[OpenAFS-devel] Re: Breaking callbacks on unlink

[Russ Allbery]

chas williams - CONTRACTOR <chas@cmf.nrl.navy.mil> writes:

> unfortunately afs meets your first requirement -- afs works -- most of
> the time.

Not for me.  This is a reality check.  There's a reason why people are
putting *so much work* into 1.6.1 as opposed to working on longer-term
priorities and improvements.  It's because 1.6.0 *doesn't work* for a lot
of people, and 1.4.14 isn't much better.

> the "breaking callbacks on unlink" is yet another drop in this bucket of
> non-deterministic behavior.  as andrew said, a user asked him why it
> sometimes worked one way and sometimes another.  while not as serious as
> the idledead problem, it still leads to questions in people's minds.

See, here's the thing: no one is suffering production outages from this.
Yes, it's one of the many things, like the lack of cross-directory hard
links, lack of device file support, and per-file ACLs, that is non-POSIX
and weird about AFS.  But it's how AFS has worked for forever, and all of
us with existing infrastructures deployed on AFS are not suffering from
this problem.  So this is simply not a problem of the type that I'm
talking about.

If AFS were working for me, then discussing what to do in corner cases
where historic AFS behavior diverges from what happens on other file
systems is possibly an interesting thing to focus on.  Improving things
like this is a way to grow the AFS community.  I get that.  But that isn't
the situation that we're in right now.  The software has to work within
its own set of semantics first.

So, from my perspective, doing anything about this problem puts me into
one of two cases: either an AFS semantic that has been there for as long
as Stanford has been running AFS changes on me and I have to wonder if
anything I don't know about is going to break and trust people who say
that it shouldn't matter, when I already have serious reliability and
stability issues, or the source gains a new option.  And a new option,
particularly this sort of option, might improve life for the one person
who asked for this to be configurable (if anyone's really asking), while
making life a little bit worse and a little bit more complicated and a
little bit less reliable for everyone else who's using OpenAFS.

So no, this is not something I want.

-- 
Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>