Harald, you reported many different things, and I'm a bit lost here. Let's pick one issue.
> After a nfs-server restart, rpc-svcgssd.service - RPC security service for NFS server isn’t
> started again
Why isn't it running? What does status say?
About service restarts, with so many individual pieces it can get complicated how to properly restart the service. I wrote this table in the server guide [1], in the "Restarting NFS services" section.
> @Andreas Hasenack (ahasenack):
> the rpc_pipefs error is always present, nfs runs anyway
What do you mean? The rpc_pipefs is an absolute requirement, and there are two mount units for it: var-lib-nfs-rpc_pipefs.mount and the generated one which mounts it in /run when the nfs.conf has a path different than the built-in default. One of the two mount units must have been ran.
> nfsconf --dump # working system,
and
> # failing system /etc/nfs.conf
you used nfsconf --dump in one case, and straight /etc/nfs.conf in the other, it's not clear if the failing system has some override in a /etc/nfs.conf.d/*.conf snippet.
For blkmap, a core dump is definitely a bug, it should error clearly when it doesn't find the rpc_pipefs mountpoint instead of crashing, but fixing that looks like will just switch the error from a core dump to a "sorry, couldn't find rpc_pipefs", and it still won't run.
Harald, you reported many different things, and I'm a bit lost here. Let's pick one issue.
> After a nfs-server restart, rpc-svcgssd.service - RPC security service for NFS server isn’t
> started again
Why isn't it running? What does status say?
About service restarts, with so many individual pieces it can get complicated how to properly restart the service. I wrote this table in the server guide [1], in the "Restarting NFS services" section.
> @Andreas Hasenack (ahasenack):
> the rpc_pipefs error is always present, nfs runs anyway
What do you mean? The rpc_pipefs is an absolute requirement, and there are two mount units for it: var-lib- nfs-rpc_ pipefs. mount and the generated one which mounts it in /run when the nfs.conf has a path different than the built-in default. One of the two mount units must have been ran.
> nfsconf --dump # working system,
and
> # failing system /etc/nfs.conf
you used nfsconf --dump in one case, and straight /etc/nfs.conf in the other, it's not clear if the failing system has some override in a /etc/nfs. conf.d/ *.conf snippet.
For blkmap, a core dump is definitely a bug, it should error clearly when it doesn't find the rpc_pipefs mountpoint instead of crashing, but fixing that looks like will just switch the error from a core dump to a "sorry, couldn't find rpc_pipefs", and it still won't run.
1. https:/ /ubuntu. com/server/ docs/service- nfs