A capability is an abstraction or representation of a non-functional requirement; some functionality required by your actor that is not considered part of the core business logic. As you write an actor that exposes some data over a RESTful endpoint, the HTTP server is not part of your business logic, it is a service used by your actor–a capability.

In wasmcloud, capability providers are dynamic libraries that implement a capability contract. A capability contract is a unique name that identifies the interface or abstraction. By convention, these capability contract IDs are prefixed by a vendor ID (the vendor of the contract, not necessarily the specific implementation). For example, the following is a list of the first-party capability contract IDs supported by default wasmcloud providers:

  • wasmcloud:httpserver
  • wasmcloud:httpclient
  • wasmcloud:messaging
  • wasmcloud:telnet
  • wasmcloud:keyvalue
  • wasmcloud:graphdb
  • wasmcloud:logging
  • wasmcloud:blobstore

In our case, all of the contracts created by the wasmcloud team are prefixed with the wasmcloud prefix. This does not mean that other organizations cannot provide an implementation of a wasmcloud contract. On the contrary, we have multiple implementations of wasmcloud contracts for things like an S3 blob store, multiple key-value providers, etc.

Capability Provider FFI

In the current version of wasmcloud, capability providers must be compiled into what are called “shared objects”. These take the form of .so files on Linux, .dylib files on macOS, and .dll files on Windows. You can implement a capability provider in any language, so long as the target output can be one of these shared object files and it conforms to the FFI requirements for a capability provider.

The wasmcloud host runtime will instantiate a new capability provider by invoking the following function:

pub extern "C" fn __capability_provider_create() -> *mut CapabilityProvider 

where CapabilityProvider is a trait implemented in Rust. If you’re writing a capability provider in another language other than Rust, you can feel free to return a pointer to that object, provided the associated function calls match the signature required by the trait.

When implementing your own capability provider in Rust, you can use the wasmcloud-provider-core crate’s capability_provider! macro to provide a developer-friendly way of implementing the FFI function.

Capability Provider Trait Requirements

The following is the Rust definition for the CapabilityProvider trait:

pub trait CapabilityProvider: CloneProvider + Send + Sync {
    /// This function will be called on the provider when the host runtime is ready and has configured a dispatcher. This function is only ever
    /// called _once_ for a capability provider, regardless of the number of actors being managed in the host
    fn configure_dispatch(
        dispatcher: Box<dyn Dispatcher>,
    ) -> Result<(), Box<dyn Error + Send + Sync>>;
    /// Invoked when an actor has requested that a provider perform a given operation
    fn handle_call(
        actor: &str,
        op: &str,
        msg: &[u8],
    ) -> Result<Vec<u8>, Box<dyn Error + Send + Sync>>;
    /// This function is called to let the capability provider know that it is being removed
    /// from the host runtime. This gives the provider an opportunity to clean up any
    /// resources and stop any running threads
    fn stop(&self);

All capability providers MUST respond to the following operations (invoked via handle_call):

  • BindActor - Called with a serialized version of the CapabilityConfiguration struct when an actor is being linked to a capability provider. This operation must be idempotent, and so must simply return an Ok with an empty vector as the response if multiple linking attempts are performed for the same actor.
  • RemoveActor - The opposite of linking an actor to a provider, this is called by the host runtime to indicate that a given actor is no longer actively communicating with (and using the resources of) the provider. This also must be an idempotent operation, and providers must happily return an empty byte vector for attempts to remove non-existent or already-removed actors.
  • HealthRequest - The wasmcloud host runtime will periodically (providers are not privy to the frequency) poll a provider to check to see if it is healthy. If it responds without error, but indicates that it is unhealthy, it will be stopped and removed from the runtime. Return an Ok with a HealthResponse struct to indicate a properly running provider.

⚠️ You must also be sure that you do nothing in the stop function call that might panic, which includes writing to STDOUT when that facility might be unavailable. In most cases, you probably want to simply do nothing in this function, because per-actor resources should be cleaned up when handling the RemoveActor operation.

Building a Provider

Building a provider is most easily done in Rust. For a hands-on tutorial and walk-through guide, check out our creating a capability provider guide.

WASI-Based Capability Providers 🔮

In the (hopefully) near future, the WASI specification will support regular socket-based networking, and multiple low-level WebAssembly engines (such as wasm3 and wasmtime) will support WASI-based networking functionality. When this functionality is available, reliable, and well-tested in the WASI spec and supporting engines, then wasmcloud will automatically support the ability to create capability providers as portable, “privileged” WebAssembly modules that use WASI to provide networking services to their linked actors.

This section of the documentation will change when that functionality is ready to use.