Re: [lvs-users] two box lvs with keepalived: the backup gets horrible sl

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Re: [lvs-users] two box lvs with keepalived: the backup gets horrible slow
From: svensven <svensven@xxxxxxxxx>
Date: Sun, 28 Mar 2010 17:09:37 +0200
On 2010-03-28 00:44, Simon Horman wrote:
> On Thu, Mar 25, 2010 at 05:17:15PM +0100, Harald Dunkel wrote:
>> I am trying to setup a two box lvs as described in section 9.3 of
>> the howto (the version with a single nic and 2 marks). For testing
>> I have setup a telnet service.
>> The first telnet session picks up the master, the second telnet is
>> sent to the backup, as expected. Problem: If the telnet session to
>> the backup is terminated, then the backup gets horrible slow. The
>> master is as fast as usual (as it seems).
> Strange. What exactly do you mean by fast and slow?

Could it be it's the same issue as in [1], where master node will sync
connection state to the backup after n packets (default 3 incoming),
and the backup then thinking that the following packets should be
handled by ipvs, while they actually should not? This killed my
virtual test machine, at least -- it got so busy with looping the
packet and writing logs, that all connections to it timed out.

Depending on how Harald tests this, it could be that the packet sent
to terminate the connection is actually the 3rd incoming packet,
assuming he simply does a connect, then disconnect.


[1]: Connection sync breaks fwmark-based localnode setup

Please read the documentation before posting - it's available at: mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to

<Prev in Thread] Current Thread [Next in Thread>