We have four servers, in each rack two. With set-clusterfaultdomain I set the two rack-ids of the four servers. Created the s2d storage pool (FaultDomainAwarenessDefault=StorageRack). Added cloud witness disk. Created four volumes with mirror (parity not available). Put on each volume a vhdx of a test server, and started this on every hyperv-host one. All looks good so far.
Broke the network connection between the two racks, so the cluster goes into isolation mode. After some time, just one of the failed virtual server returns to the surviving side of the cluster (with the internet connection). The other enters failed state. After investigating the config, it looks like one of the volumes is failed, on which the failed server had his vhdx stored. I waited at least 15 minutes before helping the cluster to repair. it should have repaired itself in some way. Repair of volume in console give "network path not found".
Why is one volume (with mirror resiliency) failed ? It should allways been available on one side of the cluster because we just have two racks, and with mirror it should be available in each rack. How can I determine which volume will fail ? How to prevent the failure of the cluster-shared-volume, thus the virtual servers running on it ?