Erik McCormick <emccorm...@cirrusseven.com> wrote:
Looping the list back in since I accidentally dropped it yet again :/
On Mon, Mar 19, 2018 at 8:45 AM, Torin Woltjer
<torin.wolt...@granddial.com> wrote:
That's good to know, thank you. Out of curiousity, without
pacemaker/chorosync, does haproxy have the capability to manage a floating
ip and failover etc?
HAProxy can't do that alone. However, using Pacemaker just to manage a
floating IP is like using an aircraft carrier to go fishing. It's best
to use Keepalived (or similar) to do that job. It only does that one
thing, and it does it very well.
Just for balance: from my experience talking with a lot of HA experts
over the last few years, I don't think there is broad consensus on
this point. With the loud disclaimer that I don't personally have a
strong background of experience with keepalived, I've heard multiple
experts point out that the simplicity of keepalived can occasionally
cause problems, especially related to its lack of fencing and the way
it determines which node should be the master and how that can cause
problems when there are network partitions.
I don't mean to spread FUD, so apologies if it comes across that way,
but I recommend doing some in-depth reading so you can make up your
mind either way.
Additionally take a look at the bug queue for Neutron's L3 HA feature
which uses keepalived:
https://bugs.launchpad.net/neutron/+bugs?field.tag=l3-ha
Certainly many of these are not related directly to keepalived, but
equally there are still some issues with this feature which need to be
addressed.
In summary: HA is *hard*. Anything which claims to be both simple and
a full solution deserves to be scrutinised very carefully.
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
