Skip to main content
A pod spec defines the containers and pod-level requirements for runnable work. Deployments, jobs, cron jobs, pods, and replica sets all use pod specs. The path differs by workload type, but the container and requirement model is shared.

​
Containers

A container spec defines the image and process to run. Common container settings include:
  • imageURI for the container image.
  • imageDigest when an image digest should be pinned.
  • imagePullPolicy for pull behavior.
  • command and args to override image defaults.
  • env for environment variables.
  • resources.requests and resources.limits for CPU, memory, and disk.

​
Requests and limits

IDYL uses resource requests for placement. If a request is omitted for a resource but a limit is present for the same resource, the limit is used as the effective request for placement.

​
GPU requirements

GPU requirements belong on the pod spec as gpu. GPU is a host device requirement for the pod, not a per-container setting. All containers in the pod share the pod-level GPU access. 
spec:
  gpu: nvidia:24Gi

​
Host requirements

Use require when a workload needs host capabilities beyond CPU, memory, disk, or GPU selection. 
require:
  cuda: ">=13.0"