在线上故障排查期间,仅靠肉眼扫视原始反向代理日志,很难将客户端限流激增(429)或客户端提前断开连接(499)与真正的上游服务崩溃(502 和 504)区分开来。内置的日志提取器可在浏览器本地扫描粘贴的日志行或拖入的 .log 文件,精准提取带引号请求行之后的 HTTP 响应状态码或 JSON 状态字段,并计算出精确的频率分布明细。
以下是一段来自出现延迟毛刺的 API 网关的 Nginx Combined 格式混合日志示例:
198.51.100.42 - - [01/Nov/2026:14:22:01 +0000] "GET /v2/orders?limit=50 HTTP/2.0" 200 4812 "-" "curl/8.7.1"
198.51.100.42 - - [01/Nov/2026:14:22:02 +0000] "POST /v2/orders HTTP/2.0" 308 0 "-" "curl/8.7.1"
203.0.113.19 - - [01/Nov/2026:14:22:03 +0000] "POST /v2/orders/ HTTP/2.0" 429 118 "-" "python-httpx/0.27.0"
192.0.2.88 - - [01/Nov/2026:14:22:05 +0000] "GET /v2/reports/export HTTP/2.0" 499 0 "-" "Mozilla/5.0"
192.0.2.88 - - [01/Nov/2026:14:22:09 +0000] "GET /v2/reports/export HTTP/2.0" 504 569 "-" "Mozilla/5.0"
203.0.113.55 - - [01/Nov/2026:14:22:10 +0000] "GET /v2/inventory HTTP/2.0" 200 1940 "-" "Go-http-client/2.0"
提取器会同时汇总单个状态码和 RFC 9110 响应类别,帮助您立即评估 SLI 错误预算消耗情况:
{
"totalResponses": 6,
"byClass": {
"2xx": { "count": 2, "share": "33.33%" },
"3xx": { "count": 1, "share": "16.67%" },
"4xx": { "count": 2, "share": "33.33%" },
"5xx": { "count": 1, "share": "16.67%" }
},
"byCode": [
{ "code": 200, "phrase": "OK", "count": 2, "standard": "RFC 9110" },
{ "code": 308, "phrase": "Permanent Redirect", "count": 1, "standard": "RFC 9110" },
{ "code": 429, "phrase": "Too Many Requests", "count": 1, "standard": "RFC 6585" },
{ "code": 499, "phrase": "Client Closed Request", "count": 1, "standard": "Nginx Extension" },
{ "code": 504, "phrase": "Gateway Timeout", "count": 1, "standard": "RFC 9110" }
]
}
结合 /v2/reports/export 接口上成对出现的 499 和 504,可以立即洞察故障模式:缺乏耐心的客户端在上游工作进程响应前主动中断了连接(499),而持续保持到超过 proxy_read_timeout 阈值的连接则以 504 Gateway Timeout 告终。