[BUG] APM injection blocked when container has no CPU/memory limit — request incorrectly used as limit

[satvamthakkar]

Agent version

7.74.0

Bug Report

When a container has no CPU limit but a CPU request below 50m, the Admission Controller blocks APM auto-injection with the annotation:
apm.datadoghq.com/injection-error: "The overall pod's containers limit is too low, cpu pod_limit=10m needed=50m"

The container has no CPU limit, meaning unlimited CPU is available. However, PodSumResourceRequirements() in pkg/clusteragent/admission/mutate/autoinstrumentation/libraryinjection/resources.go copies the request value into the imit when no limit is set (line ~198), causing the minimum check to fail against the request value instead of recognizing that no limit means unlimited resources.

Not setting CPU limits is a widely recommended Kubernetes practice to avoid CPU throttling. This bug blocks APM injection for any pod following the practice with a low cpu request.

Reproduction Steps

1. Deploy a pod with CPU request below 50m and no CPU limit:
   containers:
   - name: app
   image: nginx
   resources:
   requests:
   cpu: 10m
   memory: 128Mi
   limits:
   memory: 256Mi
   # No CPU limit
2. APM auto-injection is skipped
3. Check pod annotation: kubectl get pod -o jsonpath='{.metadata.annotations.apm.datadoghq.com/injection-error}'
4. Returns: The overall pod's containers limit is too low, cpu pod_limit=10m needed=50m

Agent configuration

APM auto-instrumentation enabled with default settings. No DD_ADMISSION_CONTROLLER_AUTO_INSTRUMENTATION_INIT_RESOURCES_CPU override.

Operating System

Ubuntu 22.04.05 LTS Kernel Version 5.15.0-1102-azure

Other environment details

AKS (Azure Kubernetes Service), Kubernetes 1.34, Datadog Cluster Agent 7.74.0 managed by Datadog Operator (Helm chart datadog-operator 2.17.0)