I have collection containing date field. I\'m Grouping records by week and other related fields.
This is my aggregation query:
For mongo >= v3.4, look at weekStart. The idea is to substruct milliseconds from given Timestamp
db.raw.aggregate([
// stage 1
{ "$match" : {
"Timestamp":{
"$gte": ISODate("2012-05-30"),
"$lt": ISODate("2014-07-31")
}
}},
// stage 2
{ "$project" : {
ApplicationId: 1,
Country: 1,
week: {$isoWeek: "$Timestamp"},
// [TRICK IS HERE] Timestamp - dayOfWeek * msInOneDay
weekStart: { $dateToString: { format: "%Y-%m-%d", date: { // convert date
$subtract: ["$Timestamp", {$multiply: [ {$subtract:[{$isoDayOfWeek: "$Timestamp"},1]}, 86400000]}]
}}},
// stage 3
{ "$group" : {
"_id":{
"ApplicationId": "$ApplicationId",
"Country": "$Country",
"week": "$week"
},
"Date":{ "$first": "$weekStart" },
"Visits": { "$sum": 1 }
}}
])
You seem to want a "date value" representing the date at the start of the week. Your best approach is "date math" with a little help from the aggregation operator $dayOfWeek:
db.raw.aggregate([
{ "$match" : {
"Timestamp":{
"$gte": new Date("2012-05-30"),
"$lt": new Date("2014-07-31")
}
}},
{ "$group" : {
"_id":{
"ApplicationId": "$ApplicationId",
"Country": "$Country",
"weekStart":{
"$subtract": [
{ "$subtract": [
{ "$subtract": [ "$Timestamp", new Date("1970-01-01") ] },
{ "$cond": [
{ "$eq": [{ "$dayOfWeek": "$Timestamp" }, 1 ] },
0,
{ "$multiply": [
1000 * 60 * 60 * 24,
{ "$subtract": [{ "$dayOfWeek": "$Timestamp" }, 1 ] }
]}
]}
]},
{ "$mod": [
{ "$subtract": [
{ "$subtract": [ "$Timestamp", new Date("1970-01-01") ] },
{ "$cond": [
{ "$eq": [{ "$dayOfWeek": "$Timestamp" }, 1 ] },
0,
{ "$multiply": [
1000 * 60 * 60 * 24,
{ "$subtract": [{ "$dayOfWeek": "$Timestamp" }, 1 ] }
]}
]}
]},
1000 * 60 * 60 * 24
]}
]
}
},
"Date":{ "$first": "$Timestamp" },
"Visits": { "$sum": 1 }
}}
])
Or a little cleaner with $let from MongoDB 2.6 and upwards:
db.raw.aggregate([
{ "$match" : {
"Timestamp":{
"$gte": new Date("2012-05-30"),
"$lt": new Date("2014-07-31")
}
}},
{ "$group" : {
"_id":{
"ApplicationId": "$ApplicationId",
"Country": "$Country",
"weekStart":{
"$let": {
"vars": {
"dayMillis": 1000 * 60 * 60 * 24,
"beginWeek": {
"$subtract": [
{ "$subtract": [ "$Timestamp", new Date("1970-01-01") ] },
{ "$cond": [
{ "$eq": [{ "$dayOfWeek": "$Timestamp" }, 1 ] },
0,
{ "$multiply": [
1000 * 60 * 60 * 24,
{ "$subtract": [{ "$dayOfWeek": "$Timestamp" }, 1 ] }
]}
]}
]
}
},
"in": {
"$subtract": [
"$$beginWeek",
{ "$mod": [ "$$beginWeek", "$$dayMillis" ]}
]
}
}
}
},
"Date":{ "$first": "$Timestamp" },
"Visits": { "$sum": 1 }
}}
])
The resulting value in the "grouping" is the epoch milliseconds that represents the start of the day at the start of the week. The "start of the week" is generally considered to be "Sunday", so if you intend another day then you would need to adjust by the appropriate amount. The $add operator with the $dayMillis
variable value can be used here to apply "Monday" for example.
It's not a date object, but something that you can easily feed to another method to construct a date object in post processing.
Also note that other things you are using such as $first usually require that the documents are sorted in a particular order, or generally by your "Timestamp" values. If those documents are not already ordered then you either $sort first or use an operator such as $min to get the first actual timestamp in the range.
With MongoDB 3.6
{
'$project' : {
'firstDateOfWeek': {
'$dateFromString': {
'dateString': {
'$concat': [
{
'$toString': '$_id.year'
},
'-',
{
'$toString': '$_id.week'
}
]
},
'format': "%G-%V"
}
}
}
}