Failover DNS Replies

[dan.shalinsky]

Hello:

We have 2 CSS's in a geographic failover configuration. Traffic generally goes only to our primary location and would only failover to the secondary location if any services at our primary location go down. We are using the APP session btw the CSS's.

The issue we're seeing (it may not be a new issue) is that as soon as the service hits the dying state, our primary CSS starts giving out the secondary IP address. My understanding was that this should only occur once the service was actually down, but not during the dying state.

Is this normal expected behaviour or is it configuration?? We have just upgraded to 6.10.405 code.

Thanks and regards,

Dan

[Gilles Dufour]

Dan,

how did you configure this ?

Did you simply configure a dns domain under the content rule [old version] or did you configure a dns-record [new and best way] ?

With the dns-record solution you can configure dns keepalive.

If you are using this method, what keepalive are you using ?

Regards,

Gilles.

[dan.shalinsky]

Hi Gilles:

I think we're using the old version as this config is from the 5.02 days.

owner XX

dns both

content rule1

vip address xx.xx.xx.xx

add service serv1

dnsbalance preferlocal

add dns www.domain.com 5

active

Am I correct that the IP specified in the dns-record should be the VIP of the content rule?

I already tried changing the TTL using the old add dns command but it seemed to have no effect. Does this TTL behave differently? We always have used a 5 second TTL.

Further, what concerns me most is that a rule is being considered down only because the services are "dying". Can you confirm if my understanding of a dying service is correct: Traffic is still routed to a service that is dying and is only re-routed once the service is actually down.

Regards,

Dan

[dan.shalinsky]

Hi:

I now understand what the problem is. We're using the "dnsbalance preferlocal" command on our primary CSS's content rules.

The definition of this command is to reply with the VIP of the remote CSS's rule if all local services exceed their load threshold. This must occur by default when a service misses a keepalive and is in the DYING state.

So, is there any way around this behaviour with the service only DYING and not actually DOWN? Ideally, we'd only like the remote VIP replied once all local services are DOWN, not just DYING.

Any help appreciated.

Regards,

~Dan