Rds operation review

RDS / Aurora Operational Review

Conduct a comprehensive operational review of Amazon RDS instances and Amazon Aurora clusters aligned with the AWS Well-Architected Framework and the Amazon RDS and Aurora best-practices guides.

This skill uses the AWS RDS, Aurora, CloudWatch, CloudWatch Logs, EC2, and Cost Explorer APIs only — no Kubernetes (k8s) or eks MCP tools. All data is collected through native AWS APIs available to the DevOps Agent's primary cloud source role.

When to Use

Activate this skill when the user asks to:

• Review, audit, or assess RDS instances or Aurora clusters
• Check RDS / Aurora best-practices compliance
• Evaluate database security, cost, reliability, performance, or backups
• Perform an RDS / Aurora operational readiness review (ORR)
• Investigate database health or configuration drift

Step 1: Identify Target Resources

Ask the user which databases to review. Accept:

• Specific instance / cluster identifiers and regions
• "all instances" / "all clusters" in specific regions
• "all RDS in all regions"

If no scope is given, default to all configured account regions.

Step 2: Discover RDS / Aurora Resources

Per region (skip empty regions):

rds.DescribeDBInstances           # RDS instances + Aurora cluster member instances
rds.DescribeDBClusters            # Aurora clusters (skip if user said "Provisioned only")
rds.DescribeDBClusterEndpoints    # Aurora reader/writer endpoints
rds.DescribeGlobalClusters        # Aurora Global Database (cross-region)
rds.DescribeDBProxies             # RDS Proxy

Capture per resource:

• Engine, engineVersion, dbInstanceClass / serverlessV2 capacity
• Multi-AZ, AvailabilityZones, subnetGroup, VpcId
• AllocatedStorage, MaxAllocatedStorage, StorageType, Iops, StorageThroughput, StorageEncrypted, KmsKeyId
• BackupRetentionPeriod, PreferredBackupWindow, PreferredMaintenanceWindow, AutoMinorVersionUpgrade
• DeletionProtection, PubliclyAccessible, IAMDatabaseAuthenticationEnabled
• PerformanceInsightsEnabled, PerformanceInsightsRetentionPeriod, MonitoringInterval (Enhanced Monitoring), MonitoringRoleArn
• EnabledCloudwatchLogsExports, AssociatedRoles
• ReadReplicaSourceDBInstanceIdentifier / ReadReplicaDBInstanceIdentifiers
• Tags (rds.ListTagsForResource — param name is ResourceName and value is the resource ARN)

Step 3: Discover Configuration Dependencies

For each database, collect:

rds.DescribeDBParameterGroups           # custom vs default
rds.DescribeDBClusterParameterGroups    # Aurora
rds.DescribeOptionGroups                # RDS option groups
rds.DescribeDBSubnetGroups              # subnet AZ spread
ec2.DescribeSecurityGroups              # vpcSecurityGroupIds → ingress rules
kms.DescribeKey                         # for KmsKeyId (encryption at rest)

Step 4: Collect Backups, Snapshots, Maintenance

rds.DescribeDBSnapshots                            # manual + automated, per instance
rds.DescribeDBClusterSnapshots                     # Aurora
rds.DescribeDBInstanceAutomatedBackups
rds.DescribeDBClusterAutomatedBackups
rds.DescribeDBEngineVersions                       # current vs latest minor/major
rds.DescribePendingMaintenanceActions
rds.DescribeEvents                                 # last 14 days; duration in MINUTES (20160)
rds.DescribeEventSubscriptions

For each instance/cluster, also fetch:

• rds.DescribeDBLogFiles — list of logs available locally on the DB
• logs.DescribeLogGroups with prefix:
  • Instances: /aws/rds/instance/<dbInstanceIdentifier>/
  • Aurora: /aws/rds/cluster/<dbClusterIdentifier>/

Step 5: Collect CloudWatch Metrics (7-Day Historical)

One cloudwatch.GetMetricData call per resource. Period: 21600 (6 hours). StartTime: 7 days ago. EndTime: now.

5.1 RDS Instances (dimension DBInstanceIdentifier)

id	metricName	stat	unit
cpu	CPUUtilization	Average	%
freeMem	FreeableMemory	Minimum	bytes
freeSpace	FreeStorageSpace	Minimum	bytes
readIops	ReadIOPS	Average	count/s
writeIops	WriteIOPS	Average	count/s
readLat	ReadLatency	Average	s
writeLat	WriteLatency	Average	s
dbConn	DatabaseConnections	Maximum	count
swap	SwapUsage	Average	bytes
netIn	NetworkReceiveThroughput	Average	bytes/s
netOut	NetworkTransmitThroughput	Average	bytes/s
binLog	BinLogDiskUsage	Average	bytes
replLag	ReplicaLag	Maximum	s

5.2 Aurora Clusters (dimension DBClusterIdentifier)

id	metricName	stat	unit
cpu	CPUUtilization	Average	%
freeMem	FreeableMemory	Minimum	bytes
dbConn	DatabaseConnections	Maximum	count
replLag	AuroraReplicaLag	Maximum	ms
bufCache	BufferCacheHitRatio	Average	%
commitLat	CommitLatency	Average	ms
readLat	ReadLatency	Average	s
writeLat	WriteLatency	Average	s
volReadIops	VolumeReadIOPs	Sum	count
volWriteIops	VolumeWriteIOPs	Sum	count
serverlessAcu	ServerlessDatabaseCapacity	Average	ACU

5.3 Per-Instance Aurora Members

Same instance metrics as 5.1, dimensioned by DBInstanceIdentifier of each cluster member.

Fetch cloudwatch.DescribeAlarmsForMetric for the key metrics (CPUUtilization, FreeStorageSpace, DatabaseConnections, ReadLatency, WriteLatency, ReplicaLag / AuroraReplicaLag) per resource so the report can flag missing alarms.

Step 6: Collect Logs (7-Day)

Per database log group, scan with logs.FilterLogEvents for:

Pattern	What it indicates
ERROR / FATAL	engine errors
Out of memory / OOM	memory pressure
connection limit exceeded / too many connections	connection saturation
deadlock / lock wait timeout	contention
slow query / duration: (Postgres slow log)	query performance
aborted connection (MySQL)	client/network issues
replication has stopped / IO_THREAD errors	replication health
checkpoint warnings / archiver failed	I/O / WAL issues
failed to connect / authentication failed	auth/network

Record occurrence counts per pattern over the 7-day window.

Step 7: Collect CloudTrail-Visible RDS Events

Use rds.DescribeEvents (last 14 days) per resource — covers the same operational signal as CloudTrail for RDS without an extra API permission. Look for: failover, restart, parameter group apply, out-of-memory, storage-full, low-storage, automated-backup-failed, maintenance-related events, read replica error, instance stopped.

Step 8: Cost Data

Once per review (not per resource):

costexplorer.GetCostAndUsage     # 3 months, by USAGE_TYPE,
                                 # filter Service = "Amazon Relational Database Service"
rds.DescribeReservedDBInstances  # RI inventory

Estimate per-database monthly cost as a proportional split of the latest month total (Cost Explorer doesn't break down per DB):

1. Get the most recent full month total RDS spend.
2. Per resource, weight by vCPUs(instance class) × (Multi-AZ ? 2 : 1) + allocatedStorageGB.
3. For Aurora clusters, sum across cluster member instances; storage is cluster-shared.
4. Note the estimate in the report with an "estimated" badge.

Step 9: Analyze Against Best Practices

Evaluate ALL collected data across the five Well-Architected pillars and assign a severity to every finding: CRITICAL, HIGH, MEDIUM, LOW, or INFO.

9.1 Security

Ref: Security in Amazon RDS

• Network exposure: PubliclyAccessible=true → CRITICAL for production. Security groups with 0.0.0.0/0 ingress on the DB port → CRITICAL.
• Encryption at rest: StorageEncrypted=false → HIGH (or CRITICAL for regulated workloads). Customer-managed KMS key preferred over AWS-managed.
• Encryption in transit: parameter rds.force_ssl=1 (Postgres) / require_secure_transport=ON (MySQL/MariaDB) → MEDIUM if not set.
• IAM auth: IAMDatabaseAuthenticationEnabled=false → MEDIUM. Master password in app code instead of Secrets Manager → HIGH.
• Secrets Manager rotation: not enabled → MEDIUM.
• Audit logging: engine audit logs not in EnabledCloudwatchLogsExports → MEDIUM (HIGH for PCI/HIPAA scope).
• CMK rotation: KMS key without automatic rotation → LOW.

9.2 Reliability

Ref: High Availability for Amazon RDS

• Multi-AZ: production single-AZ → HIGH (Aurora: ≥1 reader in a different AZ).
• Subnet group AZ spread: subnets span <2 AZs → HIGH.
• Backups: BackupRetentionPeriod=0 → CRITICAL. <7 days for production → HIGH.
• Deletion protection: production with DeletionProtection=false → HIGH.
• Auto minor version upgrade: AutoMinorVersionUpgrade=false for non-prod → MEDIUM.
• Custom parameter group: instance using default.<engine> parameter group → MEDIUM (can't tune).
• Pending maintenance: required actions not applied within window → MEDIUM, escalating with age.
• Replication health (read replicas): ReplicaLag 7-day max > 30s → HIGH; > 300s → CRITICAL.
• Aurora replicas: clusters with no reader → MEDIUM; single-AZ readers → MEDIUM.
• DR: cross-region read replica or Aurora Global Database absent for tier-1 workloads → MEDIUM.

9.3 Performance

Ref: DB Instance Performance

7-day metric thresholds (full table in references/metrics-thresholds.md):

• CPUUtilization avg > 70% → MEDIUM, > 90% → HIGH.
• FreeStorageSpace < 20% allocated → HIGH, < 10% → CRITICAL.
• FreeableMemory < 10% instance class memory → HIGH; sustained SwapUsage > 0 → MEDIUM.
• DatabaseConnections max > 80% of max_connections parameter → HIGH.
• ReadLatency / WriteLatency avg > 20ms (10ms for io1/io2) → MEDIUM.
• BufferCacheHitRatio (Aurora) < 95% → MEDIUM, < 90% → HIGH.
• BinLogDiskUsage growth without retention bounds → MEDIUM.
• Performance Insights disabled → MEDIUM (free 7-day tier should always be on).
• Enhanced Monitoring disabled or interval > 60s for production → MEDIUM.
• Connection pooling: high connection churn / no RDS Proxy → MEDIUM for serverless / many-client workloads.

9.4 Cost Optimization

Ref: Cost-Optimized Architectures

• Stopped instances: DBInstanceStatus=stopped (still pays storage) → HIGH.
• Previous-generation classes: db.m4 / db.r4 / db.t2 → MEDIUM.
• gp2 → gp3: any StorageType=gp2 → MEDIUM (≈20% cheaper, equal/better performance).
• Over-provisioned IOPS: io1/io2 with avg IOPS used / IOPS provisioned < 50% → MEDIUM.
• Reserved Instance coverage: production On-Demand without RIs → MEDIUM, with estimated savings.
• Storage autoscaling: MaxAllocatedStorage not set → LOW, increases stockout risk.
• Idle databases: DatabaseConnections 7-day max < 5 + CPUUtilization 7-day avg < 5% → MEDIUM (candidate to stop / delete).
• Manual snapshots: many > 90 days old → LOW (review retention).
• Cost-allocation tags: missing Environment, Owner, CostCenter → LOW.
• Graviton migration: x86 (db.r5, db.m5, db.t3) where Graviton is supported → MEDIUM (~20% saving).

9.5 Operational Excellence

Ref: Monitoring Amazon RDS

• CloudWatch alarms: missing alarms on CPUUtilization, FreeStorageSpace, DatabaseConnections, FreeableMemory, ReplicaLag → MEDIUM each.
• Event subscriptions: no rds.DescribeEventSubscriptions covering this resource → MEDIUM.
• Engine version currency: minor version not latest available → LOW; major version EOL/within 6 months → HIGH.
• Log exports: engine error / slow / audit logs not exported to CloudWatch → MEDIUM.
• Maintenance window: not configured / overlaps business hours → LOW.
• Tagging: missing operational tags (Environment, Owner, Runbook, OnCall) → LOW.
• Drift: parameters changed in default parameter group (impossible by design — use as a flag for non-default usage check).

Step 10: Generate Report

Generate a separate shareable report artifact for each resource reviewed.

Artifact naming: rds-review-<resource-name>-<YYYY-MM-DD>.md Example: rds-review-prod-aurora-2026-04-29.md

For each resource, create the artifact as a Markdown document with:

Report Header

# RDS / Aurora Operational Review — <resource-name>
Account: <account-id> | Region: <region> | Date: <YYYY-MM-DD>
Engine: <engine> <engineVersion> | Class: <dbInstanceClass / serverlessV2 ACU> | Multi-AZ: <yes/no>

Executive Summary

• Health: ✅ HEALTHY / ⚠️ WARNINGS / ❌ CRITICAL
• Finding counts by severity
• Top 3 critical/high items

Configuration Snapshot

| Item | Value | | Engine / version | … | | Storage | type, allocated, max, IOPS, throughput, encrypted (KMS) | | Network | VPC, subnet group, AZs, public access, security groups | | Backup | retention, window, automated, deletion protection | | HA / DR | Multi-AZ, replicas, Global DB | | Auth | IAM auth, Secrets Manager, master user | | Observability | Performance Insights, Enhanced Monitoring, log exports |

Findings by Pillar

For each of Security, Reliability, Performance, Cost, Operational Excellence:

| # | Finding | Severity | Current State | Recommendation |

CloudWatch Metrics (7-Day)

| Metric | Stat | 7-Day Avg | 7-Day Max / Min | Status | Finding |

Log Pattern Analysis (7-Day)

| Pattern | Occurrences | Severity | Finding |

RDS Events (14-Day)

Notable events (failovers, restarts, low-storage, OOM, maintenance) with timestamps.

Pending Maintenance

List from rds.DescribePendingMaintenanceActions with action type, target window, age.

Engine Version

Current vs latest minor / latest major. Flag EOL.

Cost Summary

• Latest-month estimated cost for this resource (proportional split, with "estimated" badge)
• RI coverage status
• Top 3 cost-optimization opportunities (linked to findings)

Priority Matrix

| # | Finding | Severity | Pillar | Effort | Impact |

Next Steps

• Immediate (CRITICAL/HIGH — 7 days)
• Short-term (MEDIUM — 30 days)
• Long-term (LOW — 90 days)

Appendix — Reference Links

• Amazon RDS Best Practices
• Amazon Aurora Best Practices
• Monitoring RDS
• Performance Insights
• Encrypting RDS Resources
• Multi-AZ DB Instance Deployments
• Aurora Global Database
• RDS Proxy
• Well-Architected Framework

Severity Definitions

Severity	Definition	SLA
CRITICAL	Immediate risk to availability, security, or data integrity	Fix within 24–48 hours
HIGH	Significant gap that could lead to incidents	Fix within 1 week
MEDIUM	Notable improvement opportunity	Plan within 30 days
LOW	Minor optimization or hardening	Address when convenient
INFO	Observation, no action required	N/A

Engine-Specific Considerations

• MySQL / MariaDB — innodb_buffer_pool_size, max_connections, slow query log, binlog retention, deprecated query_cache_type on 8.0+.
• PostgreSQL — shared_buffers, work_mem, autovacuum, pg_stat_statements, log_min_duration_statement, replication slots.
• Oracle — BYOL tracking, SGA/PGA sizing, AWR, tablespace autoextend.
• SQL Server — tempdb file count (1 per vCPU up to 8), MAXDOP, cost threshold for parallelism, Always On AGs vs Multi-AZ.
• Aurora — Serverless v2 ACU floor/ceiling, Global Database for DR, fast clones, Backtrack (MySQL only), cluster cache management.

Known API Quirks (recorded so the agent doesn't trip on them)

• rds.ListTagsForResource — parameter is ResourceName, value is the resource ARN.
• rds.DescribeEvents — Duration is in minutes (20160 = 14 days).
• applicationautoscaling.DescribeScalableTargets — ServiceNamespace="rds" covers Aurora replica autoscaling.
• logs.DescribeLogGroups prefix differs:
  • Instances: /aws/rds/instance/<id>/
  • Aurora clusters: /aws/rds/cluster/<id>/
• Some Aurora metrics are cluster-only (AuroraReplicaLag, BufferCacheHitRatio, VolumeReadIOPs), others are instance-only (FreeableMemory, CPUUtilization per member). Query the right dimension or both.
• rds.DescribeDBSnapshots requires pagination for accounts with many snapshots — use MaxRecords and Marker.

Data Source Boundaries

This skill explicitly does not call:

• Kubernetes API / k8s MCP — RDS is fully managed; no cluster API exists.
• eks MCP — outside scope.
• Dante or any non-AWS scripts — keep the skill self-contained on the AWS DevOps Agent's primary cloud-source IAM role.

If a customer needs deeper SQL-level analysis (e.g. Performance Insights db.load breakdowns, top SQL by wait event), call pi.GetResourceMetrics and pi.DescribeDimensionKeys directly — both are in the same rds permission family when Performance Insights is enabled.