LVS
lvs-devel
Google
 
Web LinuxVirtualServer.org

Re: Multiple load balancers problem

To: Hans Schillstrom <hans@xxxxxxxxxxxxxxx>
Subject: Re: Multiple load balancers problem
Cc: Julian Anastasov <ja@xxxxxx>, lvs-devel@xxxxxxxxxxxxxxx
From: Dmitry Akindinov <dimak@xxxxxxxxxxx>
Date: Mon, 03 Sep 2012 11:54:52 +0400
Hello,

On 2012-08-31 12:21, Hans Schillstrom wrote:

Hello,

On 2012-08-28 00:43, Hans Schillstrom wrote:
Hello

[snip]


        No loop? Because S2 will receive DR method in
the SYNC messages for its stack. You see traffic flow
or just that there is no reset?

Yes, after the failover, all connections that were open on the S2 (which
now becomes the active balancer) do not reset and continue to function
just fine (traffic flow in both directions).

We do understand (we think) your explanation about the sync table
problem  - the connections to the actual balancer are marked in a
special way, which causes problems when these connection table records
are used on a new balancer.

It looks like updating the kernels is the only way, if (as you and Hans
Schillstrom outlines) even the latest CentOS/RedHat kernels do not
contain necessary patches. It's a pity, as the idea was to provide an
"out of the box" solution for our customers, and asking them to update
to some Linux kernel is not what they like to hear.

But thank you very much in any case, we will update the kernels on our
test systems and will see if this (last?) problem disappears.


There is newer kernels floating around in the 3.x range
Have a look at ELRepo Project there you can find a 3.5.x kernel
that have all necessary changes.

Installed CentOS 6.3 and the kernel 3.5.3 from elrepo.org

[root@fe3 ~]# uname -a
Linux fe3.msk 3.5.3-1.el6.elrepo.x86_64 #1 SMP Sun Aug 26 14:05:15 EDT
2012 x86_64 x86_64 x86_64 GNU/Linux


But we still see the same problem: connections to the old balancer are
RSET'ed when the new balancer takes over. Any idea which particular
kernel we need to use? Or, do we need to apply some specific patches and
build our own kernel?

A silly question,  old and new in this case
I hope it's a 3.5.x kernel in both cases ...

Yes, sure. :-) The kernels were upgraded on both frontends.
And in fact, that helped. My previous report that the problem was still observed with the new kernel - well, it was not correct. The test environment on virtual machines was not set up correctly.

We will do more extensive testing this week, but for now it appears that problems we reported initially have gone when we upgraded the kernel to 3.5.3.

Thank you!

[]

--
Best regards,
Dmitry Akindinov


--
To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html

<Prev in Thread] Current Thread [Next in Thread>
  • Re: Multiple load balancers problem, Dmitry Akindinov <=