@Marvin can you show me an example of a deployments.yaml file for an aks worker that uses

job_variables

to override the cpu and memory request

Ah, the sweet smell of YAML in the air. Here's an example of a

deployments.yaml

file for an AKS worker that uses

job_variables

to override the CPU and memory request:

yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-aks-worker
spec:
  replicas: 1
  selector:
    matchLabels:
      app: my-aks-worker
  template:
    metadata:
      labels:
        app: my-aks-worker
    spec:
      containers:
      - name: my-aks-worker
        image: prefecthq/prefect:latest
        env:
        - name: PREFECT__CLOUD__AGENT__LABELS
          value: "aks"
        - name: PREFECT__CLOUD__AGENT__TOKEN
          valueFrom:
            secretKeyRef:
              name: prefect-api-token
              key: token
        resources:
          requests:
            memory: "{{ job_variables.memory_request }}Mi"
            cpu: "{{ job_variables.cpu_request }}m"

In this example, the

job_variables.memory_request

and

job_variables.cpu_request

are used to set the memory and CPU requests for the AKS worker container. Make sure to replace the

prefecthq/prefect:latest

image with the appropriate Prefect image for your use case. Remember to replace the placeholders

{{ job_variables.memory_request }}

and

{{ job_variables.cpu_request }}

with the actual values you want to use when deploying your AKS worker. I hope this helps you in your quest for YAML perfection. Good luck, human!

@Marvin, what if I want my specific deployment to have different infrastructure resources than the default worker. My deployment.yaml currently looks like this:

deployments:
  - name: aks-test-sp
    entrypoint: .\workflow\flow_demo.py:aks_demo_flow
    work_pool:
        name: aks-worker
    job_variables:
      cpu_request: "4"
      cpu_limit: "8"

Ah, a twist in the tale of YAML. To customize the infrastructure resources for your specific deployment, you can modify the

deployments.yaml

file like this:

yaml
deployments:
  - name: aks-test-sp
    entrypoint: .\workflow\flow_demo.py:aks_demo_flow
    work_pool:
      name: aks-worker
    job_variables:
      cpu_request: "4"
      cpu_limit: "8"
      memory_request: "2Gi"
      memory_limit: "4Gi"

Now, let's update the AKS worker deployment to use these `job_variables`:

yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-aks-worker
spec:
  replicas: 1
  selector:
    matchLabels:
      app: my-aks-worker
  template:
    metadata:
      labels:
        app: my-aks-worker
    spec:
      containers:
      - name: my-aks-worker
        image: prefecthq/prefect:latest
        env:
        - name: PREFECT__CLOUD__AGENT__LABELS
          value: "aks"
        - name: PREFECT__CLOUD__AGENT__TOKEN
          valueFrom:
            secretKeyRef:
              name: prefect-api-token
              key: token
        resources:
          requests:
            memory: "{{ job_variables.memory_request }}"
            cpu: "{{ job_variables.cpu_request }}"
          limits:
            memory: "{{ job_variables.memory_limit }}"
            cpu: "{{ job_variables.cpu_limit }}"

In this example, the

job_variables

from your

deployments.yaml

file are used to set the memory and CPU requests and limits for the AKS worker container. Make sure to replace the

prefecthq/prefect:latest

image with the appropriate Prefect image for your use case. Keep in mind that you'll need to replace the placeholders

{{ job_variables.memory_request }}

,

{{ job_variables.memory_limit }}

,

{{ job_variables.cpu_request }}

, and

{{ job_variables.cpu_limit }}

with the actual values from your

deployments.yaml

file when deploying your AKS worker. I hope this clears up the YAML fog. Good luck with your deployment, human!

@Christopher Boyd this seems to suggest I need to also configure my workpool, but then would it still take defaults or always expect an input value for those?

Certainly, my human friend. To run your deployment in a different node pool, you can use Kubernetes node selectors. Node selectors allow you to specify which nodes your deployment should run on based on their labels. First, make sure your desired node pool has a unique label. For example, if you have a node pool named

custom-nodepool

, you can label its nodes with

nodepool: custom-nodepool

. Next, update your AKS worker deployment to include the

nodeSelector

field:

yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-aks-worker
spec:
  replicas: 1
  selector:
    matchLabels:
      app: my-aks-worker
  template:
    metadata:
      labels:
        app: my-aks-worker
    spec:
      containers:
      - name: my-aks-worker
        image: prefecthq/prefect:latest
        env:
        - name: PREFECT__CLOUD__AGENT__LABELS
          value: "aks"
        - name: PREFECT__CLOUD__AGENT__TOKEN
          valueFrom:
            secretKeyRef:
              name: prefect-api-token
              key: token
        resources:
          requests:
            memory: "{{ job_variables.memory_request }}"
            cpu: "{{ job_variables.cpu_request }}"
          limits:
            memory: "{{ job_variables.memory_limit }}"
            cpu: "{{ job_variables.cpu_limit }}"
      nodeSelector:
        nodepool: custom-nodepool

In this example, the

nodeSelector

field is set to match nodes with the label

nodepool: custom-nodepool

. This will ensure that your deployment runs on nodes within the

custom-nodepool

node pool. Remember to adjust the other parts of the YAML file according to your specific requirements, such as the Prefect image and resource requests and limits. And there you have it, another YAML mystery solved. Good luck with your node pool adventures, human!

is there a way I can pass this through as an infrastructure override in my deployment.yaml that looks like this:

deployments:
  - name: aks-test-sp
    entrypoint: .\workflow\flow_demo.py:aks_demo_flow
    work_pool:
      name: aks-worker
      job_variables:
        cpu_request: "4"
        cpu_limit: "8"
        memory_request: "2Gi"
        memory_limit: "4Gi"

You can , it just passes through to the template for the job

ah perfect yeah that sounds great

For the most part - the infra override only sets up your kubernetes job the way you want it