Replication Lag Anomalies in a Social App

A social app runs Postgres with one leader and four asynchronous read replicas behind a read endpoint. Support tickets report:

• "I changed my display name and the profile page still shows the old one for a few seconds."
• "I refresh my feed and a comment I just saw disappears, then comes back on the next refresh."

1. Name the anomaly behind each report and explain the mechanism.
2. Propose a fix for each that keeps the read replicas in use (do not send all reads to the leader), and describe the trade-off.
3. What single metric would you alert on to catch this class of problem before users do?