Monitoring and SRE glossary
Short, exact definitions of the terms that come up when you run production systems: availability targets such as SLA, SLO and error budget, incident metrics such as MTTR and MTTD, and performance terms such as Apdex, p95 latency and TTFB. Every entry has the formula, a worked example with real numbers and the common mistakes.
A
- Alert fatigue
- When so many alerts fire that people stop reading them and miss the real ones.
- Apdex
- A 0 to 1 score of user satisfaction with response times, based on a threshold T.
- APM (application performance monitoring)
- Measuring how fast and how reliably your code serves requests, down to queries and outbound calls.
D
- Distributed tracing
- Following one request across services as a tree of timed spans that share a trace ID.
- DNS propagation
- The delay before resolvers everywhere see a changed DNS record, set by the old record's TTL.
E
- Error budget
- The amount of unreliability an SLO allows: (1 − SLO) × the period.
- Escalation policy
- Rules for who is notified next when an alert is not acknowledged in time.
F
- Flapping
- A check that keeps switching between up and down, sending a stream of alerts.
H
- Health check endpoint
- A URL your app exposes so load balancers and monitors can ask whether it is working.
- Heartbeat monitoring (cron monitoring)
- Alerting when a scheduled job fails to check in on time, instead of waiting for it to report an error.
I
- Incident
- An unplanned event that disrupts or degrades a service and needs someone to respond.
M
- MTBF (mean time between failures)
- Average operating time between one failure and the next.
- MTTD (mean time to detect)
- Average time from the start of a problem until monitoring or a person notices it.
- MTTR (mean time to recovery)
- Average time from the start of an incident until service is restored.
N
- N+1 query
- One query for a list, then one more query per item: a classic ORM performance bug.
O
- Observability
- How well you can understand a system's internal state from the data it emits.
- On-call
- The rota of people responsible for responding to alerts, including outside working hours.
- OpenTelemetry
- The open standard and SDKs for producing traces, metrics and logs, sent over OTLP.
P
- p95 latency
- The response time that 95% of requests beat; the slowest 5% are slower.
- Postmortem
- A written review after an incident: what happened, why, and what will change.
R
- Real user monitoring (RUM)
- Measuring performance and errors in your real visitors' browsers.
S
- SLA (service level agreement)
- A contract that promises a level of service and says what happens if it is missed.
- SLI (service level indicator)
- The measured number behind an SLO, such as the share of successful requests.
- SLO (service level objective)
- An internal reliability target for an SLI over a time window, such as 99.9% over 28 days.
- Span
- One timed operation inside a trace, such as a query or an HTTP call.
- Status page
- A public or private page showing whether your services are up, with incidents and maintenance.
- Synthetic monitoring
- Scripted checks that act like a user on a schedule, from outside, even with no traffic.
T
- TLS certificate expiry
- The notAfter date after which browsers and clients reject your HTTPS certificate.
- TTFB (time to first byte)
- Time from starting a request until the first byte of the response arrives.
U
- Uptime
- The share of time a service is available, usually shown as a percentage.
How the terms fit together
- Targets. An SLI is what you measure, an SLO is the internal target, an SLA is the contract, and the error budget is the gap between the SLO and 100%.
- Incidents. An incident is detected (MTTD), routed by on-call and an escalation policy, fixed (MTTR) and reviewed in a postmortem.
- Performance. APM, traces and RUM explain where time goes; p95 latency, Apdex and TTFB summarise it.