Atlassian Rebuilds Metrics Pipeline on OpenTelemetry Across 100,000 Hosts

Atlassian detailed its migration from gostatsd to OpenTelemetry for a metrics platform ingesting data from about 100,000 hosts across 14 regions at a 99.95% SLO. The company kept the StatsD-over-UDP interface and rebuilt the pipeline: aggregation cut data volume from 4.8 billion to 220 million points per minute (~96%), halving CPU usage.

• Platform ingests 4.8B points per minute, stores 220M — a ~96% reduction
• Folding metrics into the tracing sidecar saved an average 3.9% CPU
• Sidecar cost cut by roughly 30% across the fleet
• gostatsd and nomad account for ~38% of CPU requests in metrics clusters