An observability platform is one system that collects an app's metrics, logs and traces, stores them, searches them, draws graphs from them and raises alerts. It is also called an observability tool, or an APM platform, where APM stands for application performance monitoring. Engineers use it to find out why a live app is slow or failing.
Why is one system needed? A modern app is not one program. It is usually tens or hundreds of small programs, called services, and each does one job: logins, search, payments or notifications. Each service runs on many servers, the computers in a data centre that answer requests over a network. One customer's request can pass through ten services in under a second. So when something fails, nobody can find the fault by looking at one machine.
The platform collects three kinds of data, together called telemetry. Metrics are numbers measured over time, such as requests per second, or the share of requests that failed. Logs are lines of text that a program writes as it runs, such as "seat lock failed for booking 88213". Traces follow one request through every service it touched, with the time it spent in each. The data comes from agents, which are small programs installed beside the app, or from open-source libraries such as OpenTelemetry, built into the app's code.
Take, for example, a bus ticket booking app in Coimbatore. On a Friday evening before a long weekend, customers complain that their bookings hang at the payment screen. The engineer opens the platform. The graph of bookings per minute has dropped from 900 to 300. She opens a trace of one failed booking. It shows the request passing through six services. Five of them took less than 50 milliseconds each. The seat-lock service took 9.8 seconds, because it was waiting for its database. She opens the logs of that service for the same minute. She finds thousands of lines saying that the database connection pool is full. A connection pool is a fixed number of open connections that a program keeps to its database. A setting changed that afternoon had cut the pool from 100 connections to 10. She changes it back, and bookings recover within minutes.
The platform also raises alerts. An engineer writes a rule, such as "call the on-call engineer if more than 2% of bookings fail for five minutes". The alert reaches the engineer's phone. It can also go to an AI SRE, which starts investigating at once.
Besides the app itself, the platform receives data from the systems around it. The Kubernetes platform reports on the containers in which the app's services run. The API gateway, the single entry point through which requests reach the services, reports how long each request took. Mobile app performance testing reports how fast the app feels on real phones. In the other direction, logs that matter for security also go to the SIEM, the security team's own search system.
How is it different from infrastructure monitoring? Infrastructure monitoring watches the machines: their processors, memory and disks. An observability platform watches the app's own requests. Many companies buy both from the same seller.