Posting this idea at the request of one of the Jira users. You can find more technical details about this in the Jira issue:

https://jira.mongodb.org/browse/SERVER-61613

The gist of it is that I may have two collections, b2 and b3, that are not distinguishable in the Mongo Shell, like this:

db.b2.find()

[

  { _id: 1, created: ISODate("2021-11-18T23:16:33.149Z") },

  { _id: 2, created: ISODate("2021-11-18T23:16:33.149Z") }

] 

db.b3.find()

[

  { _id: 1, created: ISODate("2021-11-18T22:53:02.113Z") },

  { _id: 2, created: ISODate("2021-11-18T22:53:02.113Z") }

] ```

When I merge each into a collection `pg` with this syntax:

db.pg.aggregate([{$merge: {into: "b3", whenMatched: "merge", whenNotMatched: "fail"}} ]);
```

, it reports a cryptic message:

MongoServerError: $merge failed to update the matching document, did you attempt to modify the id or the shard key? :: caused by :: Performing an update on the path 'id' would modify the immutable field '_id '

The pg collection is just an example that in real life is a an aggregation stage before $merge.

The underlying issue is that b2._id is BSON int and b3._id is BSON double, which is not visible in Mongo Shell, and that Mongo DB server fails to recognize numeric types that would work just fine in other comparisons, like queries.

Please report an error or a warning that _id fields in merged collections are of different numeric types, so it would be clear what the problem is, which will save people a lot of time trying to figure out why the type of _id is being changed.

Given that _id is immutable, people would have to drop the collection and recreate it, which may be quite tricky for large databases. If there was a tool to do that or if the server would recognize compatible numeric values and report the error only when numeric types cannot be coalesced, it would also be helpful.