Pod to pod communication not working with Cilium

smarshie73

Hi. I installed Cilium as per instructions in Dec 2023 lab exercises (using the supplied manifest cilium-cni.yaml).

I've noticed that after creating the nginx deployment and service in Exercise 3.4 that it can only be reached from the node the pod is running on. If I scale up the deployment then it seems to be hit and miss whether curl succeeds (I assume depending on which endpoint the request is handed off to by the service).

Running cilium connectivity tests as advised on the cilium website (using kubectl apply -n cilium-test -f https://raw.githubusercontent.com/cilium/cilium/1.15.0/examples/kubernetes/connectivity-check/connectivity-check.yaml) confirms the problem, with failure on all tests of the type pod-to-b-multi-node.... or host-to-b-multi-node.

Why would this be happening and what should be done to put it right?

I should point out that my two nodes are both in the same VPC in AWS and both have completely open security groups.

Cheers

Steve

smarshie73

Ahh I worked it out. Turns out that opening up TCP was not enough. Cilium relies on UDP encapsulated packets. Opening up everything on UDP fixed it.

chrispokorni

Hi @smarshie73,

The introductory chapter has a demo video on AWS EC2 lab environment, which covers critical details such as SG rules, instance size, and recommended OS.

Regards,
-Chris

sgaede

@smarshie73 said:
Ahh I worked it out. Turns out that opening up TCP was not enough. Cilium relies on UDP encapsulated packets. Opening up everything on UDP fixed it.

+1
Thanks so much!