svn commit: r331880 - stable/11/etc

[Rodney W. Grimes]

> On 04/10/18 04:28, Kyle Evans wrote:
> > On Mon, Apr 9, 2018 at 11:59 AM, Warner Losh <imp at bsdimp.com> wrote:
> >>
> >>
> >> On Mon, Apr 9, 2018 at 10:09 AM, Kyle Evans <kevans at freebsd.org> wrote:
> >>>
> >>> Right- so, back out this MFC (and the subsequent FreeBSD_version bump)
> >>> and fix the ports to do the right thing for 12.x while that's still
> >>> not a technically supported branch?
> >>
> >>
> >> Don't back out the version bump. Other things may be riding along on it 'for
> >> free'. Better to bump it again when you unMFC (if it's been more than a few
> >> days since we've had one), and then yet-again when a fixed MFC happens.
> >> Unless there's something you can ride along on for free :)
> >>
> >> Otherwise, that's a great plan.
> > 
> > Ok, I think the result of this thread and discussion with 0mp is the
> > following set of actions:
> > 
> > 1.) One (1) commit to stable/11 to revert the MFC and bump
> > FreeBSD_Version again for the removal
> > 2.) One (1) commit to doc to document the new FreeBSD_Version
> > 3.) Fixing ports to use the "new" behavior on 12, both the
> > yet-to-be-patched ports and the ports that had already been patched
> > under the assumption that it would still land first in 11.1-stable
> > 4.) Documenting the original commit?
> > 
> > The hard part of point #3 has already been done by 0mp, who has
> > submitted patches for all of the ports using this behavior. His
> > patches will just need a bump of the version they're testing to the
> > 12.x FreeBSD_Version and a fix-up on the patches that already landed.
> > 
> > For point #4, this seems like the type of breakage we should be
> > documenting in release notes or something for the eventual upgrading
> > of systems to 12.0. All usage of _limits stuff in custom rc scripts
> > need to be audited, and all rc.conf(5)'s need to be scrubbed for
> > ${name}_limits usage that doesn't make sense with the new context. I'm
> > not sure what the most appropriate action here is, or what we should
> > do this far ahead of time for such a thing.
> > 
> > If this sounds like a good path forward, I'll execute #1 and #2 in the
> > morning (CST, so ~11 hours from this e-mail being sent).
> > 
> 
> This still doesn't fix the issue of some early start up scripts 
> depending on stuff that's not available yet, when for instance /usr is 
> on a separate FS (which was the normal way to set up a system way back 
> when).

The words "way back when" is become less needed.  With the advent of
hypervisors, 100GbE and NVMe over fabric the idea of sharing /usr across
a whole cluster of VM's is become more desireable.   1 copy of /usr
shared from a memory file system via fast network technologies has
advantages.

> This issue was first noticed more than 2 years ago, so someone did 
> notice the breakage.  It just hasn't been fixed for an entire release cycle.
> Regards

One thing that is not helping the bit rot is that some reworking of
/etc/rc* stuff has caused it to now ignore most errors from my things,
and that just causes a message on the console that no one is reading
so when something gets broke it a) doesnt get noticed cause thier
machine just boots, and b) doesnt get fixed cause few are actually
effected by the caused error.

If you are willing to help me weed out the failure modes, I am willing
to try and mount an effort, along with others, to try and get us back
on track.

-- 
Rod Grimes                                                 rgrimes at freebsd.org