DevOps & Cloud · Kubernetes right-sizing
PodRight
Pay for the Kubernetes you use.
Two weeks of real usage turned into right-sized CPU and memory for every workload, the monthly saving in dollars, the pods at risk, and the exact commands to fix them.
$ k8s-rightsize report --prometheus http://prometheus:9090
8 containers checked; requests could drop by about $673 a month; 2 are at risk
shop/api/api x6 saves $286/month
cpu request 2 -> 360m (p95 310m) memory 4Gi -> 736Mi
shop/worker/worker x4
RS003 memory peaked at 992Mi, 97% of its 1Gi limit: close to OOM kills
The problem
Most clusters request two to five times the CPU and memory they use, and the cloud bill charges for every requested core. Others are the opposite: memory at the limit, OOM kills at night, CPU throttled into latency. Nobody has time to size 200 containers by hand.
What it does
Reads real usage from Prometheus (cAdvisor and kube-state-metrics) or a CSV from Datadog, New Relic or CloudWatch
Right-sizes each container from its 95th-percentile CPU and 99th-percentile memory, with headroom you choose
Puts a monthly dollar figure on every over-requested workload, at your own node prices
Flags the risks next to the savings: OOM kills coming, CPU throttling, eviction, pods with no requests or limits
Writes the fixes: kubectl commands (including removing CPU limits) and strategic-merge patches for Kustomize or Helm
Groups pods into Deployments, StatefulSets and DaemonSets and sizes for the busiest replica
Read-only: no agent, nothing changed in the cluster
How it works
- Point it at Prometheus or export two weeks of usage
- Review the savings and risks with the team
- Apply the generated changes in staging, then production
Works with
Runs in your CI
# Monthly right-sizing review
k8s-rightsize report --prometheus $PROM --format md --out rightsize.md --kubectl fixes.sh
Your data
- Reads metrics only; never changes the cluster.
- No agent to install.
- Runs on your laptop or in CI.
Who it's for
Questions
Will smaller requests cause outages?
Requests follow real peaks plus headroom, memory limits track the peak, and risky workloads are flagged separately. Roll out to staging first and watch for a week.
Why remove CPU limits?
CPU limits throttle a container even when the node has spare CPU; the request already guarantees its share. You can keep limits with one option if your policy requires them.
Related
Often used with PodRight
BlastRadar
Change impact analysisKnow what a change will touch. Before a release, see every service a change can affect and the line of code that links it. After the release, confirm those services are still healthy.
$300–600 (₹10,000–20,000 in India)View product → DevOps & CloudSLOStudio
SLOs and error budgetsReliability you can measure. Define each SLO in a few lines and get correct burn-rate alerts, an error budget everyone understands, and a release gate, on the monitoring you already run.
$1,500–3,000 (₹40,000–80,000 in India)View product → DevOps & CloudTestScope
Test impact selectionRun the tests that matter. Every pull request runs the tests its change can affect, each with a stated reason, and the full suite whenever a change is too broad to trust.
$1,000–2,500 (₹25,000–60,000 in India)View product →See PodRight on your own data
A 30-minute call. We run the product on your export, repository or document and walk you through the results.