Debugging MongoDB’s 2dsphere geoIndex

When I first learned about MongoDB’s geospatial capabilities similar to PostGIS, I was beyond excited because it meant I would not have to write helper functions to find data along geohash borders. Currently my data is indexed by geohash, which has great ordering properties, but is a little unwieldy for the “find me x near y”-type queries.

Unfortunately, it seems as though the 2dsphere index doesn’t cover geoWithin queries. As my collection of geoJSON documents grew from GBs to TBs, I noticed my tools taking extraordinary amounts of time to perform find queries, which caused me to investigate what was happening under the hood.

First, I double-checked that my geospatial index was present:

mongos> db.my_collection.getIndexes()
[
	{
		"v" : 1,
		"key" : {
			"_id" : 1
		},
		"name" : "_id_",
		"ns" : "telemetry.********"
	},
	{
		"v" : 1,
		"key" : {
			"value.avg_location" : "2dsphere"
		},
		"name" : "geoindex",
		"ns" : "telemetry.*******",
		"2dsphereIndexVersion" : 2
	}
]

So, the index was clearly there. I also verified that its size was appropriate — it was around 50mb, which seemed fine for this amount of test data.

Next, I ran .explain() on a query to see if it was hitting the index:

var query = {
    "value.avg_location": {
        "$geoWithin": {
            "$polygon": [
                [-122.07496047019957,
                    37.65882781135178
                ],
                [-122.07496047019957,
                    37.663906822384874
                ],
                [-122.0871913433075,
                    37.663906822384874
                ],
                [-122.0871913433075,
                    37.65882781135178
                ],
                [-122.07496047019957,
                    37.65882781135178
                ]
            ]
        }
    }
};
db.my_collection.find(query).explain("executionStats");

Nope:

"winningPlan" : {
	"stage" : "COLLSCAN",
	"filter" : {
		"value.avg_location" : {
			"$geoWithin" : {
				"$polygon" : [
					[
						-122.07496047019957,
						37.65882781135178
					],
					[
						-122.07496047019957,
						37.663906822384874
					],
					[
						-122.0871913433075,
						37.663906822384874
					],
					[
						-122.0871913433075,
						37.65882781135178
					],
					[
						-122.07496047019957,
						37.65882781135178
					]
				]
			}
		}
	},
	"direction" : "forward"
},
"rejectedPlans" : [ ]
"executionStats" : {
	"nReturned" : 164,
	"executionTimeMillis" : 341340,
	"totalKeysExamined" : 0,
	"totalDocsExamined" : 10760184,
	"executionStages" : {
		"stage" : "SINGLE_SHARD",
		"nReturned" : 164,
		"executionTimeMillis" : 341340,
		"totalKeysExamined" : 0,
		"totalDocsExamined" : 10760184,
		"totalChildMillis" : NumberLong(341334),
		"shards" : [
			{
				"shardName" : "rs0",
				"executionSuccess" : true,
				"executionStages" : {
					"stage" : "COLLSCAN",
					"filter" : {
                                           ....
					},
					"nReturned" : 164,
					"executionTimeMillisEstimate" : 334786,
					"works" : 10760186,
					"advanced" : 164,
					"needTime" : 10760021,
					"needFetch" : 0,
					"saveState" : 84076,
					"restoreState" : 84076,
					"isEOF" : 1,
					"invalidates" : 0,
					"direction" : "forward",
					"docsExamined" : 10760184
				}
			}
		]
	}
},
"ok" : 1

So, from the “stage” param COLLSCAN, we can clearly see it’s doing a collection scan to find the documents. We want it to say something like “IXSCAN” to indicate that it’s using the index.

I tried deleting and rebuilding the index:

mongos> db.mycoll.dropIndex("geoindex")
true
mongos> db.mycoll.createIndex({"value.avg_location" : "2dsphere" }, {name: "geoindex"});
mongos>

Still, no dice!

Okay, well Mongo’s docs says that the most efficient queries will query on AND return only the indexed fields. So, I tried limiting (“projecting” in Mongo-parlance) the query to just return the “value.avg_location” field, but we ended up still doing a “COLLSCAN.”

UPDATE: I ended up fixing this by re-structuring the query. It turns out the original syntax I used for the polygon query is the older syntax, which doesn’t hit the index. The newer syntax looks like this:

{
  "location": {
    $geoWithin: {
      $geometry: {
        type: "Polygon",
        coordinates: [ [ <the coords> ] ]
      }
    }
  }
}