Daniel Gryniewicz

NFS doesn't have auto-failover on the client side in the protocol. Instead, the server fails over, and the clients re-connect to the new server. It was designed originally (40+ years...

I do not believe this is possible with NFS. The issue is that some state is stored on the server, and some state is stored on the client, and those...

That's not a clean transition, it's a failure. The server failed, the client tore down it's connection, and created a new one. All outstanding changes were (potentially) lost, corrupting data...

The problem with this is that it assumes you wrote the client. If you're using a pre-existing client (such as the Linux kernel client), then it does what it does,...

The v4 stats cover v4.x. There aren't separate stats for 4.1 or 4.2.

There's no (interesting) errors in that log. Does this log include the failed mount? If so, can you turn on FULL_DEBUG and try again?

I'm not a proxy expert, but it looks like responses from AWS are timing out. @boulos any ideas?

It would be possible, if someone is interested, to add some kind of mapping in FSAL_VFS between Ganesha handles and underlying FS handles. This would allow arbitrary sized FS handles....

You're right, this seems like a good QoL patch.

That's what Grace is for, to force clients to reclaim locks.