Yes — with an important distinction. In Prefect 3, you can use

DaskTaskRunner

for Prefect tasks, and those tasks can execute on pods in your EKS cluster if the Dask workers are running on EKS. So the architecture is:

text
Prefect flow run
  -> uses DaskTaskRunner
  -> submits tasks to Dask scheduler
  -> Dask workers execute those tasks
  -> those workers can be Kubernetes pods on EKS

Key point: Prefect itself does not turn each task into its own Kubernetes pod automatically just because you're using a task runner. The

DaskTaskRunner

offloads task execution to a Dask cluster. If that Dask cluster is deployed on EKS, then the task code runs on those EKS pods. Example:

python
from prefect import flow, task
from prefect_dask import DaskTaskRunner

@task
def work(x):
    return x * 2

@flow(
    task_runner=DaskTaskRunner(
        address="<tcp://dask-scheduler.your-namespace.svc.cluster.local:8786>"
    )
)
def my_flow():
    futures = [work.submit(i) for i in range(10)]
    return [f.result() for f in futures]

In that setup: - the flow run executes wherever your Prefect worker starts it - the tasks execute on the Dask workers - if those Dask workers are pods on EKS nodes, then yes, the tasks run on your cluster pods You can also have Prefect create/connect to a Dask cluster dynamically via

cluster_class

/

cluster_kwargs

.

DaskTaskRunner

supports:

text
DaskTaskRunner(
    cluster=None,
    address=None,
    cluster_class=None,
    cluster_kwargs=None,
    adapt_kwargs=None,
    client_kwargs=None,
    performance_report_path=None,
)

So for Kubernetes-backed Dask, the typical pattern is to use

dask-kubernetes

and pass a Kubernetes cluster class / config there. A couple of important caveats: -

ThreadPoolTaskRunner

and

ProcessPoolTaskRunner

run on the *same machine/pod as the flow run*; they do not spread tasks across EKS pods. - If you want work on separate pods, you need an external distributed executor like Dask or Ray. - Prefect deployments/work pools control *where the flow run starts*; the Dask cluster controls where the tasks execute. If your goal is specifically:

“I want Prefect tasks to run on pods deployed to our EKS cluster, not Fargate”

then the answer is: - Yes, by running Dask workers as pods on your EKS cluster - and configuring those pods with your usual Kubernetes node selectors / tolerations / affinity so they land on EC2-backed nodes, not Fargate Useful references: - Prefect Dask integration docs - prefect-dask usage guide - Dask on Kubernetes If you want, I can sketch the recommended EKS architecture for this in one of two ways: 1. Prefect worker on Kubernetes + existing Dask cluster on EKS 2. Prefect worker on Kubernetes + ephemeral Dask cluster created per flow run