docker registry罕见原因导致的故障dial tcp 127.0.0.1:5000: connect: connection refused

背景

系统环境：k8s+docker+cri-dockerd

因为我不想把镜像通过Docker hub公开，以及将来在不联网的生产环境部署，自己运行一个docker存储库，在k8s部署工作负载时从中拉取镜像。

问题

没有修改环境配置，进行了一些k8s和docker相关操作后，再推送镜像时突然发生错误。

复制代码

Get "http://localhost:5000/v2/": dial tcp 127.0.0.1:5000: connect: connection refused

解决

一开始我按一般排查故障的方法，检查 registry 容器日志，docker 服务日志，重启docker服务，重新部署 registry 容器等等，均未解决问题，百思不得其解。

后续我进行k8s操作，部署时发现问题，大概是在k8s部署的容器可以分配一个Node端口，同一个Node的同一个端口只能分配一次，导致只有一个Node时不能部署第二份。表现如下：

复制代码

root@vmi1640551:~# kubectl -n test-cinema-2 get po
NAME                            READY   STATUS    RESTARTS   AGE
a-bookings-1-756694bb6b-sdqbg   0/1     Pending   0          8m4s
a-movies-1-66785d95ff-6jp27     0/1     Pending   0          8m4s
a-showtimes-1-fcb9d8bc6-9txh5   0/1     Pending   0          8m4s
a-users-1-59bb6845cf-zb7xw      0/1     Pending   0          8m4s
proxy                           1/1     Running   0          8m14s
root@vmi1640551:~# kubectl -n test-cinema-2 describe po a-bookings-1-756694bb6b-sdqbg  
Name:             a-bookings-1-756694bb6b-sdqbg
Namespace:        test-cinema-2
Priority:         0
Service Account:  default
Node:             <none>
Labels:           app=a-bookings-1
                  pod-template-hash=756694bb6b
Annotations:      kompose.cmd: kompose --file docker-compose.yml convert
                  kompose.version: 1.32.0 (HEAD)
Status:           Pending
IP:               
IPs:              <none>
Controlled By:    ReplicaSet/a-bookings-1-756694bb6b
Containers:
  bookings:
    Image:      localhost:5050/cinema-2/bookings
    Port:       5003/TCP
    Host Port:  5003/TCP
    Limits:
      cpu:  100m
    Requests:
      cpu:        100m
    Readiness:    http-get http://:5003/health-check delay=0s timeout=1s period=3s #success=1 #failure=2
    Environment:  <none>
    Mounts:
      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-pmx4b (ro)
Conditions:
  Type           Status
  PodScheduled   False 
Volumes:
  kube-api-access-pmx4b:
    Type:                    Projected (a volume that contains injected data from multiple sources)
    TokenExpirationSeconds:  3607
    ConfigMapName:           kube-root-ca.crt
    ConfigMapOptional:       <nil>
    DownwardAPI:             true
QoS Class:                   Burstable
Node-Selectors:              <none>
Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
  Type     Reason            Age                   From               Message
  ----     ------            ----                  ----               -------
  Warning  FailedScheduling  3m9s (x2 over 8m20s)  default-scheduler  0/1 nodes are available: 1 node(s) didn't have free ports for the requested pod ports. preemption: 0/1 nodes are available: 1 No preemption victims found for incoming pod.
root@vmi1640551:~#

我猜测可能是有k8s中容器占用了5000端口。修改 registry 绑定的本地端口后（比如改为5050），推送成功了。

检查确实如此，有一个工作负载的配置如下：

yaml 复制代码

apiVersion: apps/v1
kind: Deployment
metadata:
  annotations:
    kompose.cmd: kompose --file docker-compose.yml convert
    kompose.version: 1.32.0 (HEAD)
  labels:
    io.kompose.service: users
  name: users
spec:
  replicas: 1
  selector:
    matchLabels:
      io.kompose.service: users
  template:
    metadata:
      annotations:
        kompose.cmd: kompose --file docker-compose.yml convert
        kompose.version: 1.32.0 (HEAD)
      labels:
        io.kompose.network/cinema-2-default: "true"
        io.kompose.service: users
    spec:
      containers:
        - image: localhost:5000/cinema-2/users
          name: users
          ports:
            - containerPort: 5000
              hostPort: 5000 # ! 注意这里 !
              protocol: TCP
          readinessProbe:
            httpGet:
              path: /health-check
              port: 5000
            periodSeconds: 3 # 默认 10
            failureThreshold: 2 # 默认 3
            successThreshold: 1
            timeoutSeconds: 1
      restartPolicy: Always

删除实际未使用的 hostPort 后恢复正常。

后续疑问

有些疑问还没来得及解决：

k8s、docker的网络原理是怎样的？
特别的，由k8s pull镜像、从主机docker push镜像和curl localhost:5000 的请求会被如何路由？是否有区别？
registry 和 k8s中部署的工作负载应该只有一个能监听唯一Node的5000端口，为什么看起来似乎都部署成功了，看不到错误？