[Rebuild] Strategy for running the bot in "regions"

Andy Allan gravitystorm at gmail.com
Mon Jun 25 11:42:08 BST 2012


On 25 June 2012 11:28, Frederik Ramm <frederik at remote.org> wrote:

> I'd just crawl over the world in a grid, e.g. 1° by 1° or smaller (small
> enough to have reason to assume that, on average, one changeset will be
> enough to make all modifications required in the area).

If you have the time, can you run some stats for us from wtfe? It
would be great to find out how many red+yellow nodes/ways/rels there
are in 1° boxes, and what the dispersion looks like.

> I'd be loading the
> data from the database with code that mimics the map call.

Does anyone have a map-call-but-with-history algorithm, or preferably
an implementation? :-) Wishful thinking, I guess!

> I think it would be ok, if performance considerations make that attractive,
> to choose a method that would possibly omit some cases (e.g. nested
> relations), and then collect them in a couple broad strokes after the main
> job is done, even if that would lead to a few world-spanning changesets in
> the end.

Whichever algorithm we use, I expect to log all the entity ids and
compare that against a db-selection at the end of the job. I'm sure
there's some relations floating around that have no foreign-key links
to nodes whatsoever.

I guess also the bot will need to deal with making its selection,
doing the processing, and finding that the changeset gets rejected due
to version-mismatches as people continue editing as it does its job.

Cheers,
Andy



More information about the Rebuild mailing list