forrest-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Crossley <cross...@apache.org>
Subject Re: Profiling Forrest [FOR-572]
Date Sun, 04 Sep 2005 23:29:02 GMT
Ron Blaschke wrote:
> David Crossley wrote:
> > David Crossley wrote:
> >> Ron Blaschke wrote:
> >> > Ron Blaschke wrote:
> 
> [snip]
> >> > NN  Components(Role)    Average time   Lastest times
> >> > 1   file ([1])           213 ms         30 ms    30 ms   581 ms
> >> > 2   idgen                  6 ms         10 ms     0 ms    10 ms
> >> > 3   xinclude            4202 ms       6219 ms  6389 ms     0 ms
> >> > 4   linkrewriter ([2])  2333 ms          0 ms    10 ms  6990 ms
> >> > 5   xslt ([3])            13 ms         20 ms    20 ms     0 ms
> >> > 6   xslt ([4])            66 ms         20 ms    10 ms   170 ms
> >> > 7   xslt ([5])             6 ms          0 ms    10 ms    10 ms
> >> > 8   html                   0 ms          0 ms     0 ms     0 ms
> >> >     Total time          6269 ms       6269 ms  6429 ms  7311 ms
> [snip]
> 
> >> Let us interpret that. The body-index.html was requested
> >> three times. The right-hand column is the initial request.
> >>
> >> Other columns are the subsequent requests. These should all
> >> come from the cache, and that is true for most components.
> >>
> >> However, "xinclude" does not. Its first processing is fast
> >> (there are no xinclude statements in the source). On every
> >> subsequent occasion it is being processed and taking too
> >> much time.
> 
> Right.
> 
> >> Comment-out that xinclude transformer in sitemap.xmap
> >> and the "extra" time then gets listed on the "idgen".
> >> Comment-out that and the time moves to "linkrewriter".
> >> 
> >> I wondered if the stylesheet that does the profiler
> >> presentation might have some bug. So i added the
> >> prettier stylesheet that is provided in the Cocoon block.
> >> Same effect. (The stylesheet seems better so i left that
> >> in place).
> 
> Quite interesting.  There seems to be something odd going on
> here.
> 
> > Compared with revision 227321 ...
> 
> This revision number is from Cocoon, right?

No, Forrest. This was just before the upgrade
of Forrest's Cocoon.

> > body-index.html
> > NN Component    Av.  Other times
> > 1 file           252  16  16  489
> > 2 idgen           3    1   1    5
> > 3 xinclude        9    7   7   11
> > 4 linkrewriter 1011    9   9 2014
> > 5 xslt           85   15  15  156
> > 6 xslt          115   13  13  218
> > 7 xslt           10    1   1   19
> > 8 html           11    0   0   22
> > 9 TOTAL        1567   88  64 3497
> 
> Nice job, David.  These numbers look quite good.  This should help
> narrow the problem down. A comparison of two CPU profiles might bring
> something up.
> 
> Before digging down to the problem, I thought I'd start with the
> profiler plugin, so don't wait for me looking for the problem.
> 
> Second, I am investigating how to tell Forrest that index.html and
> /index.html really are the same thing.

That would certainly help. These issues might be related:

http://issues.apache.org/jira/browse/FOR-268
"relative URIs with absolute paths are broken in the site.xml linking mechanism"

http://issues.apache.org/jira/browse/FOR-271
"extra leading slash when absolute path URIs used in site.xml"

webapp/resources/stylesheets/absolutize-linkmap.xsl
webapp/resources/stylesheets/dotdots.xsl

-David


Mime
View raw message