perf: reduce heap residency for large base64 relay requests

Three layered optimizations targeting Gemini-style 5MB base64 payloads where
RSS could balloon to tens of GB under concurrent load:

1. Byte-based param override (relay/common/override.go)
   - Switch legacy/operations hot paths from common.Marshal round-trips and
     map[string]any conversions to gjson/sjson on []byte directly.
   - Avoids cloning 5MB strings during each Set/Delete operation.

2. strings.Builder for Gemini response markdown (relay/channel/gemini/relay-gemini.go)
   - Replace string concatenation + strings.Join when assembling
     "![image](data:...;base64,DATA)" content for inline image responses.
   - Pre-allocates capacity from inline_data byte sizes.

3. Outbound BodyStorage + streaming Decoder (this commit's core)
   - New relay/common/outbound_body.go helper wraps marshaled upstream bodies
     in common.BodyStorage, allowing disk-cache mode to offload jsonData to
     a temp file while waiting for upstream TTFB. The original []byte can
     then be GC'd, removing ~5MB/req of heap residency during the longest
     window of a request.
   - All 7 relay handlers (gemini/claude/responses/embedding/image/compatible/
     rerank) plus chat_completions_via_responses adopt the helper with
     defer closer.Close() and explicit jsonData = nil.
   - relay/common/relay_info.go: new UpstreamRequestBodySize so
     relay/channel/api_request.go can populate req.ContentLength (lost when
     body becomes a type-erased io.Reader).
   - common/gin.go UnmarshalBodyReusable: when storage is disk-backed and
     content-type is JSON, decode via DecodeJson(storage) instead of
     storage.Bytes()+Unmarshal, removing one transient 5MB copy per request.
     memory mode and form/multipart paths unchanged.
This commit is contained in:
CaIon
2026-05-22 19:08:38 +08:00
parent b9bc6f0e21
commit fddf54ccc5
15 changed files with 407 additions and 169 deletions
+20
View File
@@ -25,6 +25,23 @@ import (
"github.com/gorilla/websocket"
)
// applyUpstreamContentLength populates req.ContentLength when the upstream
// body is wrapped in a BodyStorage (see relay/common/outbound_body.go).
//
// net/http.NewRequest only auto-detects ContentLength for *bytes.Reader,
// *bytes.Buffer and *strings.Reader. When the body is a type-erased io.Reader
// (which is the case for ReaderOnly(BodyStorage)), the Content-Length header
// would otherwise be omitted, forcing chunked transfer encoding and breaking
// some upstreams that require an explicit Content-Length.
func applyUpstreamContentLength(req *http.Request, info *common.RelayInfo) {
if info == nil {
return
}
if info.UpstreamRequestBodySize > 0 && req.ContentLength <= 0 {
req.ContentLength = info.UpstreamRequestBodySize
}
}
func SetupApiRequestHeader(info *common.RelayInfo, c *gin.Context, req *http.Header) {
if info.RelayMode == constant.RelayModeAudioTranscription || info.RelayMode == constant.RelayModeAudioTranslation {
// multipart/form-data
@@ -297,6 +314,7 @@ func DoApiRequest(a Adaptor, c *gin.Context, info *common.RelayInfo, requestBody
if err != nil {
return nil, fmt.Errorf("new request failed: %w", err)
}
applyUpstreamContentLength(req, info)
headers := req.Header
err = a.SetupRequestHeader(c, &headers, info)
if err != nil {
@@ -326,6 +344,7 @@ func DoFormRequest(a Adaptor, c *gin.Context, info *common.RelayInfo, requestBod
if err != nil {
return nil, fmt.Errorf("new request failed: %w", err)
}
applyUpstreamContentLength(req, info)
// set form data
req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
headers := req.Header
@@ -522,6 +541,7 @@ func DoTaskApiRequest(a TaskAdaptor, c *gin.Context, info *common.RelayInfo, req
if err != nil {
return nil, fmt.Errorf("new request failed: %w", err)
}
applyUpstreamContentLength(req, info)
req.GetBody = func() (io.ReadCloser, error) {
return io.NopCloser(requestBody), nil
}