Don't Panic — Application Performance Monitoring

Application performance monitoring, or APM, is how you follow a piece of application work through production without interrogating every server as though it were personally responsible for checkout. It watches the speed, reliability, and availability of requests, jobs, and dependencies. The goal is modest and valuable: turn “something is slow” into evidence worth testing.

The useful shape is a feedback loop. A user notices a symptom. A service signal says which operation is affected. A trace, the record of one operation across components, leads to its spans, the timed pieces of that operation. From there, you form one hypothesis, change one thing, and measure the original signal again. A dashboard is merely where the clues wait.

The mildly rude surprise is that healthy infrastructure can coexist with a broken application path. CPU may be calm while one payment call consumes most of a checkout trace. An average latency can look polite while a smaller group of requests waits far too long. That is why APM begins with operations and user-facing latency, errors, traffic, and saturation rather than a general hunt for a guilty machine.

Instrumentation produces the telemetry. Zero-code instrumentation gives broad coverage of familiar frameworks; code-based instrumentation supplies the business meaning a framework cannot guess. Context propagation carries a trace identity across services, proxies, and queues. Lose it at one boundary and the request becomes several disconnected stories, which is rather like receiving a mystery novel with the middle pages filed in another building.

The evidence comes in several forms. Metrics reveal changing rates and latency distributions. Traces show an individual path. Logs retain detailed events. Profiles explain which code consumed CPU or memory after a trace has narrowed the search. They are collaborators, not substitutes. Sampling and attribute choices decide what remains visible and what becomes expensive or sensitive, so they deserve deliberate budgets.

Read the Intro when you need the full data path and the limits of APM. Use the Slides for the relationships between signals, spans, propagation, and sampling. Keep the Cheatsheet nearby during an investigation. Then try the Exercise: it makes the feedback loop tangible by comparing a local baseline, a known scenario, and the evidence after reversal.