[Tilesathome] Render Queue getting bigger

[Craig]

Dermot McNally wrote:
> At the risk of advocating based on ignorance, would it not seems safe
> to push that figure up a bit? Also, why should the behaviour have
> changed so suddenly a week ago? Could there be a bunch of jobs getting
> wedged on some broken clients?
>
>   

I've only been running t at h for a couple of weeks, and it's always been 
pretty slow for me, regardless of the number of active jobs. It's either 
waiting for the server to issue a job, and on the rare occasions when it 
does get a job, it rarely manages to download the OSM data to perform 
the rendering.

This makes me think that the bottleneck isn't in t at h at all, but in the 
API server.

If everyone's client is performing like mine, the t at h server will issue 
80 jobs to 80 clients (or however many simultaneous jobs are allowed). 
Very few of those clients will fetch their OSM data successfully - the 
rest will sit waiting trying to download from the API server, and will 
take 30 minutes before they time out and fail, thereby decreasing the 
number of completed jobs.

You can see this from the log page - at the moment, it is reporting 7 
jobs completed in the last hour - but the t at h server must have issued at 
least 160 jobs in this time (80 simultaneous with 30 minute timeouts on 
each) - more if some clients have been successful.

Presumably it isn't feasible to reduce the 30 minute timeout, as this 
will push up the load on the API server even further. I've also been 
using Potlatch a lot recently, and it seems to "hang up" waiting for 
data from the API at roughly the same time as t at h slows down.

Having said that, it has been better over the last couple of days, and 
this evening has returned to a state of being unusable (and I'm talking 
about Potlatch here, not t at h). Does this coincide with the occurrence of 
Tiger uploads, or other bulk data imports which are being performed at 
the moment? If so, is there a way to throttle the bulk import scripts to 
allow slightly better performance for Potlatch (and JOSM, and t at h, and 
everything else)?

Cheers,

    Craig