lbaas backup member flag

[Kevin Fox]

Haproxy supports a flag on members to tell it not to ever use the member unless all others have failed. This allows you to setup apology servers. I don't see any way to set this flag using the neutron lbaas member settings.

[Eugene Nikanorov (enikanorov)]

Can you tell if such feature is supported by other load balancing solutions?
If so, it might be worth introducing it into LBaaS member API in the scope of some blueprint.

LBaaS API currently is a very basic and we would like to extend it with generic (e.g. supported by many vendors) features.

However there are plans to make LBaaS api extensible in more flexible way when such feature as you are proposing could be added easily.
Meanwhile I'll mark this as wishlist.

[Kevin Fox (kevin-fox-y)]

It looks like a common feature. A quick search shows the following links

Cisco seems to support it:
http://www.cisco.com/en/US/docs/app_ntwk_services/data_center_app_services/ace_appliances/vA3_1_0/configuration/slb/guide/classlb.html#wp1306327

IBM:
http://pic.dhe.ibm.com/infocenter/wasinfo/v7r0/index.jsp?topic=%2Fcom.ibm.websphere.edge.doc%2Flb%2Ftprf_rules.html

Some more references:
http://kemptechnologies.com/en/faq/faq-8-how-do-i-setup-sorry-server-lm
http://www.badllama.com/content/lvs-load-balancing-cluster

[Eugene Nikanorov (enikanorov)]

Ok, I think this could be added someday when corresponding vendors will have interest in integrating their solutions into Neutron LBaaS.
Or you could add this for Haproxy driver as soon as lbaas extension framework is ready: https://blueprints.launchpad.net/neutron/+spec/lbaas-extensions

[Micheal Thompson (mthompson-n)]

Eugene,

This would be a common feature in the ADC space.

Typical cases are:

1. All members have failed and as a result sorry page is implemented.
2. A given number of members fail and as a result capacity is not adequate issues sorry page.

[John Dewey (retr0h)]

FYI, the Cisco ACE is set for end-of-life in 2015.

We need to determine a clean user-facing API to support this. Most vendors should understand the notion of a backup pool, and leave it up to the vendor to implement the appropriate way.

Netscaler would likely do the bulk of the work in the Netscaler Control Center which sits between the LBaaS API and the Netscaler(s).

I have implemented this very feature with F5 BigIP in the past. It required another backup pool be defined, and performing the selection via an iRule. This is all do-able though the iControl API [1].

[1] https://gist.github.com/retr0h/9925465