I am using kubernetes 1.9.2 created but kubeadm. this kubernetes cluster is running in 4 ec2 nodes.
I have a deployment that requires using cache in every pod. in order to accomlish that we used session affinity from ClusterIP.
since I was ELB in front of my Kubernetes cluster I wonder how the session affinity is behaving.
the natural behavior would be that for every client IP a different will get the requests but given the traffic is transferred via ELB , whoch IP does the session affinity recognizes , the ELB IP or the actual Client IP?
when I check the traffic to the pods I see that 102 pods get all the requests and the 2 other pods are just waiting.
many thanks for any help.
SessionAffinity recognizes Client IP and ELB should pass the Client IP.
I think you should work with HTTP Headers and Classic Load Balancers and setup X-Forwarded-For: client-ip-address
Also, this seems to be a know issue enabling Session affinity goes to a single pod only #3056.
It was reported for 0.18.0 and 0.19.0 version of NGINX Ingress controller.
Issue was closed and commented that is was fixed in version 0.21.0, but in December initial author said it still doesn't work for him.