How to aggregate on a non-standard week? - database

Question

I have a collection with data like below:
{
"_id": ObjectId("95159a08a27971c35a2683f"),
"Date": ISODate("2018-04-03T07:00:00Z"),
"Employee": "Bill",
"Hours": 7.5
}
{
"_id": ObjectId("372c6be4912fdd32398382f"),
"Date": ISODate("2018-04-05T07:00:00Z"),
"Employee": "Bill",
"Hours": 2
}
And I would like to get the total hours per week, but the week needs to start on Saturday and end on Friday. If I were working with a standard week I would just do:
db.myCollection.aggregate(
{$match: {
Employee: "Bill",
Date: {
$gte: ISODate("2018-03-15T07:00:00Z"),
$lte: ISODate("2018-04-06T07:00:00Z")
}
}},
{$group: {
_id: {$week: "$Date"},
hours: {$sum: "$Hours"}
}
}
)
which works fine for standard weeks that start on Sunday and end Saturday.
How would I modify this to work with the non-standard week that I described? Would I be better off querying the data and aggregating it manually at the code level? Could I aggregate by week for Date - 1 day or would that be too weird?
UPDATE:
If anyone needs to make this work across years they can use the updated version of Neil's answer below:
{ "$group": {
"_id": {
"week":{
"$let": {
"vars": {
"satWeek": {
"$cond": {
"if": { "$eq": [ { "$dayOfWeek": "$Date" }, 7 ] },
"then": { "$add": [ { "$week": "$Date" }, 1 ] },
"else": { "$week": "$Date" }
}
}
},
"in": {
"$cond": {
"if": { "$gt": ["$$satWeek", 52] },
"then": 0,
"else": "$$satWeek"
}
}
}
},
"year": { "$year": "$Date" }
},
"hours": { "$sum": "$hours" }
}}

Answer 1

Well $week returns based on the "input" date, so Date - 1 day would still be in the "previous week" based on what it would return from that input even if you adjusted it by one day.
It's certainly always best to make the "server" do this type of thing, otherwise you're just pulling a lot of data "over the wire" which kind of obviates the point of using a database in the first place.
In brief, "starting on a Saturday" simply means that what was Week 1 becomes Week 2, etc when the day is Saturday. So week + 1. And the only real caveat is that anything above Week 52 becomes Week 0.
Therefore:
{ "$group": {
"_id": {
"$let": {
"vars": {
"satWeek": {
"$cond": {
"if": { "$eq": [ { "$dayOfWeek": "$Date" }, 7 ] },
"then": { "$add": [ { "$week": "$Date" }, 1 ] },
"else": { "$week": "$Date" }
}
}
},
"in": {
"$cond": {
"if": { "$gt": ["$$satWeek", 52] },
"then": 0,
"else": "$$satWeek"
}
}
}
},
"hours": { "$sum": "$hours" }
}}
The main adjusting point there being based around the test for $dayOfWeek, which returns 7 for Saturday.
Of course whether it's 52 or 53 as the boundary depends on whether it's a leap year or not, but since you want to aggregate per "week" then I presume you only want a "year at most", and then can adjust that as an input parameter based on whether your date selection is within a leap year or not.
Or of course, adjust the coding to be even more considerate of that. But the basic principle is adjust the "output week" instead of the "input Date"
As an alternate case, then I guess Date plus 1 day would actually get you the same result be skewing all dates forward:
{ "$group": {
"_id": { "$week": { "$add": [ "$Date", 1000 * 60 * 60 * 24 ] } },
"hours": { "$sum": : "$hours" }
}}
And if you needed local timezone adjustment, then from MongoDB 3.6 you can simply include the timezone information to adjust by:
{ "$group": {
"_id": {
"$week": {
"date": { "$add": [ "$Date", 1000 * 60 * 60 * 24 ] },
"timezone": "America/Chicago"
}
},
"hours": { "$sum": : "$hours" }
}}
Also note as of MongoDB 3.4 there is $isoWeek and like functions, where you get some different handling:
Returns the week number in ISO 8601 format, ranging from 1 to 53. Week numbers start at 1 with the week (Monday through Sunday) that contains the year’s first Thursday.
So all math would be based from Monday instead of Sunday and in consideration of which day of the year is counted as the "first week", as well as starting from 1 instead of 0.