Keepalived vrrp problem

[Sal Tepedino]

On Fri, 2007-02-16 at 00:07 +0100, Shaun Mccullagh wrote:
> Couple of suggestions:
>  
> You don't need /usr/local/bin/keepalived-transition.sh. Keepalive adds 10.1.1.110/24 on the node when it transistions to Master by default. The notify_* scripts are intended for other things.

The transition script is because this is a localnode configuration. It's
more for moving the .110 address on and off the loopback, and to take it
off of eth1 when the system transitions to backup (as I noticed it was
hanging around) and to clear it out when the system is shut down (also
because it was hanging around). Keepalive adds the address to eth1 fine,
but it doesn't seem to handle the localhost config, as far as I can
tell.

> Try increasing advert_int to 5 and garp_master_delay  to 5

Tried this (and did it again just now, just to be sure). This only
causes the backup to transition to master slower, but it still does it.

> Also AH authentication does not work well with sync_groups, use auth_type PASS. 

Tried this also. (and again just now, just to be sure). I tried AH,
PASS, and no auth, and the results are the same.
 
> In fact, in your config you do not need a sync_group as you only have one vrrp_instance

I figured this, but I wasn't sure if keepalived was doing connection
syncing between the master and backup without this. I'll leave it out
till I figure this problem out and can test one way or the other. I
know, it's not supposed to, but I'm fairly sure when I set an LVS up a
few months ago (non-localnode) that I needed this statement for the
directors to fail over the connection info, but I don't have the config
to look at now. I know, it's supposed to be so the two instances
failover together. I was just being thorough ;)

 
> I would like to reaasure you that the latest few releases of KL (and even before) are very stable and reliable

Oh, I know. I had no problems with it the last time I set it up. I'm
guessing that it's the localnode configuration that's causing problems,
but others have said they set it up this way, so I'm fairly sure it's
possible. I've done localnode configs with ldirectord/heartbeat, but not
with keepalived. 

-- 
Sal Tepedino <stepedino at bhsoftware.com>