[OpenAFS] [1.2.7] Strange file server meltdown

[Nickolai Zeldovich]

Russ Allbery <rra@stanford.edu> wrote:

> > ...  On my own client builds, I use this change that makes things
> > time out much faster:
> 
> This seems like an extremely useful modification.  Currently, we've been
> using the prebuild clients.  Is this the sort of modification that would
> make sense to go into the standard build, or should we be looking at
> switching over to building our own clients?

It does make sense to include something along these lines into the
release.  Unfortunately, doing it right requires a fair amount of
work, so it might be a bit of time until it's done.  In particular,
the hard timeout isn't really appropriate, so instead an idle dead
time needs to be implemented for clients (pretty easy, I have that
done).  The AFS client connection/retry management code also needs
to be changed to recognize the difference between different types
of timeouts (eg you don't want to time out all calls to a server
just because one volume is busy and a call to it timed out).

-- kolya