Service Health
Service Health glossary
Plain-language definitions of the reliability terms used in Service Health. Follow each guide for a worked example and the app's counting rules.
| Term | Meaning |
|---|---|
| API or endpoint | An operation a service handles, usually a method and path, such as GET /orders/{id} |
| Service | An application component that handles requests or work and has an owning team |
| Reliability | How consistently the service provides the behaviour its users need |
| SRE | Site reliability engineering, an engineering approach to operating reliable services |
| SLI | Service level indicator: a measured aspect of service behaviour, such as the share of successful calls |
| SLO | Service level objective: a target for an indicator over a stated window |
| SLA | Service level agreement: an agreement with a customer that may specify consequences if a promised level is missed |
| Good call | A call meeting the particular objective being evaluated |
| Slow line | The duration beyond which a call spends speed budget |
| Server error | An HTTP 5xx response; OpenObserve also counts server spans with no HTTP status and a span status of ERROR. These spend success budget |
| Error budget | The bad-call allowance implied by an objective, such as 1% for a 99% target |
| Speed budget | The allowance for calls exceeding the speed threshold |
| Budget left | The percentage of that allowance still available, rather than the percentage of good calls |
| Burn rate | The recent bad-call share divided by the allowed bad-call share |
| Rolling window | A lookback period that moves with time, rather than resetting on a calendar boundary |
| Breached | The allowance has been exhausted |
| At risk | Less than half the allowance remains, or the current pace projects exhaustion in less than seven days |
| Burndown | The budget line showing allowance spent through the window |
| Burn spike | A concentrated period of budget spending identified from retained detail |
| Latency | How long a call takes |
| Percentile | A duration describing how much of a set of calls finishes within it |
| p50, p95, p99 | Percentiles describing the middle, 95th and 99th parts of the duration distribution; combined figures in the app are weighted summaries |
| Tail latency | The slow end of the duration distribution |
| Percentage point | An absolute difference between percentages: 99% to 98% is one percentage point |
| Tier | The service's critical, standard or best-effort category, supplying default objectives |
| Override | A deliberate value replacing an inherited setting |
| Templating | Grouping request names by replacing IDs in paths with placeholders |
| Exclusion | A rule leaving traffic outside health objectives and related actions |
| Unrouted calls | Calls with no named API route, counted separately per service |
| Dependency | Another service or external system a service calls |
| Wait | Time spent on a dependency while doing work |
| Likely cause | A suggested place to begin investigation, inferred from troubled call relationships |
Guides
Google's SRE references
The definitions above use the distinction between indicators, objectives and agreements explained in Google's Service level objectives. Service Health's particular defaults and rules are described in these guides.
For more context, start with the official SRE Workbook table of contents, especially Implementing SLOs and Alerting on SLOs.