Home > Storage > PowerStore > Data Protection > Dell PowerStore: Metro Volume > Polarization
One of the most critical situations for a Metro Volume is a split-brain scenario when both volumes of a Metro Volume are active simultaneously. This situation can lead to having different data on both volumes and may require manual intervention. To prevent a split-brain situation, PowerStore uses polarization for failure handling. For each Metro Volume, one volume is assigned with a preferred role while the peer side is configured with a non-preferred role. The initial preferred side of a Metro Volume is the side where the Metro Volume replication session configuration was performed. The following figure shows an example of the distributed, preferred side of Metro Volumes.
You can change the preferred role to non-preferred, and conversely, in PowerStore Manager. This action is non-disruptive for the hosts when a Metro Volume is in the status Operating Normally. While the Metro Volume session is running with the active-active status, the PowerStore cluster must ensure I/O can be replicated to the peer PowerStore appliance. Polarization occurs when a side is not responding, and it is leading to a status of Fractured for affected Metro Volumes. When polarization is invoked and the preferred side determines that the non-preferred is not available, the replication traffic is stopped and host I/O on the preferred side continues. On the non-preferred side, hosts do not receive acknowledgments for their writes, the I/O stops, and paths become ALUA unavailable. In this situation when paths are unavailable, the ESXi server shows an All Path Down (APD) event for affected datastore volumes. Depending on the host connectivity, the affected VMs continue or VMware HA must restart VMs on hosts with paths to the preferred volume.
After the connection is reestablished, PowerStore initiates self-healing of the Metro Volume with data on the most current volume. Without manual intervention, the preferred volume is seen as the most current, and self-healing initiates the synchronization from the preferred side to the non-preferred side. Self-healing uses the last common base and does not require a full synchronization.
During polarization, only hosts with paths to the preferred side can access the volume. When a Metro Volume is not in an expected state for host access, you can use promote and demote operations to override the preferred or non-preferred role for a Metro Volume. The promote operation enables host I/O on the non-preferred side of a Metro Volume, and the demote operation disables host I/O on the preferred side. After a promote operation, PowerStore determines the data on the promoted side to be more current than the data on the preferred side and starts self-healing from the promoted volume to the demoted volume. An unwanted manual intervention with the promote operation could lead to a split-brain situation. It could lead to data loss because production data on the preferred could be overwritten by data on promoted side during self-healing. PowerStore helps to mitigate the potential risk of data loss by performing a snapshot before running the promote and demote actions, and when self-healing begins.