[EKS] [regression] excessive etcdserver throttling
- Dominant language
- Shell
- Stars
- 5.4k
- Forks
- 334
- PR merge metrics
- No merged PRs in 30d
Description
Ever since upgrading our EKS clusters to 1.35, it is becoming almost impossible to apply resources to newly created EKS clusters due to excessive throttling from etcd.
```
rpc error: code = ResourceExhausted desc = etcdserver: throttle: too many requests
```
After creating a new EKS cluster and applying manifests, helm-charts, etc, we almost always see throttling. This breaks helm install/upgrade actions and leads to broken installations.
This behaviour was not present in previous versions of EKS and I would like to request that new etcdservers are provisioned with more generous rate limits.
Contributor guide
Research direction
No repository file, test, or entry point is identified. Start by reproducing the throttling on a newly created EKS 1.35 cluster while applying manifests and Helm charts, then investigate the cluster's etcd rate-limit configuration. Done means new clusters accept these applications without recurring `etcdserver: throttle: too many requests` errors.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- aws, kubernetes
- Domain
- cloud, distributed-systems, infrastructure
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 30/100