Glossary
Service level indicator (SLI)
An SLI (service level indicator) is the measurement a reliability target is expressed in, usually the ratio of good events to valid events.
Also called SLI.
A service level indicator is the measurement itself, and the useful ones are almost always a ratio: good events divided by valid events. "Requests answered without a server error, over all requests that were not client errors" is an SLI. Out of 1,000,000 requests, with 20,000 client errors excluded as not valid and 980 server errors counted as bad, the SLI is 979,020 / 980,000 = 99.9%. It is deliberately mechanical: anyone reading the definition would compute the same number from the same data.
SLI stands for service level indicator, and it is the first of three terms that are easy to blur. The SLI is the measurement: 99.9% of valid requests succeeded over the last 28 days. The SLO is the target the team sets on it, such as 99.5% over 28 days. The SLA is the contract with customers, often looser again at 99%, with service credits if it is missed. The indicator is the only one of the three that is a fact rather than a promise.
Choosing one is harder than it looks, because the definition decides what you will and will not notice. An SLI measured at the load balancer misses everything that never reached it — a DNS failure, an expired certificate, a routing problem — which are precisely the outages that affect all customers at once. An SLI measured from outside, by a monitor, catches those and adds its own noise from the monitor's network instead.
The best indicators track something a user would recognise as working: a request answered correctly and quickly, a job that finished, a page that loaded. Indicators that track internal health — CPU, queue depth, replica lag — are worth alerting on, but they belong in diagnosis rather than in the objective, because a service can be perfectly healthy by all of them and still be broken for its users.
| SLI | Good event | Measured at |
|---|---|---|
| Availability | A request answered without a server error | The load balancer, or a check from outside |
| Latency | A request answered faster than a threshold, such as 300ms | The server, or a check from outside |
| Correctness | A response with the right content, not just a 200 | A check that asserts on the response body |
| Freshness | Data updated within its expected age | The pipeline that writes it |
| Job success | A scheduled job that ran and finished | A heartbeat the job sends when it succeeds |
Questions
What does SLI stand for?
SLI stands for service level indicator: the number a reliability target is measured with, usually the share of events that went well, such as the percentage of requests answered without an error.
What is an example of an SLI?
The proportion of HTTP requests answered without a server error, leaving out requests the client got wrong. Latency is the other common one: the proportion of requests answered in under 300 milliseconds. Both are ratios of good events to valid events.
What is the difference between an SLI, an SLO and an SLA?
The SLI is the measurement, the SLO is the target the team sets on it, and the SLA is the contract with customers, usually looser than the SLO and backed by service credits. For example: an SLI of 99.93%, against an SLO of 99.9%, under an SLA of 99.5%.
How do you calculate an SLI?
Count the good events, count the valid events, and divide. With 980,000 valid requests of which 979,020 succeeded, the SLI is 979,020 / 980,000 = 99.9%. Deciding what counts as valid, such as leaving out client errors, is the part that needs writing down.
Should an SLI be measured inside or outside the service?
Both have blind spots. Measuring on the server sees every request that arrived but none that never did, such as one lost to a DNS failure or an expired certificate. A check from outside catches those but adds noise from its own network. Many teams set the SLO on server data and use outside checks as a cross-check.
Uptimely checks your services from outside and records whether each check passed and how fast it answered, so an availability SLI comes from real measurements.
Start monitoring free →Related terms
Part of the Uptimely monitoring glossary — plain definitions of the vocabulary uptime monitoring, incident response and status pages are discussed in. Something here read wrong to you? Email support@getuptimely.com.