Infrastructure & Operations › Observability

APM

Application performance monitoring tools.

Also known as: application performance monitoring, apm tool, apm

APM (application performance monitoring) is a class of tools that instrument an application to show how it behaves for users: how long requests take, how often they fail, which database queries or downstream calls are slow, and where time is spent. Where infrastructure monitoring answers “is the machine healthy?”, APM answers “is the application healthy, and why is this request slow?”

You install an agent or SDK in the app. It records each request as a trace, broken into spans for the work inside — the handler, the database call, the call to another service (see distributed tracing).

GET /checkout            420 ms
 ├─ auth.verify            30 ms
 ├─ db.query orders       260 ms   ← the slow part
 └─ payment.charge        120 ms

The breakdown is the point. A dashboard says “p95 latency is up”; a trace shows the one slow query responsible, so you can fix the actual cause instead of guessing.

The classic mistakes:

• Confusing APM with host monitoring. The CPU can look fine while a single endpoint is slow because of a downstream dependency. Both views are needed.
• Turning it on and ignoring the volume. Request-level data is huge and can be expensive; use sampling to keep costs and overhead reasonable (see telemetry sampling).
• Assuming zero overhead. Instrumentation costs something. Usually worth it, but measure rather than assume.
• Treating the tool as the goal. APM is one of the logs, metrics and traces pillars — it helps most when correlated with the others by time and correlation ID.

Frontend engineers use APM for page and API timing; backend engineers for request and dependency latency; data engineers for pipeline and job runtime. Modern setups often export through a common standard like OpenTelemetry so you aren’t locked into one vendor.