I'm trying to get a list of current holders at specific times from a collection. My collection looks like this:
[
{
"time": 1,
"holdings": [
{ "owner": "A", "tokens": 2 },
{ "owner": "B", "tokens": 1 }
]
},
{
"time": 2,
"holdings": [
{ "owner": "B", "tokens": 2 }
]
},
{
"time": 3,
"holdings": [
{ "owner": "A", "tokens": 3 },
{ "owner": "B", "tokens": 1 },
{ "owner": "C", "tokens": 1 }
]
},
{
"time": 4,
"holdings": [
{ "owner": "C", "tokens": 0 }
]
}
]
tokens show the current holdings of an owner if the holdings have changed to the last document. I would like to change the collection so that holdings always includes the full current holdings for any point in time.
At time: 1, the holdings are: A: 2, B: 1.
At time: 2, the holdings are: A: 2, B: 2. The collections does not include A's holdings however, because they haven't changed. So what I'd like to get is:
[
{
"time": 1,
"holdings": [
{ "owner": "A", "tokens": 2 },
{ "owner": "B", "tokens": 1 }
]
},
{
"time": 2,
"holdings": [
{ "owner": "A", "tokens": 2 }, // merged from prev doc.
{ "owner": "B", "tokens": 2 }
]
},
{
"time": 3,
"holdings": [
{ "owner": "A", "tokens": 3 },
{ "owner": "B", "tokens": 1 },
{ "owner": "C", "tokens": 1 }
]
},
{
"time": 4,
"holdings": [
{ "owner": "A", "tokens": 3 }, // merged from prev
{ "owner": "B", "tokens": 1 }, // merged from prev
{ "owner": "C", "tokens": 0 }
]
}
]
From what I understand $mergeObjects does that, but I don't understand how I can merge all previous docs in order up to the current doc for each doc. So I'm looking for a way to combine setWindowFields with mergeObjects I think.
CodePudding user response：
This is a nice challenge.
So far, I got this complicated solution:
- Get all of our timestamps in all of our documents. This is the purpose of the first 4 steps.
$setWindowFieldsis used to accumulate this data. $groupby owner and calculate the empty timestamps aswantedTimes- next 5 steps.$setempty timestamps withtokens: nullto be filled with actual data and$unwindto separate - next 3 steps- Use
$setWindowFieldsto find the last known token for each owner at each timestamp. - Fill this last known state for documents with unknown token - 2 steps
$groupand format answer:
db.collection.aggregate([
{
$setWindowFields: {
sortBy: {time: 1},
output: {
allTimes: {$addToSet: "$time", window: {documents: ["unbounded", "current"]}
}
}
}
},
{
$setWindowFields: {
sortBy: {time: -1},
output: {
allTimes: {$addToSet: "$allTimes", window: {documents: ["unbounded", "current"]}
}
}
}
},
{
$set: {
allTimes: {
$reduce: {
input: "$allTimes",
initialValue: [],
in: {"$concatArrays": ["$$value", "$$this"]}
}
}
}
},
{$set: {allTimes: {$setIntersection: "$allTimes"}}},
{$unwind: "$holdings"},
{$sort: {time: 1}},
{$group: { _id: "$holdings.owner",
tokens: {$push: {tokens: "$holdings.tokens", time: "$time"}},
times: {$push: "$time"}, firstTime: {$first: "$time"},
allTimes: {$first: "$allTimes"}}
},
{
$addFields: {
wantedTimes: {
$filter: {
input: "$allTimes",
as: "item",
cond: {$gte: ["$$item", "$firstTime"]}
}
}
}
},
{
$project: {
tokens: 1,
wantedTimes: {$setDifference: ["$wantedTimes", "$times"]}
}
},
{
$set: {
data: {
$map: {
input: "$wantedTimes",
as: "item",
in: {time: "$$item", tokens: null}
}
}
}
},
{$project: {tokens: {"$concatArrays": ["$tokens", "$data"]}}},
{$unwind: "$tokens"},
{
$setWindowFields: {
partitionBy: "$_id",
sortBy: {"tokens.time": 1},
output: {
lastTokens: {
$push: "$tokens.tokens",
window: {documents: ["unbounded", "current"]}
}
}
}
},
{
$set: {
lastTokens: {
$filter: {
input: "$lastTokens",
as: "item",
cond: {$ne: ["$$item", null]}
}
}
}
},
{
$set: {
"tokens.tokens": {$ifNull: ["$tokens.tokens", {$last: "$lastTokens"}]}
}
},
{
$group: {
_id: "$tokens.time",
holdings: {$push: {owner: "$_id", tokens: "$tokens.tokens" }}
}
},
{$project: {time: "$_id", holdings: 1, _id: 0}},
{$sort: {time: 1}}
])
CodePudding user response：
From a performance perspective I recommend you split it into 2 calls, the first will be a quick findOne just to get the maximum time value in the collection.
Once you have that value the pipeline can be much leaner:
const maxItem = await db.collection.findOne({}).sort({ time: -1 });
db.collection.aggregate([
{
$unwind: "$holdings"
},
{
$group: {
_id: "$holdings.owner",
times: {
$push: {
time: "$time",
tokens: "$holdings.tokens"
}
},
minTime: {
$min: "$time"
}
}
},
{
$addFields: {
times: {
$reduce: {
input: {
$range: [
"$minTime",
maxItem.time 1 // this is max time
]
},
initialValue: {
values: [],
lastIndex: 0
},
in: {
values: {
"$concatArrays": [
"$$value.values",
[
{
$cond: [
{
$in: [
"$$this",
"$times.time"
]
},
{
"$arrayElemAt": [
"$times",
"$$value.lastIndex"
]
},
{
"$mergeObjects": [
{
tokens: 0
},
{
"$arrayElemAt": [
"$times",
{
$subtract: [
"$$value.lastIndex",
1
]
}
]
},
{
time: "$$this"
}
]
}
]
}
]
]
},
lastIndex: {
$cond: [
{
$in: [
"$$this",
"$times.time"
]
},
{
$sum: [
"$$value.lastIndex",
1
]
},
"$$value.lastIndex"
]
}
}
}
}
}
},
{
$unwind: "$times.values"
},
{
$group: {
_id: "$times.values.time",
holdings: {
$push: {
owner: "$_id",
tokens: "$times.values.tokens"
}
}
}
},
{
$project: {
_id: 0,
time: "$_id",
holdings: 1
}
},
{
$sort: {
time: 1
}
}
])
This is still quite a heavy query as it requires to $unwind and $group the entire collection, however there is no workaround this due to the requirements. if the collection is too big for this approach I recommend iteration owner by owner, or time by time and doing separate updates accordingly.
If you don't care about performance at all and want it in a single query you can still use the same pipeline, you will have to first extract the max time in the collection, this will require you to add an initial $group stage, like so:
db.collection.aggregate([
{
$group: {
_id: null,
maxTime: {
$max: "$time"
},
roots: {
$push: "$$ROOT"
}
}
},
{
$unwind: "$roots"
},
{
$replaceRoot: {
newRoot: {
"$mergeObjects": [
"$roots",
{
maxTime: "$maxTime"
}
]
}
}
},
... same pipeline ...
])