continuum-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brett Porter <br...@apache.org>
Subject Re: [Discussion] Continuum 2.0 Roadmap
Date Tue, 05 Feb 2008 23:08:47 GMT
This looks very exciting, and agree with most of the thread that  
follows. I'm just going to reply in summary - most of my thoughts are  
actually non-technical :)

Regarding databases: I don't think xml files are the solution (except  
for the configuration where it makes a lot more sense :) - the data  
needs to be queryable. I think Andy made a good point in his comment  
on the roadmap - we need to look at the actual problems. Here's what I  
think needs to be improved:
- better centralisation of access. The architecture of Continuum  
bleeds JDO decisions all through the code since you access lazy stuff  
for the first time in obscure places.
- I think this might be that the model is too complicated (sorry, my  
fault) - it assumes complex relationships are handled easily. It seems  
to be going ok these days, but I feel it would be hard to modify.
I haven't looked at Rahul's branch yet, but I think we should consider  
a more decoupled database (ie, lookup build results for a project but  
keep them separate in the model to avoid the need to lazyload 90% of  
the time), and more centralised database logic. I would consider JPA  
just because it gives more options in terms of an implementation. It  
is quite easy to use from a development standpoint. But we also need  
to consider what functionality is needed up front - I think high on  
the list needs to be migrations between versions. Also, We are  
probably going to need to store more data in the future, and be able  
to query it (particularly historical datapoints).

On the container: I would prefer to move off Plexus simply because  
it's a moving target and it's a barrier to entry for new developers.

Now, my more general observations. Firstly, the roadmap doesn't appear  
to have any features - these are all technology changes. Some of that  
might be cool and a feature in itself, but I think there needs to be a  
balance between evolution, features, and bugfixing. I would also  
emphasise that features should be creative new things Continuum can do  
(for which we've had plenty of ideas), not just catching up to other  
CI servers :)

I think the first part of the roadmap is key - separating the layers  
out, and basically building Continuum to be lightweight and  
distributed from the ground up. I hope that's the focus of the  
development. Note this also impacts the database as it should store  
much less information on builder machines (it can ship history back to  
the main server).

I also think that supporting plugins is a good idea - it has been a  
huge bonus in other apps and in Maven itself. I'd like to investigate  
using OSGi for this.

But by far the biggest question I have is what happened to 1.2? I  
think Continuum needs to set a target to achieve, but get there in  
gradual steps that at each stage sees a production release. The long  
1.1 cycle really set Continuum back - a lot of it was changing  
features, but there was also a lot of changing technologies :) I don't  
think Continuum will survive another year-and-a-half release cycle. So  
the start could be to break all the actions out (plexus, not webwork)  
into services and add some features, then the next release could  
adjust the database model and add some other features. And as we split  
these things out we make sure they are nicely documented and tested.

That's my thoughts :)

Cheers,
Brett

On 30/01/2008, at 9:34 AM, Emmanuel Venisse wrote:

> Hi
>
> I started a document [1] with my ideas about Continuum 2.
> As you can see in this doc, I want to add lot of things in the next  
> version.
>
> Feel free to comment on it.
>
>
> [1]
> http://docs.codehaus.org/display/CONTINUUM/Continuum+2.0+Design+Discussion
>
> Emmanuel


Mime
View raw message