Easy
A DNS record change and a long-running client
A service's A record has a 300-second TTL. Ops repoints the name at a new host, waits an hour, and decommissions the old one. A long-running client process, which resolved the name once at startup and keeps a pool of persistent connections, is still sending requests to the old IP and failing hours later. What explains it?
30 of 45 in Networking