Service Level AOI

Service Level [AOI] - DRAFT

Service Level Objectives

External

http://eapd.cms.gov/ is available 99.9% of the year ( all but ~8 hours) a. 2 hours for major infrastructure upgrades that are cost-prohibitive to blue green b. 6 hours per year for AWS outages and other management reserve c. Site fully loads in < 5 seconds

Service Level Indicators

The primary goal here is to define indicators (metrics) that are useful and actionable. MACPro eAPD’s mission is to support the states creation, submission, and response process for APDs and contracts.

In order to support the states, the system needs to be available and responding in a reasonable amount of time. As such, the most important primary indicators will be the ratio of successful responses and the latency in those responses. All other indicators can help indicate or predict other deeper issues, but do not directly indicate the ability to serve information.

As such, monitoring efforts should focus on:

1. Availability Monitoring
2. Rate of HTTP 400/500 responses from the ELB
3. Duration of HTTP 200 responses ELB Targets

Indicators that are generally NOT useful for availability monitoring, but may help with troubleshooting, proper provisioning and configuration changes:

1. CPU usage
2. Memory Usage

(High levels of CPU and memory usages generally indicate proper provisioning of resources and therefore should be carefully considering when investigating a reported issue)

Availability

Endpoint	Service	Status	Alert
https://eapd.cms.gov/	Primary	Not Available	Alert

Health

Service	Metric	Stat	Period	Threshold	# of Datapoints	Actions	Notes
Application ELB	HTTPCode_Target_5XX_Count	Average	5 minutes	> 200	2	Alert
	HTTPCode_Target_5XX_Count	Average	5 minutes	> 50	5	Warn
	TargetResponseTime	p95	5 minutes	> 2.5 seconds	2	Alert
	TargetResponseTime	p95	5 minutes	> 0.5 seconds	5	Warn
Instances	HealthyHostCount	Average	5	1	1	Alert
	CPUUtilization	n/a	n/a	< 20%	n/a	Alert
RDS	CPUUtilization	Average	5	> 50%	2	Warn
Lambdas	Errors	n/a	n/a	> 0	n/a	Alert

• Alerts would trigger an alerting mechanism for investigation by on-call
• Warning would generate a report that should be investigated