[rdo-users] [TripleO] Long OpenStack deployment times with RDO recent releases

[Alex Schultz]

On Wed, Sep 22, 2021 at 9:53 AM Guido Langenbach <
guido.langenbach at cloudseeds.de> wrote:

> Hey guys,
>
> we migrated an OpenStack installation of one of our customers to TripleO
> RDO when Rocky was released. Back then deployment times were fine, as the
> whole cluster didn't have that many nodes yet.
>
> In the meantime we upgraded to Ussuri and use 3 controllers and 44 compute
> nodes by now. Our deployment times exponentially increased with each set of
> compute nodes we added. With our current setup, a complete deployment run
> takes about 15 to 16 hours. In our clusters we count the following
> ressources right now:
>
> VMs: ~ 2100
> Networks: ~500
> Ports: ~6100
> Volumes: ~5700
>
>
This sounds like it's related to
https://bugs.launchpad.net/tripleo/+bug/1915761.  This should be fixed in
the latest versions and we did backport it to Train.  So it'd be
interesting to see if you're missing some of these patches. We've had
reports of updates taking about 4.5 hours with newer versions of train to
update so the numbers you have seem to point to possibly missing patches
related to that bug or an execution configuration problem.

Additionally in newer versions we switched the execution strategy to try
and speed things up as well which should be available in the latest
versions of train/ussuri due to
https://review.opendev.org/q/topic:%22strategy-improvements%22+(status:open%20OR%20status:merged)


> We implemented ARA for now so we can get exact measures of each
> ansible-playbook runtime to see what is taking the most time. My question
> is: How big are your production OpenStack environments and how long does it
> take you to deploy?
>
>
Are you running ansible-playbook by hand? And do you have a ansible.cfg?
We added an `openstack tripleo config generate ansible` command that'll
generate a starting ansible.cfg that's similar to what we have in the
mistral execution.


> Which methods do you guys use to scale-up compute nodes? (spoiler:
> --skip-deploy-identifier doesn't seem to work properly)
> Is Blacklisting all other Compute nodes the right move? Do you even
> blacklist the Controllers as well?
>
> Best Regards,
>
> Guido
>
> _______________________________________________
> users mailing list
> users at lists.rdoproject.org
> http://lists.rdoproject.org/mailman/listinfo/users
>
> To unsubscribe: users-unsubscribe at lists.rdoproject.org
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rdoproject.org/pipermail/users/attachments/20210922/24ebfdbd/attachment-0001.html>