Observability costs tend to grow faster than traffic, and the usual response, renegotiating with the vendor or cutting retention, treats the symptom rather than the cause. The real driver is almost always a small set of design decisions: unbounded label cardinality, logs carrying work that belongs to metrics, and instrumentation added by default rather than by intent. This talk breaks down where the money actually goes in a modern telemetry pipeline, from the moment a signal is emitted to the moment someone queries it, and shows which changes cut cost without cutting visibility. Drawing on patterns common across platform teams, it covers which changes pay off, which turn out not to be worth the effort, and which trade-offs are worth accepting deliberately. It also covers the point at which sampling stops being safe.
Hanna Novikava is a Site Reliability / DevOps Engineer working on infrastructure, delivery and observability for financial platforms and beyond. Her work focuses on Kubernetes, infrastructure as code, and keeping platform teams effective as systems and their operating costs grow. She believes CI should stand for continuous improvement as much as continuous integration - pipelines are only worth as much as the practices around them. Based in Warsaw.
