Before DevOps was involved our streaming platform had no complete visibility into what was happening across live and on-demand video workflows, and the result was predictable: errors were caught late, incidents escalated before anyone noticed, and the commercial monitoring tools the industry recommended cost more than the team could justify.

In this talk we'll see how a DevOps engineer approached that problem with open source tooling and what we learned along the way, including what went wrong.

We'll cover:

• What makes streaming infrastructure observability different from a typical web application, from player metrics, origin health, CDN behavior, encoding pipelines, to real-time stream health where all need to be correlated simultaneously

• The challenge we didn't see coming: metrics in streaming are not universal. A value that signals a critical error on one stream configuration is completely normal on another  and getting that wrong nearly derailed the whole project, but forced us to build a more flexible and configurable solution

• How we designed and deployed a monitoring stack using Grafana, Prometheus, Kubernetes, Helm, ArgoCD and extended with video-specific tooling like FFmpeg and TSDuck to capture stream health signals at the transport layer

• The concrete outcomes: catching errors earlier, improving real-time visibility into live streams, and replacing expensive commercial alternatives with a maintainable open source solution

• What any DevOps engineer can take from this approach, regardless of their  industry

This is not a theoretical talk. Every decision was driven by a real operational problem, and every outcome, including our mistakes, will be shared honestly.