LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: ipvs failback patch - still have failback problems

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Re: ipvs failback patch - still have failback problems
From: Horms <horms@xxxxxxxxxxxx>
Date: Wed, 30 Nov 2005 01:40:31 +0000 (UTC)
Ranga Nathan <kairanga@xxxxxxx> wrote:
> Horms:
> Due to many iterations I could not figure out what your last response was.

Sorry about thati, I think it was a missfire and didn't contain
any new information at all.

> I have now done everything that I am aware of. Bringing up both daemons 
> on both LDs with reversed sync ids and a delay in heartbeat of 30 
> seconds. The failback is still dropping connections. FYI, bother LDs run 
> almost identical SuSE 9.3

Ok, as I mentioned in an earlier post, there is no code that I am aware
of that forces the active to send all its connection data over to the
standby node. Which is basically what you need. Perhaps Karl has
information about such a patch. Perhaops it out there and I am just
missing it. But I suspect that he was just talking about the code that
was merged into 2.6 ages ago, and you are already running.

If you are worried about connections being dropped when your master
plays reboot games you have two options.

1) Find, make, or sweektalk someone else into making a patch that
   allows a syncd to request the other side to send all its data,
   and hopefully notify the requesting side when it is done.

   This will still be a bit racy, as very new connections on
   the active node might not have been synchronised when a failover 
   occurs. But this isn't any worse than the situation when an
   active node goes down, and should help your problem.

2) Make it so when a failover occurs, the node that takes over becomes
   the master, and stays the master until it fails. If you use
   heartbeat set auto_failback=off

-- 
Horms


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