1. 13
  1.  

  2. 3

    Just have your clients reconnect every 10 minutes or so?

    1. 2

      Just restart whichever server has the highest load every 10 minutes. Solves memory leaks too.

      1. 1

        Or listen to a certain response from the server and only then reconnect. That way you only need to reconnect on demand. Still this is a hackish kind of way to “solve” the problem

      2. 3

        Pardon my naivety but isn’t this a case of your load balancer not… balancing load?

        What is the load balancer actually for, then?

        How do other load balancers successfully handle stateful protocols like SSH?

        1. 3

          Who load balances SSH?

          1. 1

            What is the load balancer actually for, then?

            For giving you a chance to use your resources, and justifying your on call team.

          2. 1

            This seems like it would be a general problem for load balancers that are written with request-oriented protocols in mind. It might be interesting to look at how other protocols that use long streams of request/response pairs have ended up solving this. Perhaps look at database load balancers?