SCaLE 20X: Kubernetes Cloud Cost Monitoring with OpenCost & Optimization Strategies
0 likes136 views
This document outlines strategies for Kubernetes cloud cost monitoring using OpenCost, which is an open-source solution designed to help organizations manage and optimize their cloud costs effectively. It discusses the complexities of Kubernetes resource management, presents the OpenCost specification and implementation, and emphasizes the importance of financial operations (FinOps) practices within organizations to optimize cloud usage and costs. Additionally, the document details various optimization strategies for resource allocation, usage, and management to maximize efficiency in cloud environments.
SCaLE 20X: Kubernetes Cloud Cost Monitoring with OpenCost & Optimization Strategies
- 1. Kubernetes Cloud Cost Monitoring with OpenCost & Optimization Strategies Matt Ray Senior Community Manager for OpenCost [email protected] SCaLE 20x
- 2. Who am I? Matt Ray • Senior Community Manager for OpenCost at Kubecost. • Co-host of the Software Defined Talk podcast for 8+ years. • Living in Sydney, Australia for 6+ years after relocating from Austin, Texas. • Active in Open Source for much, much longer. • [email protected] • https://www.linkedin.com/in/mhray/ • @mattray on GitHub, Mastodon, too many Slacks
- 3. The complexity of operating Kubernetes efficiently is real 3 Technical COMPLEXITY Metal VMs K8s Higher-level abstractions, more shared resources, and increasingly dynamic Behavioral COMPLEXITY Decentralized releases means any engineer can increase spend quickly Abstraction Any engineer Any org Any time $ $ $ $ $ $
- 4. OpenCost Specification and Implementation Cloud Native Computing Foundation Sandbox Project FinOps Certified Solution • https://opencost.io • https://github.com/opencost • https://www.cncf.io/projects/opencost/ Open source Kubernetes cost monitoring
- 5. OpenCost Specification • Adobe • Armory • AWS • D2IQ • Google Cloud • Kubecost • Mindcurv • New Relic • Pixie • Red Hat • SUSE Created by a community of Kubernetes practitioners https://github.com/opencost/opencost/blob/develop/spec/opencost-specv01.md
- 6. OpenCost Specification https://github.com/opencost/opencost/blob/develop/spec/opencost-specv01.md Kubernetes workloads are often transient and vary in the resources they consume. How do we measure who is responsible for what and how much? • Management fees • Expenses from nodes • Persistent volumes • Attached disks • Load balancers • Network ingress/egress
- 7. Total Cluster Costs = Total Cluster Costs = Cluster Asset Costs + Cluster Overhead Costs Total Cluster Costs Cluster Asset Costs Cluster Overhead Costs
- 8. Cluster Asset Costs Cluster Asset Costs = Resource Allocation Costs + Resource Usage Costs = Total Cluster Costs Cluster Asset Costs Cluster Overhead Costs Resource Allocation Costs Resource Usage Costs Cluster Overhead Costs
- 9. Cluster Asset Costs: Node Costs Cluster Asset Costs = Resource Allocation Costs + Resource Usage Costs = Total Cluster Costs Resource Allocation Costs Resource Usage Costs Cluster Overhead Costs Network Egress Cluster Management Fees Persistent Volume Load Balancer Node (CPU, RAM, GPU)
- 10. We've got the cost of our Kubernetes assets Now let's distribute them across Workloads
- 11. Workload Costs Inside the Kubernetes Cluster What is Measured • CPU • Memory • GPU • Storage Volume • Load Balancer Aggregations • Container • Pod • Deployment • StatefulSet • Job • Controller Name • Controller Kind • Label • Annotation • Namespace • Node • Cluster
- 12. Workload Costs + Cluster Idle Costs Total Cluster Costs = Workloads + Cluster Idle Costs + Cluster Overhead Costs = Total Cluster Costs Resource Allocation Costs Resource Usage Costs Cluster Overhead Costs Total Cluster Costs Workload Costs Cluster Overhead Costs Cluster Idle Costs
- 13. Workload Costs + Cluster Idle Costs Total Cluster Costs = Workloads + Cluster Idle Costs + Cluster Overhead Costs = Total Cluster Costs Resource Allocation Costs Resource Usage Costs Cluster Overhead Costs Workload Costs Cluster Overhead Costs Cluster Idle Costs
- 14. Workload Costs + Cluster Idle Costs Total Cluster Costs = Workloads + Cluster Idle Costs + Cluster Overhead Costs = Total Cluster Costs Resource Allocation Costs Resource Usage Costs Cluster Overhead Costs Cluster Idle Costs Allocated Costs Cluster Overhead Costs Usage Costs
- 15. Cluster Idle Cost Idle costs can be calculated at the Asset/Resource and at the Workload level. Cluster Idle Cost = Cluster Asset Costs – Workload Costs Allocated Idle Total CPU Capacity
- 16. I understand what we're measuring How do I get at the metrics?
- 18. Deploying OpenCost https://www.opencost.io/docs/install Prometheus • Prom community Helm chart the default OpenCost Manifest • kubectl apply --namespace opencost -f https://raw.githubusercontent.com/opencost/opencost/develop/kubernetes/opencost.yaml OpenCost Helm Chart • https://github.com/opencost/opencost-helm-chart/ • Configurable settings (Prometheus, namespaces, etc.)
- 19. Accessing OpenCost • API • Web UI • kubectl cost • Prometheus
- 21. What's the Future of OpenCost?
- 22. What's the Future of OpenCost? What do you want it to be?
- 23. Near-Term Roadmap • External Asset Costs • Backstage integration • More Clouds • More Documentation • More Integrations
- 24. Get Involved with OpenCost https://www.opencost.io Slack • https://slack.cncf.io/ #opencost GitHub • https://github.com/opencost/opencost • https://github.com/opencost/opencost-helm-chart • https://github.com/opencost/opencost-website OpenCost Working Group • https://bit.ly/opencost-calendar • https://bit.ly/opencost-meeting LinkedIn • https://www.linkedin.com/showcase/opencost/
- 25. Kubernetes Optimization Strategies We've got the numbers, now what do we do?
- 26. FinOps https://finops.org The FinOps Foundation provides guidance on cloud financial management through best practices, education, and standards. Establish a FinOps practice within your organization.
- 27. 27 Domains Personas FinOps Practitioner Executive Business/Product Owner Finance/Procurement Engineering/Operations Principles ▸ Teams need to collaborate ▸ Everyone takes ownership for their cloud usage ▸ A centralized team drives FinOps ▸ Reports should be accessible and timely ▸ Decisions are driven by business value of cloud ▸ Take advantage of the variable cost model of the cloud Understanding Cloud Usage and Cost Performance Tracking & Benchmarking Real-Time Decision Making Cloud Rate Optimization Cloud Usage Optimization Organizational Alignment FinOps Framework Maturity Phases Rates & Usage FinOps is an evolving cloud financial management discipline and cultural practice that enables organizations to get maximum business value by helping engineering, finance & business teams to collaborate on data-driven spending decision Crawl Walk Run Slide provided by the FinOps Foundation under the CC BY 4.0
- 28. Optimization Strategy Start at the top Efficiencies and cost savings compound Coordinate your savings across the org This is an iterative process
- 29. Abandoned Workloads • Deleting abandoned pods, controllers, or even entire namespaces Right Sizing Containers • Updating pod manifests to reflect observed usage • Providing requests and possibly LimitRanges for default resource allocations • Always provide CPU requests, probably do not use CPU limits • Always use memory requests and make limits equivalent Managing Unclaimed Volumes • Delete volumes that are unused by any pods or move them to a cheaper storage tier Workload Strategies Applications Running on Kubernetes
- 30. Kubernetes Strategies Cluster Configurations Right Sizing Cluster Nodes • Adjust the number and size of your cluster's nodes to stop overspending on unused capacity • AMD CPUs may be less expensive than Intel for some workloads Underutilized Nodes • Adjust the number and type of your cluster's nodes to stop overspending on unused capacity • Check CPU, memory, storage class, and network requirements Managing Unclaimed Volumes • Delete volumes that are unused by any pods or move them to a cheaper storage tier
- 31. Operating System Optimizations Under the Kubernetes Clusters Delete Unassigned Resources • Disks and IP addresses that are not being used by any clusters may continue to incur charges Resize Local Disks • Resize local disks with low utilization Switch to Arm architecture • Arm CPUs are generally less expensive than Intel across cloud providers for similar performance
- 32. Cloud Infrastructure Optimizations Cloud FinOps Reserved Instances • Consider purchasing reserved instances based on historical resource usage patterns Spot Instances • Identify workloads ready for spot (preemptible) nodes and resize your cluster to realize the savings of migrating workloads to spot Savings Plans • Talk to your cloud vendor about all your options This is why you have a FinOps team.
- 33. Optimization Strategy Start at the top Efficiencies and cost savings compound Coordinate your savings across the org This is an iterative process