Open Stack

On Fri, Sep 09, 2016 at 09:29:46AM +0200, mathieu bultel wrote:
> On 09/09/2016 12:42 AM, Emilien Macchi wrote:
> > On Thu, Sep 8, 2016 at 4:18 PM, David Moreau Simard <dms at redhat.com> wrote:
> >> How long does the upgrade job take ?
> > undercloud upgrade is a matter of 10 min max.
> > overcloud upgrade will be much more, I don't have metrics right now.
> > matbu maybe?
> It really depend on the infra which run the upgrade. I don't know much
> about the upstream infra but
> on my local box, with a SSD, 8 cores and 32Go of RAM, It could take
> around 1h30 2hours for a full upgrade.
> On centos ci infra and with RDO, some jobs can takes 4hours or so ...
> 
> I'm really curious to see how long a full upgrade will take with upstream.
> 
> Right now, the full upgrade job didn't go far from the controller
> upgrade (step 2).
> AFAIK, the timeout in upstream is 3 hours minus 10 minutes ...
> I think if we keep a 2 nodes deployment with only pacemaker, it would be
> enough... I will keep you in touch of my progress here..
> 
> But, even if the jobs takes 2 or 3 hours to vote, I think it would be a
> real huge gain for the tripleo work.

Today we use 8vCPU 8GB RAM, 80GB HDD our nodepool nodes. Obviously how the job
performs is subjective to which cloud it is one.

With that said, it might be worth moving these tests into 3rd party CI where
more powerful servers would live.  Even though tripleo jobs upstream today
support 3 hour 10 minutes as the time out, some of our longest running jobs.