[OSM-talk] Is tile rendering having a crisis?

[Jon Burgess]

On Tue, 2016-08-09 at 14:41 -0700, Ben Discoe wrote:
> The tile renderer, "renderd", has been heavily overloaded for quite a
> while, like 90% of the time it is dropping, and the dirty queue is
> entirely ignored.  However, in the past day, something has happened
> so
> that it is even more overloaded, even the standard request queue is
> not getting handled, and the dropped is out of control:
> 
> http://munin.openstreetmap.org/openstreetmap/orm.openstreetmap/render
> d_processed.html
> 
> Does anyone know what's going on?  This is a large frustration for
> me,
> and I'd happily donate money to beefing up our render server(s).

There was an update to the rendering style a couple of days ago and
this triggered all of the existing tiles on Orm to be marked as needing
a fresh render the next time someone requests them. This means there
are far more requests hitting the render daemon than normal. It
typically takes about a week for the request queues to return to normal
when this happens. If you look at the "by-year" chart you can see
spikes in the dropped counts where this has happened about a dozen
times in the past year.

The OSM operations team are aware of the increased load on the tile
servers. There was a recent jump in usage when one of the other tile
providers stopped unlimited free access to their service. There was
also a site using OSM tiles for a popular Pokemon Go map causing lots
of additional traffic, once this was identified the site was blocked.

	Jon 

(member of the OSM operations team)