Amazon CloudWatch

Definition

Amazon CloudWatch is AWS’s unified monitoring and observability service. It collects and visualizes real-time metrics, logs, and traces from AWS services and your own applications — then enables you to set alarms, trigger automated actions, and investigate operational issues. CloudWatch is the foundation for operational excellence on AWS.

Core Capabilities

Metrics

CloudWatch automatically collects metrics from 70+ AWS services:

• EC2: CPU utilization, network in/out, disk I/O
• RDS: database connections, read/write IOPS, freeable memory
• Lambda: invocations, errors, duration, throttles
• Custom metrics: publish application-level metrics (business KPIs, response times) via the CloudWatch API or Embedded Metric Format (EMF)

Metrics are stored for 15 months. Standard resolution is 1-minute granularity; high-resolution is 1-second.

Logs

CloudWatch Logs ingests log streams from any source:

• Lambda function logs (automatic)
• EC2 application logs (via CloudWatch Agent)
• API Gateway access logs
• RDS audit logs, PostgreSQL/MySQL slow query logs
• VPC Flow Logs

CloudWatch Logs Insights provides interactive SQL-like queries against log data — essential for debugging production issues.

Alarms

Set thresholds on any metric and trigger actions:

• Send notification via SNS (email, Slack, PagerDuty)
• Execute Auto Scaling policy
• Trigger Lambda remediation function
• Create composite alarms (multiple conditions, single alert)

Dashboards

Custom visualizations combining metrics from multiple services and accounts. Share dashboards across teams.

CloudWatch Container Insights

Specialized monitoring for containerized workloads (ECS, EKS, Kubernetes):

• Automatic collection of CPU, memory, network, storage metrics per container/pod
• Flame graphs and resource utilization breakdowns
• Enhanced observability mode (EKS): node-level, pod-level, and container-level metrics
• No code changes — deployed via DaemonSet (EKS) or task definition (ECS)

Application Signals (New 2024/2025)

Application Signals is CloudWatch’s application performance monitoring (APM) capability:

• Auto-instrument applications (Java, Python, Node.js, .NET) via CloudWatch Agent
• Collect RED metrics (Rate, Errors, Duration) per service and API endpoint
• Service map showing dependencies and latency between services
• Define SLOs and track error budget burn rates
• No third-party APM tool required for standard AWS workloads

CloudWatch vs Third-Party Tools

Use CloudWatch as the baseline for all AWS monitoring. Add third-party tools for cross-cloud visibility, advanced UX, or compliance reporting requirements.

Common Mistakes

Mistake 1: Relying on default metrics without enabling detailed monitoring. EC2 default metrics are 5-minute resolution; enable detailed monitoring (1-minute) for production instances where rapid response matters.

Mistake 2: Storing logs indefinitely. CloudWatch Logs charges for storage; set retention policies (7, 30, 90 days) to control costs. Archive older logs to S3 for long-term retention.

Mistake 3: Building alarms without runbooks. An alarm that fires without a documented response procedure creates alert fatigue. Link every alarm to a runbook describing how to investigate and resolve.

Related AWS Services

• AWS CloudTrail: Audit log of API calls (who did what) — complements CloudWatch (what is happening)
• AWS X-Ray: Distributed tracing for request flows across microservices
• Amazon EventBridge: Trigger responses to CloudWatch alarm state changes
• AWS Systems Manager: Automated remediation triggered by CloudWatch alarms

Related FactualMinds Content

• AWS Cloud Security Services
• AWS Managed Services
• AWS Well-Architected Review Checklist