816889 - IPv6 not consistently used for the same site

Status: RESOLVED DUPLICATE of bug 725587

(Core :: Networking, defect)

Description

[Wolfgang Rosenauer [:wolfiR]]

I'm trying to explain as exactly as possible what I observe since some weeks (unfortunately I cannot tell when it started and therefore what changed to make it break for me).
Using Firefox 18.0b1 at the moment.

I'm on openSUSE 12.2 and my home network is connected via a SIXXS tunnel to the IPv6 network. The involved webserver is natively connected to IPv6.
This worked for quite some time but at some point it began to break for my webmail access.

The webmail software I'm using creates some traffic during login to get all the JS code, icons and data into Firefox but after quite some data transfer I get immediately logged out of the webmail session just when it really opened. When trying to find out why (I control the webmail app on the server) I found that my IP address is changing and therefore the session management decides that this is breakin attempt closing the session.

Looking at the logs of the apache instance shows that a single request is performed via IPv4 instead of IPv6 which logs me out of the session because it's an IP change.
I have no real idea what to check why that happens.
It's not always reproducible but pretty often so I might be able to deliver debug data.

I usually only use Firefox but have tested with chromium a few times now and never hit that issue so I just guess it might be related to Firefox.

Comment 1

[Matthias Versen [:Matti]]

We have implemented http://en.wikipedia.org/wiki/Happy_Eyeballs.
You could try to disable it via "network.http.fast-fallback-to-IPv4"

Comment 2

[Wolfgang Rosenauer [:wolfiR]]

Disabled it now and will run with that for some time to see if it really stopped. First tests were successful.

But that sounds very dangerous to me in principle!
Dualstack is or actually will get very common during the transition phase and there are most likely a lot of webapps out there binding a session to an ip address.
Is there a concept to mitigate this?

Comment 3

[Matthias Versen [:Matti]]

dupe of bug 725587 (but our Happy Eyeballs implementation changed in the meantime) ?

Comment 4

[Honza Bambas (:mayhemer)]

(In reply to Matthias Versen (Matti) from comment #3)
> dupe of bug 725587 (but our Happy Eyeballs implementation changed in the
> meantime) ?

I think this is dup of that bug, yes.