Trait CallInterceptor

Source

pub trait CallInterceptor:
    Send
    + Sync
    + 'static {
    // Required methods
    fn before<'a>(
        &'a self,
        req: &'a mut ClientRequest,
    ) -> impl Future<Output = Result<(), ClientError>> + Send + 'a;
    fn after<'a>(
        &'a self,
        resp: &'a ClientResponse,
    ) -> impl Future<Output = Result<(), ClientError>> + Send + 'a;

    // Provided method
    fn on_error<'a>(
        &'a self,
        req: &'a ClientRequest,
        err: &'a ClientError,
    ) -> impl Future<Output = ()> + Send + 'a { ... }
}

Expand description

Hooks called before every A2A request and after every response.

Implement this trait to add cross-cutting concerns such as authentication, logging, or metrics. Register interceptors via crate::ClientBuilder::with_interceptor.

§Object-safety note

This trait uses impl Future return types with explicit lifetimes, which is not object-safe. Internally the SDK wraps implementations in a boxed-future shim. Callers implement the ergonomic trait API.

Required Methods§

Source

fn before<'a>( &'a self, req: &'a mut ClientRequest, ) -> impl Future<Output = Result<(), ClientError>> + Send + 'a

Called before the request is sent.

Mutate req to modify parameters or inject headers.

Source

fn after<'a>( &'a self, resp: &'a ClientResponse, ) -> impl Future<Output = Result<(), ClientError>> + Send + 'a

Called after a successful response is received.

Not called when the request fails; see on_error.

Provided Methods§

Source

fn on_error<'a>( &'a self, req: &'a ClientRequest, err: &'a ClientError, ) -> impl Future<Output = ()> + Send + 'a

Called when the transport returns an error for a request this chain’s before hooks saw, in reverse registration order like after.

It observes; it cannot change the error the caller receives. By then the params have moved to the transport, so req.params is null; req.method and req.extra_headers are as before left them. It is not called when a before hook itself fails, nor for errors that arrive later inside an open stream.

Not overriding it (the default does nothing) means the interceptor never learns that the agent rejected what it attached. For an interceptor that attaches credentials that is a real cost: BearerAuthInterceptor overrides it so a token the agent answered with 401 is dropped from its provider’s cache rather than sent again until it expires.