Simon Horman a écrit :
> On Tue, Apr 28, 2009 at 09:43:18AM +0200, Fabien Duchêne wrote:
>> Hello Simon,
> Hi Fabien,
>> I made the same fix and tested it for 4 days, it seem to be stable.
> Thanks, that is very useful information.
For information, I used slamd to inject HTTP/S and LDAP load.
I tested with and without the fix, no any significant difference (20 tests).
> Does this ensure that the trailing 3 octets of .all and .ip6 are 0?
No, you're right, they are uninitialized, and this shouldn't be the same.
In fact, I thought about using all (so exactly your solution), but
when I saw ->ip working I stopped (lazy man, I know).
Your solution is much cleaner, I will test it ASAP.
> I haven't run any tests to confirm the problem, but I am a Dev,
> so I can get it fixed :-)
> I am a little concerned that this fix may not be correct in
> some bizzare case where the packing means that .ip, .ip6 and
> .all are not the same thing.
> I have sent an email to lvs-devel and netfilter-devel to try
> and get some discussion on this. The subject of that email is
> "[bug] FWMARKs and persistence in IPVS: The Use of Unions".
I just subscribe to this list to see what's going on..
Btw, thanks for your help!
Please read the documentation before posting - it's available at:
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users