Observability
Subject: Computer and Technology noun /əbˌzɜːr.vəˈbɪl.ə.ti/ 1 definition
- distributed-systems
- software-engineering
- site-reliability
- cloud-computing
-
Definition (noun)
Entry #252 · UpdatedOriginating in control theory and central to modern distributed systems, Observability is the degree to which a system's internal states can be inferred solely from its external outputs. Unlike traditional monitoring—which tracks predefined thresholds for known failure modes—observability relies on rich instrumentation that emits high-cardinality telemetry, including structured logs, metrics, and distributed traces. This allows engineers to query arbitrary dimensions across service boundaries and diagnose novel, unforeseen failures without deploying new code.
Common Types
Application Observability: Instruments service code and runtimes with distributed traces, structured logs, and custom metrics to expose request latency, exception paths, and dependency failures.
Infrastructure Observability: Collects host, container, and orchestrator telemetry (such as CPU saturation, memory pressure, and kernel events) to pinpoint resource bottlenecks and node degradation.
Network Observability: Captures packet flows, DNS resolution, and service-mesh traffic via mechanisms like eBPF to isolate routing faults, packet loss, and cross-zone latency.
Data Observability: Monitors data pipelines for freshness, schema drift, volume anomalies, and lineage so teams can detect broken transformations before they corrupt downstream analytics.
Example
When checkout latency spikes even though CPU and error-rate alerts remain green, an on-call engineer queries OpenTelemetry trace spans grouped by
regionandpayment_provider. Correlating the outlierPOST /v2/chargespans with structured logs and container metrics reveals TLS connection-pool exhaustion in one newly deployed EU microservice revision.