Nginx 入口：向 url 添加参数时出现 502

Question

我在 k8s 中部署了一个 Nodejs Rest 服务，通过 nginx 入口公开。它响应基本的获取，但当我传递 URL 参数时，我得到 502。

import express from "express";
const app = express();
app.get("/service-invoice", async (req, res) => {
  res.send(allInvoices);
}
app.listen(80);

其中 allInvoices 只是从 MongoDB 加载的发票对象的集合。

我使用以下入口配置将其部署到 k8s：

---
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: service-invoice-read
  namespace: ctx-service-invoice
  annotations:
    kubernetes.io/ingress.class: addon-http-application-routing
    nginx.ingress.kubernetes.io/rewrite-target: /$2
spec:
  rules:
  - http:
      paths:
      - path: /service-invoice-read(/|$)(.*)
        pathType: Prefix
        backend:
          service:
            name: service-invoice-read
            port:
              number: 80

用curl 调用它：

curl localhost:30000/service-invoice-read/service-invoice

我得到一个有效的 json 响应。到目前为止，一切都很好。

但是，我也想通过 Id 访问这些对象。为此，我有以下代码：

app.get("/service-invoice/:id", async (req, res) => {
  try {
    const id = req.params.id;
    const invoice = // code to load invoice by id from mongo
    res.send(invoice);
  } catch (e) {
    res.status(sc.NOT_FOUND).send(error);
  }
});

用curl调用此：

curl localhost:30000/service-invoice-read/service-invoice/e98e03b8-b590-4ca4-978d-270986b7d26e

导致 502 - Bad Gateway 错误。

我在 pod 日志中看不到任何错误，所以我很确定这是来自 nginx。

我真的不明白这是从哪里来的。我尝试过不使用 try/catch 在日志中查看它是否爆炸，但仍然没有乐趣。

这是我的入口日志，按照评论中的要求：

2022/03/03 18:45:21 [error] 847#847: *4524 upstream prematurely closed connection while reading response header from upstream, client: 10.42.1.1, server: _, request: "GET /service-invoice-read/service-invoice/6220d042a95986f58c46356f HTTP/1.1", upstream: "http://10.42.1.100:80/service-invoice/6220d042a95986f58c46356f", host: "localhost:30000"
2022/03/03 18:45:21 [error] 847#847: *4524 connect() failed (111: Connection refused) while connecting to upstream, client: 10.42.1.1, server: _, request: "GET /service-invoice-read/service-invoice/6220d042a95986f58c46356f HTTP/1.1", upstream: "http://10.42.1.100:80/service-invoice/6220d042a95986f58c46356f", host: "localhost:30000"
2022/03/03 18:45:21 [error] 847#847: *4524 connect() failed (111: Connection refused) while connecting to upstream, client: 10.42.1.1, server: _, request: "GET /service-invoice-read/service-invoice/6220d042a95986f58c46356f HTTP/1.1", upstream: "http://10.42.1.100:80/service-invoice/6220d042a95986f58c46356f", host: "localhost:30000"
10.42.1.1 - - [03/Mar/2022:18:45:21 +0000] "GET /service-invoice-read/service-invoice/6220d042a95986f58c46356f HTTP/1.1" 502 150 "-" "curl/7.68.0" 140 0.006 [ctx-service-invoice-service-invoice-read-80] [] 10.42.1.100:80, 10.42.1.100:80, 10.42.1.100:80 0, 0, 0 0.004, 0.004, 0.000 502, 502, 502 b78e6879fabe2d5947525a2b694b4b9f
W0303 18:45:21.529749       7 controller.go:1076] Service "ctx-service-invoice/service-invoice-read" does not have any active Endpoint.

有人知道我在这里做错了什么吗？

Answer 1

问题并不像看上去的那样。在这种情况下，配置工作正常。真正的问题是代码中存在一个错误，该错误被全局异常处理程序抑制而没有被记录。由于某种原因，这导致了 502——尽管我仍然不明白为什么我得到了这个确切的响应，但我并不是特别感兴趣。

全局异常处理程序的目的是在服务可能终止时保持服务运行。鉴于 k8s 中的服务终止是完全可以接受的，我删除了该处理程序并允许 pod 终止，这为我提供了有关正在发生的情况的更多信息。