Skip to Content
Tuan Tran
Scheduling & Storage

Platform Operations Exercises

Model scheduling and horizontal scaling

Filter by uncommitted requests and compare scores of eligible nodes. The HPA model uses the basic ratio and replica bounds. It assumes metrics are present and omits stabilization, tolerance, readiness handling, and scaling policies.

Filter and score
  • Node A · 500m · 4096 MiB · score 90 · Eligible
  • Node B · 2000m · 2048 MiB · score 75 · Eligible
  • Node C · 1500m · 512 MiB · score 95 · Filtered out

Selected Node A

HPA replica proposal

Raw proposal 5 · bounded to 3–20 replicas: 5

Estimate a recovery timeline

Add the three phases as if they happened sequentially. Compare elapsed time with the RTO. The model omits parallel work, retries, and external dependency delays.

Sequential recovery phases

Estimated 25 minutes · 5 minutes of headroom

Calculate tenant CPU quota

This model assumes every Pod has the same total CPU request and that other workloads use none of the quota. Memory, object count, and other budgets can constrain actual admission.

CPU request quota

16 Pods fit the budget · 4 exceed it

Check Lessons 15 through 19

What does resource filtering use to evaluate CPU fit?
What is the denominator for a CPU-utilization HPA?
Does a toleration require placement on a tainted node?
Which controller can add eligible nodes for unschedulable replicas?
Can a low CPU limit raise latency on an idle node?
What does a PVC represent?
Why use WaitForFirstConsumer for zonal storage?
Can multiple Pods on one node use an RWO volume?
Does a StatefulSet configure PostgreSQL replication?
What can follow PVC deletion when the PV reclaim policy is Delete?
Does a regional control plane provide database HA by itself?
Can existing workloads continue through an API server outage?
Does disk replication prevent an accidental data deletion from being replicated?
What does a five-minute RPO measure?
Why fence a partitioned database writer before replacement?
Does a valid signature prove the image has no vulnerabilities?
What permissions should a workload creation role be reviewed for?
Does Restricted Pod Security replace RBAC?
What does an image digest provide?
How does Binary Authorization relate to Argo CD?
Does a namespace alone isolate network traffic?
What is the scope of ResourceQuota?
Does denying direct Secret reads fully bound users with workload-write access?
Does AppProject restrict direct kube-apiserver access?
When does a separate cluster make sense?

On this page