Abysmal IO delay inside containers

[inDane]

Dear Proxmoxers,

i have two boxes, one is running truenas with ZFS RaidZ1 and exporting via NFS.
The other one is running Proxmox 7.2. (pve-host)

If I am running i.e.
fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --bs=4k --iodepth=64 --readwrite=randrw --rwmixread=75 --size=4G --filename=./testfile.fio

on the pve-host, on the nfs share, everything is fine. the IO delay wont spike much <15%. But if I am using an ubuntu container with a raw disk and use the above command, it will immediately spike to 99% and it will eventually timeout the share and everything is hanging.

I seem to be missing something significant here, can you give me a hint on this?

EDIT: Writing IO seems to be the problem. Reading is just fine.

Best
inDane

 

[inDane]

i can reliably crash my system if i put IO load inside a container that has its image as raw on the nfs share...

 

[ness1602]

You can limit BW on a container.

 

[inDane]

You can limit BW on a container.

on the network device you mean? How should that affect the underlying NFS?

Client -> SSH -> Container -> fio-command -> IOs towards NFS.

If you are talking about the network device, then it would only affect the "client -> ssh" part.