Monitoring and Maintenance
Know which tool answers which question
CloudWatch holds the built-in database metrics at one-minute granularity: CPUUtilization, FreeableMemory, DatabaseConnections, ReadIOPS/WriteIOPS, ReplicaLag, and for Aurora AuroraReplicaLag, VolumeBytesUsed, and BufferCacheHitRatio. Enhanced Monitoring runs an agent in the instance and delivers OS-level process and memory detail at 1–60 second granularity. Performance Insights measures database load and attributes it to SQL statements, wait events, and users - this is the tool for "which query is slow." CloudTrail records control-plane API calls, answering "who deleted the table." Export engine logs (error, slow query, general, audit) to CloudWatch Logs for searching and alerting.
DynamoDB and other engines
DynamoDB's critical metrics are ThrottledRequests, ProvisionedThroughputExceededException, ConsumedReadCapacityUnits, and SystemErrors; Contributor Insights surfaces the partition keys driving hot traffic. Redshift exposes QueryDuration, WLMQueueLength, and system tables for skew and vacuum monitoring.
Maintenance duties
Watch replica lag before promoting, rotate logs, review storage headroom and enable storage autoscaling, run index and statistics maintenance, and validate that maintenance windows do not collide with batch peaks. Subscribe to RDS event subscriptions through EventBridge so failovers, low storage, and maintenance events trigger automation.
Traps
Performance Insights does not show per-process OS CPU - that is Enhanced Monitoring. CloudWatch's standard one-minute metrics are not real-time; detailed/1-second monitoring is a separate setting. Throttling on DynamoDB is often a partition-key design problem, not a capacity problem - adding capacity to a hot partition does not fix it. Cross-Region read replicas show higher ReplicaLag and are not failover targets for Multi-AZ.
Rule of thumb
Map the question: slow query → Performance Insights; OS or process spike → Enhanced Monitoring; API audit → CloudTrail; capacity or throttling trend → CloudWatch metrics plus Contributor Insights.