Question

Your app writes to a Postgres primary and routes reads to replicas, with a rule: for ~2s after a user writes, route *their* reads to the primary so they see their own change (read-your-writes). After a deploy at 16:00 that moved this routing decision behind a new load balancer, users intermittently report that a setting they just saved 'didn't save' — then it's fine on the next reload. Dashboards: replica lag is normal (~200ms), error rate flat, write success rate 100%. How do you triage, restore read-your-writes, and confirm no writes were actually lost?