Stale DNS cache after pod reschedule

After a routine deploy that rescheduled many pods, your checkout service starts intermittently connecting to dead backends. Dashboards: ~15% of requests to the payments pods fail with connection-refused / connection-timeout; the failing connections target pod IPs that no longer exist (old pods that were replaced during the deploy); CoreDNS query latency is elevated and its cache hit-rate is unusually high. Recent context: someone recently increased the in-cluster DNS TTL and enabled aggressive client-side DNS caching in the service's HTTP client to 'reduce DNS load.' How do you triage and mitigate?