Use redundancy between multiple controller nodes [kolla]
I've deployed openstack xena using *kolla-ansible* on *centos 8 stream*, I've used multiple controller nodes, a bunch of compute nodes and a storage cluster. Controller nodes include network services. I want to know how can I successfully set up redundancy between these controller nodes, in a way that some nodes are active while others are on standby state, and would only become active in case when the active nodes are to fail. Thank you. Regards
On Tue, 2022-07-19 at 13:51 +0100, A Monster wrote:
I've deployed openstack xena using *kolla-ansible* on *centos 8 stream*, I've used multiple controller nodes, a bunch of compute nodes and a storage cluster. Controller nodes include network services.
I want to know how can I successfully set up redundancy between these controller nodes, in a way that some nodes are active while others are on standby state, and would only become active in case when the active nodes are to fail. in generall that is not needed with openstack.
most service are partly or mostly stateless using the db to store any relevent persitent state. that generally menas that for example you can have n nova api process active at any time and you do not need to use external managment fo the serivce to have some be active and other be passive. in general such active/standby configurtation is an antipatter and should not need to be done with openstack. in kolla keepalived and haproxe shoudl migrate the contol plane vips and include/exclude donw services automaiticaly. if you want to manually take over that function its likely possibel but not how HA is intended to work in general. i dont work on kolla anymore so perhapse there perspectiive has change but in generaly i would advocat for active active ha in preference ot active standby.
Thank you. Regards
Thank you for the clarification. So I should let the handling of redundancy to keepalive and haproxy instead of using external tools. On Tue, 19 Jul 2022 at 15:21, Sean Mooney <smooney@redhat.com> wrote:
On Tue, 2022-07-19 at 13:51 +0100, A Monster wrote:
I've deployed openstack xena using *kolla-ansible* on *centos 8 stream*, I've used multiple controller nodes, a bunch of compute nodes and a storage cluster. Controller nodes include network services.
I want to know how can I successfully set up redundancy between these controller nodes, in a way that some nodes are active while others are on standby state, and would only become active in case when the active nodes are to fail. in generall that is not needed with openstack.
most service are partly or mostly stateless using the db to store any relevent persitent state.
that generally menas that for example you can have n nova api process active at any time and you do not need to use external managment fo the serivce to have some be active and other be passive.
in general such active/standby configurtation is an antipatter and should not need to be done with openstack.
in kolla keepalived and haproxe shoudl migrate the contol plane vips and include/exclude donw services automaiticaly.
if you want to manually take over that function its likely possibel but not how HA is intended to work in general.
i dont work on kolla anymore so perhapse there perspectiive has change but in generaly i would advocat for active active ha in preference ot active standby.
Thank you. Regards
The main thing you have to enable in Kolla Ansible is Neutron L3 HA, which is disabled by default. See enable_neutron_agent_ha. Other services should be redundant when using multiple controllers. On Tue, 19 Jul 2022 at 16:48, A Monster <amonster369@gmail.com> wrote:
Thank you for the clarification. So I should let the handling of redundancy to keepalive and haproxy instead of using external tools.
On Tue, 19 Jul 2022 at 15:21, Sean Mooney <smooney@redhat.com> wrote:
On Tue, 2022-07-19 at 13:51 +0100, A Monster wrote:
I've deployed openstack xena using *kolla-ansible* on *centos 8 stream*, I've used multiple controller nodes, a bunch of compute nodes and a storage cluster. Controller nodes include network services.
I want to know how can I successfully set up redundancy between these controller nodes, in a way that some nodes are active while others are on standby state, and would only become active in case when the active nodes are to fail. in generall that is not needed with openstack.
most service are partly or mostly stateless using the db to store any relevent persitent state.
that generally menas that for example you can have n nova api process active at any time and you do not need to use external managment fo the serivce to have some be active and other be passive.
in general such active/standby configurtation is an antipatter and should not need to be done with openstack.
in kolla keepalived and haproxe shoudl migrate the contol plane vips and include/exclude donw services automaiticaly.
if you want to manually take over that function its likely possibel but not how HA is intended to work in general.
i dont work on kolla anymore so perhapse there perspectiive has change but in generaly i would advocat for active active ha in preference ot active standby.
Thank you. Regards
participants (3)
-
A Monster
-
Pierre Riteau
-
Sean Mooney