How to spot an outlier in MongoDB - mongodb

Question

Assume the following records in mongodb
{
_id: // primary key
age: // some age.
}
The system generates primary key and is guaranteed to be increasing monotonically.
The business logic provides value for age. Age should be increasing, however due to a bug, under some remote cases, the age could be decreasing.
Eg: age could go from 1 yr, 2 yr, 3yr, "2 yr", 4yr, 5yr etc.
How to write a query to spot the outlier in the age ?

Answer 1

Assuming your collection is called 'junk' (sorry, no bad intentions here) I think this might work...
db.junk.aggregate([
{$lookup: {
from: "junk",
let: { age: "$age", id: "$_id" },
pipeline: [
{ $match :
{ $expr:
{ $and:
[
{$gt: ["$_id", "$$id"]},
{ $lt: ["$age", "$$age"] }
]
}
}
}
],
as: "data"
}},
{ $project: { _id: 1, "age": 1, "data": 1, "found": { $gt: [{ $size: "$data" }, 0] } } },
{ $match : { found: true }}
])
The intent is to self join on the same collection where the id is greater than another document, but the age is less for the same document. Count how many records are in this collection, and if the count is greater than 0 output.
Example Collections:
So, for testing this I populated a collection called 'junk' with 7 documents...
> db.junk.find()
{ "_id" : ObjectId("5daf4700090553aca6da1535"), "age" : 0 }
{ "_id" : ObjectId("5daf4700090553aca6da1536"), "age" : 1 }
{ "_id" : ObjectId("5daf4700090553aca6da1537"), "age" : 2 }
{ "_id" : ObjectId("5daf471b090553aca6da1538"), "age" : 3 }
{ "_id" : ObjectId("5daf471e090553aca6da1539"), "age" : 4 }
{ "_id" : ObjectId("5daf4721090553aca6da153a"), "age" : 3 }
{ "_id" : ObjectId("5daf4724090553aca6da153b"), "age" : 5 }
Results:
Here is what my results look like after running this query...
{ "_id" : ObjectId("5daf471e090553aca6da1539"), "age" : 4, "data" : [ { "_id" : ObjectId("5daf4721090553aca6da153a"), "age" : 3 } ], "found" : true }
It found a record having a later outlier (ObjectId 5daf471e090553aca6da1539 precedes the outlier, ObjectId 5daf4721090553aca6da153a is the outlier). Obviously this could be projected differently to show just the outlier, but I wanted to first verify the query works as expected and not invest more time on a inadequate approach.