perf(queue): 限制大媒体任务内存并消除准入惊群

将同步与异步非文本任务的准入唤醒改为按任务和全局队首推进,批量续租等待者,避免大量请求同时争抢 advisory lock 和数据库连接。

对 Base64 请求解析、素材解码、上游媒体执行和结果物化增加分层并发限制,复用请求素材并释放重复 Gemini wire 数据;生产 API 默认入口解析并发 16,媒体物化并发 8。

新增 1000 个同步 Gemini 图像编辑请求的模拟上游压力验收,10 MiB 输入和输出下全部成功,Heap 峰值增长约 2.58 GiB,并验证共享上传哈希、单次 attempt 与本地零落盘。

验证:go test ./... -count=1;go vet ./...;gofmt -l 无输出;kubectl kustomize deploy/kubernetes/production;10 MiB Gemini Base64 千任务压力测试通过。
This commit is contained in:
2026-07-30 13:13:32 +08:00
parent fd3b6bf042
commit ea58d21d03
26 changed files with 1980 additions and 74 deletions
+7
View File
@@ -65,6 +65,13 @@ func (c GeminiClient) Run(ctx context.Context, request Request) (Response, error
}
return Response{}, annotateResponseError(err, requestID, responseStartedAt, responseFinishedAt)
}
if wire != nil && geminiImageModelType(request.ModelType) {
// Gemini image responses can contain multi-megabyte Base64 fields. The
// decoded wire body is still used for native synchronous passthrough, so
// retaining the original JSON bytes would keep a second full copy alive
// until the client response finishes.
wire.RawJSON = nil
}
if requestID == "" {
requestID = requestIDFromResult(output)
}