Why $set in aggregation pipeline does not overwrite embedded object

Working with Data
,
1

let’s say my database is like

{ foo: [
  { _id: 0, meta: { embed: { a: 1, b: 2 } } }
]}

when i do

db.foo.updateMany({_id: 0}, {$set: { _id: 0, 'meta.embed': { b: 3 } } })

the result is

{ _id: 0, meta: { embed: { b: 3 } } }

However, if i use aggregation pipeline

db.foo.updateMany({_id: 0}, [{$set: { _id: 0, 'meta.embed': { b: 3 } } }])

the result is

{ _id: 0, meta: { embed: { a: 1, b: 3 } } }

So what’s the difference here?
I have to use aggregation pipeline because the real case contains other operators, so how can i use aggregation pipeline to achive the same result?

Environment: Windows 10, MongoDB 6.0.6

1 Like
3

I had written a post about working with arrays but that’s not the case here, but looking (again) through the documentation it seems that within an aggregation pipeline $set is an alias for $addFields, in this case I guess it makes sense that it does not replace the existing field but adds a new element to the location specified.
In non-aggregation mode, it’s running so that it’ll replace what’s there.

If you wanted to do the same in the aggregation you need to add a new stage to the pipeline as far as I can tell to do this:

db.Test.updateMany(
    {_id: 1}, 
    [
        {
            $unset:'meta.embed'
        },
        {
            $set: {'meta.embed': { b: 3 } } 
        }
    ]
)

You could also simply (subjectively) it to:

db.Test.updateMany(
    {_id: 1}, 
    [
        {
            $unset:'meta.embed'
        },
        {
            $set: {'meta.embed.b':3} 
        }
    ]
)
4

I managed to work around this by writing a 4 stage pipeline to fully emulate the behavior. The pseudo is like

    collection.updateMany(query, [
      { $set: preset }, // Copy all involved fields of inner operators to a temporary field by walking through $set 
      { $unset }, // Remove all objectlike fields of $set
      { $set }, // Actual operation
      { $unset: [tempKey] }, // Remove the temporary field in first stage
    ])

This was painful but works. Though i hope if we could do it better.