Following upgrade from ESXi 7.0.x to 8.0.x, direct-attached iSCSI datastores, with the following configuration, are not mounted:
Software iSCSI. Multiple iSCSI vmkernel ports, all in the same subnet and each connected without a switch to separate iSCSI targets. Port binding enabled.
/var/log/syslog.log reports repeated login timeouts
/var/log/vmkernel.log reports iSCSI repeatedly going offline and online.
hexdump of iSCSI LUN fails to read the LUN data
ESXi vSphere 8.0.x
This appears to be similar to an invalid port binding configuration, where each vmk port group cannot access all targets.
While this does not appear to be problematic in ESXi 7.x, it causes the symptoms described above in ESXi 8.x.
Implement one of the following:
1) connect the hosts to the iSCSI targets via a switch, so that each iSCSI vmkernel port can connect to each target.
2) configure each iSCSI vmkernel port and its connected target to be in a separate subnet, and do not enable port binding.
3) simplify the configuration to have one iSCSI vmkernel port connected to a single target (useful for testing purposes, but does not provide path redundancy)