23 releases

0.11.0-rc1 Apr 22, 2024
0.10.4 Aug 14, 2023
0.10.3 Mar 18, 2023
0.10.0 Sep 27, 2022
0.4.0 Jul 31, 2021

#65 in Asynchronous

Download history 458/week @ 2024-07-20 498/week @ 2024-07-27 174/week @ 2024-08-03 537/week @ 2024-08-10 741/week @ 2024-08-17 419/week @ 2024-08-24 285/week @ 2024-08-31 252/week @ 2024-09-07 733/week @ 2024-09-14 781/week @ 2024-09-21 364/week @ 2024-09-28 484/week @ 2024-10-05 438/week @ 2024-10-12 867/week @ 2024-10-19 430/week @ 2024-10-26 573/week @ 2024-11-02

2,558 downloads per month
Used in 2 crates

MIT license

190KB
4K SLoC

fang

Crates.io docs page test style

Fang

Background task processing library for Rust. It can use PostgreSQL, SQLite or MySQL as an asyncronous task queue.

Key Features

Here are some of the fang's key features:

  • Async and threaded workers. Workers can be started in threads (threaded workers) or tokio tasks (async workers)
  • Scheduled tasks. Tasks can be scheduled at any time in the future
  • Periodic (CRON) tasks. Tasks can be scheduled using cron expressions
  • Unique tasks. Tasks are not duplicated in the queue if they are unique
  • Single-purpose workers. Tasks are stored in a single table but workers can execute only tasks of the specific type
  • Retries. Tasks can be retried with a custom backoff mode

Installation

  1. Add this to your Cargo.toml

the Blocking feature

[dependencies]
fang = { version = "0.11.0-rc1" , features = ["blocking"], default-features = false }

the Asynk feature

  • PostgreSQL as a queue
[dependencies]
fang = { version = "0.11.0-rc1" , features = ["asynk-postgres"], default-features = false }
  • SQLite as a queue
[dependencies]
fang = { version = "0.11.0-rc1" , features = ["asynk-sqlite"], default-features = false }
  • MySQL as a queue
[dependencies]
fang = { version = "0.11.0-rc1" , features = ["asynk-mysql"], default-features = false }

the Asynk feature with derive macro

Substitute database with your desired backend.

[dependencies]
fang = { version = "0.11.0-rc1" , features = ["asynk-{database}", "derive-error" ], default-features = false }

All features

fang = { version = "0.11.0-rc1" }

Supports rustc 1.77+

  1. Create the fang_tasks table in the database. The migration of each database can be found in fang/{database}-migrations where database is postgres, mysql or sqlite.

Migrations can be also run as code, importing the feature migrations-{database} being the database the backend queue you want to use.

[dependencies]
fang = { version = "0.11.0-rc1" , features = ["asynk-postgres", "migrations-postgres" ], default-features = false }
use fang::run_migrations_postgres;
run_migrations_postgres(&mut connection).unwrap();

Usage

Defining a task

Blocking feature

Every task should implement the fang::Runnable trait which is used by fang to execute it.

If you have a CustomError, it is recommended to implement From<FangError>. So this way you can use ? operator inside the run function available in fang::Runnable trait.

You can easily implement it with the macro ToFangError. This macro is only available in the feature derive-error.

use fang::FangError;
use fang::Runnable;
use fang::typetag;
use fang::PgConnection;
use fang::serde::{Deserialize, Serialize};
use fang::ToFangError;
use std::fmt::Debug;


#[derive(Debug, ToFangError)]
enum CustomError {
    ErrorOne(String),
    ErrorTwo(u32),
}

fn my_func(num : u16) -> Result<(), CustomError> {
    if num == 0 {
        Err(CustomError::ErrorOne("is zero".to_string()))
    }

    if num > 500 {
        Err(CustomError::ErrorTwo(num))
    }

    Ok(())
}

#[derive(Serialize, Deserialize)]
#[serde(crate = "fang::serde")]
struct MyTask {
    pub number: u16,
}

#[typetag::serde]
impl Runnable for MyTask {
    fn run(&self, _queue: &dyn Queueable) -> Result<(), FangError> {
        println!("the number is {}", self.number);

        my_func(self.number)?;
        // You can use ? operator because
        // From<FangError> is implemented thanks to ToFangError derive macro.

        Ok(())
    }

    // If `uniq` is set to true and the task is already in the storage, it won't be inserted again
    // The existing record will be returned for for any insertions operaiton
    fn uniq(&self) -> bool {
        true
    }

    // This will be useful if you want to filter tasks.
    // the default value is `common`
    fn task_type(&self) -> String {
        "my_task".to_string()
    }

    // This will be useful if you would like to schedule tasks.
    // default value is None (the task is not scheduled, it's just executed as soon as it's inserted)
    fn cron(&self) -> Option<Scheduled> {
        let expression = "0/20 * * * Aug-Sep * 2022/1";
        Some(Scheduled::CronPattern(expression.to_string()))
    }

    // the maximum number of retries. Set it to 0 to make it not retriable
    // the default value is 20
    fn max_retries(&self) -> i32 {
        20
    }

    // backoff mode for retries
    fn backoff(&self, attempt: u32) -> u32 {
        u32::pow(2, attempt)
    }
}

As you can see from the example above, the trait implementation has #[typetag::serde] attribute which is used to deserialize the task.

The second parameter of the run function is a struct that implements fang::Queueable. You can re-use it to manipulate the task queue, for example, to add a new job during the current job's execution. If you don't need it, just ignore it.

Asynk feature

Every task should implement fang::AsyncRunnable trait which is used by fang to execute it.

Be careful not to call two implementations of the AsyncRunnable trait with the same name, because it will cause a failure in the typetag crate.

use fang::AsyncRunnable;
use fang::asynk::async_queue::AsyncQueueable;
use fang::serde::{Deserialize, Serialize};
use fang::async_trait;

#[derive(Serialize, Deserialize)]
#[serde(crate = "fang::serde")]
struct AsyncTask {
    pub number: u16,
}

#[typetag::serde]
#[async_trait]
impl AsyncRunnable for AsyncTask {
    async fn run(&self, _queueable: &mut dyn AsyncQueueable) -> Result<(), Error> {
        Ok(())
    }
    // this func is optional
    // Default task_type is common
    fn task_type(&self) -> String {
        "my-task-type".to_string()
    }


    // If `uniq` is set to true and the task is already in the storage, it won't be inserted again
    // The existing record will be returned for for any insertions operaiton
    fn uniq(&self) -> bool {
        true
    }

    // This will be useful if you would like to schedule tasks.
    // default value is None (the task is not scheduled, it's just executed as soon as it's inserted)
    fn cron(&self) -> Option<Scheduled> {
        let expression = "0/20 * * * Aug-Sep * 2022/1";
        Some(Scheduled::CronPattern(expression.to_string()))
    }

    // the maximum number of retries. Set it to 0 to make it not retriable
    // the default value is 20
    fn max_retries(&self) -> i32 {
        20
    }

    // backoff mode for retries
    fn backoff(&self, attempt: u32) -> u32 {
        u32::pow(2, attempt)
    }
}

In both modules, tasks can be scheduled to be executed once. Use Scheduled::ScheduleOnce enum variant.

Datetimes and cron patterns are interpreted in the UTC timezone. So you should introduce the offset to schedule in a different timezone.

Example:

If your timezone is UTC + 2 and you want to schedule at 11:00:

let expression = "0 0 9 * * * *";

Enqueuing a task

the Blocking feature

To enqueue a task use Queue::enqueue_task

use fang::Queue;

// create a r2d2 pool

// create a fang queue

let queue = Queue::builder().connection_pool(pool).build();

let task_inserted = queue.insert_task(&MyTask::new(1)).unwrap();

the Asynk feature

To enqueue a task use AsyncQueueable::insert_task.

For Postgres backend:

use fang::asynk::async_queue::AsyncQueue;
use fang::AsyncRunnable;

// Create an AsyncQueue
let max_pool_size: u32 = 2;

let mut queue = AsyncQueue::builder()
    // Postgres database url
    .uri("postgres://postgres:postgres@localhost/fang")
    // Max number of connections that are allowed
    .max_pool_size(max_pool_size)
    .build();

// Always connect first in order to perform any operation
queue.connect().await.unwrap();

Encryption is always used with crate rustls. We plan to add the possibility of disabling it in the future.

// AsyncTask from the first example
let task = AsyncTask { 8 };
let task_returned = queue
    .insert_task(&task as &dyn AsyncRunnable)
    .await
    .unwrap();

Starting workers

the Blocking feature

Every worker runs in a separate thread. In case of panic, they are always restarted.

Use WorkerPool to start workers. Use WorkerPool::builder to create your worker pool and run tasks.

use fang::WorkerPool;
use fang::Queue;

// create a Queue

let mut worker_pool = WorkerPool::<Queue>::builder()
    .queue(queue)
    .number_of_workers(3_u32)
    // if you want to run tasks of the specific kind
    .task_type("my_task_type")
    .build();

worker_pool.start();

the Asynk feature

Every worker runs in a separate tokio task. In case of panic, they are always restarted. Use AsyncWorkerPool to start workers.

use fang::asynk::async_worker_pool::AsyncWorkerPool;

// Need to create a queue
// Also insert some tasks

let mut pool: AsyncWorkerPool<AsyncQueue> = AsyncWorkerPool::builder()
        .number_of_workers(max_pool_size)
        .queue(queue.clone())
        // if you want to run tasks of the specific kind
        .task_type("my_task_type")
        .build();

pool.start().await;

Check out:

Configuration

Blocking feature

Just use TypeBuilder for WorkerPool.

Asynk feature

Just use TypeBuilder for AsyncWorkerPool.

Configuring the type of workers

Configuring retention mode

By default, all successfully finished tasks are removed from the DB, failed tasks aren't.

There are three retention modes you can use:

pub enum RetentionMode {
    KeepAll,        // doesn't remove tasks
    RemoveAll,      // removes all tasks
    RemoveFinished, // default value
}

Set retention mode with worker pools TypeBuilder in both modules.

Configuring sleep values

Blocking feature

You can use use SleepParams to configure sleep values:

pub struct SleepParams {
    pub sleep_period: Duration,     // default value is 5 seconds
    pub max_sleep_period: Duration, // default value is 15 seconds
    pub min_sleep_period: Duration, // default value is 5 seconds
    pub sleep_step: Duration,       // default value is 5 seconds
}

If there are no tasks in the DB, a worker sleeps for sleep_period and each time this value increases by sleep_step until it reaches max_sleep_period. min_sleep_period is the initial value for sleep_period. All values are in seconds.

Use set_sleep_params to set it:

let sleep_params = SleepParams {
    sleep_period: Duration::from_secs(2),
    max_sleep_period: Duration::from_secs(6),
    min_sleep_period: Duration::from_secs(2),
    sleep_step: Duration::from_secs(1),
};

Set sleep params with worker pools TypeBuilder in both modules.

Contributing

  1. Fork it!
  2. Create your feature branch (git checkout -b my-new-feature)
  3. Commit your changes (git commit -am 'Add some feature')
  4. Push to the branch (git push origin my-new-feature)
  5. Create a new Pull Request

Running tests locally

  • Install diesel_cli.
cargo install diesel_cli --no-default-features --features "postgres sqlite mysql"
  • Install docker on your machine.

  • Install SQLite 3 on your machine.

  • Setup databases for testing.

make -j db
  • Run tests. make db does not need to be run in between each test cycle.
make -j tests
  • Run dirty/long tests.
make -j ignored
  • Take down databases.
make -j stop

The -j flag in the above examples enables parallelism for make, is not necessary but highly recommended.

Authors

  • Ayrat Badykov (@ayrat555)

  • Pepe Márquez (@pxp9)

Dependencies

~5–22MB
~358K SLoC