I have noticed that containers in a pod can use localhost to talk to each other as advertised. For example one container starts a server socket on localhost:9999 and a second container can connect to that address. This fails if I expose the server container's port. Also it fails if I create a TCP liveness probe on that port. It appears that the liveness probe uses the pod IP address and cannot connect to localhost:9999 unless it is exposed. If both containers use the pod IP, i.e., $HOSTNAME:9999, and the port is exposed then everything works. Does any one have an example that works where each container uses localhost and the TCP probe works?
Here is an example deployment using TCP liveness probe, TCP readiness probe and networking between containers in a pod with the server containers port exposed:
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
name: test
spec:
template:
metadata:
labels:
app: test
spec:
containers:
- name: server
image: alpine
command:
- '/bin/sh'
- '-c'
- 'nc -p 8080 -kle echo pong'
livenessProbe:
tcpSocket:
port: 8080
readinessProbe:
tcpSocket:
port: 8080
ports:
- containerPort: 8080
- name: client
image: alpine
command:
- '/bin/sh'
- '-c'
- 'while true; do echo -e | nc localhost 8080; sleep 1; done'Creating and verifying the deployment:
> kubectl create -f test.yml
> kubectl get pod -l app=test
NAME READY STATUS RESTARTS AGE
test-620943134-fzm05 2/2 Running 0 1m
> kubectl log test-620943134-fzm05 client
pong
pong
pong
[…]