Neutron-LBaaS v2: mapping HAproxy process to cpu core

[Bartek Żurawski]

HAproxy in default configuration is spawned on one CPU core. This could be changed by adding 'nbproc' option which provide possibility to chose how many cores should be used by HAproxy and which frontend/backend should use more than one core. That kind of configuration will provide better performance to load balancer.

[Michael Johnson (johnsom)]

Hi Bartek,

With each release of haproxy we evaluate the improvements in the multi-process capabilities of HAProxy.
Enabling nbproc > 1 brings with it a number of caveats and features that are no longer available.
Examples include issues with the stats sockets/collection, issues with stick tables for session persistence, and issues doing non-disruptive configuration reload.

Currently, for HAProxy 1.6, nbproc > 1 is still highly discouraged as mentioned here in the documentation:
http://cbonte.github.io/haproxy-dconv/1.6/configuration.html#3.1-nbproc

So, based on advice we have received from folks at HAProxy technologies, we have been reluctant to enable multi-process mode by default.

We do understand some environments may benefit from using this setting. It is pretty straight forward to locally modify our jinja templates to enable multi-process mode for your deployment.

[Bartek Żurawski (bartekzurawski1)]

Hi Michael,

I just want to add it as an option because IMO it could help especially when there is big traffic that is going through LB. So by default HAproxy will be spawned in one thread (like it's now) but if you want you will be able to spawn it also with multi-process mode.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Michael Johnson (<email address hidden>) on branch: master
Review: https://review.openstack.org/415034
Reason: Abandoning this patch because it has been in merge conflict for four months without comment. Neutron-lbaas is also now deprecated and not accepting new features.

[Gregory Thiemonge (gthiemonge)]

Abandoned after re-enabling the Octavia launchpad.