Mongodb best schema design for key value pairs indexation - mongodb

Question

I experience some difficulties to find an answer to this question by myself :
What is the best way to store Key/Value Pair in a mongo document AND put an index on it in order to perform multi-criteria searches.
I have documents stored in a mongo database which look like that :
{
_id:xxxxxxxxxx,
name : "x1",
tags : [
{key:"color", value:"blue"},
{key:"size", value:"L"},
{key:"weight", value:5}
]
}
Each key of the "tags" property is unique (I cannot have more than 1 color or size per document) and each key is optionnal (I can have document without "color" specified).
According to the "Indexes FAQ", I created an index like this :
{
"tags.key":1, "tags.value":1
}
To perform a query against this index I use this query (to return all blue items) :
.find({
tags:{$elemMatch:{"key":"color","value":"blue"}}
})
An explain() shows that the query uses the index. Good!
But now, if I want to make a multi-criteria search, I use this query (to return all blue items with an "L" size) :
.find({
$and:[
{tags:{$elemMatch:{"key":"color","value":"blue"}}},
{tags:{$elemMatch:{"key":"size","value":"L"}}}
]
})
This query is working, but the execution plan shows an IXSAN that scans more documents than the number returned :
{
"cursor" : "BtreeCursor Keys",
"isMultiKey" : true,
"n" : 22,
"nscannedObjects" : 53,
"nscanned" : 53,
"nscannedObjectsAllPlans" : 53,
"nscannedAllPlans" : 109,
"scanAndOrder" : false,
"indexOnly" : false,
"nYields" : 0,
"nChunkSkips" : 0,
"millis" : 2,
"indexBounds" : {
"tags.key" : [
[
"color",
"color"
]
],
"tags.value" : [
[
"blue",
"blue"
]
]
},
"server" : "BLABLABLA:27017",
"filterSet" : false,
"stats" : {
"type" : "PROJECTION",
"works" : 55,
"yields" : 0,
"unyields" : 0,
"invalidates" : 0,
"advanced" : 22,
"needTime" : 0,
"needFetch" : 0,
"isEOF" : 1,
"children" : [
{
"type" : "KEEP_MUTATIONS",
"works" : 55,
"yields" : 0,
"unyields" : 0,
"invalidates" : 0,
"advanced" : 22,
"needTime" : 31,
"needFetch" : 0,
"isEOF" : 1,
"children" : [
{
"type" : "FETCH",
"works" : 54,
"yields" : 0,
"unyields" : 0,
"invalidates" : 0,
"advanced" : 22,
"needTime" : 31,
"needFetch" : 0,
"isEOF" : 1,
"alreadyHasObj" : 0,
"forcedFetches" : 0,
"matchTested" : 22,
"children" : [
{
"type" : "IXSCAN",
"works" : 53,
"yields" : 0,
"unyields" : 0,
"invalidates" : 0,
"advanced" : 53,
"needTime" : 0,
"needFetch" : 0,
"isEOF" : 1,
"keyPattern" : "{ tags.key: 1.0, tags.value: 1.0 }",
"boundsVerbose" : "field #0['tags.key']: [\"color\", \"userid\"], field #1['Keys.v']: [\"4189\", \"4189\"]",
"isMultiKey" : 1,
"yieldMovedCursor" : 0,
"dupsTested" : 53,
"dupsDropped" : 0,
"seenInvalidated" : 0,
"matchTested" : 0,
"keysExamined" : 53,
"children" : []
}
]
}
]
}
]
}
}
It seems that only the first $elemMatch is considered in IXSCAN.
So, is there a way to improve this or no?
Either by changing the query, either by changing the design of my document?
Your help is welcome!
Fred