Introduction

The dynamic parts of most Plone pages are:

• Breadcrumbs

• Navigation tree

• Collection portlets.

Moving these page fragments to Edge Side Includes would allow caching of the page indefinitely in a caching proxy.

In this experiment I demonstrate how XSLT can generate a navtree for a large site very quickly.

Generating a navtree

Create some test data (100,000 entries)::

$ python testdata.py > navtree.xml

Execute navtree queries with timing:

$ xsltproc --timing navtree.xsl navtree.xml > navtree.html
Parsing stylesheet navtree.xsl took 0 ms
Parsing document navtree.xml took 12657 ms
Applying stylesheet took 37 ms
Saving result took 0 ms

Points to consider

• The test data is a 17M file, so parsing takes a long time (12s when not in the FS cache, 1s when FS cached.) Clearly this can't be done on every request so a custom python based server would be used instead of Apache or Nginx XSLT modules.

• The navigation portlet may not always be rendered, for instance when there is no content for it to display. This may lead to a whole portal column being hidden. If the tree is generated in an ESI then we cannot know at page rendering time whether the navigation portlet should be displayed or not. This could possibly be fixed up with Javascript.

• The site structure data in navtree.xml should be updated whenever the portal_catalog changes. Parsing is so slow that content editors may not see their changes immediately reflected in the navigation tree. Parsing could be speeded up by partitioning the data (e.g. per site section) though this would require additional bookkeeping by the portal_catalog. It might be sufficient just to ensure that the current item's title is taken from the page which contains the ESI include, passing it in the ESI querystring.

• No security filtering is performed yet, though it should be possible and relatively quick. Logged in users could have their roles and groups stored in the data section of the auth tkt cookie, making it available to the navtree renderer.