-
Type:
Bug
-
Resolution: Incomplete
-
Priority:
Major - P3
-
None
-
Affects Version/s: 2.6.3, 2.6.4
-
Component/s: Geo
-
Labels:None
-
ALL
-
geoIntersect queries are getting very long to complete (~1300ms) against a collection with 5530 polygons / multi polygons (154.41mb)
The query generally returns 2 or 3 documents.
Sample Query:
db.locations.find({"l": {"$geoIntersects": {"$geometry": {"type": "Point", "coordinates": [-47.81302261352539, -21.18204307556152]}}}}, {"m":1,"s":1,"n":1, "c":1})
Query Explain Output:
{
"cursor" : "BtreeCursor l_2dsphere",
"isMultiKey" : true,
"n" : 2,
"nscannedObjects" : 2,
"nscanned" : 5,
"nscannedObjectsAllPlans" : 2,
"nscannedAllPlans" : 5,
"scanAndOrder" : false,
"indexOnly" : false,
"nYields" : 1,
"nChunkSkips" : 0,
"millis" : 1368,
"indexBounds" : {
"l" : [
[
"4f2211303",
"4f2211303"
],
[
"4f22113031",
"4f22113031"
],
[
"4f221130313",
"4f221130313"
],
[
"4f2211303133",
"4f2211303133"
],
[
"4f22113031331",
"4f22113031332"
]
]
},
"server" : "c0:27017",
"filterSet" : false
}
Collection Stats:
{
ns: "pinion.locations",
count: 5530,
size: 80728672,
avgObjSize: 14598,
storageSize: 161910784,
numExtents: 4,
nindexes: 4,
lastExtentSize: 90599424,
paddingFactor: 1,
systemFlags: 1,
userFlags: 1,
totalIndexSize: 11250176,
indexSizes: {
_id_: 188048,
l_2dsphere: 10358992,
t_1: 335216,
n_1: 367920
},
ok: 1
}
Collection Indexes:
[
{
"v" : 1,
"key" : {
"_id" : 1
},
"name" : "_id_",
"ns" : "pinion.locations"
},
{
"v" : 1,
"key" : {
"l" : "2dsphere"
},
"name" : "l_2dsphere",
"ns" : "pinion.locations",
"background" : true,
"safe" : true,
"2dsphereIndexVersion" : 2
},
{
"v" : 1,
"key" : {
"t" : 1
},
"name" : "t_1",
"ns" : "pinion.locations",
"background" : true,
"safe" : true
},
{
"v" : 1,
"key" : {
"n" : 1
},
"name" : "n_1",
"ns" : "pinion.locations",
"background" : true,
"safe" : true
}
]
The documents are large polygon areas like states, districts and municipalities.
Some are regular Polygons, other are MultiPolygons.
The database is running on a Replica Set managed by Compose.io (ex MongoHQ) with a lot of idle hardware.
I have MMS monitoring enabled should more stats are needed.
Locks and Queues are very very low, all other queries returns very fast, including geoNear queries on other collections.