[Building Sakai] Windows 8 Compatability & Load Balancing

[David Adams]

We had exactly this problem with our BIG-IP load balancer and trying to set actual expiration dates on the injected load-balancer-session cookies. I could never tie it exactly to a specific browser, although Windows seemed to have the problems more than Mac, and some browsers more than others. Overall, it only affected about 2% or fewer of our users, but that was enough to cause dozens of complaints every hour.

I definitely want a solution to this, and I've experimented with rules to manipulate the cookies on their way out, but I've been extremely cautious about this to try to avoid a recurrence of the problem. However, the lack of the ability to put an expiration date on the F5's injected cookie does cause a problem when we need to take a particular machine out of the pool, as lots of users never actually close their browser for days or weeks at a time, and so we could never drain off everyone.

To get around that, we have come up with a workaround of a new rule that injects another cookie with an expiration timestamp, and if that timestamp is in the past, the rule will blank out the load balancer cookie, which will then force the load balancer to select a new server from the pool. With this new rule we get a lot faster drains that we did before, though we still have problems with users who log out and back in within the cookie timeout windows sticking around longer than they should--any time a user logs out they should get assigned to a new server at that time as well.

I'm happy to share the code for this rule if you or anyone else is interested.

-dave

Kevin Pittman wrote:
> On Wed, Nov 14, 2012 at 03:30:37PM -0500, Kevin Pittman wrote:
> > With Windows 8, however, there doesn't seem to be any way to get either
> > browser to work.  Both exhibit symptoms that we've seen before when
> > cookies are disabled, except that I've verified on both browsers that
> > cookies are fully enabled.  Those symptoms are that the user tries to
>
> Just to close this out, I actually found the problem with a little help
> from our BIG-IP expert.  It turns out that our BIG-IP boxes do something
> non-standard in generating a session cookie with an expiration date, so
> we switched them to generate a standard session cookie that expires when
> the browser session ends, and all of the Windows 8 browsers are happy now.
>
> I'm still not sure why IE, Chrome, and Opera on Windows 8 react so badly
> to the malformed cookie, while on other Windows O/Ss they usually worked
> okay, and Chrome & Opera never had any trouble with the cookie on Macs.
> Whatever the case, at least I know the source of the problem now.
>
> Thanks,
> Kevin
> Georgia Tech Sakai Application Administrator
>
> P.S. It would still be interesting to hear what other schools are using
> for load balancing.
>
-- 
David Adams
Director, Learning Systems Integration and Support
Virginia Tech Learning Technologies