[Mondrian] Re: Hudson

Aaron Phillips aphillips at pentaho.com
Mon Feb 2 16:59:55 EST 2009

My 2 centavos are as follows:

On Mon, 2009-02-02 at 13:29 -0800, Julian Hyde wrote:
> Build guys,
> Now we have Hudson online, I have a few questions about Hudson:
> 1. We still have one failure. How important is it to fix this? (I.e.
> is the failure preventing anything from happening, such as generation
> of binaries or code coverage?)

It is important if you want to have notifications and code coverage.
Notifications (email) happen when you go from stable to unstable and
vice-versa, so if the project remains unstable (aka tests are failing),
then there will be no notifications.  This is the typical configuration,
but we can change this to notify on every unstable build.
> 2. The cobertura report
> ( http://ci.pentaho.com/job/mondrian/lastStableBuild/cobertura ) is no
> longer available. What's up there?

Cobertura reports for some reason are only generated when there are no
failing tests which explains why it has disappeared.

> 3. How difficult would it be to test one or more other configurations?
> Important aspects of the configuration are JDK { jdk14, jdk15,
> jdk16 }, database { MySQL, Derby, Oracle, ... }, native evaluation
> { false, true }. One other configuration (say (jdk16, Derby,
> native-eval=false)) would give us good extra coverage.

Pretty sure that it will not be trivial to do this.  Hudson does have a
notion of "matrix" builds which allow for manipulation of axes like you
mention, however I'm sure this will be a time-consuming effort.  We also
have the issue of h/w resources which will have to be bolstered greatly
for this.
> 4. The 'megatest' shell script tests multiple configurations. How
> difficult would it be to run a shell script from hudson? If that's
> possible, megatest could be our long regress, and we could just have
> one config for the short regress. (Yeah, I used to run this on
> marmalade, but interpreting the build results was a manual process and
> I got behind in this process.)

I would recommend a separate Hudson slave to run megatest on a regular
but nearly as frequent interval.  How difficult?  I personally don't
know as I've never run them before.  But basically anything you can do
on a command line can be wrapped in a Hudson job, so it is very
> 5. What are your opinions about having a long & a short regress? If
> so, is the ideal time for the short regress?

No opinion on this.  I think the answer will be constrained by the
amount of available CPU.
> By the way, there are a couple of options for making the short regress
> faster: 1. Run against Oracle-XE (it's free-as-in-beer and easy to
> install on Linux) not MySQL, 2. Don't generate code coverage.  (I note
> that it's ~1 hr right now, whereas build+regress against Oracle takes
> ~7 minutes on marmalade.)
> Julian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.pentaho.org/pipermail/mondrian/attachments/20090202/164c1635/attachment.html 

More information about the Mondrian mailing list