https://prefect.io
Join Slack
Hi, does anyone know how to set the service account for the dask workers in an ephemeral cluster? Th...
# ask-community
b
Hi, does anyone know how to set the service account for the dask workers in an ephemeral cluster? This may be more of a dask question but the k8s pods are being set to 
default
and was wondering if anyone ran into the same issue. Example executor code is in the thread.
Copy code
DaskExecutor(
        cluster_class=lambda: KubeCluster(
            name=slugify(prefect.context.flow_name),
            pod_template=make_pod_spec(
                image=prefect.context.image,
                cpu_request="1000m",
                memory_request="2Gi",
                threads_per_worker=2,
                env=env_dict,
                extra_pod_config={
                    "serviceAccount": "dask-workflows-sa",
                    "serviceAccountName": "dask-workflows-sa",
                    "priority": 1,
                }
            ),
            scheduler_pod_template=make_pod_spec(
                image=prefect.context.image,
                cpu_request="1000m",
                memory_request="1Gi",
            ),
        ),
        adapt_kwargs={
            "minimum": 1,
            "maximum": 4,
        },
    )
I tried to also set 
priority
but it doesn’t seem like anything in the 
extra_pod_config
block makes it to the worker pod.
k
Hey @Brad I, did you try through the RunConfig?
KubernetesRun(service_account_name="")
? I went through KubeCluster docs and this looks right. I don’t know why the 
extra_pod_config
wouldn’t be respected
b
Hi @Kevin Kho, thanks for the suggestion, but unfortunately the scheduler and worker k8s yaml still shows 
default
as the serviceAccount and serviceAccountName. We also have the service account set as the default in our agent. The dask documents mention a 
config.yaml
or 
~/.kube/config
file, is that something I can pass through the 
DaskExecutor
or should I try setting it inside our docker image?
@Kevin Kho, weird, so I commented out the separate 
scheduler_pod_template
and everything seems to work. So I needed to make your change plus let dask use the same pod template for both workers and the scheduler. As long as it’s working 🤷 , thanks!
k
What!? I am confused by that (and how you figured it out) by glad you got it working!
5 Views
Open in Slack
PreviousNext
Powered by

Prefect Community

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.

Powered by