xt/agcore
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat(llm): 完成 Phase 5 热身准备（Ollama / non_exhaustive / ProviderConfig）

Browse Source 
Phase 5 三个 Step 全部落地：

Step 5.2 — Ollama Provider
- 新增 OllamaProvider newtype 包装（默认 localhost:11434/v1，零 API key）
- ProviderType 新增 Ollama 变体与 FromStr 解析

Step 5.3 — #[non_exhaustive] 前置标记
- ProviderType / StopReason / FinishReason / EvictionPolicy 加 #[non_exhaustive]
- 编译期兼容护栏，避免下游 silent break

Step 5.1 — ProviderConfig 扩展
- 加 timeout_secs / max_retries 字段、Default、from_env(prefix)
- create_provider 各分支通过 pub(crate) from_parts 一次性构造并注入 timeout
  （同时避开 Anthropic 的 default_headers 与双重 client 构造）
- map_reqwest_error 改为方法读取 self.timeout_secs（移除硬编码 120s）
- AnthropicProvider::with_timeout 同值短路，with_client 标 #[deprecated]
- DeepSeek / Qwen 加公开 with_client，new_with_client 走代理
- 7 个新测试：5 个 from_env 单元测试 + 3 个 timeout 传导 wiremock
  （OpenAI Chat / DeepSeek / Anthropic）
- Cargo.toml 加 temp-env dev-dep
This commit is contained in:
徐涛
2026-07-05 08:12:40 +08:00
parent 9e476e79bb
commit 98dfe6c1ed
10 changed files with 807 additions and 199 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llm/provider/openai.rs
+92 -55
View File
@@ -52,31 +52,63 @@ pub struct GenericOpenaiProvider {
model: String,
provider_name: &'static str,
extra_headers: Vec<(String, String)>,
/// HTTP 请求超时秒数。由 `ProviderConfig::timeout_secs` 传入，
/// 在 `LlmError::Timeout { duration }` 中回显。`reqwest::Client` 不暴露 timeout getter，
/// 因此单独存储以便错误消息与配置保持一致。
timeout_secs: u64,
}
impl GenericOpenaiProvider {
/// 基础构造器。
pub fn new_with_name(
/// 一次性构造 —— `create_provider` 路径专用，避免 `new_with_name` + `with_client` 的双重 client 构造。
///
/// 调用方负责预先构造好带正确 timeout 的 `http_client`。`extra_headers` 与 `timeout_secs`
/// 一并设置字段，避免后续修改。
pub(crate) fn from_parts(
base_url: String,
api_key: String,
model: String,
provider_name: &'static str,
http_client: Client,
extra_headers: Vec<(String, String)>,
timeout_secs: u64,
) -> Self {
let http_client = Client::builder()
.timeout(Duration::from_secs(120))
.build()
.expect("创建 HTTP 客户端失败");
Self {
http_client,
base_url,
api_key,
model,
provider_name,
extra_headers: Vec::new(),
extra_headers,
timeout_secs,
}
}
/// 基础构造器。
///
/// `timeout_secs` 应用于 `reqwest::Client` 的请求超时配置。
/// 应由 `ProviderConfig::timeout_secs` 传入（调用方如不知道，可传 30）。
pub fn new_with_name(
base_url: String,
api_key: String,
model: String,
provider_name: &'static str,
timeout_secs: u64,
) -> Self {
let http_client = Client::builder()
.timeout(Duration::from_secs(timeout_secs))
.build()
.expect("创建 HTTP 客户端失败");
Self::from_parts(
base_url,
api_key,
model,
provider_name,
http_client,
Vec::new(),
timeout_secs,
)
}
/// 带额外请求头的构造器（如 Qwen 需要 SSE 启用头）。
pub fn new_with_name_and_headers(
base_url: String,
@@ -84,10 +116,21 @@ impl GenericOpenaiProvider {
model: String,
provider_name: &'static str,
extra_headers: Vec<(String, String)>,
timeout_secs: u64,
) -> Self {
let mut base = Self::new_with_name(base_url, api_key, model, provider_name);
base.extra_headers = extra_headers;
base
let http_client = Client::builder()
.timeout(Duration::from_secs(timeout_secs))
.build()
.expect("创建 HTTP 客户端失败");
Self::from_parts(
base_url,
api_key,
model,
provider_name,
http_client,
extra_headers,
timeout_secs,
)
}
pub fn with_client(mut self, client: Client) -> Self {
@@ -119,10 +162,10 @@ impl GenericOpenaiProvider {
Ok(builder.json(body))
}
fn map_reqwest_error(e: reqwest::Error) -> LlmError {
fn map_reqwest_error(&self, e: reqwest::Error) -> LlmError {
if e.is_timeout() {
LlmError::Timeout {
duration: Duration::from_secs(120),
duration: Duration::from_secs(self.timeout_secs),
}
} else if e.is_connect() {
LlmError::Other(format!("连接失败: {}", e))
@@ -135,9 +178,7 @@ impl GenericOpenaiProvider {
///
/// ponytail: Qwen 等部分 OpenAI-compatible 提供方可能返回非标准 error body
/// （无法解析为 JSON），此处直接用 status code + 原始 body 兜底。
async fn handle_error_response(
response: reqwest::Response,
) -> LlmError {
async fn handle_error_response(response: reqwest::Response) -> LlmError {
let status = response.status().as_u16();
let body = response.text().await.unwrap_or_default();
@@ -148,10 +189,7 @@ impl GenericOpenaiProvider {
// 与 OpenAI 完全一致；DeepSeek/Qwen 通常遵循。
LlmError::RateLimit { retry_after: None }
}
_ if status >= 500 => LlmError::Request {
status,
body,
},
_ if status >= 500 => LlmError::Request { status, body },
_ if status == 400 && body.contains("context_length_exceeded") => {
LlmError::ContextLength {
actual: 0,
@@ -200,7 +238,9 @@ impl GenericOpenaiProvider {
stop_sequences[0].clone(),
))
} else {
Some(crate::llm::types::shared::StopSequence::Multiple(stop_sequences))
Some(crate::llm::types::shared::StopSequence::Multiple(
stop_sequences,
))
};
let frequency_penalty = request.get_extra_opt("frequency_penalty");
@@ -245,9 +285,7 @@ impl GenericOpenaiProvider {
let stop_reason = match choice.finish_reason {
Some(FinishReason::Stop) => StopReason::Stop,
Some(FinishReason::Length) => StopReason::Length,
Some(FinishReason::ToolCalls) | Some(FinishReason::FunctionCall) => {
StopReason::ToolUse
}
Some(FinishReason::ToolCalls) | Some(FinishReason::FunctionCall) => StopReason::ToolUse,
Some(FinishReason::ContentFilter) => StopReason::ContentFilter,
Some(FinishReason::Other) | None => StopReason::Stop,
};
@@ -263,7 +301,10 @@ impl GenericOpenaiProvider {
}
/// 非流式 `chat()` 入口。
pub async fn chat_blocking(&self, request: MessageRequest) -> Result<MessageResponse, LlmError> {
pub async fn chat_blocking(
&self,
request: MessageRequest,
) -> Result<MessageResponse, LlmError> {
let req = self.convert_request(request)?;
let url = format!("{}/chat/completions", self.base_url.trim_end_matches('/'));
@@ -280,7 +321,7 @@ impl GenericOpenaiProvider {
.await
.map_err(|e| {
error!(error = %e, "请求失败");
Self::map_reqwest_error(e)
self.map_reqwest_error(e)
})?;
let status = response.status();
@@ -303,8 +344,7 @@ impl GenericOpenaiProvider {
pub async fn chat_stream_inner(
&self,
request: MessageRequest,
) -> Result<Pin<Box<dyn Stream<Item = Result<StreamEvent, LlmError>> + Send>>, LlmError>
{
) -> Result<Pin<Box<dyn Stream<Item = Result<StreamEvent, LlmError>> + Send>>, LlmError> {
let mut req = self.convert_request(request)?;
req.stream = Some(true);
req.stream_options = Some(StreamOptions {
@@ -322,7 +362,7 @@ impl GenericOpenaiProvider {
.await
.map_err(|e| {
error!(error = %e, "流式请求失败");
Self::map_reqwest_error(e)
self.map_reqwest_error(e)
})?;
let status = response.status();
@@ -331,9 +371,9 @@ impl GenericOpenaiProvider {
}
let byte_stream: std::pin::Pin<Box<dyn Stream<Item = Result<Bytes, LlmError>> + Send>> = {
let s = response.bytes_stream().map(|r| {
r.map_err(|e| LlmError::Other(format!("流式读取失败: {}", e)))
});
let s = response
.bytes_stream()
.map(|r| r.map_err(|e| LlmError::Other(format!("流式读取失败: {}", e))));
Box::pin(s)
};
@@ -368,8 +408,7 @@ impl LlmProvider for GenericOpenaiProvider {
async fn chat_stream(
&self,
request: MessageRequest,
) -> Result<Pin<Box<dyn Stream<Item = Result<StreamEvent, LlmError>> + Send>>, LlmError>
{
) -> Result<Pin<Box<dyn Stream<Item = Result<StreamEvent, LlmError>> + Send>>, LlmError> {
self.chat_stream_inner(request).await
}
@@ -389,12 +428,13 @@ impl LlmProvider for GenericOpenaiProvider {
pub struct OpenaiChatProvider(pub GenericOpenaiProvider);
impl OpenaiChatProvider {
pub fn new(base_url: String, api_key: String, model: String) -> Self {
pub fn new(base_url: String, api_key: String, model: String, timeout_secs: u64) -> Self {
Self(GenericOpenaiProvider::new_with_name(
base_url,
api_key,
model,
"openai",
timeout_secs,
))
}
@@ -412,8 +452,7 @@ impl LlmProvider for OpenaiChatProvider {
async fn chat_stream(
&self,
request: MessageRequest,
) -> Result<Pin<Box<dyn Stream<Item = Result<StreamEvent, LlmError>> + Send>>, LlmError>
{
) -> Result<Pin<Box<dyn Stream<Item = Result<StreamEvent, LlmError>> + Send>>, LlmError> {
self.0.chat_stream(request).await
}
@@ -436,7 +475,10 @@ enum BlockState {
/// 正在累积 text block。
InText { block_index: u32 },
/// 正在累积 tool_call block。
InTool { block_index: u32, tool_call_index: u32 },
InTool {
block_index: u32,
tool_call_index: u32,
},
/// 正在累积 refusal block。
InRefusal { block_index: u32 },
}
@@ -456,9 +498,7 @@ pub struct ChunkToEventStream {
}
impl ChunkToEventStream {
fn new(
chunks: Pin<Box<dyn Stream<Item = Result<Bytes, LlmError>> + Send>>,
) -> Self {
fn new(chunks: Pin<Box<dyn Stream<Item = Result<Bytes, LlmError>> + Send>>) -> Self {
Self {
chunks,
buffer: String::new(),
@@ -500,12 +540,7 @@ impl ChunkToEventStream {
let mut events = Vec::new();
// 元信息：MessageStart（仅在第一次见到 role=assistant 时）。
if self.partial.id.is_none()
&& chunk
.choices
.iter()
.any(|c| c.delta.role.is_some())
{
if self.partial.id.is_none() && chunk.choices.iter().any(|c| c.delta.role.is_some()) {
events.push(StreamEvent::MessageStart {
id: chunk.id.clone(),
model: chunk.model.clone(),
@@ -650,9 +685,7 @@ impl ChunkToEventStream {
self.partial.stop_reason = Some(match fr {
FinishReason::Stop => StopReason::Stop,
FinishReason::Length => StopReason::Length,
FinishReason::ToolCalls | FinishReason::FunctionCall => {
StopReason::ToolUse
}
FinishReason::ToolCalls | FinishReason::FunctionCall => StopReason::ToolUse,
FinishReason::ContentFilter => StopReason::ContentFilter,
FinishReason::Other => StopReason::Other,
});
@@ -692,7 +725,9 @@ impl ChunkToEventStream {
match self.partial.clone().finalize() {
Ok(full) => {
events.push(StreamEvent::MessageComplete { full_response: full });
events.push(StreamEvent::MessageComplete {
full_response: full,
});
}
Err(e) => {
events.push(StreamEvent::Error {
@@ -707,10 +742,7 @@ impl ChunkToEventStream {
impl Stream for ChunkToEventStream {
type Item = Result<StreamEvent, LlmError>;
fn poll_next(
mut self: Pin<&mut Self>,
cx: &mut Context<'_>,
) -> Poll<Option<Self::Item>> {
fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
loop {
// 先尝试从 buffer 取一行处理
if let Some(line) = self.next_line() {
@@ -814,6 +846,7 @@ mod tests {
"sk-test".into(),
"gpt-4o".into(),
"openai",
30,
);
let response = provider
.chat_blocking(MessageRequest {
@@ -842,6 +875,7 @@ mod tests {
"sk-test".into(),
"gpt-4o".into(),
"openai",
30,
);
let err = provider
.chat_blocking(MessageRequest {
@@ -868,6 +902,7 @@ mod tests {
"sk-test".into(),
"gpt-4o".into(),
"openai",
30,
);
let err = provider
.chat_blocking(MessageRequest {
@@ -906,6 +941,7 @@ data: [DONE]\n\n";
"sk-test".into(),
"gpt-4o".into(),
"openai",
30,
);
let mut stream = provider
.chat_stream_inner(MessageRequest {
@@ -974,6 +1010,7 @@ data: [DONE]\n\n";
"k".into(),
"gpt-4o".into(),
"openai",
30,
);
let ir = provider.convert_response(resp).unwrap();
assert_eq!(ir.stop_reason, StopReason::ToolUse);