"and it turned out this wasn’t driven by memory pressure. Rather, the new capacity had caused all of the servers in the fleet to exceed the maximum number of threads allowed by an operating system configuration."

An auto scaling irony for AWS! We seem to be back to the late 1990s :)