I'm working on a project where I want to use Kubernetes and Docker. The microservice I'm about to implement must create a permanent HTTP/2 connection to another service (provided by others and I can't modify anything in that service) pro user / client in order to send asynchronous and cloud initiated messages to that user. Also, each following request from that client must use the same connection.

Obviously that is a challenge in terms of scalability, because every request from a client must be routed to the same instance of my microservice, which created the permanent connection to the other service. What makes things worse is the fact that my clients can change the IPs and that they can't use cookies. But what they can do is to send a custom header value which identifies them.

I thought about HAProxy and nginx, but can't find an option in either of them to load balance requests by a header value. Is there really no way to do that? How would you approach that issue? Any ideas?

Thanks!