Re: [PATCH RFC 0/2] SUNRPC: dispatch ready transports round-robin across clients

["Chuck Lever" <cel@kernel.org>]

The extra good numbers redacted for brevity ...

Also, thanks Daire for your user stories, those details are interesting.


On Tue, Oct 6, 2026, at 2:07 PM, Benjamin Coddington wrote:
> On the question in your second mail, an admin setting one table for
> more than one fairness behavior: the posted kernel keys on source
> address, so I can emulate "all movers are one class" today by giving
> the movers one address.

OK, that helps me understand the motivation for your earlier question
regarding whether I would accept such an approach. You were referring,
in particular, to the proposal to introduce a new netlink API for
creating tables that configure server fairness, after Neil, Jeff, and
I all rejected tuneables in June.

The design requirements have evolved to need some kind of configuration,
because a single NFS server is frequently deployed in environments
where several different workloads are active at once. The requirement
to protect fairness between NFSv3, NFSv4, and presumably LOCALIO
client workloads makes this challenging.

I'm not terribly excited at adding this configuration to exports.
Today, netgroups are defined separately, for instance, and exportfs
is just a consumer of those groups. Making it more like iptables,
controlled via nfsdctl, resonates with me more.

We also have the problem of using raw IP addresses, which is a)
not user interface-friendly (hostnames might be nicer); and b)
not entirely secure.


-- 
Chuck Lever (Come to NFS bake-a-thon! https://nfsv4bat.org)