I have a situation.
I have EKS cluster running, with two pods handling incoming udp traffic, load balanced using NLB.
The external server/client will send udp packets to NLB ip and NLB load balances to udp pods.
The peer here only knows about NLB ip only.
So, now while sending messages to peer, I am not able to use NLB ip address, the packet goes out with src ip as EC2 instance public ip, which peer does not recognizes and drops it.
As per application requirement, the src ip should be same as NLB ip, as peer is configured with it.
Can someone help me here, whether it is possible in aws while using NLB ?
AWS 官方已更新 2 年前
We want to use client-ip for this UDP use case. Moreover, client ip preservation cannot be disabled for UDP packets for Network load balancer.
From https://docs.aws.amazon.com/elasticloadbalancing/latest/network/load-balancer-target-groups.html#client-ip-preservation :
"By default, client IP preservation is enabled (and can't be disabled) for instance and IP type target groups with UDP and TCP_UDP protocols."
Yes that is correct. I miss read the type. Hmm You need the return packets to be sourced from the NLB still? The only way I’ve done this in the past was with F5s on prem with return packets flow around the load balancer and layer 2 arping with loop backs on the target NICS.
You need some kind of source NAT firewall.
Actually return packet works fine, if it is sent back immediately. But if it is sent after sometime, it does not go via NLB, it source ip/port shows up as ec2 instance ip/port at peer, which I do not want !!