CPU-based autoscaling fails telemetry pipelines. Collectors sit at 40% CPU while queues overflow and latency spikes—yet HPA won't scale. Prometheus scrape intervals and KEDA polling add 60+ seconds of delay during traffic bursts.

This session shows how to use KEDA with OpenTelemetry metrics as scaling signals—queue depth, ingestion rate, and latency percentiles instead of CPU. We compare Prometheus-based approaches with patterns that connect scaling directly to telemetry in the OTel Collector pipeline.

You'll see reference architectures for scaling OTel Collectors: which signals work, how to set thresholds aligned to SLOs, and how to avoid false scale events and cardinality pitfalls. Attendees leave with a decision framework, working ScaledObject configurations, and production-tested patterns.