Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Prefect Community

Hi everyone,
I'm running Prefect in a Kubernetes environment using the official Helm chart. I am using `selfHostedServer`
All deployments share a common `jobTemplate`.
Some deployments are more resource-intensive than others.
Is there a way to define different `resources.requests/limits` per deployment, without duplicating the entire job template?
Thanks!

Yes, it is possible to customize different request limits per deployment by adding variables to the base job template. There's an incomplete example hidden in the docs <https://reference.prefect.io/prefect_kubernetes/worker/|here>. I'm realizing that we don't have good docs for this, so let me know if you run into issues and I can help!

<@U085GQBPNN4> are you using `prefect.yaml` to define your deployments?

FWIW, YAML supports "<https://yaml.org/type/merge.html|merge keys>" which allow you to override key/value pairs. You can pair this with a definitions block in your `prefect.yaml` file. <https://docs.prefect.io/v3/how-to-guides/deployments/prefect-yaml#reuse-configuration-across-deployments>

```definitions:
  test: &amp;test
    foo: bar
    bar: baz

deployments:
- name: one
  value: *test
- name: two
  value:
    &lt;&lt;: *test
    foo: not_bar```
This would output:
```deployments:
- name: one
  value:
    foo: bar
    bar: baz
- name: two
  value:
    foo: not_bar
    bar: baz```
Note the special `&lt;&lt;: *` syntax which allows you to override a specific setting within that block. You could apply something similar  to your `prefect.yaml` file to override deployment-specifc job resource limits.

image.png

Hey <@U08K5Q2HMM0>, <@U02GE39K551> We are not using `prefect.yaml` to define deployments. I’ve added the `resources` snippet to our base job template, and all Prefect jobs use it without any issues.
The problem is with one specific deployment: it requires higher CPU and memory requests. I’ve added the updated `job variables` configuration in the deployment configuration, and I can see it correctly when I edit the deployment in the UI.
However, when the job actually runs, the pod still uses the CPU and memory values from the base job template instead of the deployment configuration.
I tried a few things, including removing the `resources` snippet from the base job template, but I still don’t see the deployment-specific configuration applied at runtime.

Example snippet from the base job template:
```"
spec": {
  "containers": [
    {
      "env": "{{ env }}",
      "args": "{{ command }}",
      "name": "prefect-job",
      "image": "{{ image }}",
      "imagePullPolicy": "{{ image_pull_policy }}",
      "resources": {
        "requests": {
          "cpu": "100m",
          "memory": "256Mi"
        }
      }
    }
  ]
}```


k8s-work-pool-base-config-with-limits.json

Screenshot 2025-08-13 at 12.11.15 PM.png

<@U085GQBPNN4> ok, I understand what's going on now. You'll need to modify the Work Pool Base Job template to make this work correctly. I've included an example above. You can set the defaults in your work pool base config. The schema makes 4 new options available for CPU and Memory requests/limits.

This will allow you to then override the config on a per-deployment basis and have the correct values passed to your Job/Pod configs.

Screenshot 2025-08-13 at 12.17.09 PM.png

To edit the Work Pool config JSON schema, go to your work pool, click on the three dots in the top right corner of the screen. Then click edit from the dropdown list. Click the Advanced tab under the Base Job Template section on the edit page.

Thanks <@U08K5Q2HMM0> much appreciated. That solution worked for me.