If you don't know what this is, then you don't need to know what this is :D Nothing to see here. Move along.
This is an implementation of the Atlas API that is tuned for read performance. If it was hosted on local servers, nearly all responses would be sub-second. It is currently hosted on Heroku, meaning most responses are served in 1-2 seconds.
Give me all restaurants within 1km of Southbank
pois?where[close_to]=-37.821836,144.960029,1000&where[type]=eat
Atlas: Not implemented
FastAtlas: 1-2 seconds
Give me just the name and id of all Italy POIs that aren't geocoded
pois?select=name&where[place]=359845&where[geocoded]=false
Atlas: Not implemented
FastAtlas: 1-2 seconds
Give me all places that contain the lat long of Eiffel Tower
places?where[contains_point]=48.8582493546056,2.294511795044
Atlas: Not implemented
FastAtlas: 1 second
Give me 1000 places
places?limit=1000
Atlas: 30-40 seconds
FastAtlas: 1-2 seconds
Give me 1000 POIs in Australia
pois?where[place]=362249&limit=1000
Atlas: 30-40 seconds
FastAtlas: 1-2 seconds
Give me only the names and ids of ALL the POIs in Australia
pois?select=name&where[place]=362249&limit=12000
Atlas: 12 pages of 1000 POIs takes around 6-8 minutes
FastAtlas: 2-3 seconds
Give me all POIs in a bounding box
pois?where[contained_in]=38.079598,37.479598,-122.120143,-122.720143
Atlas: 20-25 seconds
FastAtlas: 1-2 seconds
Get 1000 POIs in a place, with reviews
pois?where[place]=362494&limit=1000&detailed=true
Atlas: 40-50 seconds
FastAtlas: 3-5 seconds
- It is implemented only in JSON. XML should be considered for deprecation in the existing API.
- It is implemented in Ruby 1.9.2 and Rails 3.0.6, taking advantages of the performance boosts over Ruby 1.8.7 and Rails 2.3.2
- The database it uses for its backend is a denormalized version of the existing Atlas data.
- The JSON returned has all whitespace removed, making it less human readable but much smaller and faster.
- It strives to be as simple as possible. There are two tables (pois & places), two controllers (pois & places), and 3 model classes (Poi, Place and LatLong for the funky LatLong math). Along with this page, that's it.
- "Show me all POIs within a certain distance of a place" - so you can find all restaurants within 1km of Southbank
- "Let me choose the fields I want returned" - you might want 'name' but not 'etyhl_id', so you can omit ethyl_id to improve performance more.
- "Show me all places that contain this point" - give a lat/long and you can find out all places that contain that point
- Getting all POIs for a place now uses
pois/where[place]=1006242instead ofplaces/1006242/pois - If a field is null, then it is not returned in the JSON. So if a field is not returned, you can be sure it is null. This is to reduce response size.
- No search for POIs by ISBN. To get this data denormalised was going to take longer than a weekend.
- No posting data. The data is read-only.
- No support for multiple languages
- No versioning (draft/published). The latest version (in its default language) of each POI/Place in Atlas was used to seed the database.
- No tests. No good error handling. Bit of a hack really.
GET /places
- Exact name match: where[name]=Elbow Cay
- Pattern name match: where[name]=Elb*
- Limit response to 10 results: limit=10
- Skip the first 20 results: offset=20
- By id: where[id]=362494
- By parent: where[parent]=362494
- Contains point: where[contains_point]=lat,long
GET /pois
- Exact name match: where[name]=Eiffel Tower
- Pattern name match: where[name]=Eif*
- Limit response to 10 results: limit=10
- Skip the first 20 results: offset=20
- Find all POIs in a place where[place]=362494
- Bounding box: where[contained_in]=n,s,e,w
- Poi Type (Eat | Sleep | Night | See | Shop | Do | General | Go): where[type]=Eat
- Full detail. Default is summary: detailed=true
- Within a certain number of metres to a point: where[close_to]=lat,long,metres
- If geocoded (has lat/long): where[geocoded]=true/false