Monitoring, Operations and Management
Core concepts
CloudMonitor collects metrics (CPU, memory via agent, disk, network), supports alarms with notification contacts, and integrates with Auto Scaling. ActionTrail records control-plane API calls for audit and compliance. Log Service (SLS) centralizes logs for query and analysis. Operation Orchestration Service (OOS) automates routine tasks like patching and runbooks.
Operational practices: tagging for cost allocation and inventory, resource groups for organization, snapshots and backups for recovery, and health checks for availability. Distinguish metrics (numeric time series) from logs (event records) from traces (request flow).
Traps in questions
- CloudMonitor vs. ActionTrail: metrics/alarms vs. API audit trail. A question about "who deleted the instance" is ActionTrail, not CloudMonitor.
- Agent-dependent metrics: memory and process metrics often require the CloudMonitor agent; hypervisor-level metrics do not.
- Alarm vs. auto-recovery: an alarm notifies; remediation requires Auto Scaling, OOS, or manual action.
- Log retention: SLS retention settings affect cost and compliance; logs are not kept forever by default.
Rule of thumb
For any production workload, define what to measure, when to alarm, and what action fires. An alarm with no runbook is incomplete.
Worked example
An instance's CPU spikes at night and no one notices. Add a CloudMonitor alarm on CPU over a threshold for a sustained period, route notifications to an on-call contact, and attach an Auto Scaling rule to add capacity. Use ActionTrail separately to investigate configuration changes.