The content in this section will help Flyte users isolate the most probable causes for some of the common issues that could arise while getting started with the project.
Before getting started, collect the following information from the underlying infrastructure:
Statuscolumn from the output of:
$ kubectl describe pod <PodName> -n <namespace>
Where <PodName> will typically correspond to the node execution string that you can find in the UI.
Pay close attention to the Events section in the output.
Also, collect the logs from the Pod:
$ kubectl logs pods -n <namespace>
Where <namespace> will typically correspond to the Flyte <project>-<domain>, e.g. flytesnacks-development.
Depending on the contents of the logs or the Events, you can try different things:
Debugging common execution errors#
message: '0/1 nodes are available: 1 Insufficient cpu. preemption: 0/1 nodes are available: 1 No preemption victims found for incoming pod.'#
This issue is more common on MacOS devices. Make sure that your Docker daemon has allocated a minimum of 4 CPU cores and 3GB of RAM
terminated with exit code (137). Reason [OOMKilled]#
For single binary environment deployed with Helm chart, make sure you are using the most recent charts
For EKS deployments, you cand adjust resource limits and requests in the inline section of the
inline: task_resources: defaults: cpu: 100m memory: 100Mi storage: 100Mi limits: memory: 1Gi
Also, the default container resource limits are can be overriden from the task itself:
from flytekit import Resources, task @task(limits=Resources(mem="256Mi") def your_task(...
If your environment requires the use of a network proxy use the
--envoption when starting the sandbox and pass the proxy configuration:
$ flytectl demo start --env HTTP_PROXY=<your-proxy-IP>
If you’re building a custom Docker image, make sure to use a tag other than
latest. Otherwise, the Kubernetes default pull policy will be changed from
Always, forcing an image pull with every Pod deployment.
Issues running workloads#
flyte-binary: make sure that the endpoint name you have set in your
config.yamlfile, is included in the DNS names of the SSL certificate installed (be it self signed or issued by a Certificate Authority)
sandbox: verify the
FLYTECTL_CONFIGenvironment variable has the correct value by running:
$ export FLYTECTL_CONFIG=~/.flyte/config-sandbox.yaml
If you’re using a custom container image and using Docker, make sure your
Dockerfileis located at the same level of the
flytedirectory and that there is an empty
__init__.pyfile in your project’s folder :
myflyteapp ├── Dockerfile ├── docker_build_and_tag.sh ├── flyte │ ├── __init__.py │ └── workflows │ ├── __init__.py │ └── example.py └── requirements.txt
An error occurred (AccessDenied) when calling the PutObject operation in an EKS deployment#
Make sure that the Kubernetes service account Flyte is using has the annotation that refers to the IAM Role is connected to:
$ kubectl describe sa <my-flyte-sa> -n <flyte-namespace>
Name: <my-flyte-sa> Namespace: flyte Labels: app.kubernetes.io/managed-by=eksctl Annotations: eks.amazonaws.com/role-arn: arn:aws:iam::<aws-account-id>:role/flyte-system-role Image pull secrets: <none> Mountable secrets: <none> Tokens: <none> Events: <none>
Otherwise, obtain your IAM role’s ARN and manually annotate the service account:
$ kubectl annotate serviceaccount -n <flyte-namespace> <http://eks.amazonaws.com/role-arn=arn:aws:iam::xxxx:role/<flyte-iam-role>eks.amazonaws.com/role-arn=arn:aws:iam::xxxx:role/<flyte-iam-role>
Refer to this community-maintained guides for further information about Flyte deployment on EKS