Openstack version upgrade pre-req

asked 2020-07-22 20:30:48 -0500

SRIG gravatar image

Hi Team, We have an OpenStack environment as I) One Controller Server + II) One Block Storage Server + III) 10 Compute Node Servers + IV) With OpenStack 3.8(Octa) in Ubuntu 16 Operating System in all the 12 Hypervisors/Servers. V) Components - Horizon + Neutron + Nova + Heat + Swift + Cinder Vi) we have around 100VM's in 10 compute nodes

Now we are planning to upgrade the Operating system (Ubuntu 16 to Ubuntu 20) + Openstack version (Octa to Pike/Victoria - latest greatest )

So now what are the steps we need to follow first?
Like which server need to upgrade first in this environment

  1. Is it Controller machine / Block storage / Compute node --> Which server need to install first?
  2. For upgrading do I need to update Operating system first or Openstack version first ?
  3. How to validate the Openstack version upgrade without any issues? (Like Nove-manage, Neutron...etc)
  4. How to ensure the availability of the 100VM's in 10 compute nodes?

Thanks in advance really appreciate your response.

Regards Sri

Thanks in advance

edit retag flag offensive close merge delete

Comments

Start here: https://docs.openstack.org/operations.... Then search for OpenStack summit presentations that share upgrade experiences.

Warning: Victoria is not released yet. Also, you can't go directly to Victoria or Ussuri. You need to go through all (?) intermediate releases.

Bernd Bausch gravatar imageBernd Bausch ( 2020-07-22 21:55:48 -0500 )edit

Depending on your situation, it might be an option to create an equivalent Ussuri or Victoria cloud, then copy all VMs to that cloud by snapshotting them. How easily you can copy Cinder volumes probably depends on the backend. How easily you can copy the virtual networks depends on their complexity.

Bernd Bausch gravatar imageBernd Bausch ( 2020-07-22 22:00:10 -0500 )edit

Off-topic: The fact you have a single controller tells me that you don't care for down time that much. On the other hand you have ten compute nodes and 100 VMs, which doesn't look like a toy cloud. I think you should investigate high availability.

Bernd Bausch gravatar imageBernd Bausch ( 2020-07-22 22:04:04 -0500 )edit

Thank you , I will review the above and try to apply.

SRIG gravatar imageSRIG ( 2020-07-24 16:02:02 -0500 )edit