[OpenAFS] best way to debug windows 1.7.11 cache/rx/NAT problems?

Derrick Brashear shadow@gmail.com
Tue, 8 May 2012 14:27:41 -0400


On Sat, May 5, 2012 at 3:21 PM, Lars Schimmer <l.schimmer@cgv.tugraz.at> wr=
ote:
> Hi!
>
>
>
> We have apployed 1.7.11 on =F3ur Windows 7 x64 system.
>
> Now we still do have problems of clients not seeing the recent state of
> files.
>
>
>
> E.g. User change a file on machine 1, check it 10 sec later on machine 2,=
 he
> does not see the changes made on machine 1.
>
>
>
> Machine 1 is on the same subnet of servers, full routeable,
>
> Machine 2 is on a private subnet with a Cisco style NAT in between.
>
>
>
> Now I do see a lot of possibilities to debug the problem, but which way w=
ill
> find the best solution?
>
> E.G. could be the NAT issue, or cache problems or callbacks?

NAT issue with port expiring seems likely. assuming you have access to
the fileserver there are actually log messages for this condition:
we expect the client to be reachable and it is not. without looking at
source, you get a -1 error in RCallback, istr. might not be that, but
it's a -1 (timeout) because the port mapping expires and so until the
client talks to the fileserver again, nothing realizes the callback is
gone

the surprising thing is NAT ping should keep the port open unless
1) the timeout interval is aggressively short, like, 20 seconds.
2) there's a hard timeout on NAT at least for UDP where regardless of
use, you lose your timeout.

if you can recreate, tcpdump on the fileserver, ideally, to the IP of
the NAT will make it obvious what is happening.



--=20
Derrick