As mentioned in the thread 

https://lore.kernel.org/linux-nfs/601285843.50695650.1748800817824.JavaMail.zimbra@desy.de/T/#u (local)


We observe that interrupted batch processing jobs put the client into an unrecoverable state that requires
the client host reboot. Finally, I was able to build a custom kernel with all required third-party drivers to prove
my assumption. So indeed, marking pNFS device unavailable fixes the issue. Thus, please consider the proposed
change and backport it to older kernels. I did testing with (which is not part of the patch) and will try to
add a trace point as soon as I find out how to implement one.

Tigran Mkrtchyan (1):
  pNFS/flexfiles: mark device unavailable on fatal connection error

 fs/nfs/flexfilelayout/flexfilelayoutdev.c | 4 ++++
 1 file changed, 4 insertions(+)

-- 
2.49.0