No probe if arp failed

patrik.spiess · ‎07-11-2011

Hi All

We use an ACE-Module [build 3.0(0)A2(3.2)] in our Datacenter.

Today I recognized a useless behaviour on this ACE (in my eyes).

I configured a serverfarm with 4 rservers. 1 is up an 3 of them are not yet connected. So if I issue 'sh serverfarm xyz' three of the rservers are in the state 'ARP_FAILED' which is true at this moment.

rserver: RSERVER_SMTP02

[ipadress]:0 8 ARP_FAILED 0 0 0

But if I issue 'sh logging' I see a lot of failed probes for theese rservers.

Jul 11 2011 16:05:51 ZONE1: %ACE-6-251008: Health probe failed for server abc on port 25, connectivity error: server open timeout (no SYN ACK)

Does this make sense?

Wyh does the ACE send probes to ARP_FAILED rservers?

Is it possible to stop the probing of ARP_FAILED rservers? (without manually issuing 'no inservice')

Thanks for your ideas

Patrik

Nicolas Fournier · ‎07-11-2011

Hi Patrick,

If you don't define any probe towards a rserver, the ACE will still mark it as down within a serverfarm if he is unable to ARP form it (or for the next-hop towards it). This can be seen as some sort of default probe.

Now regarding your questions:

Why does the ACE send probes to ARP_FAILED rservers?

It doesn't send probe but it tries to. If the server is still in ARP_FAILED, this will trigger an ARP request for the rserver IP and it will indicate to the ACE if the server was introduced into the network in case of success. If we didn't use the probes to detect this, we should still ARP at regular intervals or wait for the rserver to send a gratuitous ARP to detect it is now there.

Is it possible to stop the probing of ARP_FAILED rservers? (without manually issuing 'no inservice')

The only way I see would be to disable the probe on the serverfarm config and to rely on ARP only to verify the server health.

Regards,

Nicolas