Re: [Maposmatic-dev] Re: [cairo] Cairo & libpng/zlib parameters

[David Decotigny]

Hello planet,

Jeroen van Rijn wrote:
> On Sun, Jan 17, 2010 at 15:00, David MENTRE <address@hidden> wrote:
> > I can't say if for those jobs Cairo is the bottleneck, let me know. Let
> > me know also if you need more information.
>
> Chris,
>
> Bear in mind that these times include postgis querying and some
> additional set up. You could time more accurately as far as

Indeed, these overall rendering times also include the time it takes to 
perform our own queries /before/ mapnik does its job. And these queries 
can be very loooong, much longer than the time it takes for mapnik/cairo 
to do the actual rendering. All in all, I do believe that cairo is not 
the bottleneck for us. And if mapnik were still considered too slow, I'd 
say it's more because of the sql queries on a loaded postgis DB than 
because of cairo, but for this I'm not totally sure, not at all.

> mapnik/cairo is concerned by limiting yourself to the time spent in
> ocitysmap/street_index.py:def _render_map_into_files
>
> More specifically the timestamps between these messages:
> LOG.debug('rendering from %s to %s.%s...' % (osm_map_file,
>                                                    out_prefix,
>                                                    out_formats))
>
> LOG.debug('saving %s.%s...' % (out_prefix, fmt))
>
> The emitted log messages are not currently timestamped, but that can
> be arranged. I think it's a good idea for us to add that anyway so we
> can track down improvements and regressions better on our end as well.

Well, the timestamps are in fact being saved, but only on our dev server 
and not yet on the prod server. However, even if the timestamps were 
available in the prod server, we do not log the debug mesages anyway (we 
are stopping at INFO level). Maybe we could change this, so that this 
timing information is available for all the mapnik renderings...

Best regards,