Operations, Monitoring, and Cost Management
What to know
Operational excellence relies on CloudMonitor for metrics, alarms, and dashboards, plus log collection for troubleshooting. Define alarms on key signals (CPU, memory where available, disk, latency) and route notifications. Use tagging for cost allocation and resource governance. Cost levers include right-sizing instances, reserved/commitment pricing for steady workloads, pay-as-you-go for spiky ones, storage lifecycle rules, and releasing idle resources (unattached disks, unused EIPs). Auto Scaling aligns spend with demand. Backups and snapshots support recovery objectives: RTO (time to recover) and RPO (data loss tolerance).
Traps
Questions test reserved vs. on-demand: commitments cut cost for predictable steady load but waste money if the workload disappears; on-demand suits variable load. Watch for idle-resource waste - unattached disks and idle EIPs still bill. Alarms without notification actions are useless. Confusing RTO and RPO is common: RTO is downtime tolerance, RPO is acceptable data loss. Also, tags only help cost allocation if applied consistently.
Rule of thumb
Right-size first, then apply commitments to the stable baseline and leave the variable peak on pay-as-you-go. Tag every resource with owner and environment.
Worked example
A workload runs at a steady 60% baseline with seasonal spikes. Reserve capacity for the baseline, let Auto Scaling cover spikes on pay-as-you-go, and set CloudMonitor alarms with notifications for CPU and latency.