Re: check with SGML::Parser::OpenSP

On Fri, 2005-08-12 at 15:41 +0200, Bjoern Hoehrmann wrote:

> I get 0.70-0.75 for the CGI version regardless of whether 5 or 20
> concurrent requests on my machine [...]
> and mod_perl 2.01 (with a minimal default configuration): [...]
>   Requests per second:    2.56 [#/sec] (mean)

That sounds promising, and expected too.  It would probably be even
better with HTML::Template's "cache" option as well as preloading the
error messages to the global var inside BEGIN (so that it's loaded only
once at initial script parse time).

> With constant memory usage after the first x requests (with OpenSP
> 1.5.1 it probably would not constant as it has a huuuge memory leak).

BTW, any news about when 1.5.2 could be out?

...

Regarding v.w.o currently, watching top shows sort of constant load
averages of > 100 (!), with the bottleneck clearly being CPU resources.
And practically all of CPU is consumed by "check" (onsgmls consumes
nothing compared to that).

If someone wants to fiddle with the live instance, this:

   mkdir -m 755 /var/cache/validator
   chown nobody:nogroup /var/cache/validator

...and applying the attached patch could help a bit.  Also,
HTML::Template could be upgraded to 2.7 which has some caching related
improvements.  The current code which doesn't use caching at all would
most likely not benefit from that, though.