use std::time::Duration;
use tracing::{debug, warn};
use super::auth::MantleAuth;
use crate::providers::bedrock::provider::ModelTarget;
use crate::providers::http::{SseData, http_client, post_sse, responses_code};
use crate::providers::openai::responses::mappers::{ResponsesRequestPolicy, build_request};
use crate::providers::openai::responses::streaming::decode_responses;
use crate::providers::response_stream::{OpenedStream, response_stream};
use crate::{Context, LlmError, LlmResponseStream, Result};
use utils::variables::Vars;
#[derive(Clone)]
pub(crate) struct MantleClient {
http: reqwest::Client,
template_vars: Vars,
auth: MantleAuth,
base_url_override: Option<String>,
}
impl MantleClient {
pub(crate) fn new(region: String, auth: MantleAuth, base_url_override: Option<String>) -> Self {
Self {
http: http_client(),
template_vars: Vars::new().with("AWS_REGION", region.clone()).with("AWS_DEFAULT_REGION", region),
auth,
base_url_override: base_url_override.map(|url| url.trim_end_matches('/').to_string()),
}
}
pub(crate) fn with_auth(mut self, auth: MantleAuth) -> Self {
self.auth = auth;
self
}
pub(crate) fn stream(
&self,
target: ModelTarget,
base_url_template: &'static str,
context: Context,
idle_timeout: Duration,
) -> LlmResponseStream {
if let Some(arn) = &target.inference_profile_arn {
warn!(
model = %target.model,
inference_profile_arn = %arn,
"Ignoring inferenceProfileArn: Responses requests use the selected model ID"
);
}
let client = self.clone();
response_stream(
async move { client.send(&target, base_url_template, &context).await },
decode_responses(),
idle_timeout,
)
}
pub(crate) fn endpoint(&self, base_url_template: &str) -> Result<String> {
let base = match &self.base_url_override {
Some(base) => base.clone(),
None => {
self.template_vars.expand(base_url_template).map_err(|e| LlmError::InvalidArgument(e.to_string()))?
}
};
Ok(format!("{}/responses", base.trim_end_matches('/')))
}
async fn send(
&self,
target: &ModelTarget,
base_url_template: &str,
context: &Context,
) -> Result<OpenedStream<SseData>> {
let url = self.endpoint(base_url_template)?;
let model = target.model.as_str();
let body = build_request(target.identity.as_ref(), model, context, &ResponsesRequestPolicy::MANTLE)?;
debug!(model, url, auth = %self.auth, "Sending Bedrock responses request");
let encoded = serde_json::to_vec(&body)?;
let headers = self.auth.headers("POST", &url, &encoded).await?;
post_sse(&self.http, &url, headers, &body, responses_code).await
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn endpoint_expands_the_region_from_the_model_template() {
let endpoint = client(None).endpoint("https://bedrock-mantle.${AWS_REGION}.api.aws/openai/v1").unwrap();
assert_eq!(endpoint, "https://bedrock-mantle.us-west-2.api.aws/openai/v1/responses");
}
#[test]
fn endpoint_preserves_a_models_own_api_path() {
let endpoint = client(None).endpoint("https://bedrock-mantle.${AWS_REGION}.api.aws/v1").unwrap();
assert_eq!(endpoint, "https://bedrock-mantle.us-west-2.api.aws/v1/responses");
}
#[test]
fn base_url_override_replaces_the_catalog_template() {
let endpoint = client(Some("http://127.0.0.1:9999/".to_string()))
.endpoint("https://bedrock-mantle.${AWS_REGION}.api.aws/openai/v1")
.unwrap();
assert_eq!(endpoint, "http://127.0.0.1:9999/responses");
}
fn client(base_url_override: Option<String>) -> MantleClient {
MantleClient::new("us-west-2".to_string(), MantleAuth::None, base_url_override)
}
}