How Sennder reduced EKS cluster size by up to 43% without manual tuning
Customer Brief
Sennder is a leading digital freight forwarder operating a marketplace model that connects shippers with carriers. Headquartered in Berlin, Sennder serves a variety of customers across Europe, from small, family-run carriers to enterprise shippers such as Nike, P&G, and Amazon.
On the engineering side, Sennder runs 8 clusters and 150+ nodes on AWS EKS, using Karpenter and KEDA to enable highly dynamic scaling throughout the day. A Cloud Platform team supports 100+ engineers by providing shared tools and standards for deployment, automation, and observability across Kubernetes-based services.
Sennder controlled costs through commitments, but Kubernetes efficiency remained a challenge. Teams routinely over-requested CPU and memory to “stay safe,” leading to low average utilization and periodic performance issues, including CPU throttling cascades.
Sennder reduced cluster size by up to 43%, eliminated CPU throttling caused by inaccurate resource requests, removed the need to continuously manage resource tuning, and enabled an easy, safe rollout across environments through opt-in selectors.
The Challenge
Sennder runs multiple EKS clusters across separate domains such as testing, production, and third-party tools. Services are deployed via a unified Helm chart and automated CI/CD pipelines, and observability is handled through Datadog.
Sennder had been using Zesty Commitment Manager for four years to maximize commitment coverage and control costs at the cluster level, but efficient node utilization remained a challenge.
Even with observability in place, educating teams on CPU throttling, memory limits, and regular request adjustments was a painful task. The platform team tried to support developers with dashboards and repeated reminders, but it didn’t scale. Teams couldn’t keep CPU/memory requests aligned with dynamic usage over time.
The result was chronic over-scaling and low average utilization. In practice, workloads reserved up to 90% of a node’s CPU while actually using around 10%. This “requests vs. usage” gap inflated node counts.
When CPU requests were too low, or services hit node limits, Kubernetes throttling could trigger chain reactions: slower responses and failed health checks led to user-facing issues such as slow-loading operations boards, missed notifications, or blocked offers, disrupting the operations team’s daily workflow and halting logistics work.
Platform Engineering Lead at Sennder
Zesty’s Solution
Sennder explored right-sizing solutions on the market, but most of them conflicted with their KEDA-based horizontal autoscaling or were too limited to handle real usage patterns. When Zesty introduced a pod rightsizing solution designed to be compatible with KEDA and their specific setup, it finally felt like the right fit, and they decided to move forward.
Onboarding was straightforward: it was easy to integrate as a Helm chart, flexible enough to adapt to their different environments, and installed like any other EKS add-on. Within a couple of hours, it started collecting data. The solution was safe by design, as Zesty only ingests CPU and memory usage metrics and runs with no noticeable resource overhead or security concerns.
The platform team managed Pod Rightsizing centrally, guiding engineers while keeping adoption simple and controlled. They enabled it by default in non-critical environments and then gradually introduced it in production through a simple opt-in flag in their Helm chart. Engineers could enable automation very selectively, choosing exactly which workloads to right-size. This approach allowed Pod Rightsizing to be rolled out safely and incrementally at the workload level, without adding complexity.
Platform Engineering Lead at Sennder
The Result
In non-critical environments, Sennder saw 30–40% reductions in cluster size. In environments and services where Pod Rightsizing was more fully applied, reductions reached 43%, a meaningful improvement.
Once the solution was in place, Sennder no longer had to worry about constantly chasing teams to adjust resource requests. The solution simply did the work in the background.
Platform Engineering Lead at Sennder
By relying on ongoing usage monitoring rather than manual, point-in-time checks, right-sizing became more effective. With clear visibility in the UI and full workload-level controls for limits and rollout, Pod Rightsizing gave Sennder the flexibility and granularity they needed.
Zesty made optimization transparent for engineers, automatically correcting outdated or poorly configured resource values without breaking deployments or requiring code changes, freeing the team from repetitive work.
Platform Engineering Lead at Sennder
