[OpenAFS-devel] Soft mounts for /afs root

Harald Barth haba@pdc.kth.se
Thu, 14 Jun 2001 18:42:25 +0200


> Is this learning cells "on the fly" explicit, via some command that 
> tells the client that there should now be an entry in /afs for 
> /afs/some-new-cell or implicit, such that a readdir of /afs may show 
> nothing yet accessing /afs/my-favorite-cell both works and creates 
> /afs/my-favorite-cell?

In my current test code it is implicit, you start with the cells that
you have in CellServDB and then additional cells populate when you
access the /afs/your-favourite-cell. Yet another approach is to show
the new cells in /afs as soon as someone uses the new cell somehow on
the client. This might be through a moutpoint in the allready visible
part of the afs space or by "afslog -c some-new-cell" Of course you
can make a command for that, too. But the implicit way is more
surprising, so naturally I had to test that first. I have not tested
it on unsuspicious users, yet ;-)

> I have a lot of trouble picturing how the implicit approach would 
> make sense to a user.

I can imagine that this might be troublesome to some users and support
people alike, but if you don't run "fs newcell" on all your hosts when
CellServDB changes, different hosts will have read different
CellServDB at boot anyway and the view in /afs will look different.
This feature might not be useful to everyone, but I think it is very
useful on my laptop where I should know what is going on:

habariff:/usr/src/arla$ ls /afs/        
#dynrootcell#     arla.e.kth.se  isk.kth.se   nada.kth.se   stacken.kth.se
admin.kth.se      e.kth.se       it.kth.se    pdc.kth.se    telge.kth.se
alfvenlab.kth.se  extundo.com    mech.kth.se  sipb.mit.edu
habariff:/usr/src/arla$ ls /afs/club.cc.cmu.edu
Public  andre       archive  data     public   system  user  www
admin   andre-beta  common   lazarus  service  tmp     usr
habariff:/usr/src/arla$ ls /afs/
#dynrootcell#     arla.e.kth.se    extundo.com  mech.kth.se  sipb.mit.edu
admin.kth.se      club.cc.cmu.edu  isk.kth.se   nada.kth.se  stacken.kth.se
alfvenlab.kth.se  e.kth.se         it.kth.se    pdc.kth.se   telge.kth.se

Meanwhile on the console:

2001-06-14 18:12:25: arlad: Inserted cell club.cc.cmu.edu with id 14

Together with a gui to add new cells to /afs, this might be useful to
users of broken file managers which are prone to stat() everything in
their way and thus hanging while stat()ing every entry in /afs. Such a
gui is included the the latest release of Arla for MaxOS X. That one
is still very explicit (You can't go beyond the cells mentioned in
CellServDB).

Harald.