[Bug 284616] nfsd thread count auto-tune do not respect MAXNFSDCNT

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=284616

Rick Macklem <rmacklem@FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rmacklem@FreeBSD.org

--- Comment #3 from Rick Macklem <rmacklem@FreeBSD.org> ---
(In reply to florian.millet from comment #0)
A few comments.

I would consider this behaviour (ignoring MAXNFSDCNT)
a feature and not a bug.  MAXNFSDCNT is cruft left
over from decades ago when nfsd kernel threads were
processes.
You'll notice some ancient cruft where children[MAXNFSCNT]
is defined, but only children[0] is actually used.
(That cruft should be cleared out, but it is harmless.)

You'll also notice that MAXNFSDCNT is defined as 256 and I
am pretty sure that there are servers that want to use ore
than that limit now (and have been able to do so for more
than a decade, due to the commit you mention.
To change that now would be a POLA violation, imho.

Why don't you just use the #maxthreads option to set the
limit to what you need?
(If this doesn't work, that is a bug that needs to be fixed.)

I'll admit that "man nfsd" is misleading, in that it states
that "-n" is the equivalent of setting both #minthreads and
#maxthreads to the same value.
This is inaccurate, as you note. It should say that "-n" is
deprecated in favor of #minthreads, #maxthreads.

Note that having too many kernel threads does not impact the
system much. An inactive kernel thread mostly just uses a
few pages for its kernel stack. (Minimal comapred to the
"processes" used decades ago.)

I will ask on freebsd-current@freebsd.org to see what others
think w.r.t. this.

-- 
You are receiving this mail because:
You are the assignee for the bug.