Transferred volume landed above plan, with an escalation open claiming our stack could not carry it. I spent two days on call watching every query crossing the Presto engine while reading CPU, memory, disk and network on 21 servers in Splunk, vROps and ExtraHop, deciding live which long-running queries to kill. With another engineer, I modeled six months of ingress and egress for every server. The stack had carried heavier load before, which retired the escalation.