UEC/Eucalyptus installation test: 20090927.1

Matt Zimmerman mdz at canonical.com
Mon Sep 28 04:18:45 BST 2009


On Sun, Sep 27, 2009 at 10:31:31PM +0100, Matt Zimmerman wrote:
> On Sun, Sep 27, 2009 at 10:00:05PM +0100, Colin Watson wrote:
> > The cloud controller doesn't actually run at package installation time
> > when we're doing a full system install from the CD. This is why we do it
> > at init time instead.
> 
> I see.  I guess that's our best bet at this point, then, and I'll stick with
> the init method.  I'll just fix it to go properly in lock-step with the
> startup of the services so that it doesn't fail.

I've pushed a branch to:

https://edge.launchpad.net/~mdz/eucalyptus/auto-registration

which implements automatic registration of SC, CC and Walrus using upstart.

It will (I think) enforce the necessary ordering constraints of startup and
registration.  I've tested it with simple "dpkg -i *.deb" kind of scenarios,
and it seems to work, so I think it's ready for inclusion in the beta.

It's only designed for the scenario where the CLC, CC, SC and Walrus are all
on the same host, which is what I think matters most for beta as that's how
the installer sets it up.

I think we should be able to accommodate other configurations as well with
minor tweaks now that the infrastructure is there.  For example, we could
relax some of the event triggers from the registration jobs, and instead
just have them respawn until they succeed, with trivial changes to the
upstart jobs.

The only thing missing from the cluster installation is the configuration of
a public IP address pool.  The logic is all there, and we would just need to
raise the priority of the question so that it gets asked.  Alternatively,
Eucalyptus could be configured to use private addresses by default if no
public addresses are available.

I haven't touched the node controller code yet.  When Dustin and I were
testing last week, it looked like the node installer was working fine,
detecting the cluster controller and so on.  My understanding is that it's
still necessary to go back to the cluster host and run euca_conf
--discover-nodes before the node will actually be live.

Testing of both cluster and node would be greatly appreciated.  I'd like to
get this uploaded in the morning once someone else has validated it, and get
it onto server ISOs ASAP after that.

Is there anything else missing for beta at this stage?

-- 
 - mdz



More information about the ubuntu-devel mailing list