[393] in athena10

home help back first fref pref prev next nref lref last post

Re: Simulating an AFS outage

daemon@ATHENA.MIT.EDU (Bill Cattey)
Wed Aug 6 14:16:54 2008

From: Bill Cattey <wdc@MIT.EDU>
To: Jonathan D Reed <jdreed@mit.edu>
Cc: athena10@mit.edu
In-Reply-To: <Pine.LNX.4.64L.0808060922020.13701@infinite-loop.mit.edu>
Content-Type: text/plain
Date: Wed, 06 Aug 2008 14:16:06 -0400
Message-Id: <1218046566.16937.3.camel@localhost.localdomain>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

I agree with jdreed that getting failsafe to work correctly is very
important.  We're trying to eliminate work tasks by de-supporting the
"create login with homedir in /tmp" functionality by relying upon a
working failsafe login environment.

-- 
-Bill

----

William Cattey
Linux Platform Coordinator
MIT Information Services & Technology

N42-040M, 617-253-0140, wdc@mit.edu
http://web.mit.edu/wdc/www/


On Wed, 2008-08-06 at 09:41 -0400, Jonathan D Reed wrote:
> I've been attempting to test what happens in the event of the user's AFS 
> homedir not being available.  So far, it's not great:
> 
> After login, it will grind for a while, and then eventually report that my 
> home directory could not be found.  It then asks if I want to log in 
> anyway with root (/) as my homedir (!), but says things are unlikely to 
> work unless I select a failsafe session.  If I click "Yes", then get an 
> error about the ".dmrc" file being ignored.  Then, I get the "Your session 
> lasted less than 10 seconds" error.  Viewing xsession errors shows a bunch 
> of errors for being unable to create the standard Nautilus directories in 
> /mit/jdreed (Desktop, Templates, Public (namespace collision here?!), 
> Documents, Music, etc).  seahorse-agent complains via libgnomevfs that it 
> can't create /mit/jdreed/.gnome2, and then the session fails.
> 
> So, I try again, and select a failsafe session (which is not intuitive). 
> My xsession still only lasts 10 seconds, and gnome-session complains about 
> being unable to create ~/.gnome2.
> 
> I also tried to use the "Secure remote connection" gdm option, but that 
> doesn't actually seem to do anything (it behaves identically to me 
> attempting to log in normally).
> 
> N.B. Simulating an AFS outage was accomplished by using iptables to send 
> all packets from my homedir's file server to the DROP target.  If that's 
> not the correct way to do it, I'll happily retest.
> 
> I also tested this with my test account to ensure that there wasn't 
> anything weird with my account.
> 
> This kind of sucks.   Is there any better way we can deal with this that 
> doesn't involve heavy customization of the login code?  I'd like to see 
> one of the following:
> 
> - gdm says something like "Your AFS home directory is unavailable, please 
> try again later."
> - gdm offers to give the user a tty session to athena.dialup
> - Failsafe works correctly (see below)
> 
> And while we're at it, Failsafe itself seems a little broken.   While it 
> didn't run .startup.X, it still seems to run .environment and .cshrc.mine. 
> At least, after logging in with failsafe, and opening a Terminal window, 
> all my aliases (from .cshrc.mine) are there, along with some environment 
> variables I set in .environment.
> 
> -Jon
> 


home help back first fref pref prev next nref lref last post