Skip to content

Relocation

Relocation moves a running workload to another location. Simplyblock distinguishes three kinds of relocation. A DR relocation moves entire applications between Kubernetes clusters and restarts them on the target site. An online migration across clusters would move virtual machines and volumes without a restart and is planned for the future. Within a single storage cluster, volumes already move between storage nodes online today.

DR Relocation (Restart-Based)

A DR relocation is a planned move of a protected application, including its Kubernetes objects and its volumes, along a DR path. It is requested as a RecoveryAction of kind Relocate on the DR hub and applies to a single application or to a recovery plan with several applications in priority order.

The relocation runs in the following steps:

  1. Stop and demote: The application is stopped on the source site, and its volumes are demoted.
  2. Final synchronization: The remaining changes are replicated to the target site, so no data is lost.
  3. Promote: The volumes are promoted on the target site.
  4. Restart on the target: The Kubernetes objects are restored on the target site in the tier order of the application, and the application starts. Health probes confirm that it is up.
  5. Clean up: The remaining workload objects on the old site are removed.

The application is unavailable between the stop on the source and the successful start on the target. If the target does not become healthy within the action timeout while the source is still healthy, the relocation rolls back and the application runs on the source again.

Failback is a relocation along the reverse DR path. The differences between a relocation and an unplanned failover are described in Failover.

Online Migration Across Clusters

Coming soon

Online migration across clusters is not designed yet. It is intended to cover the live migration of KubeVirt virtual machines and the live migration of volumes from one simplyblock cluster to another, without restarting the workload. Until it is available, workloads move between clusters with a restart-based DR relocation.

Volume Cutover Between Clusters

Below the application level, the storage-level asynchronous replication of simplyblock can prepare a planned cutover of individual volumes to another storage cluster of the same control plane. A replication policy in migration mode replicates the volumes to the target cluster while they keep serving I/O on the source. The cutover itself is then committed per volume: the volume is frozen briefly, the remaining delta is transferred, and the target cluster serves the volume from then on. The workload has to reconnect to the volume on the target cluster, so this is not an online migration of the workload. See Asynchronous Replication.

Volume Migration Within a Cluster

Within one simplyblock storage cluster, volumes can already be migrated online between storage nodes without service interruption, for example, to follow a workload for data locality, to balance I/O, or to drain a node before its removal. See Volume Migration.

Further Reading