After a datastore is extended from host level, a storage rescan must be completed on all ESXi hosts accessing the datastore so that they detect the new size of the datastore.
If a storge rescan is not run, or if it fails on a host, then, when that host next queries the volume metadata it will find a mismatch between the extended volume size recorded there (i.e. "stored size 8589819905 blocks") and its cached record of the original smaller device size (i.e. "actual size 6442448863"). As the device appears to the host to be smaller than its stored size, a "Device Shrank" warning will be returned.
Subsequently, if a write is attempted to the extended part of the volume, the host will trigger a warning "Address temporarily unmapped" because it has determined that the device shrank and that the addressed part of the device is unavailable.