KCNA case studies
KCNA · Put your knowledge to work in context.
The Kubernetes and Cloud Native Associate (KCNA) is CNCF's entry-level, vendor-neutral certification for people who want to prove they understand Kubernetes and the surrounding cloud native ecosystem before moving on to hands-on, role-based credentials. The source describes it as a pre-professional certification with no prerequisites and a beginner experience level, aimed at students, career changers and early-career engineers. The KCNA case studies are 5 cases and about 3,976 words.
1Eviction Order on a Memory-Pressured Node
A four-node cluster runs nodes with 16 GiB RAM each; DaemonSets already request about 5 GiB per node, leaving roughly 11 GiB of allocatable memory for application pods. Three Deployments are running: web : 6 replicas, requests 2 GiB, limits 4 GiB, currently using 3.5 GiB each cache : 2 replicas, requests 3 GiB, limits 3 GiB, currently using 3.0 GiB each batch : 4 replicas, no requests and no limits, currently using 2.5 GiB each During a nightly batch spike one node reports MemoryPressure and the kubelet begins evicting pods.
816 words, about 4 min2Liveness Probe Kills a Cold-Starting Service
A payments API runs as a Deployment with 8 replicas. The pod spec has a liveness probe on GET /healthz:8080 with initialDelaySeconds: 8 , periodSeconds: 5 , failureThreshold: 3 , timeoutSeconds: 1 , and a readiness probe on the same path with initialDelaySeconds: 3 .
738 words, about 4 min3Mutable Image Tags Break the Rollout and the Rollback
A team delivers a Go service to dev, staging and prod from one pipeline. The build pushes registry.example.com/payments:v1.4 - the tag is rewritten on every merge to main; no digest is recorded anywhere.
761 words, about 4 min4Graceful Shutdown and a Disruption Budget That Blocks a Drain
A checkout service runs 12 replicas behind an ingress. Requests take up to 9 seconds.
911 words, about 5 min5An HPA That Cannot See, and Should Not Be Looking at CPU
An order-processing worker consumes from a queue. Each pod is single-threaded and spends most of its time waiting on the database, so CPU sits near 20%.
750 words, about 4 min