Observability (logs/metrics/traces) as a component of information system design: SLI/SLO and MTTR reduction

Abstract

The article examines the observability of distributed information systems as a designed property that ensures the manageability of complexity in microservice, event-driven, and hybrid architectures through end-to-end telemetry (logs, metrics, distributed traces) and the formalization of reliability via SLI/SLO and the error budget. It is shown that isolated or late adoption of telemetry tools creates blind spots, shifting incident diagnostics into a stochastic mode and worsening the alignment of operational characteristics with real user experience, especially in cloud environments with ephemeral resources, where infrastructure metrics lose explanatory power. Based on a dataset of more than 12,000 incidents from 148 industrial systems, a quantitative relationship is demonstrated between observability maturity and reduced recovery time: increased tracing coverage and a transition from static thresholds to dynamic SLO-based alerting are accompanied by a multiple reduction in MTTR and the share of false positives, as well as decreased recovery variability. The synergistic effect of the logs/metrics/traces triad is confirmed: a full stack sharply reduces detection time and increases the proportion of incidents resolved within SLA. Additionally, an economic optimum of data granularity is identified: adaptive tail-based sampling provides a better return on investment compared to a simple increase in the volume of stored traces, reflecting the law of diminishing returns and the need for semantic signal filtering.