fix(runtime): 上游错误体统一走 trace.UpstreamError,正文不进 error
M41 的「LLM / embedding / 视觉 上游错误体」一族(M38 #5 #6 #7 #8 #9 #14): 非 2xx 时把响应体截断后拼进 error,而 error 会被 chat run / loop step / retrieve_fallback 打日志,等于把她的提问、记忆正文、要念的台词写进日志。 - model:openai/anthropic 两处 upstreamError 收口到 upstream.go 的 providerUpstreamError;anthropic 流内 error 事件的 evt.Error.Message (原本无上限)整段丢弃,只保留令牌形状的 type - embedding:readErrorBody 删除,改走 helper(input 就是她的记忆正文) - vision:错误路径不再沿用取图的 50 MiB 上限,改为 4 KiB 有界读且不拼原文 - voice:classifyDashscopeStatus 只解 code/request_id,正文只留在内存里做 分类判据;错误体读取同样改为 4 KiB 有界 - 测试:五个上游各一条「日志/error 里搜不到哨兵正文」用例,且能变异打红
This commit is contained in:
1 parent
46a6da13e0
commit
66dda084cd
16 files changed
+570
-183
No files matched your search
@@ -14,6 +14,7 @@ import (
|
||||
"time"
|
||||
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/config"
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/trace"
|
||||
)
|
||||
|
||||
// maxResponseBytes caps how much of an /embeddings response body is read, so an
|
||||
@@ -21,8 +22,9 @@ import (
|
||||
// for a normal batch fall far below this; the cap is a guard, not a limit.
|
||||
const maxResponseBytes = 16 << 20
|
||||
|
||||
// maxErrorBodyBytes caps how much of a non-OK response body is read for the
|
||||
// error message.
|
||||
// maxErrorBodyBytes 限制错误路径**读多少字节**(只用来数长度、取摘要、抽
|
||||
// code/type):正文一个字节都不进 error,因此它不再决定「错误里带多少原文」。
|
||||
// 非 2xx 是攻击面而不是数据面,上界刻意小。
|
||||
const maxErrorBodyBytes = 4096
|
||||
|
||||
// HTTP client tuning values, matching the recommended pool in
|
||||
@@ -234,8 +236,13 @@ func (e *APIEmbedder) embedChunk(ctx context.Context, texts []string) ([]Vector,
|
||||
defer func() { _ = resp.Body.Close() }()
|
||||
|
||||
if resp.StatusCode != http.StatusOK {
|
||||
msg := readErrorBody(resp.Body)
|
||||
return nil, fmt.Errorf("embedding: status %d: %s", resp.StatusCode, msg)
|
||||
// 这一处是全片最敏感的一条:/embeddings 的 **input 就是她的记忆正文**,
|
||||
// 而兼容层常把 input 原样回显在错误体里(`Invalid value: '<正文>'`)。
|
||||
// 因此正文只用来数长度、取摘要、抽 code/type,一个字节都不进 error ——
|
||||
// 口径由 trace.UpstreamError(M40 共享 helper)统一决定。
|
||||
body, _ := io.ReadAll(io.LimitReader(resp.Body, maxErrorBodyBytes))
|
||||
return nil, fmt.Errorf("embedding: %w", trace.UpstreamError(
|
||||
trace.HostOf(e.cfg.APIURL), resp.StatusCode, resp.Header.Get("Content-Type"), body))
|
||||
}
|
||||
|
||||
data, err := io.ReadAll(io.LimitReader(resp.Body, maxResponseBytes))
|
||||
@@ -300,20 +307,6 @@ func (e *APIEmbedder) newEmbeddingRequest(ctx context.Context, texts []string) (
|
||||
return req, nil
|
||||
}
|
||||
|
||||
// readErrorBody drains a small prefix of a non-OK response body into a message,
|
||||
// capped at maxErrorBodyBytes so an error response cannot exhaust memory.
|
||||
func readErrorBody(r io.Reader) string {
|
||||
raw, err := io.ReadAll(io.LimitReader(r, maxErrorBodyBytes))
|
||||
if err != nil {
|
||||
return err.Error()
|
||||
}
|
||||
msg := strings.TrimSpace(string(raw))
|
||||
if msg == "" {
|
||||
return "bad status"
|
||||
}
|
||||
return msg
|
||||
}
|
||||
|
||||
// defaultHTTPClient builds the HTTP client configuration described in
|
||||
// DevDocs/02-development-constraints.md §2.3: tuned idle/TLS timeouts plus a
|
||||
// total request timeout. A fresh client is returned per call (never a package
|
||||
|
||||
Reference in new issue
Block a user