Great question — and the exact approach depends on which worker you’re using. Are you running on Kubernetes or Docker (or something else)? I’ll outline both below. If you confirm your worker type, I can tailor an exact snippet for your setup. Assumption: Prefect 3.x Key concept - Per‑deployment overrides are done via deployment

job_variables

. - Those variables must be defined in your work pool’s base job template. If the template doesn’t expose resource fields, you’ll need to add them once to the pool; then you can set them per deployment. Helpful docs - Customize job variables: Customize job variables - Manage work pools and base job templates: Manage work pools - Kubernetes worker guide: Kubernetes worker - Docker worker guide: Docker worker - Get default base job template (so you can customize it): -

prefect work-pool get-default-base-job-template --type kubernetes

-

prefect work-pool get-default-base-job-template --type docker

Kubernetes worker: set container.resources.requests/limits 1) Update your pool’s base job template to expose resources. - Start from the default template:

prefect work-pool get-default-base-job-template --type kubernetes > k8s-template.json

- Edit the first container to include a resources placeholder and add a matching variable in the template’s variables schema. Minimal example change (only showing the key parts): k8s-template.json (snippet)

{
  "job_configuration": {
    "job_manifest": {
      "spec": {
        "template": {
          "spec": {
            "containers": [
              {
                "name": "prefect-job",
                "image": "{{ image }}",
                "imagePullPolicy": "{{ image_pull_policy }}",
                "args": "{{ command }}",
                "env": "{{ env }}",
                "resources": "{{ resources }}"
              }
            ],
            "serviceAccountName": "{{ service_account_name }}"
          }
        }
      }
    }
  },
  "variables": {
    "properties": {
      "resources": {
        "type": [
          "object",
          "null"
        ],
        "default": null,
        "description": "Kubernetes container resources (requests/limits).",
        "examples": [
          {
            "requests": {
              "cpu": "500m",
              "memory": "512Mi"
            },
            "limits": {
              "cpu": "1",
              "memory": "1Gi"
            }
          }
        ]
      }
    }
  }
}

- Apply the updated template to your pool:

prefect work-pool create "my-k8s-pool" --type kubernetes --base-job-template ./k8s-template.json --overwrite

2) Set resources per deployment via Python or prefect.yaml - Python deploy:

from prefect import flow

  @flow
  def my_flow():
      ...

  my_flow.from_source(
      source="<https://github.com/org/repo>",
      entrypoint="flows/my_flow.py:my_flow",
  ).deploy(
      name="k8s-with-resources",
      work_pool_name="my-k8s-pool",
      job_variables={
          "resources": {
              "requests": {"cpu": "500m", "memory": "512Mi"},
              "limits":   {"cpu": "1",    "memory": "1Gi"}
          }
      },
  )

- prefect.yaml deployment block:

deployments:
    - name: k8s-with-resources
      work_pool:
        name: my-k8s-pool
      entrypoint: flows/my_flow.py:my_flow
      job_variables:
        resources:
          requests:
            cpu: "500m"
            memory: "512Mi"
          limits:
            cpu: "1"
            memory: "1Gi"

Notes for Kubernetes - The effective key path in the pod spec is: spec.template.spec.containers[0].resources.requests.cpu (and .memory) and resources.limits.cpu (.memory).

- If your base template doesn’t include the

resources

placeholder, per-deployment overrides will not take effect — add it once as shown above. Docker worker: set memory and CPU per deployment - The Docker worker exposes memory directly and CPU via container create kwargs. From the Docker worker configuration, these keys are available: -

mem_limit

(e.g., "2g" or "2048m") -

memswap_limit

(optional) -

container_create_kwargs

(pass through to docker-py, e.g.,

nano_cpus

,

cpu_shares

, etc.) - Python deploy:

from prefect import flow

  @flow
  def my_flow():
      ...

  my_flow.from_source(
      source="<https://github.com/org/repo>",
      entrypoint="flows/my_flow.py:my_flow",
  ).deploy(
      name="docker-with-limits",
      work_pool_name="my-docker-pool",
      job_variables={
          "mem_limit": "2g",
          "container_create_kwargs": {
              "nano_cpus": 2000000000  # ~2 CPUs
          }
      },
  )

- prefect.yaml deployment block:

deployments:
    - name: docker-with-limits
      work_pool:
        name: my-docker-pool
      entrypoint: flows/my_flow.py:my_flow
      job_variables:
        mem_limit: "2g"
        container_create_kwargs:
          nano_cpus: 2000000000

CLI usage tip - The most reliable path is to put

job_variables

in

prefect.yaml

and run:

prefect deploy

This avoids guessing CLI flags and ensures the values are captured in your deployment definition. Would you confirm which worker you’re using (Kubernetes or Docker)? If it’s Kubernetes, I can share a complete, ready-to-paste base job template including the variables schema so you can drop it into your pool and start overriding per deployment.