NFS behaviors

[mokaz]

Hi there all,

I've tried to research on that subject although came empty net.

How would VMs that are running on NFS storage behaves in case of "momentary" storage loss?
I'm trying to figure indeed if Proxmox would buffer IOPS for some times before dead'ing paths.

I bet the safest would be to shutdown each VMs of course.
I am in need to update a FW on my NFS provider, on which IOPS are on the low end really..

Let me know,
Kind regards,
Boris

 

[aaron]

Chances are that you end with a hanging NFS mount. So the best way to handle this would be to shut down the affected VMs and once the NFS server is back up, check if the network share is mounted correctly and is working. If that is not the case, you can unmount it on the PVE nodes with umount -f -l /mnt/pve/<nfs storage> and once the pvestatd service wants to check it again (usually within a few seconds), it should get mounted again.

 

[6uellerbpanda]

it depends

if you use async for NFS then you might get an filesystem/io error for the vm's and need to repair it with fsck at the next boot if it is recoverable - depending when dirty flush happens.
with sync param you might (flush) also get io errors but filesystem will be fine at next boot.

at least this is my experience with NFS storage and different sync params...

I always use sync io for NFS backend.

 

[mokaz]

Nice thanks a lot for the infos guys. I think I'll go for the safe thing, power down the VMs.
I'll sure check my NFS setup for sync params.

Cheers,
m.