Observability
lastmod 2025-12-02created 2025-12-021 edits
In control theory, observability is a term describing whether information about the state variables of a system can be determined from its output variables. When information about a particular state variable can be determined from the system's output variables, that state variable is said to be observable, and when all state variables of the system are observable, the system is said to be observable.
Categories
Observability in computing
Observability is the ability to understand the internal state or condition of a complex system based only on knowledge of its external outputs, specifically its telemetry.
Observability plays an important role in maintaining the availability, performance, and security of modern software systems and cloud computing environments.
The term "observability" comes from control theory, an engineering theory concerned with the automated control of dynamic systems, such as regulating the flow of water through a pipe based on feedback from a flow control system.
Observability provides deep visibility into modern distributed applications, supporting automated, real-time problem identification and resolution. The more observable a system is, the faster and more accurately IT teams can move from an identified performance problem to its root cause without additional testing or coding.
Achieving observability in IT operations (ITOps) and cloud computing requires software tools that aggregate, correlate, and analyze a continuous stream of performance data from applications and the hardware and networks they run on. Teams can then use the data to monitor, troubleshoot, and debug apps and networks, and ultimately optimize the customer experience and meet service level agreements (SLAs) and other business requirements.
Observability is often confused with application performance monitoring and network performance management (NPM). However, observability tools are a natural evolution of application performance monitoring and NPM data collection methods. These solutions are better suited to addressing the increasingly distributed and dynamic nature of cloud-native application deployments.
Observability does not replace other monitoring approaches; it improves and extends them.
The three pillars of observability
Observability platforms continuously discover and collect performance telemetry by integrating with instrumentation built into app and infrastructure components and by adding capabilities and instrumentation to these components.
Observability focuses on three main types of telemetry.
- Logs
- Logs are granular, timestamped, complete, and immutable records of application events. Among other things, logs can be used to create high-fidelity, millisecond-by-millisecond records of every event, including surrounding context. Developers use logs for troubleshooting and debugging.
- Traces
- Traces record the end-to-end "journey" of every user request, from the user interface or mobile app through the entire architecture and back to the user.
- Metrics
- Metrics (also called time-series metrics) are fundamental measures of application and system health over time. For example, metrics are used to measure how much memory or CPU capacity an application uses over five minutes, or how much latency an application experiences during a usage spike.
Observability tools also generate dependency maps that show how each application component depends on other components, applications, and IT resources.
After collecting telemetry, the platform correlates the data in real time, giving DevOps teams, site reliability engineering (SRE) teams, and IT staff complete contextual information. Teams learn the "what, where, and why" of any event that might indicate, cause, or resolve an application performance problem.
Many observability platforms also automatically discover new sources of telemetry as they appear in the system (for example, when a new application programming interface (API) is added to the network). Leading platforms include artificial intelligence for operations (AIOps) capabilities that can separate signals indicating real problems from "noise," data unrelated to current or potential problems.