Difficulty optimizing Mongo distinct query to use indexes - mongodb

Question

I am having difficulty persuading Mongo to run a distinct query that looks like it should be covered by the indexes without fetching a large number of documents in the collection.
My documents have the general form:
{
_tenantId: 'someString',
_productCategory: 'some string from a smallish set'
...
}
I have an index on (_tenantId, _productCategory).
I want to find out what the set of distinct product categories is for a given tenant, so the query is:
db.products.distinct( '_productCategory', { _tenantId: '463171c3-d15f-4699-893d-3046327f8e1f'})
This runs rather slowly (several seconds for a collection of around half a million products against a local DB, which is Mongo 3.2.9). Against our pre-production SaaS-based Mongo (which is probably more memory constrained than my local instance which has free run of my machine) it take several 10s of seconds for the same data.
Explaining the query yields:
{
"queryPlanner" : {
"plannerVersion" : 1,
"namespace" : "engage-prod.products",
"indexFilterSet" : false,
"parsedQuery" : {
"_tenantId" : {
"$eq" : "463171c3-d15f-4699-893d-3046327f8e1f"
}
},
"winningPlan" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : {
"_tenantId" : 1,
"_productCategory" : 1
},
"indexName" : "_tenantId_1__productCategory_1",
"isMultiKey" : false,
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 1,
"direction" : "forward",
"indexBounds" : {
"_tenantId" : [
"[\"463171c3-d15f-4699-893d-3046327f8e1f\", \"463171c3-d15f-4699-893d-3046327f8e1f\"]"
],
"_productCategory" : [
"[MinKey, MaxKey]"
]
}
}
},
"rejectedPlans" : [ ]
},
"executionStats" : {
"executionSuccess" : true,
"nReturned" : 406871,
"executionTimeMillis" : 358,
"totalKeysExamined" : 406871,
"totalDocsExamined" : 406871,
"executionStages" : {
"stage" : "FETCH",
"nReturned" : 406871,
"executionTimeMillisEstimate" : 80,
"works" : 406872,
"advanced" : 406871,
"needTime" : 0,
"needYield" : 0,
"saveState" : 3178,
"restoreState" : 3178,
"isEOF" : 1,
"invalidates" : 0,
"docsExamined" : 406871,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 406871,
"executionTimeMillisEstimate" : 40,
"works" : 406872,
"advanced" : 406871,
"needTime" : 0,
"needYield" : 0,
"saveState" : 3178,
"restoreState" : 3178,
"isEOF" : 1,
"invalidates" : 0,
"keyPattern" : {
"_tenantId" : 1,
"_productCategory" : 1
},
"indexName" : "_tenantId_1__productCategory_1",
"isMultiKey" : false,
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 1,
"direction" : "forward",
"indexBounds" : {
"_tenantId" : [
"[\"463171c3-d15f-4699-893d-3046327f8e1f\", \"463171c3-d15f-4699-893d-3046327f8e1f\"]"
],
"_productCategory" : [
"[MinKey, MaxKey]"
]
},
"keysExamined" : 406871,
"dupsTested" : 0,
"dupsDropped" : 0,
"seenInvalidated" : 0
}
}
},
"serverInfo" : {
"host" : "Stevens-MacBook-Pro.local",
"port" : 27017,
"version" : "3.2.9",
"gitVersion" : "22ec9e93b40c85fc7cae7d56e7d6a02fd811088c"
},
"ok" : 1
}
Note that even though it runs an IXSCAN it still returns over 400K documents (nReturned).
If I create a compound field _tenantAndProductCategory containing a lexical concatenation (with a : separator) and index that so it's a single field index, then the query:
db.products.explain('executionStats').distinct( '_productTenantAndCategory', { _productTenantAndCategory: {$gte: '463171c3-d15f-4699-893d-3046327f8e1f',$lt: '463171c3-d15f-4699-893d-3046327f8e1g'}})
works entirely within the index and yields:
{
"queryPlanner" : {
"plannerVersion" : 1,
"namespace" : "engage-prod.products",
"indexFilterSet" : false,
"parsedQuery" : {
"$and" : [
{
"_productTenantAndCategory" : {
"$lt" : "463171c3-d15f-4699-893d-3046327f8e1g"
}
},
{
"_productTenantAndCategory" : {
"$gte" : "463171c3-d15f-4699-893d-3046327f8e1f"
}
}
]
},
"winningPlan" : {
"stage" : "PROJECTION",
"transformBy" : {
"_id" : 0,
"_productTenantAndCategory" : 1
},
"inputStage" : {
"stage" : "DISTINCT_SCAN",
"keyPattern" : {
"_productTenantAndCategory" : 1
},
"indexName" : "_productTenantAndCategory_1",
"isMultiKey" : false,
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 1,
"direction" : "forward",
"indexBounds" : {
"_productTenantAndCategory" : [
"[\"463171c3-d15f-4699-893d-3046327f8e1f\", \"463171c3-d15f-4699-893d-3046327f8e1g\")"
]
}
}
},
"rejectedPlans" : [ ]
},
"executionStats" : {
"executionSuccess" : true,
"nReturned" : 62,
"executionTimeMillis" : 0,
"totalKeysExamined" : 63,
"totalDocsExamined" : 0,
"executionStages" : {
"stage" : "PROJECTION",
"nReturned" : 62,
"executionTimeMillisEstimate" : 0,
"works" : 63,
"advanced" : 62,
"needTime" : 0,
"needYield" : 0,
"saveState" : 0,
"restoreState" : 0,
"isEOF" : 1,
"invalidates" : 0,
"transformBy" : {
"_id" : 0,
"_productTenantAndCategory" : 1
},
"inputStage" : {
"stage" : "DISTINCT_SCAN",
"nReturned" : 62,
"executionTimeMillisEstimate" : 0,
"works" : 63,
"advanced" : 62,
"needTime" : 0,
"needYield" : 0,
"saveState" : 0,
"restoreState" : 0,
"isEOF" : 1,
"invalidates" : 0,
"keyPattern" : {
"_productTenantAndCategory" : 1
},
"indexName" : "_productTenantAndCategory_1",
"isMultiKey" : false,
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 1,
"direction" : "forward",
"indexBounds" : {
"_productTenantAndCategory" : [
"[\"463171c3-d15f-4699-893d-3046327f8e1f\", \"463171c3-d15f-4699-893d-3046327f8e1g\")"
]
},
"keysExamined" : 63
}
}
},
"serverInfo" : {
"host" : "Stevens-MacBook-Pro.local",
"port" : 27017,
"version" : "3.2.9",
"gitVersion" : "22ec9e93b40c85fc7cae7d56e7d6a02fd811088c"
},
"ok" : 1
}
Having to build single field indexes with manually compounded keys for all the aggregation queries I need is not a very desirable path to follow. Since all the information is present in the compound index I started with, why can't Mongo execute the original distinct query with cover by that index? Is there anything I can do to overcome this in the way of query optimization?
Note This is actually a sub-problem of a slightly more complex one involving an aggregation pipeline to actually count the number of occurrences of each category, but I am restricting my question for now to the simpler distinct query since it seems to capture the essence of failure to use an index that should cover things (which I was also seeing in the aggregation pipeline case), while being a simpler overall query.