https://flyte.org
Join Slack
and i could see this annotation in the pod which is created by flyte ```apiVersion: v1 kind: Pod met...
# flyte-support
e
and i could see this annotation in the pod which is created by flyte
Copy code
apiVersion: v1
kind: Pod
metadata:
  annotations:
    cluster-autoscaler.kubernetes.io/safe-to-evict: "false"
    scheduling.k8s.io/group-name: agfcbnm8t8f9vzz5gxvs-n0-0
    scheduling.volcano.sh/queue-name: training
    volcano.sh/task-spec: worker
a
woo So it works? This is a great candidate for a docs contribution 🙂
e
not it doesnt work, even with the annotations being there pod group still craede with default queue
a
so, are you setting this in a PodTemplate as part of a Flyte workflow definition?
e
yes
a
ok and the expected behavior is that it triggers the creation of a Volcano PodGroup?
e
it does create the podgroup but it always use default queue irrespective of whatever the annotation say
so we need to pass run_policy for scheduling policy which i believe is not supported atm, any plsn to support this as its there in CRD and when i pass to the pytorch job manifest manually it just works
@average-finland-92144 i have made the change and test in my localnow, any idea how can i make it run in my remote. Any idea how to update my helm chart with this binary
i am getting this in my remote
Copy code
ImportError: cannot import name 'SchedulingPolicy' from 'flytekitplugins.kfpytorch.task' (/opt/micromamba/envs/runtime/lib/python3.10/site-packages/flytekitplugins/kfpytorch/task.py)
which makes sense
5 Views
Open in Slack
PreviousNext
Powered by

Flyte

Flyte enables production-grade orchestration for machine learning workflows and data processing created to accelerate local workflows to production.

Powered by