Crate desync

source ·
Expand description

Desync

Desync provides a new synchronisation type, Desync<T>, which works by ordering operations on its enclosed data type instead of the traditional method of using mutexes to protect critical sections. This allows concurrency to be built around two basic operations:

  • desync_thing.sync(|thing| /* ... */) for synchronous access to the data
  • desync_thing.desync(|thing| /* ... */) for asynchronous access to the data - running the supplied task in the background.

If only the sync() operation is used, this is roughly equivalent to a standard Mutex, except with much stronger guarantees about which thread gets the data first. The other operation, desync() effectively replaces the need to spawn threads and move data around in order to add concurrency to a program.

Desync also provides equivalent methods for async code: future_sync() will perform an operation in the current async context and future_desync() will schedule an operation in the background. These can be freely mixed with the sync() and desync() operations so it becomes fairly easy to mix code using traditional threading and code using async futures. As Desync uses order-of-operations to guarantee exclusive access to the data, these operations can borrow the contained data across any awaits that might be needed, unlike locks created using the Mutex type, which can’t be sent between threads.

Desync provides fairly strong ordering guarantees: in particular, when any of the methods return, the ordering of the operation is guaranteed relative to any following operation. This property makes desync code quite easy to follow and less prone to race conditions than traditional threading. The ability to easily schedule updates asynchronously provides a way around common scenarios where the need to lock multiple mutexes can create deadlocks.

Quick start

There is a single new synchronisation object: Desync. You create one like this:

use desync::Desync;
let number = Desync::new(0);

It supports two main operations. desync will schedule a new job for the object that will run in a background thread. It’s useful for deferring long-running operations and moving updates so they can run in parallel.

let number = Desync::new(0);
number.desync(|val| {
    // Long update here
    thread::sleep(Duration::from_millis(100));
    *val = 42;
});
 
// We can carry on what we're doing with the update now running in the background

The other operation is sync, which schedules a job to run synchronously on the data structure. This is useful for retrieving values from a Desync.

let new_number = number.sync(|val| *val);           // = 42

Desync objects always run operations in the order that is provided, so all operations are serialized from the point of view of the data that they contain. When combined with the ability to perform operations asynchronously, this provides a useful way to immediately parallelize long-running operations.

The future_sync() action returns a boxed Future that can be used with other libraries that use them. It’s conceptually the same as sync, except that it doesn’t wait for the operation to complete:

let future_number = number.future_sync(|val| future::ready(*val));
assert!(executor::block_on(async { future_number.await.unwrap() }) == 42 );

Note that this is the equivalent of just number.sync(|val| *val), so this is mainly useful for interacting with other code that’s already using futures. The after() function is also provided for using the results of futures to update the contents of Desync data: these all preserve the strict order-of-operations semantics, so operations scheduled after an after won’t start until that operation has completed.

Pipes and streams

As well as support for futures, Desync provides supports for streams. The pipe_in() and pipe() functions provide a way to process stream data in a desync object as it arrives. pipe_in() just processes stream data as it arrives, and pipe() provides an output stream of data.

pipe() is quite useful as a way to provide asynchronous access to synchronous code: it can be used to create a channel to send requests to an asynchronous target and retrieve results back via its output. (Unlike this ‘traditional’ method, the actual scheduling and channel maintenance does not need to be explicitly implemented)

Re-exports

Modules

  • The main Desync struct
  • Desync pipes provide a way to generate and process streams via a Desync object
  • The scheduler provides the JobQueue synchronisation mechanism.