> It's also curious to me that the load balancer doesn't smooth this out. If you have ten application servers and a client makes ten requests over a single HTTP/2 connection, I'd expect each server to respond to one request each.
A lot of people set up their load balancers with session pinning (i.e. always choose the same backend based on the session id). This can improve things like cache performance.
Not sure if this is the case here, but it sounds like it.
A lot of people set up their load balancers with session pinning (i.e. always choose the same backend based on the session id). This can improve things like cache performance.
Not sure if this is the case here, but it sounds like it.