A three-service app on three nodes in three zones, watched by Prometheus, Loki and Tempo, shown in Grafana, with Alertmanager on call. Slow a service down and find it in the metrics and the trace, crash a pod and find it in the logs, then upgrade the nodes and watch PodDisruptionBudgets and topology spread keep the app up. Or press ▶ Guided tour.
Metrics tell you something is wrong (Prometheus scrapes counters and histograms; PromQL turns them into rates and percentiles), traces tell you where (each service reports a span of the same request), and logs tell you why (Fluent Bit ships every container's output to Loki). Alert rules fire only when a condition holds for a while, so a blip doesn't page anyone. Reliability during maintenance comes from spreading replicas across zones and from PodDisruptionBudgets: a drain evicts pods through the API, which refuses any eviction that would leave fewer ready replicas than the budget allows, so node upgrades proceed one pod at a time.