[ckan-discuss] Architecture Best Practice

Sat Apr 27 10:00:47 BST 2013

> I am considering two scenarios for a multi-department deployment and would
> like to hear from others on the pros/cons of each.
> 
> In the first scenario, each department has its own instance of CKAN.
> 
> In the second scenario, I run one central CKAN, with multiple organizations
> turned on and each department using a different organization.

Either way is possible, which one fits best depends on your exact needs.
A single CKAN using the organizations feature will be much simpler and
easier to setup.

> Many of my departments have significant customization needs in terms of
> fields/facets -- generally around 300 custom fields each. These custom
> fields do not overlap, so if I'm using one central CKAN, this would add
> thousands of fields to the schema. I imagine this would cause performance
> issues.

Wow! That's a lot of custom fields. I'm not sure whether this would
cause performance issues or not (I don't know of an example where
someone has had so many custom fields).

I don't think you would have to add all of the custom fields from every
organization into one massive schema. You can have a different schema
for each organization. You can use the IDatasetForm plugin interface,
which allows you to return different schemas for different datasets,
based on each dataset's package_type field. And you would use a
different package_type for each organization.

> I am also concerned that letting multiple organizations share one CKAN
> means that adding extensions or customizations that just *one* organization
> needs becomes a really tall order, and one organization deploying a
> potentially broken extension puts all other users at risk.

Potentially, I suppose. On the other hand, if you have a separate CKAN
instance for each one of your organizations, and they all deploy lots of
different extensions, this could cause you problems as well. Many CKAN
instances, many custom extensions, many things that can break.

Extensions may also cause problems with "federation" of multiple CKAN
instances, e.g. with the harvesting process. Depends on the extension, I
suppose.

It seems to me that either way you do it, if you're planning on managing
all these organizations, you probably want to keep the number of
different extensions used within control.