_id are generated client-side by the drivers when they send the documents to the cluster if one doesn’t already exists. It’s buried in the insertOne and insertMany source code somewhere - depends of which driver you are using of course.
ObjectIds are built in a very specific manner:
So first of all, the 100 processes all have a different random value in the middle. So they can’t compete with another process inserting documents because there are 5 bytes * 8 bits = 2^40 possibilities (I think?).
The timestamp and the counter are here to make sure that documents inserted during the same second have a different ObjectId within the same process.
So it’s impossible to generate the same ObjectId with a single process and very unlikely that 2 processes writing to the same cluster share the same random value in the middle. Else the incrementing counter starting at a random value has your back anyway.
So I guess the probability is very very VERYYYY close to zero.
Cheers,
Maxime.