A Service Level Indicator (SLI) in SRE is a carefully chosen metric that measures how well a service is actually performing from a user’s point of view.
In simple terms, an SLI answers this question:
👉 “Is the system working the way users expect it to?”
It is a core concept in Site Reliability Engineering (SRE) because it turns raw system data into meaningful reliability signals.
1. What is an SLI in SRE?
Service Level Indicator (SLI) is typically a ratio or percentage based on real user experience, not just internal system health.
Instead of measuring everything, SLIs focus on what matters most to users.
Example:
- % of successful requests
- Response time under a threshold
- Uptime availability
SLIs are used to define and measure SLOs (Service Level Objectives) and SLAs.
2. Common types of SLIs
1. Availability (uptime)
Measures whether the service is reachable and working.
Example:
- “99.9% of requests succeed”
This is one of the most common SLIs.
2. Latency (response time)
Measures how fast the system responds.
Example:
- “95% of requests complete in under 300ms”
Even if a system is “up,” slow performance still impacts users.
3. Error rate
Measures how often requests fail.
Example:
- HTTP 5xx errors
- Failed API calls
- Timeout errors
This directly reflects system reliability issues.
4. Throughput (traffic handling)
Measures how much load the system can handle.
Example:
- Requests per second
- Transactions per minute
This helps understand system capacity under stress.
3. Most important metrics for service performance
While all SLIs are important, the most critical ones depend on user experience.
1. Availability (most important overall)
If the service is down, nothing else matters.
A highly available system ensures users can always access it.
2. Latency (user experience quality)
Even if a system is available, slow responses create frustration.
Latency directly impacts:
- User satisfaction
- Conversion rates
- Application usability
3. Error rate (system reliability)
Errors show real failures in the system.
High error rates usually indicate:
- Bugs
- Infrastructure issues
- Dependency failures
4. Saturation (system health under load)
This is a supporting metric that shows how close the system is to breaking.
Examples:
- CPU usage
- Memory usage
- Disk I/O
4. What makes a good SLI?
A good SLI should be:
- User-focused (reflects real experience)
- Simple (easy to understand and measure)
- Actionable (helps engineers respond to issues)
- Consistent (measured the same way over time)
Which SLIs matter most?
If I had to prioritize:
- Availability – because downtime directly affects users
- Latency – because performance defines user experience
- Error rate – because it shows system failures clearly
Simple summary
An SLI is a user-focused metric that measures service performance and reliability. It helps teams understand how well a system is actually working in real conditions.