How we Attributed Cost Per Tenant in a Kubernetes Environment
summary
Describes a working pipeline that combines cloud bills (AWS CUR, GCP billing), Apache Spark ETL into a single warehouse, Prometheus/Grafana metrics, and pod labels to charge per-pod costs so teams can query “cost of data pipeline X for customer Y on date Z.” Uses pod label customer and query duration as proxy for multi-tenant subsystems, surfaces idle utilization for optimization or allocation, and produces dashboards and raw records usable by BI tools. Implementation runs in production at a sub-40 headcount company and demonstrably answers per-tenant cost questions.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:41Why would you try to attribute cost per tenant
- 4:08Cost per tenant challenges
- 5:24Cost data - why you need to get it all to 1 place
- 8:39Working with Tenant Data
- 11:49Kubernetes Metrics
- 13:52Let's talk about an example source system
- 19:08Analyze the shared part
- 22:15Main Takeaways
related talks
26mBuilding Cost Visibility For Your Kubernetes Clusters
34mGame Over on the Bill: Playing the Cost Tetris in Kubernetes
33mSponsored Session: Introducing OpenCost
32mUse AWS Application Cost Profiler to capture SaaS cost-per-tenant metrics
48mCloud FinOps & Kubernetes Optimisation at Scale
33mUncover Kubernetes cost-optimization trends
57mKubernetes Cloud Cost Monitoring with OpenCost & Optimization Strategies
29mThe Frugal k8s Engineer: Strategies for Amazon EKS