Skip to content
benoitc edited this page Nov 12, 2012 · 4 revisions

This is a spatial extension for Rcouch based on Geocouch.

Features:

  • Index GEOJSON geometries
  • Bounding Box search
  • k-nearest-neighbour (KNN) queries
  • lists functions

Usage

Create a database:

curl -X PUT http://127.0.0.1:5984/places

Add a Design Document with a spatial function:

curl -X PUT -d '{"spatial":{"points":"function(doc) {\n    if (doc.loc) {\n        emit({\n            type: \"Point\",\n            coordinates: [doc.loc[0], doc.loc[1]]\n        }, [doc._id, doc.loc]);\n    }};"}}' http://127.0.0.1:5984/places/_design/main

Put some data into it:

curl -X PUT -d '{"loc": [-122.270833, 37.804444]}' http://127.0.0.1:5984/places/oakland
curl -X PUT -d '{"loc": [10.898333, 48.371667]}' http://127.0.0.1:5984/places/augsburg

Make a bounding box request:

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/points?bbox=0,0,180,90'

It should return:

{"update_seq":3,"rows":[
{"id":"augsburg","bbox":[10.898333,48.371667,10.898333,48.371667],"geometry":{"type":"Point","coordinates":[10.898333,48.371667]},"value":["augsburg",[10.898333,48.371667]]}
]}

Or trigger a k-nearest-neighbour-query to request the n nearest geometries relative to a given query point:

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/points?n=2&q=50,10&spherical=true'

If the parameter spherical=true is set, the Haversine formula is used to calculate spherical distances for nearest-neighbour-queries. In this case the geometries are expected to use the coordinate system WGS 84.

The Design Document Function

function(doc) { if (doc.loc) { emit({ type: "Point", coordinates: [doc.loc[0], doc.loc[1]] }, [doc._id, doc.loc]); }};"

It uses the emit() from normal views. The key is a GeoJSON geometry, the value is any arbitrary JSON. All geometry types (even GemetryCollections) are supported.

If the GeoJSON geometry contains a bbox property it will be used instead of calculating it from the geometry (even if it's wrong, i.e. is not the actual bounding box).

Bounding box search and the date line

A common problem when performing bounding box searches is the date line/poles. As the bounding box follows the GeoJSON specification, where the first two numbers are the lower left coordinate, the last two numbers the upper right coordinate, it is easy to map it over the date line/poles. The lower coordinate would have a higher value than the upper one. Such a bounding box has a seems invalid at first glance, but isn't. For example a bounding box like 110,-60,-30,15 would include Australia and South America, but not Africa.

GeoCouch operates on a plane and doesn't perform spherical calculations. Therefore the bounds of the plane needs to be set explicitly with the plane_bounds parameter. If bounding boxes are flipped, a search across those bounds will be performed automatically. Give it a try (with the same Design Document as above). Insert some Documents:

curl -X PUT -d '{"loc": [17.15, -22.566667]}' http://127.0.0.1:5984/places/namibia
curl -X PUT -d '{"loc": [135, -25]}' http://127.0.0.1:5984/places/australia
curl -X PUT -d '{"loc": [-52.95, -10.65]}' http://127.0.0.1:5984/places/brasilia

And request only Australia and Brasilia:

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/points?bbox=110,-60,-30,15&plane_bounds=-180,-90,180,90'

The result is as expected:

{"update_seq":6,"rows":[
{"id":"australia","bbox":[135,-25,135,-25],"geometry":{"type":"Point","coordinates":[135,-25]},"value":["australia",[135,-25]]},
{"id":"brasilia","bbox":[-52.95,-10.65,-52.95,-10.65],"geometry":{"type":"Point","coordinates":[-52.95,-10.65]},"value":["brasilia",[-52.95,-10.65]]}
]}

The bounding with the same numbers, but different order (-30,-60,110,15) would only return Namibia:

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/points?bbox=-30,-60,110,15&plane_bounds=-180,-90,180,90'

{"update_seq":6,"rows":[
{"id":"namibia","bbox":[17.15,-22.566667,17.15,-22.566667],"geometry":{"type":"Point","coordinates":[17.15,-22.566667]},"value":["namibia",[17.15,-22.566667]]}
]}

The plane_bounds parameter is also supported for k-nearest-neighbour-queries. Note that you don't have to set the plane_bounds parameter when spherical=true is enabled.

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/points?n=3&q=175,-25&plane_bounds=-180,-90,180,90'

{"update_seq":6,"rows":[
{"id":"australia","bbox":[135,-25,135,-25],"geometry":{"type":"Point","coordinates":[135,-25]},"value":"australia"},
{"id":"oakland","bbox":[-122.270833,37.804444,-122.270833,37.804444],"geometry":{"type":"Point","coordinates":[-122.270833,37.804444]},"value":"oakland"},
{"id":"brasilia","bbox":[-52.95,-10.65,-52.95,-10.65],"geometry":{"type":"Point","coordinates":[-52.95,-10.65]},"value":"brasilia"}
]}

List function support

GeoCouch supports List functions just as CouchDB does for Views. This way you can output any arbitrary format, e.g. GeoRSS.

As an example we output the points as WKT. Add a new Design Document with an additional List function (the rest is the same as above). Make sure you use the right _rev:

curl -X PUT -d '{"_rev": "1-121efc747b00743b8c7621ffccf1ac40", "lists": {"wkt": "function(head, req) {\n    var row;\n    while (row = getRow()) {\n        send(\"POINT(\" + row.geometry.coordinates.join(\" \") + \")\\n\");\n    }\n};"}, "spatial":{"points":"function(doc) {\n    if (doc.loc) {\n        emit({\n            type: \"Point\",\n            coordinates: [doc.loc[0], doc.loc[1]]\n        }, [doc._id, doc.loc]);\n    }};"}}' http://127.0.0.1:5984/places/_design/main

Now you can request this List function as you would do for CouchDB, though with a different Design handler (_spatial/_list instead of _list ):

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/_list/wkt/points?bbox=-180,-90,180,90'

The result is:

POINT(10.898333 48.371667)
POINT(-122.270833 37.804444)
POINT(17.15 -22.566667)
POINT(135 -25)
POINT(-52.95 -10.65)

Using List functions from Design Documents other than the one containing the Spatial functions is supported as well. This time we add the Document ID in parenthesis:

curl -X PUT -d '{"lists": {"wkt": "function(head, req) {\n    var row;\n    while (row = getRow()) {\n        send(\"POINT(\" + row.geometry.coordinates.join(\" \") + \") (\" + row.id + \")\\n\");\n    }\n};"}}' http://127.0.0.1:5984/places/_design/listfunonly

curl -X GET 'http://localhost:5984/places/_design/listfunonly/_spatial/_list/wkt/main/points?bbox=-180,-90,180,90'

Other supported query arguments

stale

stale=ok is supported. The spatial index won't be rebuilt even if new Documents were added. It works for normal spatial queries as well as for the spatial List functions.

count

count is a boolean. count=true will only return the number of geometries the query will return, not the geometry themselves.

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/points?bbox=0,0,180,90&count=true'

{"count":1}

limit

limit limts the number of results returned

skip

skip allows you to start to return the results at a certain offset.

Compaction, cleanup and info

The API of GeoCouch's spatial indexes is similar to the one for the Views. Compaction of spatial indexes is per Design Document, thus:

curl -X POST 'http://localhost:5984/places/_design/main/_spatial/_compact' -H 'Content-Type: application/json'

To cleanup spatial indexes that are no longer in use (this is per database):

curl -X POST 'http://localhost:5984/places/_spatial_cleanup' -H 'Content-Type: application/json'

To get information about the spatial indexes of a certain Design Document use the the _info handler:

curl -X GET 'http://localhost:5984/places/_design/main/_spatial/_info'