Hi everyone, I deployed prefect on a single node and now aiming at deploying the same on kubernetes using helm chart. While learning from live sessions on youtube I came across this piece of code.

@task
	def say_hello():
		print("hello world !!!")

with Flow("hello world") as flow:
	say_hello()


flow.storage = Docker(registry_url = "", image name = "hello world flow")
flow.register("demo")

Can someone please explain to me 1.) Why we used container registry for storing a flow and what difference would it make compared to other script based storage like github. 2.) When we execute a flow whose server is present on K8 but no executer is defined like in above snippet, where would the flow be executed on local machine or on a pod on the cluster.

Hello! If you are using Docker storage, your flow will be stored in Docker image (it will be created for you), then this image will be pushed to the registry (which you will provide). For some users, it's preferable storage option, but you could use other storage types described here. Are you using KubernetesRun?

For 1 specifically, the pros and cons of script based storage and pickle based storage are here . Pickle based is needed if you’re packaging up other dependencies along with your code.

@Kevin Kho for 2, so if i am running 10 different flows at once , will 10 different pods be deployed on cluster.

Yes that’s right

Ok but isn't that a bad practice as in some kind wastage of resources. It might happen that a pod is capable of running the 3-4 flows but we are still deploying one for each.

The

RunConfig

will let you determine the resources for that flow like

flow.run_config = KubernetesRun(cpu_request=2, memory_request="2Gi")

Got it

RunConfig objects define where and how a flow run should be executed. Each RunConfig type has a corresponding Prefect Agent (i.e. LocalRun pairs with a Local Agent, DockerRun pairs with a Docker Agent, ...)

Yes to KubernetesRun when running on the cluster. Also use the Kubernetes Agent. Local agent for local machine is also right. I think helm would be separate from these concepts as it’s for installation

Sorry, but did not understand what you meant by

helm would be separate from these concepts as it's for installation

What are you using helm for?

I am using helm chart config from prefectHq, to deploy prefect agent along with prefect server on kubernetes

Ah ok so I meant that yes that’s right. It’s just a separate concept from the k8s RunConfig. I think you’re good.

So now all i would need to do is have two agents , one on local and one on kubernetes, and based on their labels i can decide whether the flow runs on local machine or kubernetes pods

Yes that’s right. Labels provide the separation

great , will work on this and also the storage thing. Thanks a lot Kevin and Mariia