Well you cannot project like that in the aggregation pipeline. If you want to act on array elements within an aggregation statement you first need to $unwind
the array and then either $match
the required element(s) or as in your case choose the $first
item using an additional $group
stage.
Your question does not show the structure of a document so I'll just use a sample, as my "chairs" collection:
{
"_id": 1,
"type": "chair",
"chairtype": "A",
"versions": [
{
"revision": 1,
"content": {
"name": "ABC",
"value": 10
}
},
{
"revision": 2,
"content": {
"name": "BBB",
"value": 15
}
}
]
}
{
"_id": 2,
"type": "chair",
"chairtype": "A",
"versions": [
{
"revision": 1,
"content": {
"name": "CCC",
"value": 20
}
},
{
"revision": 2,
"content": {
"name": "BAB",
"value": 12
}
}
]
}
Minimal, but enough to get the point. Now the aggregate statement:
db.chairs.aggregate([
// Normal query matching, which is good
{ "$match": { "type": "chair" } },
// Unwind the array to de-normalize
{ "$unwind": "$versions" },
// Group by the document in order to get the "first" array element
{ "$group": {
"_id": "$_id",
"chairtype": { "$first": "$chairtype" },
"versions": { "$first": "$versions" }
}},
// Then group by "chairtype" to get your average values
{ "$group": {
"_id": "$chairtype",
"MID": {"$avg": "$versions.content.value"}
}}
])
Of course if your actual document has nested arrays then you will be "unwinding" and "matching" the required elements. But that is the general process of "narrowing down" the array contents to the elements you need.