BETA You’re responsible for monitoring and scaling your self-hosted 3B deployment. Use this guide at your own risk and test changes before applying them in production.
For three default workers on one K3s node, start with 16 vCPUs and 32 GiB of RAM. With monitoring and KEDA, total requests exceed 8 vCPUs even with one Edge pod. With two Edge pods, memory requests reach about 25 GiB during an upgrade.
How much capacity should I add?
These estimates assume worker.cpus: 1 (one execution slot per worker) and include disk for builds and caches. If two workers already fit, use the first row to budget for a third. For larger workers, see the Helm values reference.
Pods to add | Extra CPU | Extra RAM | Extra local disk |
One worker | 1 vCPU | 4 GiB | 52 GiB |
Two workers | 2 vCPUs | 8 GiB | 104 GiB |
Two workers and one Edge pod | 2.3 vCPUs | 8.5 GiB | 104 GiB |
To go from one worker and one Edge pod to three workers and two Edge pods, budget 3 extra vCPUs, 10 GiB of RAM, and 104 GiB of local disk. This includes room for a replacement orchestrator during upgrades.
Check whether the extra pods will fit
Run these commands on the K3s node between upgrades. Use sudo k3s kubectl if your account needs elevated access to the kubeconfig.
export KUBECONFIG=/etc/rancher/k3s/k3s.yaml kubectl -n 3b get deployment worker edge kubectl describe nodes
Read READY for current pod counts. In the node output, subtract Requests under Allocated resources from Allocatable for both CPU and memory. For example, 16 allocatable vCPUs minus 13 requested leaves 3 for new pods. If you customized pod sizes, use their actual requests.
Check usage during a busy workflow.
kubectl top nodes df -h /var/lib/kubelet/pods
Add RAM if the node already uses most of it at peak. Under Avail, allow 52 GiB per new worker plus space for other workloads on that disk. Use sudo df if the directory is restricted.
Once the extra pods fit, follow Enable 3B autoscaling on K3s to set the worker limit to three.
