[Tilesathome] "pending tiles" and status graphs

[spaetz]

On Thu, Sep 13, 2007 at 10:06:43AM +0100, Robert Hart wrote:
> What's going on with the "pending" tile count. The munin graphs have
> gone crazy, but as far as I can tell, there isn't actually a backlog
> of render requests. Is this unrenderable tiles due to osmarender5, or
> something connected to lowzoom?

No, the graphs are correct. There are actually 7,000 pending render requests for z12 tilesets. This is partly due to some heavy bulk requesting (e.g. somebody request(ed?) 1/29 of Norway *each* day).

On the other hand, the API has been getting slower because of a) the rapid growth of data and b) because there are now only 3 dedicated API instances for t at h clients rather than having all API insatnces available for all purposes. This means that only 3 t at h clients can simultanously receive data. So I had to restrict the number of active jobs to 50 (you will see that this number is pretty constant) because a bazillion of active t at h clients were clogging the API so it did not return any data at all anymore within any reasonable time out period.

To make the situation somewhat bearable, each IP address can now only reuest high-priority renders if it has <10 unfinished render requests. (Some people were bulk requesting with high-priority). Those high-priority requests are handed out before any low-priority request, which is the lion's share of all those pending requests.

I do try to document changes to the server here: http://wiki.openstreetmap.org/index.php/Tiles%40home/Server (althogh I might forget some).

Hope that helps,
spaetz