An umbrella service consists of independent sub-services, this document refers to the Metrics sub-service (where the API is under CloudWatch name, other sub-services APIs are under different names)
For EC2, only hardware metrics like CPU Utilization, Disk I/O and Network I/O are collected, to collect system metrics like Disk Usage, Memory Usage, user has to install CloudWatch Agent on the system
Standard metrics collection granularity = 1 minutes; high resolution = 1 second 🎓
Alarms
Alarms are triggered by metrics and responded with an action
Multiple alarms can be combined into a composite alarm, e.g. trigger main alarm when all underlying alarms enters ALARM state
If X Datapoints within last Y Evaluation Periods breaches the Threshold, the alarm enters ALARM state
Standard alarm period length = multiple of 1 minute; high resolution alarm (must be set on high resolution metrics) = multiple of 10, 30 seconds, or 1 minute
The AWS Blog 