Question

A deploy ships at 11:00. Everything is fine for ~90 minutes. Then a slow-burn outage begins: pods across the fleet start failing health checks and getting evicted, and the log-ingestion pipeline backs up. Dashboards: node disk usage on the app nodes climbs steadily and crosses 95%; the eviction events correlate exactly with disk pressure; CPU/mem/request-rate are normal. The new release accidentally left a `logger.debug` on the hot request path enabled at INFO level (a level mis-set in the deploy's config), emitting ~5 KB per request to local stdout/files; log rotation is size-based but the rotation threshold was never hit fast enough. How do you triage and mitigate?