Code Room
On-callHardoc-g345
Subject Inconsistent replicasLevel Senior–Staff~40 minCommon in Databases & SQL · Concurrency · Reliability & on-call · Distributed systems interviewsIndustries Technology

Question

A managed MySQL cluster (semi/async replication, automated failover) failed over at 02:50 when the primary's host became unreachable. A replica was promoted and writes resumed in ~30s. At 09:00, reconciliation finds a few hundred rows where the *old* primary and the *new* primary disagree: the old primary has writes from 02:48–02:50 that the new primary never received, and a couple of rows were written on *both* with different values. The old host has since rejoined as a replica. Dashboards: replication healthy now, no current lag. How do you triage, prevent the divergence from spreading, and reconcile the data safely?

What a strong answer looks like

Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.

Diagram & narrate the incident
Loading whiteboard…
Run or narrate your approach, then ask the coach.