New OpenStack instance - status

[Kevin Fenzi]

On Fri, 20 Feb 2015 15:32:15 +0100
Miroslav Suchý <msuchy at redhat.com> wrote:

> Since I'm leaving for one week vacation, I think I may write down
> current status of our new OpenStack instance and write down TODO
> list. Just in case someone is desperate enough to do some fixes.

I poked at it some (with help) and made a bit more progress...

> I updated docs.git/cloud.txt - mainly which playbooks we use right
> now and where to write down IP, when you add new compute node.
> 
> Controller - should be OK. At least I see no problems there right
> now. Network is stable. I can log to EqualLogic (credentials are at
> bottom of cinder.conf). Volumes are created correctly. I can reach
> compute nodes. AMQP works and is reachable from Compute nodes (do not
> try to play with SSL&RabbitMQ it will never work on RHEL7). Horizon
> works (over https).
> 
> Compute nodes - it looks good until you try to start VM. :) I fixed
> several problems, but new ones still pop ups.
...snip...
> Right now there is some error:
>   TypeError: unsupported operand type(s) for +: 'NoneType' and
> 'str'\n" which is new to me and which I will not manage to fix before
> I will leave today. It may be last one problem or they may be dozen
> other still waiting in queue. It's hard to tell.

We got past that and git instances to spin up. Seems like it was just
needing a restart on the compute node (of compute and ovs). 

Then there was an issue of routing for the external ips. That needed an
additional rule on the compute nodes in iptables. I added that to
playbooks. 

I also added nameservers to allow instances to get dns correctly. 
 
> Smaller fixes to do:
> * playbook hosts/fed-cloud09.cloud.fedoraproject.org.yml can be
> enhanced that after packstack execution the machine should be
> restarted. Right now I am waiting for first error after packstack and
> then I restart the machine manualy and re-run playbook again. This is
> last manual workaround. Everything else was already automated.

I don't know that we want a reboot in playbook, it should be
idempotent, ie, we should be able to run it and reach the desired
state, then re-run and 0 changes. I guess it it only rebooted after
packstack first runs it could work. 

> * routing between compute nodes and controller using public IP does
> not work. Not fatal right now, but nice to have.

Yeah, not sure about that... 

Other things: 

https for keystone endpoint would be nice.

vnc consoles aren't working right. 

Need to make sure we can get our ansible to spin up and manage
instances, etc. 

Perhaps we could spin up a dev copr on it to test... and if all looks
well do another reinstall/reconfigure cycle and start using it. ;) 

kevin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://lists.fedoraproject.org/pipermail/infrastructure/attachments/20150301/1e818a2a/attachment.sig>