Extending and embedding Spin

The complete example for extending and embedding Spin can be found on GitHub.

Spin currently implements triggers and application models for:

The Spin internals and execution context (the part of Spin executing components) are agnostic of the event source and application model. In this document we will explore how to extend Spin with custom event sources (triggers) and application models built on top of the WebAssembly component model, as well as how to embed Spin in your application.

In this article we will build a Spin trigger to run the applications based on a timer, executing Spin components at configured time interval.

The current application types that can be implemented with Spin have entry points defined using WebAssembly Interface (WIT):

// The entry point for an HTTP handler.
handle-http-request: function(req: request) -> response

// The entry point for a Redis handler.
handle-redis-message: function(msg: payload) -> expected<_, error>

The entry point we want to execute for our timer trigger takes a string as its only argument (the trigger will populate that with the current date and time), and it expects a string as the only return value. This is purposefully chosen to be a simple function signature:

// examples/spin-timer/spin-timer.wit
handle-timer-request: function(msg: string) -> string

This is the function that all components executed by the timer trigger must implement, and which is used by the timer executor when instantiating and invoking the component.

Let’s have a look at building the timer trigger:

// examples/spin-timer/src/main.rs
wit_bindgen_wasmtime::import!({paths: ["spin-timer.wit"], async: *});
type ExecutionContext = spin_engine::ExecutionContext<spin_timer::SpinTimerData>;

/// A custom timer trigger that executes a component on every interval.
pub struct TimerTrigger {
    /// The interval at which the component is executed.
    pub interval: Duration,
    /// The Spin execution context.
    engine: Arc<ExecutionContext>,

A few important things to note from the start:

  • we use the WIT defined entry point with the Bytecode Alliance wit-bindgen project to generate “import” bindings based on the entry point — this generates code that allows us to easily invoke the entry point from application components that implement our new application model.
  • the new trigger has a field that contains a Application — in most cases, either CoreComponent will have to be updated with new trigger and component configuration (not the case for our simple application model), or an entirely new component can be defined and used in Application<T>.
  • the trigger has a field that contains the Spin execution context — this is the part of Spin that instantiates and helps execute the WebAssembly modules. When creating the trigger (in the new function, you get access to the underlying Wasmtime store, instance, and linker, which can be configured as necessary).

Finally, whenever there is a new event (in the case of our timer-based trigger every n seconds), we execute the entry point of a selected component:

/// Execute the first component in the application manifest.
async fn handle(&self, msg: String) -> Result<()> {
    // create a new Wasmtime store and instance based on the first component's WebAssembly module.
    let (mut store, instance) =
            .prepare_component(&self.app.components[0].id, None, None, None, None)?;

    // spawn a new thread and call the entry point function from the WebAssembly module 
    let res = spawn_blocking(move || -> Result<String> {
            // use the auto-generated WIT bindings to get the Wasm exports and call the `handle-timer-request` function.
        let t = spin_timer::SpinTimer::new(&mut store, &instance, |host| {
        Ok(t.handle_timer_request(&mut store, &msg)?)
    // do something with the result.
    log::info!("{}\n", res);

A few notes:

  • prepare_component is a function implemented by the Spin execution context, and it handles taking the Wasmtime pre-instantiated module, mapping all the component files, environment variables, and allowed HTTP domains, populating the Wasmtime store with the appropriate data, and returning the store and instance.
  • invoking the entry point handle-timer-request is done in this example in a new Tokio thread — this is an implementation choice based on the needs of the trigger.
  • the return value from the component (a string in this example) can then be used — in the case of the HTTP trigger, this is an HTTP response, which is then returned to the client.

This is very similar to how the HTTP and Redis triggers are implemented, and it is the recommended way to extend Spin with your own trigger and application model.

Writing components for the new trigger can be done by using the wit-bindgen tooling from Rust and other supported languages (see the example in Rust):

// automatically generate Rust bindings that help us implement the 
// `handle-timer-request` function that the trigger will execute.
fn handle_timer_request(msg: String) -> String {
    format!("ECHO: {}", msg)

Components can be compiled to WebAssembly, then used from a spin.toml application manifest.

Embedding the new trigger in a Rust application is done by creating a new trigger instance, then calling its run function:

// app() is a utility function that generates a complete application configuration.
let trigger = TimerTrigger::new(Duration::from_secs(1), app()).await?;
// run the trigger indefinitely

We are exploring APIs for embedding Spin from other programming languages such as Go or C#.

In this example, we built a simple timer trigger — building more complex triggers would also involve updating the Spin application manifest, and extending the application-level trigger configuration, as well as component-level trigger configuration (an example of component-level trigger configuration for this scenario would be each component being able to define its own independent time interval for scheduling the execution).

Other ways to extend and use Spin

Besides building custom triggers, the internals of Spin could also be used independently:

  • the Spin execution context can be used entirely without a spin.toml application manifest — for embedding scenarios, the configuration for the execution can be constructed without a spin.toml (see issue #229 for context)
  • the standard way of distributing a Spin application can be changed by re-implementing the loader and publish crates — all is required is that loading the application returns a valid Application that the Spin execution context can use to instantiate and execute components.