What is an SLI?
SLI, told as a story about keeping today's ride log measured in numbers, instead of by feel.
Ask how today went, and everyone answers by feel alone.So-so, pretty good, I think — it's all vague talk.
Everyone's feeling is different, so you can't compare them.You can't tell if it's better than yesterday, or close to any target.
Instead of feelings, keep a log measured in numbers.A number measured this way is called an SLI — a service level indicator.
Today: 100,000 requests, 99.2% answered within 1s, 99.95% succeeded.Measured the same way, every day, and written down as numbers.
Measure the wrong thing, and the numbers look great while guests stay unhappy.It only matters if you measure what guests actually care about.
SLI = today's log, keeping things like response time and success rate as numbers, measured the same way every single day.
A service level indicator — a numeric measure of how well a system actually performs, from the guest's point of view. Things like latency, error rate, and availability become the baseline an SLO later sets a target against.
When grown-ups say it
- SLI
- Today's log, measured in numbers. Written down as numbers, not feelings.
- Response time
- How long it takes to answer a guest's request. One of the most common SLIs.
- Success rate
- The share of attempts that finished properly. The flip side of the failure rate — same thing, viewed the other way.
- Availability
- The share of time the park stayed open. Uptime, turned into a measurable number.
- Measurement point
- Where you measure — the guest's side, or the server's. Measuring closer to the guest gets you closer to the real experience.
- vs. golden signals
- A few of the four gauges, picked out as SLIs. All four could be SLIs, but you don't have to use every one.
- SLO
- The target set using this number. You need an SLI before you can set an SLO.
- Observability
- Looking deeper than the numbers. Needed when even a great-looking log isn't the whole story. → the people who keep the park open