Service Level

• Define the metrics that matter
• Define the expected range for those metrics (what a healthy service looks like)
• Define the action to take when we can't provide the expected service

Service Level Agreement (SLA)

• An agreement between the service provider and the clients/users
• A legal contract that represents quality of the service. With explicit penalties and financial consequences
• E.g.,
• Availability
• Performance
• Data durability
• Time to respond to system failures
• SLAs are common for external paying users (always), free external users (sometimes) and internal users (occasionally)

Service Level Indicators (SLI)

• Quantitative measure (metrics!) of some aspect
• Data is collected over a measurement window and turned into a rate, average, percentile
• Quantitative measure of the compliance with a SLO
• The actual numbers, metrics measured by monitoring a service

• SLOs rely on the SLIs to measure if the goal is achieved or not

• E.g.

• request latency
• error rate
• system throughput
• availability/yield: well formed requested that succeeded. (99.95%: 3 nines and a half)

Service Level Objectives (SLO)

• Numeric thresholds to define acceptable levels for the SLIs
• Individual goals set for the system
• Each SLO represents a target value/range that the system needs to meet
• The quality attributes from the beginning of the design process will make their way to one or more SLOs

• If a system has an SLA, it will be detailed as multiple SLOs aggregated in a single legal document

• E.g.,

• p99 latency < 200ms for 5 min
• Availability of 3 nines
• Response time of 100 ms for 90th percentile