I have set up an NFS storage cluster across two Linux machines with DRBD syncing data and Heartbeat controlling a shared VIP and NFS server. This setup has been tested in multiple environments, including as KVM storage, but this configuration fails to work as a datastore for an ESXi host. The ESXi host will mount the starting node on which the NFS server is running (both nodes A and B work). When I simulate a failover and the VIP and NFS server get reattached to the secondary node however, the mount becomes greyed out and won't remount onto the second NFS server. The only way of removing the datastore is to do a full reboot of the host. Logs are below showing what the host sees as happening when there is a disconnection from Node A:
2020-05-21T14:00:07.031Z: [vmfsCorrelator] 68314582298us: [vob.vmfs.nfs.server.disconnect] Lost connection to the server 192.168.X.X mount point nfs7,
mounted as d11f0a83-92231881-0000-000000000000 ("/nfssrv7")
2020-05-21T14:00:07.031Z: [APDCorrelator] 68318122525us: [esx.problem.storage.apd.start] Device or filesystem with identifier [d11f0a83-92231881-0000-
000000000000] has entered the All Paths Down state.
When the host is initially connected to the Node B and disconnects, the error messages are the exact same except the "filesystem identifier" is different (5f29bda0-173a3be1-0000-000000000000). I have checked and these identifiers remain the same when reconnecting the nodes, which proves that it is able to distinguish between the two servers and remember which is which. The NFS servers are designed to be impossible to differentiate for an NFS client, yet ESXi obviously does so. It is my impression that if those IDs are the same, the failover would be successful, but I haven't been able a way to do this. Is anyone aware of how ESXi generates the filesystem identifier for an NFS server, how to change the identifier for an NFS mount, or allow ESXi to ignore the identifier when remounting?