Open Stack

On 6/4/2018 5:13 PM, Sean McGinnis wrote:
> Yes, exactly what I meant by the NOOP. I'm not sure what Cinder would
> check here. We don't have to see if placement has been set up or if cell0
> has been configured. Maybe once we have the facility in place we would
> find some things worth checking, but at present I don't know what that
> would be.

Here is an example from the Cinder Queens upgrade release notes:

"RBD/Ceph backends should adjust max_over_subscription_ratio to take 
into account that the driver is no longer reporting volume’s physical 
usage but it’s provisioned size."

I'm assuming you could check if rbd is configured as a storage backend 
and if so, is max_over_subscription_ratio set? If not, is it fatal? Does 
the operator need to configure it before upgrading to Rocky? Or is it 
something they should consider but don't necessary have to do - if that, 
there is a 'WARNING' status for those types of things.

Things that are good candidates for automating are anything that would 
stop the cinder-volume service from starting, or things that require 
data migrations before you can roll forward. In nova we've had blocking 
DB schema migrations for stuff like this which basically mean "you 
haven't run the online data migrations CLI yet so we're not letting you 
go any further until your homework is done".

Like I said, it's not black and white, but chances are good there are 
things that fall into these categories.

-- 

Thanks,

Matt