[JSPWIKI-592] PageViewPlugin improvements, store statistics in JCR

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 3.0
• Component/s: Plugins

Environment

All/NA

• Assignee: Unassigned
• Reporter: Harry Metske
• Created: 03/Sep/09 17:42
• Updated: 27/May/10 15:38

Description

The current PageViewPlugin implementation stores the pageview counts in a file in the work directory.
This is fine for 2.8, but for 3.0 we would like to have that in in the repository for the following reasons :

• the counts don't get lost when we clear the work directory (it is called work directory after all )
• if running in a clustered environment you have multiple places where counts are done, giving incorrect results

There are a couple of options :

• store the count as attribute of a (each) page Node
• store them all together (as binary blob ?) in one special page Node

Considerations:

Performance

There is already a memory cache in the PageViewPlugin, the pageview statistics in this cache are saved every 5 minutes to the file in the workdir.
If the stats are stored in a special single page, performance would be roughly the same.
If pagecount is stored as attribute in a page, performance will probably be worse, because we have to update every node that has a statistic entry in the memory cache.

Cluster awareness

When running in "scalable mode" , i.e. multiple wiki instances sharing the same repo, we have two options to achieve "correct" pageview counts :

• update the repo (special page or attribute of each page) on each pageview (dramatic performance penalty)
• maintain something like a memory cache like we currently do, and flush the contents each interval to the repo, reset the memory counters to zero and start counting from zero again, so we keep delta values in memory, and add them to the values in the repo. Displaying pageview counts are not more actual than the flush interval (you don't see the pageviews on other wiki members in the cluster until the flush-interval expires and you reread the total value from the repo)

My personal feeling is that the best option (trade off between accuracy/complexity/performance) is to keep an in-memory cache, store values in a single special non-versioned page that can be locked to guarantee serial access to the data.

Any other comments, suggestions, options are welcome here....