Skip to content

Latest commit

 

History

History
42 lines (30 loc) · 2.27 KB

etcd.md

File metadata and controls

42 lines (30 loc) · 2.27 KB
Raw

Etcd Upgrade

Prior to Kismatic v1.3.0, all cluster deployments were composed of two versions of etcd:

  • Kubernetes etcd with version v3.0.x
  • Calico etcd with version v2.3.x

With Kismatic v1.3.0, both versions of etcd will be upgraded to etcd v.3.1.x which is not backwards compatible with etcd v2.3.x or v3.0.x.

In addition, v1.3.0 ships Kubernetes v1.6.0 which defaults to the etcd v3 API. During the upgrade, Kismatic will migrate the kubernetes data using the etcd tooling. Before performing the migration, the Kubernetes API servers will be shut down to prevent writes to the etcd cluster. This means that your cluster will experience temporary downtime, even in the case where multiple API servers are running.

Upgrading Calico etcd v2.3.x to v.3.0.x

On an etcd cluster that has more than one member, it is not possible to directly upgrade from v2.3.x to v.3.1.x. All nodes in the cluster must be upgraded to v3.0.x, and then upgraded to v.3.1.x.

The following is performed on each node, one node at a time:

  1. Install transition-etcd package containing the etcd v3.0 binaries: etcd_v3_0, etcdctl_v3_0
  2. Backup etcd data to /etc/etcd_networking/backup/$timestamp
  3. Restart the etcd_networking service
  4. Wait until the node has caught up with the rest of the cluster
  5. Verify cluster with etcdctl_v3_0 --endpoint='https://127.0.0.1:6666 cluster-health

Upgrading Kubernetes and Calico etcd v3.0.x to v.3.1.x

The following is performed on each node, one node at a time:

  1. Remove old kismatic-etcd and transition-etcd packages
  2. Install new etcd package, which contains the following binaries: etcd_k8s, etcd_networking and etcdctl
  3. Backup etcd data to /etc/etcd_k8s/backup/$timestamp
  4. Restart the etcd_k8s service
  5. Wait until the member has caught up with rest of the cluster
  6. Verify Kubernetes etcd cluster with etcdctl --endpoint='https://127.0.0.1:2379 cluster-health
  7. Backup etcd data to /etc/etcd_networking/backup/$timestamp
  8. Restart the etcd_networking service
  9. Wait until the member has caught up with the rest of the cluster
  10. Verify Calico etcd cluster with etcdctl --endpoint='https://127.0.0.1:6666 cluster-health

Note: Downgrading etcd is not possible once all members are upgraded to v3.1.x.