What Is Application Performance Monitoring (APM)?

Last verified on Nov 8, 2026.

Application performance monitoring (APM) measures how fast and how reliably an application serves requests, and points to the code or dependency at fault when it does not. The term covers both the practice and the tools. Microsoft's documentation, for example, describes Azure Monitor Application Insights as an APM feature of Azure Monitor.

What APM Tools Typically Cover

Vendors differ, but product documentation shows a common set. Application Insights lists:

• A dashboard and an application map, giving an overview of health and a visual of how components interact.
• Live metrics, a real-time view of activity and performance.
• Transaction search, to trace and diagnose individual transactions.
• Failures and performance views, to find errors and bottlenecks.
• A profiler, to capture performance traces of application code.
• Alerts, to notify you when something crosses a threshold.

AWS offers a similar shape through CloudWatch Application Signals, which discovers your services, clients, and dependencies, shows them as a list or a visual map, reports health based on SLOs, and lets you drill into correlated X-Ray traces. These are product examples, not a standard feature list.

What You Watch

Google's SRE book names four golden signals for a user-facing system: latency, traffic, errors, and saturation. Many APM views are variations on them. A typical question is "which endpoint's latency rose, and which downstream call did the time go to?" See golden signals and latency percentiles, and for the slowest requests the guide on tail latency.

An example of what a trace-based view lets you conclude, with made-up numbers:

POST /checkout            1840 ms
  auth.verify               22 ms
  cart.load                 61 ms
  inventory.reserve       1610 ms   <- 87% of the request
    db.query (SELECT ...)  1580 ms
  payment.charge            140 ms

The request was slow because one database query inside inventory.reserve was slow. Without a trace you would see only the 1840 ms total.

How the Data Gets There

An APM product needs the application to emit telemetry. That is instrumentation: libraries or agents that record requests, calls, and errors. OpenTelemetry is the vendor-neutral standard. Its documentation describes it as an open source framework for generating, exporting, and collecting traces, metrics, and logs, and says plainly that it is not a backend. Application Insights accepts OpenTelemetry, so the same instrumentation is not tied to one product. See the OpenTelemetry integration and the traces integration, and the entries on OpenTelemetry and distributed tracing.

APM and Observability

OpenTelemetry defines observability as the ability to understand a system from the outside by asking questions about it, including questions about novel problems, and says that requires properly instrumented code. APM is one way to put that into practice, aimed at application speed and errors. What it cannot see is what you did not instrument, and what happens outside the application, such as in a user's browser. For the browser side, see real user monitoring, and for scripted external checks, synthetic monitoring.

Alerting on What APM Shows

APM data is most useful when it feeds a small number of alerts tied to user-visible symptoms. Page on error rate and latency against an SLO rather than on every resource gauge, and read the entry on alert fatigue before you add more. When a slowdown appears gradually, the guide to latency drift covers it.

APM Data in Sazabi

Sazabi is an AI-native observability and security platform. Sazabi ingests OpenTelemetry metrics (Prometheus metrics through an OpenTelemetry Collector) and OTLP traces through the same intake as your logs. It stores each metric data point and each span as a log record that its agent queries alongside your logs, so there is no separate metrics store to run. The setup pages cover how to send traces and send metrics, and the reliability issues page shows what Sazabi reports from your telemetry.