Enabled and Disabled Nodes

In the old CR, this was called “hidden”, but we felt that enabled and disabled captured the intent better.

If a node is disabled, the system should behave as if this node would not exist. This means:

you are not able to query for this node
you are not able to query for any child node: If you removed the node, all children would be removed as well.
these rules must only apply in frontend, when viewing the site. In backend, you should still see the disabled nodes, so that you can e.g. enable them again.

Cross-Cutting Concerns

Because all API methods on the read side need to implement proper handling of enabled/disabled behavior, we call this a cross-cutting concern. The complexity of the CR appears to be that there are quite some distinct concerns which somehow all interleave with each other; making it hard to build a classical layered architecture which hides some of this complexity.

#Event Structure

Modelling events for enabling/disabling of nodes is rather easy, as you only say “node X should be disabled now”; but the projection has to recursively mark all nodes below as disabled as well.

#Projector Logic

Our basic assumption is that we want to build the transitive closure (of the start node and all sub nodes), and create a “disabling node X leads to node Y being unavailable” relationship.

To implement that, we implemented a new database table called TODO; which says “because node X was explicitly disabled, we disable node Y”.

We fill this database table by directly walking the tree structure in SQL (via a recursive Common Table Expression); so that way we can avoid numerous roundtrips from PHP to SQL and back again.

Then, at query time, we join the nodes with this TODO table; to filter out disabled nodes if we are in the frontend.

Disabling a node is more complex than O(1)

You might notice that disabling a node is the first operation we explained which is not having O(1) complexity - but we believe this is OK because often, the number of subnodes affected by a hide should be relatively minor compared to all the nodes in the system.

currently we build up this table for all content streams; we could theoretically only build it for the live content stream.