Application performance monitoring: definition
Application performance monitoring (APM) focuses on how software applications behave in production. Typical APM workflows examine request or transaction latency, errors, traces, service dependencies and code-level or runtime context so teams can move from a user-facing symptom toward a cause.
What APM commonly measures
- Request and transaction throughput.
- Error rates and exception details.
- Latency distributions by service, route or operation.
- Distributed traces and downstream dependencies.
- Runtime or code-level context where supported.
- Release or deployment markers that help explain changes.
Why APM is different from infrastructure monitoring
Infrastructure monitoring tells you about hosts, containers, CPU, memory, disks or cloud resources. APM starts closer to application transactions. The two views reinforce each other: a slow endpoint may be explained by a database span, which may in turn correlate with resource saturation on the database host.
Instrumentation choices
APM can use vendor agents, language SDKs, automatic instrumentation, OpenTelemetry or a combination. The collection method affects deployment effort, data depth and portability. Test the languages and frameworks that matter in your production stack rather than relying on a general compatibility list.
Useful APM alerting
Alert on conditions that represent service impact, such as a sustained error-rate or latency objective violation. Diagnostic measurements should remain available for investigation without turning every internal metric into a page.
Common APM mistakes
- Instrumenting only a demo service instead of representative production paths.
- Ignoring async jobs, queues or downstream calls that dominate latency.
- Using averages instead of distributions.
- Failing to attach deployment and ownership context.
- Collecting detailed traces without a cost or sampling strategy.
Related concepts
Short answer
APM monitors application transactions, errors, latency and dependencies so teams can diagnose software behavior. It works best when application context connects to traces, releases and infrastructure evidence.
Practical checklist
- Instrument a representative production path, not only a demo.
- Track errors and latency by service and operation.
- Connect transaction evidence to releases and dependencies.
- Model trace volume, retention and sensitive-data exposure.