AWS CloudWatch

Overview

• Monitoring system
• Supports Dashboards

PutMetric

• Timestamp
  • (-2 weeks, +2 hours)
• Max size: GET = 8kB, POST = 40kB
• Required
  • Namespace, List<MetricDatum>
  • Delay
    • 2 minutes before get-metric-statistics returns it
      • Data timestamped < 24h may take up to 48h
    • 15 minutes before list-metrics returns it
• Number of data points
  • Single 
  • Multiple
    • Max 20 per request
  • Statistics Set (Min, Max, Sample Count,Sum)
    • Average - automatically calculated (Sum/Sample Count)
    • Percentiles need raw data
      • Exception: Sample Count == 0 OR Min=Max (but this scenario is useless)
  • None
    • Recommended "0"to avoid INSUFFICIENT_DATA but this may be expensive
      • ALARM has property how to treat Missing value

Monitoring Freqency

• Basic  (5 minutes)
  • EC2 (default)
  • SNS
• Detailed (1 minute)
  • EC2 (extra cost)
  • Auto Scaling
  • ELB
  • RDS
  • Route 53

Dimensions

• Name/Value pairs associated with metric
• Category that can be used to filter metrics
• Adding dimension creates a unique metric
  • No relationship with metrics that have different dimensions
• Examples
  • EC2 instanceId
  • ELB name

Alarm

• Can be set from CLI (simulation)
• Can be disabled from CLI
  • Alarm still changes state but no action is taken

Alarm actions

•  Types
  • Reboot
  • Recover (maintains: instanceId, private IP address)
    • Restrictions: only EBS backed instancec, newer instance types, shared tenancy
    • Max 3 attempts
  • Terminate
  • Stop
• Use cases
  • Stop impaired instance (StatusCheckFailed_System)
  • Stop instance with memory leak (custom metrics)
  • Stop idle instance (CPUUtilization)
  • Stop web servers with unusually high out traffic (NetworkOut)
  • Terminate when batch job is completed (Network Out)