Hi there,
We have an Elasticsearch application consisting of 3 nodes; each node is placed on a different host and has a separate PV bound. We’re using the following:
- Helm Chart to deploy the ES cluster: https://artifacthub.io/packages/helm/elastic/elasticsearch/7.16.3,
- K10 version is 6.5.1
- the storageclass for PV provisioning is local-path-provisioner: https://github.com/rancher/local-path-provisioner (we’re not using CSI)
- Kubernetes 1.24 (RKE2 Distro)
The Kanister Blueprint has been written to deal with the backup & restore of the ES app. The backup runs successfully, but when we want to perform a restore, the following problem occurs - executing a data-only or full restore - neither of them seems to work. The statefulset is scaled to 0, and the PVCs are deleted and recreated, but they are placed on the same host, which prevents the pod from being scheduled, since the pod anti-affinity rules cannot be fulfilled - pods would have to be scheduled on the same host.
Is it possible to alter the behavior of Kasten when re-creating the PVCs? How does the Kasten deal with HA, or multi-cluster architecture, like this?
Thanks in advance!
Best answer by Hagag
View original
