Prompt
What should I use for alerting on user-impacting errors not server health?
Latest observation
Oct 25, 2026
OpenAI APIWeb search: off
For user-impacting errors (as opposed to generic server health), you usually want alerting that is based on symptoms users feel, not just CPU/memory/disk.
Good things to alert on
Use metrics like:
- Error rate: 5xx responses, failed requests, exception counts
- Latency / slowness: p95/p99 response times, timeout rates
- Availability of critical flows: login, checkout, signup, search, API actions
- Failed jobs affecting users: email sends, payments, background tasks that directly impact UX
- Synthetic checks: scripted probes that simulate real user actions
- SLO burn rate alerts: alert when you're consuming error budget too fast
Best practice
The most effective setup is usually:
- SLIs: measures of user experience
Example: “99.9% of checkout requests succeed under 500ms” - SLOs: your targets
- Burn-rate alerts: notify when the service is on track to miss its SLO
This avoids noisy alerts from normal server fluctuations and focuses on real user pain.
What not to rely on alone
- CPU/memory/disk alerts
- Generic “host down” alerts
- Raw exception counts without context
These are useful for diagnosis, but they’re not great primary alerts for user impact.
Practical rule
If you had to pick just a few alerts:
- 5xx rate above threshold
- p95 latency above threshold
- Failed critical transaction rate
- Synthetic check failure
- SLO burn rate fast/slow alerts
If you want, I can suggest a concrete alerting setup for your stack (Prometheus/Grafana, Datadog, CloudWatch, New Relic, etc.).