NFSv4 "Courteous Server" Lands As Part Of Linux 5.19 NFSD Changes

[phoronix]

Phoronix: NFSv4 "Courteous Server" Lands As Part Of Linux 5.19 NFSD Changes

The NFS server (NFSD) changes have been merged into the Linux 5.19 kernel and a new feature this cycle is supporting the NFSv4 "Courteous Server" functionality...

https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.19-NFSD-Courteous

[MarkG]

At home, I run two ZFS servers (1 with 8x SSDs, 1 with 4x SSDs) that use kernel-based NFSv4 over 10GbE. What I don't (yet?) "see" is if/how this will affect me. Courteuos sounds good. What do I need to know? Thanks in advance.

[skeevy420]

Does this mean I have to leave a 15% tip after doing a file transfer?

[Anux]

At home, I run two ZFS servers (1 with 8x SSDs, 1 with 4x SSDs) that use kernel-based NFSv4 over 10GbE. What I don't (yet?) "see" is if/how this will affect me. Courteuos sounds good. What do I need to know? Thanks in advance.

Hm, its explained in the article. What do you not understand there?

[grung]

Does this mean I have to leave a 15% tip after doing a file transfer?

No, it is oracle - you have to pay $10k ... Per physical core

[bfields]

At home, I run two ZFS servers (1 with 8x SSDs, 1 with 4x SSDs) that use kernel-based NFSv4 over 10GbE. What I don't (yet?) "see" is if/how this will affect me. Courteuos sounds good. What do I need to know? Thanks in advance.

It won't affect you until there's a problem that causes your client and server to lose contact--e.g. if your switch goes down for more than about a minute and a half, while applications are running that use locks, you may see IO errors after the switch comes back up. With these changes, recovery from such a situation should be more graceful.

[MarkG]

... What do you not understand there?

If I understood what I (apparently) don't understand, I'm sure I'd (eventually) figure it out.

[Anux]

Before patch:
Client copys file to server gets interupted and has 90 seconds to continue before the transfer gets broken.
After patch:
Client copys file to server gets interupted and has 90 seconds to continue or even 24 hours if no one else is accessing that file.

[SigHunter]

No, it is oracle - you have to pay $10k ... Per physical core

for all the hardware in your house

[S.Pam]

Before patch:
Client copys file to server gets interupted and has 90 seconds to continue before the transfer gets broken.
After patch:
Client copys file to server gets interupted and has 90 seconds to continue or even 24 hours if no one else is accessing that file.

There are timeout mount options to change the default value to many minutes. One should also use hard mounts if having a single file continuously open, such as vm images and databases.

Mind that with vm images, the guest is will have internal io timeouts. If the guest is running Linux this can also be extended through /sys.