Resource Limits

Without limits, a misbehaving container can starve its host. Set sensible CPU and memory caps for every long-running service.

Docker Compose

services:
  my-app:
    image: ghcr.io/trueforge-org/my-app:latest
    deploy:
      resources:
        limits:
          cpus: "1.0"
          memory: 512M
        reservations:
          cpus: "0.25"
          memory: 128M

For Compose v2 outside of Swarm, the equivalent shorthand is:

services:
  my-app:
    image: ghcr.io/trueforge-org/my-app:latest
    cpus: 1.0
    mem_limit: 512m
    mem_reservation: 128m

Kubernetes

resources:
  requests:
    cpu: 250m
    memory: 128Mi
  limits:
    cpu: 1000m
    memory: 512Mi

Choosing Values

• Start by observing the application under realistic load (docker stats, kubectl top).
• Set requests to the steady-state usage; this drives scheduling.
• Set limits to a comfortable ceiling — typically 1.5×–2× the steady state.
• Avoid setting CPU limits aggressively in Kubernetes: throttling hurts latency-sensitive workloads more than it helps.

OOM Behavior

If a container exceeds its memory limit it is killed (OOMKilled). Combine limits with healthchecks and a restart policy so the workload recovers automatically.