EFA plugin enters a crash backoff loop by default install
- Dominant language
- Mustache
- Stars
- 1.3k
- Forks
- 1.1k
- Avg merge
- 22m
- Merged PRs (30d)
- 5
Description
**Describe the bug**
When using the EFA plugin installed via helm, all instances that have the ability to provision an EFA will have the daemon set create a container on the node. The container will immediately enter a crash backoff loop if no container on the node is using the EFA
**Steps to reproduce**
Install aws-efa-k8s-device-plugin via helm and using karpenter, create a node/container on an instance that supports EFA (I did this with Flyte testing on both R6a.48x and g6e.48x nodes)
**Expected outcome**
The container creates and then does nothing because there's no EFA activity
https://github.com/aws/eks-charts/issues/1233
This issue also touches on the same problem
**Environment**
* Chart name: aws-efa-k8s-device-plugin
* Chart version: latest
* Kubernetes version: 1.31
* Using EKS (yes/no), if so version? Yes -- 1.31
**Additional Context**:
I haven't tested this with manually scaled or cluster autoscaler scaled nodes, only with karpenter
Contributor guide
Research direction
Start with the aws-efa-k8s-device-plugin Helm chart and reproduce the issue on an EFA-capable R6a.48x or g6e.48x node provisioned with Karpenter. Observe the daemon set container when no workload uses EFA, then compare its behavior with the expected idle state. Done means the container remains running without entering crash backoff when EFA is unused.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- aws, helm, kubernetes
- Domain
- cloud, devops, infrastructure
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100