Overview
We have a Mongo database collecting time series data from multiple different stream channels. This data is stored using a paging pattern where each page document holds a subset of the overall timeline (currently 3600 points per page).
For example, we may have 10,000 data points divided into 3 pages with capacities: 3600, 3600, 2800 respectfully.
Below is our current update operation, it pushes a new point to the points array within a collection containing less than 3600 datapoints (defined by count).
db.stream_pages.updateOne(
{
"stream_id": ObjectId("635046f885b5ab956ba279a6"),
"group_id": ObjectId("6277597d7558ec17694ae0f8"),
"count": {
"$lt": 3600
}
},
{
"$push": {
"points": 100
},
"$inc": {
"count": 1
},
"$set": {
"end": Date()
},
"$setOnInsert": {
"stream_id": ObjectId("635046f885b5ab956ba279a6"),
"begin": Date(),
"group_id": ObjectId("6277597d7558ec17694ae0f8")
}
},
upsert: true
)
Problem Statement
We are starting to get a higher influx of data causing a significant amount of $push operations on updates. Currently, each incoming datapoint requires a $push into the latest page document. The latest document is defined as the only document containing less than 3600 points (custom indexes are used to optimize this lookup).
We would prefer to push N items into existing pages similarly like so:
db.stream_pages.updateOne(
{
"stream_id": stream_id,
"group_id": group_id,
"count": {
"$lt": 3600
}
},
{
"$push": {
"points": {
"$each": [ 100, 100, 101, 20 ],
}
},
"$inc": {
"count": N # current length of array passed
},
"$set": {
"end": Date()
},
"$setOnInsert": {
"stream_id": stream_id,
"begin": Date(),
"group_id": group_id
},
},
upsert: true
)
While this operation works, it does not respect the cap of 3600 datapoints we put on each page, causing pages to grow in excess of 3600 points. The ideal update operation would handle two cases: point capacity is respected and overflow points are upserted into a new document.
Does Mongo have a method to find and then use the queried document as input conditionals for the update? Worst case we do a query before this insertion and only push a calculated amount of points to the array, this is not atomic however which can cause issues down the line.
3
There is no option to do a single update with the current structure, while it sounds like you could benefit from a schema redesign I understand the difficulty involved in such a migration so I let’s see what are our options given the current state of things:
- I’ll start with what I think is the best approach, just executing multiple (~2) updates until all requirements are met.
Here is how the code would look like:
const newPoints = [100, 100, 101, 20];
let remaining = newPoints.length;
while (remaining > 0) {
const prev = await db.stream_pages.findOneAndUpdate(
{
"stream_id": stream_id,
"group_id": group_id,
"count": {
"$lt": 3600
}
},
[
{
"$set": {
"points": {"$concatArrays": [{"$ifNull": ["$points", []]}, {"$slice": [newPoints, {"$subtract": [3600, {"$ifNull": ["$count", 0]}]}]}]},
"count": {"$sum": [{"$ifNull": ["$count", 0]}, {"$max": [{"$subtract": [3600, {"$ifNull": ["$count", 0]}]}, {$size: newPoints}]}]},
"end": Date(),
"begin": {"$ifNull": ["$begin", Date()]},
"group_id": {"$ifNull": ["$group_id", group_id]},
"stream_id": {"$ifNull": ["$stream_id", stream_id]},
}
}
],
{
upsert: true
}
);
remanining = remanining - (3600 - (prev.value ? prev.value.count : newPoints.length))
}
The idea is to calculate the remaining size based on the update result, because each update is atomic this code can run in multiple process in parallel with no issues.
The only edge case I did not handle is if the input “newPoints” size is over 3600, if this is possible you’ll have to add it in yourself.
The second options is slightly cleaner code, it is using the aggregation pipeline with the $merge stage, the issue with this approach and also the reason why I don’t recommend it as it’s not atomic like an update should be.
Therefor this should only be used if only a single process is handling these updates, even then you’ll have to add additional logic in code to make sure only 1 update is executed at a time, or to use transactions to ensure that.
Another edge case issue is that it requires at least one document in the collection to already exist.
Either way not very optimal, If you want me to provide syntax for it then i’ll be glad to but as I mentioned I recommend against this approach.