Hands-on workshop: finding the money in your Kubernetes clusters
Three hours, hands-on, on a real cluster: attribute spend to namespaces, right-size from observed usage without causing OOMKills, and find the resources nobody owns.
- 24 September 2026 · 15:00–18:00 UTC
- Online
- 2 speakers
What this event covers
This workshop works through Kubernetes cost from first principles: why the scheduler allocates on requests rather than usage, why a cluster can be 25% utilised and completely unschedulable, and how to distribute node cost across namespaces in a way that gives idle capacity an owner.
Participants work on a provided cluster with realistic over-provisioning, so the exercises produce genuine findings rather than contrived ones.
The second half covers right-sizing without causing incidents: choosing percentiles, leaving headroom on memory limits, and the genuinely contested question of whether to set CPU limits at all.
What is Hands-on workshop: finding the money in your Kubernetes clusters?
Three hours, hands-on, on a real cluster: attribute spend to namespaces, right-size from observed usage without causing OOMKills, and find the resources nobody owns.
What happens, and when
Why usage graphs mislead and allocation is what determines the bill.
Distribute node cost across namespaces including the unallocated remainder.
Choosing percentiles, memory headroom, and the CPU limit debate.
Unattached volumes, idle load balancers, stale snapshots and their running cost.
Open questions plus a checklist to run against your own clusters.
Who is talking
Harshit Sengar
Founder and platform lead
DevOpsArk
DevOpsArk Engineering
Platform engineering team
DevOpsArk
Who should attend
- Platform engineers responsible for cluster capacity
- FinOps practitioners working with Kubernetes for the first time
- SREs asked to reduce infrastructure spend without reducing reliability
Event questions
No. A cluster with realistic over-provisioning is provided for the exercises. Bringing read-only access to your own is optional and useful for the final segment.
Comfort with kubectl and a working understanding of pods, deployments and namespaces. No prior FinOps knowledge is needed.
Yes, made available to registered participants afterwards, along with the exercise materials and the checklist.
Before or after the event
Related modules
Cost Management
Cloud and Kubernetes cost attribution
Kubernetes
Multi-cluster Kubernetes management
Monitoring
Infrastructure and application monitoring
Related reading
Kubernetes cost optimization: where the money actually goes
Why Kubernetes clusters cost more than they should, how to attribute spend to teams, and the specific changes that produce the largest savings.
Kubernetes monitoring: what to measure and why
A practical guide to Kubernetes monitoring: the cluster, node and workload signals that predict failure, the ones that only look useful, and how to alert on them without drowning.
Ask about the next one
Recordings, materials and the date of the next session.