prefect.io kubernetes agent and task execution - prefect

Question

While reading kubernetes agent documentation, I am getting confused with below line
"Configure a flow-run to run as a Kubernetes Job."
Does it mean that the process which is incharge of submitting flow and communication with api server will run as kubernetes job?
On the other side, the use case which I am trying to solve is
Setup backend server
Execute a flow composed of 2 tasks
if k8s infra available the tasks should be executed as kubernetes jobs
if docker only infra available, the tasks should be executed as docker containers.
Can somebody suggest me, how to solve above scenario in prefect.io?

Answer 1

That's exactly right. When you use KubernetesAgent, Prefect deploys your flow runs as Kubernetes jobs.
For #1 - you can do that in your agent YAML file as follows:
env:
- name: PREFECT__CLOUD__AGENT__AUTH_TOKEN
value: ''
- name: PREFECT__CLOUD__API
value: "http://some_ip:4200/graphql" # paste your GraphQL Server endpoint here
- name: PREFECT__BACKEND
value: server
#2 - write your flow
#3 and #4 - this is more challenging to do in Prefect, as there is currently no load balancing mechanism aware of your infrastructure. There are some hacky solutions that you may try, but there is no first-class way to handle this in Prefect.
One hack would be: you build a parent flow that checks your infrastructure resources and depending on the outcome, it spins up your flow run with either DockerRun or KubernetesRun run config.
from prefect import Flow, task, case
from prefect.tasks.prefect import create_flow_run, wait_for_flow_run
from prefect.run_configs import DockerRun, KubernetesRun
#task
def check_the_infrastructure():
return "kubernetes"
with Flow("parent_flow") as flow:
infra = check_the_infrastructure()
with case(infra, "kubernetes"):
child_flow_run_id = create_flow_run(
flow_name="child_flow_name", run_config=KubernetesRun()
)
k8_child_flowrunview = wait_for_flow_run(
child_flow_run_id, raise_final_state=True, stream_logs=True
)
with case(infra, "docker"):
child_flow_run_id = create_flow_run(
flow_name="child_flow_name", run_config=DockerRun()
)
docker_child_flowrunview = wait_for_flow_run(
child_flow_run_id, raise_final_state=True, stream_logs=True
)
But note that this would require you to have 2 agents: Kubernetes agent and Docker agent running at all times