What happen?
On Saturday's evening (UTC), bitcoin daemons on two independent pool servers ate all available memory (RAM & swap) and fall to massive I/O operations, which make servers extremely slow. Central server switched traffic to the last machine, but this machine was unable to handle all traffic and some RPC calls started to timeout. Then those miners getting RPC timeouts probably started to ask server even more frequently (miners don't implement any "polite waiting" in the code and send new RPC call immediately after error/timeout). Those thousands of requests per second were the last piece to shut down last working machine .