Application performance monitoring (APM) is the set of tools that records what a live app is doing, and shows engineers where it is slow or failing. APM began as a tool for one job: timing each request inside an app. Today this zone also covers the servers under the app, the business steps the app performs, and AI that investigates alerts.
To be watched, an app must first be instrumented. This means that small pieces of code are added to the app, usually from a ready-made library or a small program called an agent, installed on each server. They record every request, every error and every slow call to a database. They send these records to a central system many times a minute. Together, the records are called telemetry.
Consider an insurance company in Mumbai on 31 March. It is the last day on which customers can buy a policy and save tax for the year. At 11 p.m., customers are paying premiums at ten times the usual rate. Then payments begin to fail. Four products in this zone each answer part of the question.
The observability platform shows the app's own view. It holds metrics such as "payments per minute", the logs that each program writes, and traces of individual payments. A trace shows that each failed payment waited eight seconds for the policy service. Inside the platform, log management lets the engineer search every log line of the last hour for the word "timeout".
Infrastructure monitoring shows the machines. Its graph shows that the database server behind the policy service has been using 98% of its processor for 20 minutes.
Business observability shows the business view. It reports that 7% of premium payments were debited from customers' bank accounts, but no policy was issued. These are the customers who will call the help desk the next morning.
An AI SRE reads all of this before the engineer has opened her laptop. It suggests the cause: a sales report, scheduled for 10:45 p.m., is running heavy queries on the same database. The team stops the report, and payments recover.
The same data also shows where the machines are: in a public cloud, in the company's own data centre, or both. A company that uses both has a hybrid cloud, and its tools must see both halves. When a trace points below the app, at the network, the engineer turns to the tools in network monitoring and testing. And deciding which of a hundred alerts deserves a person at night is the problem called alert triage.