Freenas 11.1-U4 - NFS async mode?

[Sasayaki]

Hi all,

I'm running FreeNAS 11.1-U4, with a Rancher VM and a host of containers. Containers have access to the filesystem below through NFS, accessed through the Rancher-NFS plugin.

I've noticed that under heavy load with regard to disk activity, the whole system grinds to a snail's pace. For example, when users upload files to Nextcloud while streaming instructional videos by Plex, the whole system dies, even though CPU and RAM usage are extremely comfortable (40% CPU use, 3gb RAM free) and disk usage is fine, as files can be accessed through SMB at a good speed. As this wasn't a problem at all when things were in jails, and again Samba seems to work, I strongly suspect this is due to NFS being in synchronous mode.

How do I set FreeNAS to share NFS shares in async mode? For Rancher-NFS, it's just a 'async' in the mount options, but how do I make FreeNAS share things asyncronously?

Alternatively, is there any other explanation for why this happens?

 

[KrisBee]

NFS sync writes are a suspect, see https://forums.freenas.org/index.php?threads/testing-the-benefits-of-slog-using-a-ram-disk.56561/

Some options discussed here: https://forums.freenas.org/index.php?threads/nfs-share-async.63215/

I've forgotten, does rancher-nfs driver allow nfsv4 mounts? nfsv4 is supposed to be faster then nfsv3. But I'd think internal network speeds and sync writes would be the dominant factors. Client mounting async and server operating in sync mode is the default by design for NFS.

 

[Sasayaki]

I've forgotten, does rancher-nfs driver allow nfsv4 mounts? nfsv4 is supposed to be faster then nfsv3. But I'd think internal network speeds and sync writes would be the dominant factors. Client mounting async and server operating in sync mode is the default by design for NFS.

Yes, Rancher-NFS allows NSFV4 mounts, and I was running them, but I actually just now turned them back to V3 and it seems to have made a lot of difference (along with including the 'async' client parameter).

I'm reluctant to call this fixed, but it certainly seems a lot better!