When using Pod probes, I found a logical issue

General Discussions
1

my Resource manifest file:

---
kind: Pod
apiVersion: v1
metadata:
  name: web
spec:
  containers:
  - name: nginx
    image: nginx
    readinessProbe:
      tcpSocket:
        host: 127.0.0.1
        port: 9000
  - name: php
    image: php-fpm

I need to check if the PHP service is up and running within the Nginx container. The failure should be detected only if the PHP service has been confirmed to start successfully.
After debugging, I found that the tcpSocket.host parameter should have used the network namespace of the web container, but mistakenly used the host machine’s namespace. Omitting the tcpSocket.host parameter allows the probe to function correctly.
I found that startupProbe, readinessProbe, and livenessProbe all have the same issue.

Kubernetes version: v1.29.2
Cloud being used: bare-metal
Installation method: kubeadm
Host OS: Rocky-Linux-8.5
CNI and version: v3.26.4
CRI and version: containerd 1.6.28-3

2

Host is a poorly defined field. It started meaning one thing and sort of ended up meaning something else. It is relative to the host machine, not the pod. I recommend never to use it.