[dokuwiki] Re: Caching changes

On Sun, 24 Sep 2006 12:08:20 +0100
Chris Smith <chris@xxxxxxxxxxxxx> wrote:

> (a) Bug when forcing purge via query string
> 
> Two calls are made to p_cached_instructions() when the renderer cache
> is not valid.  The purge query string prevents any cache use,
> resulting in the full parsing of the wiki page twice and quite slow
> page generation.
> 
> I haven't been able to come up with a neat way of handling this - 
> excepting setting a global flag.  How often is PURGE used?  Is it
> worth coming up with a solution?

It is used very rarely and only by people knowing it exists :-) So I
guess we can live with purge being slow.

> (b) cachetime setting(Max age for cache)
> 
> Dokuwiki's cache code should now be 100% internally consistent - ie
> only by making external changes can the cache use code make incorrect 
> decisions on cache validity.
> 
> Since serving a page from cache is much faster than regenerating
> it ...
> 
> Can the default setting be increased?
> Can it be removed entirely?

The cachetime may still be useful for certain plugins. Also the RSS
inclusion currently depends on having the cache being rebuild from time
to time. But we could probably change this to use a page specific
cachetime through an RSS parameter?

Besides the cachetime, there is the purgefile (and the purgeonadd
config), too. Would this be obsolete as well? I'm asking because
I discovered a strange bug I'm not sure what causes it.

I wrote a blog entry yesterday on splitbrain.org. Today I discovered it
wasn't displayed on the frontpage. I'm not using Esther's blog plugin
but a very simple plugin written by my self. It does not use any
metadata info. But it should have picked up the new blog entry because
of the updated purgefile.

I haven't investigated yet if this is reproducable. It may have been a
rare race condition. It may also be fixed already because
splitbrain.org isn't running the most recent codebase yet. I'm just
mentioning it as an example where the purgefile still may be useful.

> (c) caching common actions for anonymous visitors
> 
> I think this would be pretty easy to implement through the cache
> object, but is it worthwhile (for the next release :-))?

I don't think it's worth the effort, because a) these lists often
change and are b) not used that much compared to viewing and editing
pages and c) are fast enough I'd say.

For b) some log analyzing would be probably a good idea to see if my
assumption is correct.

Andi

PS: Chris, would be great to see you in the IRC channel from time to
time to discuss these things live ;-)

-- 
http://www.splitbrain.org

Other related posts: