental transaction log is applied at the next restart to compact the base dump file; in this way, the heavy operation of writing the full dump of data from memory is executed only once at startup. After that, each change is written in a fast way to the append-only log.

Architecture

Usage scenario

Airone is developed to be used in situations where all of these prerequisites apply:

• no big-data: the whole dataset should fit in memory
• the dataset has to be persisted to disk
• after some data is modified, the change needs to be written to disk in a fast way
• we can slow down the program start time without any relevant consequence
• no nested objects

These limits can be a problem for some usage scenarios. Existing databases add all sort of optimizations, caching mechanisms and are very complex, in order to be able to deal with such big amount of data.
However, when the above-mentioned prerequisites apply, we can leverage these limits to simplify the source code, making it more maintanable and getting rid of all the bloatware. Moreover, limiting object nesting makes it compatible with CSV style files, ensuring you can manipulate data with standard UNIX tools such as grep and awk.

Here are two examples of good usage scenarios:

• small web server: data may change fast and can entirely fit into memory. The startup can be slow, as long as the server is performant while it’s running. I provide an external template using airone and a Rocket server
• an offline GUI/TUI program: you can store data modifications in a very fast way thanks to airone’s append-only file architecture, so the interface freeze during the saving process is barely noticeable. Here’s a very barebone template combining airone with Gtk3

Context

Airone is aimed at helping small-tech to be produced, which is a way of building technology antithetical to the Silicon Valley model. Small-tech features close-to-zero data collection, a free and open source structure to achieve good transparency, no lust to scale exponentially, no desire in tracking people’s behaviour while crunching tons of data and, overall, it has a goal to keep things simple and human.

Implementation limitations

State of the project

This project is still a Work in Progress. It has been started only out of curiosity attempting to replicate JSDB behaviour with the bare minimum number of dependencies I could achieve (only one). It has unit tests, but it has not NOT been used in production so far.

Beware that the derive macro currently does not support fields with reference and lifetime like &'a str; use owned types like String instead. If you manage to get it work, feel free to send a pull request.

Operating system

airone has not been tested on Windows and Mac. Beware that WSL is a compatibility layer and may unexpectedly break too. I encourage you to switch to an actually freedom-and-privacy respecting operating system such as GNU/Linux, where this library has been tested on more. Head to https://www.youtube.com/watch?v=Ag1AKIl_2GM for more information

Licensing

This project is licensed under an AGPL style license. Head to the COPYING file and Copyright section for detailed information. Make sure to respect the license terms to use this library.

Usage

Installation

Simply run cargo add airone.

Alternatively, add this line to your dependencies section of the Cargo.toml file.

airone = "0.7.4"

Basic operations

The crate exposes a generic struct AironeDb<T> and a convenient macro to derive custom types to be used as T.

The core lies in the AironeDbDerive derive macro. Apply it to a struct named as you wish to it will define implementations for AironeDb<Foo>. The newly creates struct acts as a proxy between you and the underlying list of data, automatically persisting changes when they happen and providing methods to interact with them.

The most basic setup you can achieve looks like this. First, we import the macros and the needed traits. Then, we initalize AironeDb<T> by using airone_init!. Eventually, we generate the implementations for the desired type by using AironeDbDerive.

use airone::prelude::*;
use airone_derive::AironeDbDerive;

airone_init!();
#[derive(AironeDbDerive)]
struct Foo
{
    pub field1: f64,
    pub field2: String,
    field3: Option<i32>
}

And here is how you can interact with your data, mainly using methods provided in AironeDb.

You can also transparently access the data in readmode using the index operator.

You can access and edit data by using common Vec methods, like push(), get() or get_mut() methods.

{
    // Open the database
    let mut db: AironeDb<Foo> = AironeDb::new();
    // Add one element using a public method
    db.push(
        Foo{
            field1: 0.5,
            field2: "Hello world".to_string(),
            field3: Some(-45)
        }
    );
    // Change a field using the generated setter method
    db.set_field3(0, None);
    // The database is closed automatically here
}
{
    // Open again, check the modified data
    // has been correctly persisted.
    let db: AironeDb<Foo> = AironeDb::new();
    assert_eq!(
        *db[0].get_field3(),
        None
    );

    // Access using index directly in read-mode
    db[0].get_field3();
}

In addition to the methods from the AironeDb struct, some getter and setters are generated for each variable to change the element at the specified index in the form of:

fn set_$field_name(&mut self, index: usize, new_value: $field_type)
fn get_$field_name(&self, index: usize)

Query chaining

You can use the Query and QueryMut structs to make queries using dot notation, chaining them one after another.

let mut db: AironeDb<QueryExample> = AironeDb::new();
// Fill in data how you want here
// …
//
// Can use dot notation chaining operations
db.build_query_mut().filter(
    |e|
    {
        e.get_my_text() == "Test string"
    }
).delete();

// Do something else with `db` object

Bulk operations

Airone does not have an actual concept of transaction. Instead, it has a concept of “autocommit”:

• when it’s enabled, each data modification is instantly flushed to disk;
• when it’s disabled, data will only be changed in-memory and flushed only when manually calling commit().

If you’re writing many changes in bulk, flushing one change to disk for each modification can degrade performance. In this situation, you may want to temporarily disable the auto_commit feature, apply modifications and then flush everything to disk in one shot.

An airone_bulk! macro is provided to encapsulate this common behaviour.

Manually changing auto_commit settings

Auto-commit is enabled by default, which means that any change will be persisted to disk instantly by internally calling commit(). You can enable or disable it through set_auto_commit() method.

When autocommit is disabled, you must manually call commit() and rollback() methods. In this situation, changes are instantly applied to the data in-memory, but they’re not written to disk until you call commit(). You can call rollback() before committing to rollback the in-memory changes to the previous valid state; nothing will be written to disk.

let mut db: AironeDb<Animal> = AironeDb::new_with_filename("rollback_example");
// This database has 3 elements inside it

// Disabling auto-commit, so
// we can now use commit() and rollback()
db.set_auto_commit(false);

// Add and rollback
db.push(Animal {
    a: 156,
    n: 1.4142135,
    testo: "test".to_string()
});
assert_eq!(db.len(), 4);
// We rollback and we have again 3 elements only
db.rollback();
assert_eq!(db.len(), 3);

// Edit a value
db.set_a(2, 10);
// The value has been changed in memory only
assert_eq!(*db.get_a(2), 10);
// Roll back
db.rollback();
// Check the object has its original value
assert_eq!(*db.get_a(2), 2);

// Deleting some objects
db.remove(1);
db.remove(0);
// Check data in memory has changed
assert_eq!(db.len(), 1);
// Rollback and make sure
// the elements are back
db.rollback();
assert_eq!(db.len(), 3);

Automatically change auto_commit settings

You can use a macro to wrap a series of instructions so that auto_commit is disabled before executing them and re-enabled afterwards.

Read airone_bulk! for more information.

airone_bulk!(
    db,
    {
        db.push(Foo {
            bar: 0
        });
        db.set_bar(0, 57);
    }
);

C language integration

You can enable the optional c feature on this crate. By doing so, the derive macro will generate C style functions too, which have already been decorated with the #[no_mangle] attribute. If you compile the project as a shared library, you can link it to a C executable. See this repository if you want to learn more about C integration.

Internal Data Format

Data is written into two files, depending of the phase of execution.

The base dump file contains the full dump of data in memory. This is recreated whenever the program starts by using the old dump data file as a base point and applying each incremental change to it. Afterwards, the data is saved to the a new dump file and the old transaction log is deleted.

From this point, the program continues its execution saving changes to a new transaction log.

Both files follow this character convention:

• the \n character is used as a newline (no carriage return)
• the \t character is used as a field separator.

Base Dump File

The base dump file is saved using a standard UNIX-style CSV text file.

Let’s use this struct as an example:

struct ExampleStruct
{
    field1: String,
    field2: f64,
    field3: i32
}

Given a list of two ExampleStruct elements, the base dump file could look like this. Notice the first line used as a column header:

field1	field2	field3
abc	3.15	57
text2	47.89	-227

If columns are reordered or renamed in the struct or in the csv, airone panics at startup to avoid corrupting data, manually fix the csv or the struct field and try to re-run

Append-only transaction log

When the program is running, changes are written to the append-only transaction log. Each line of this file is formatted as it follows, depending on the applied operation.

Adding an element

The first letter A sets the operation to Add. The new object fields are serialized as in the base dump file, by writing each field’s value in the proper order.

The index sets the position in the list where the element will be added.

Structure:

A	index	field1	field2	field3

Example:

A	3	abc	3.15	57

Deleting an element

The first letter D sets the operation to Delete. After that, it expects the index of the element to remove.

Structure:

D	index_of_element_to_remove

Example:

D	2

Setting a field value

The first letter E sets the operation to Edit. After that field, adhere to the following structure.

Structure:

E	index_of_element	field_to_change	new_value

Example

E	0	field2	-57.5

Extending and supporting custom types

You can serialize and deserialize your custom types by implementing LoadableValue and PersistableValue traits on each field type. The serialized string must be on a single line and must escape any \t, \r and \n character to ensure CSV style compatibility. For most types, you can simply use Rust’s format!() and parse() features.

Copyright

This is NOT public domain, make sure to respect the license terms. You can find the license text in the COPYING file.

Copyright © 2022,2023 Massimo Gismondi

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU Affero General Public License along with this program. If not, see https://www.gnu.org/licenses/.

External libraries used in this project

• paste crate, made by David Tolnay. Licensed under either of Apache License, Version 2.0 or MIT license at your option.

Re-exports

• pub extern crate airone_derive;

Modules

• database
• prelude
• query

Macros

• airone_bulk
  Wraps multiple instructions as a single bulk operation
• airone_init
  Generates a trait to implement a constructor