Optimizing GeoJSON MongoDB queries - mongodb

Question

I have some polygons stored in GeoJSON in my mongoDB.
The client sends a list of boxes to load (the boxes are cells of a regular grid).
The normal way to retrieve them is to make a GeoJSON query with each box, but it is slow when I have many boxes.
I do not want to retrieve duplicates (polygons which lie on two cells are returned twice), so I make a list of retrieved polygons pks to ignore them in the next queries.
Given:
box = [ [ [ 0, 0 ], [ 1, 0 ], [ 1, 1 ], [ 0, 1 ], [ 0, 0 ] ] ] // the box to load
pks = [ ObjectId("54cf535cfe022e01ab4932f5"), ObjectId("54cf535cfe022e01ab4932f6") ] // the list of polygons already retrieved
With mongoDB I would have something like this:
for box in boxes:
db.places.find( { points:
{ $geoIntersects: { $geometry: { type: "Polygon" , coordinates: box } } },
{ _id: { $nin: pks } }
} )
I use MongoEngine, so I have the following:
pks = []
for box in boxes:
p = Polygon.objects(points__geo_intersects=box, pk__nin=pks)
if len(p)>0:
pks += p.scalar("id")
I have three questions:
1. Is there a more efficient way to query the polygons with this method?
2. Would it be faster to use a Cell object containing the list of reference of polygons which lie on the cell?
In MongoEngine I would have the following model:
class Cell(Document):
x = DecimalField()
y = DecimalField()
polygons = ListField(ReferenceField('Polygon'))
meta = {
'indexes': [[ ("x", 1), ("y", 1) ]]
}
The list of boxes to load would be the coordinates corresponding to the cells to load.
This would give with MongoEngine:
polygons = {}
for b in boxes:
cell = Cell.objects.get(x=b['x'], y=b['y'])
for polygon in cell.polygons:
if not polygons.has_key(polygon.pk):
polygons[polygon.pk] = polygon.to_json()
3. Is there a more efficient way to query the polygons with this method? (I think I should use select_related(), and maybe it is possible to filter the polygons directly in the mongoDB query to avoid retrieving duplicates)

Answer 1

After making a few benchmarks, I can answer my own questions: it is not noticeably faster to use an intermediate object such as Cell.
However, using a list of primary keys in the query, like this:
pks = []
for box in boxes:
p = Polygon.objects(points__geo_intersects=box, pk__nin=pks)
if len(p)>0:
pks += p.scalar("id")
is much slower than using a dictionary to ignore duplicates, like this:
polygons = {}
for box in boxes:
p = Polygon.objects(points__geo_intersects=box)
for polygon in p:
if not polygons.has_key(polygon.pk):
polygons[polygon.pk] = polygon.to_json()