-
Type:
Bug
-
Resolution: Duplicate
-
Priority:
Major - P3
-
None
-
Affects Version/s: 6.0.9
-
Component/s: None
-
None
-
Server Triage
-
ALL
collection.countDocuments is converted in an aggregate with $math + $group by _id.
db.getCollection("test").countDocuments({"type":"test","auditEnabled":{"$ne":true}})
becomes:
db.getCollection("test").aggregate( [ { "$match" : { "type" : "test", "auditEnabled" : { "$ne" : true } } }, { "$group" : { "_id" : NumberInt(1), "n" : { "$sum" : NumberInt(1) } } } ], { "allowDiskUse" : false } );
Problem is that if documents in the collection are large, they are fully loaded in memory (apparently) to aggregate them.
Why don't you add a $project by _id after $match? This will greatly improve (10x in our case) performance. Something like:
db.getCollection("test").aggregate( [ { "$match" : { "type" : "test", "auditEnabled" : { "$ne" : true } } }, { "$project" : { "_id" : 1 } }, { "$group" : { "_id" : NumberInt(1), "n" : { "$sum" : NumberInt(1) } } } ], { "allowDiskUse" : false } );
- backported by
-
SERVER-57518 16% performance loss switching from Count to CountDocuments
-
- Backlog
-