Skip to main content
Version: v0.37 Stable
Supported Configurations
Running the control plane as a container with:

Resolve ingress update conflicts with Rancher

When deploying the Rancher agent in both vCluster and the control plane cluster with a single shared ingress, both agents repeatedly update the ingress object. This causes excessive load on the database and API server.

You might observe the following signs of ingress update conflicts:

  • Endless updates to ingress resources in the control plane cluster
  • Rapid growth of the API server database (may reach several GB)
  • API server slowdown (simple commands like kubectl get pods take several seconds)

Cause​

This issue occurs due to a conflict between Rancher agents deployed in both vCluster and the control plane clusters. Both agents attempt to update annotations on the ingress using different naming conventions, creating a continuous update loop.

Solution​

This issue is resolved in vCluster version 0.19.8 and later. To fix the problem:

  1. Upgrade your vCluster installation to version 0.19.8 or later
  2. If your API server is slow, compact your etcd database to reduce its size (specific to your Kubernetes distribution)

Upgrade vCluster using Helm​

  1. Check your current vCluster version.

    helm list -A | grep vcluster

  2. Upgrade to version 0.19.8 or later.

    If you installed vCluster using Helm, run the following command:

    helm upgrade <mycluster> loft/vcluster --reuse-values --version 0.19.8

    Replace <mycluster> with the name of your vCluster release.

  3. Verify the upgrade.

    helm list -A | grep vcluster

    Confirm the version is now 0.19.8 or later.

Verify the fix​

After upgrading, monitor your cluster to ensure:

  • Ingress updates have stabilized
  • The database size growth has slowed or stopped
  • API server performance has returned to normal
note

You must compact the etcd database before seeing performance improvements.

Check API server performance​

Run a simple command and measure response time:

time kubectl get pods -A

The command should complete in under a second if the API server is performing normally.

Additional resources​

For more details about this fix and other changes in vCluster 0.19.8, refer to the official release notes.