Accounting badly broken in v1.0.11

[wlan at mac.com]

If you are having issues with the current release, it is probably  
best to try the svn version.

I can't imagine why chilli would be sending out odd RADIUS half way  
during a session. Is it perhaps not getting any Accounting-Responses  
coming back (and hence you see the retries)?

Accounting-On/Off is a new(er) feature and is per RFC - useful for  
clearing out stale sessions when you know chilli rebooted.

David

On Apr 17, 2008, at 11:42 PM, Gunther Mayer wrote:

> Hi guys,
>
> I just wanted to share my experiences with v1.0.11 with everyone  
> thinking of upgrading to that "latest" release. At first I noticed  
> our radius/db accounting entries got all messed up for some  
> sessions and when I started investigating in more detail I stumbled  
> across a whack load of spurious radius packets that chilli sends to  
> the radius server halfway through the session or even after it ended:
> replays of entire previous sessions (from accounting start to all  
> interim updates, sometimes even all the way to accounting stop)
> replay of current session (from accounting start to zero or more  
> interim updates)
> accounting-on packets
> This can get really hairy to try and work around in the radius  
> server code and is giving me grey hair. It seems like quite a  
> serious bug to me but others here may have more robust radius code  
> than mine so perhaps haven't noticed this yet.
>
> David if this is all new to you I'll gladly follow up with a whole  
> bunch of tcpdump's that I've prepared which isolate the problems.  
> Reproducing this is fairly easy too, just fire up a session and  
> watch the radius packets that flow past...
>
> Has anyone been bitten by this too? Has anyone worked around this  
> successfully? Has anyone experienced something similar at least in  
> post svn r147 (1.0.11)?
>
> I'm stuck here trying to release an upgrade to our system which  
> involves JSON (I'm currently based on 1.0.6 just before JSON got  
> introduced) but clearly cannot use latest release, am a bit wary of  
> using bleeding edge trunk in production but also don't want to  
> successively test each earlier release (1.0.10 has known issues,  
> 1.0.7-9 anyone?) in the light of this.
>
> Gunther