Daniel Gryniewicz

So, the recovery backend for Ceph is designed to be used in a containerized environment, such as kuberneties, where the crashed Ganesha can be restarted quickly on the cluster, and...

You can, I believe, start the new ganesha before moving the IP. That way clients won't fail to connect. The first ganesha needs to be gone, though, or I think...

Did pacemaker put the target Ganesha into Grace before it moved the IP? Do the two ganesha instances share a recovery backend? Both of these are necessary for failover to...

Okay, so if NFSv4 { Graceless = True; } is not set on the target, then it should be entering grace on startup, and should start recovery. At that point,...

If you only need it for a one-off problem, then you can always go back and get an old version of Ganesha and try to build that. I don't know...

As far as I know, no one is working on RDMA support. Periodically, someone will ask about it, but no one seems to want to work on it.

That's a POSIX semantic, not an NFS semantic, so ganesha shouldn't do it. If the underlying filesystem is POSIX (as FSAL_VFS filesystems are), then that will take care of it....

I think Frank covered it. The handle also has a metadata cache in it, which can have a different lifetime than the handle itself, but that shouldn't affect these operations....

There's not enough information in that message, all it tells us is that it crashed somewhere. We'd need a backtrace with symbols to be able to tell.

Debugging in containers is hard, as I understand it. I'd start by installing all the debug packages inside the container, and reproduce, and see if that includes symbols and line...