Detailed steps to reproduce the problem?

Driver version 2.1 introduces a regression with collection.BulkWrite with very large BSON documents. This program reproduces the problem: this program passes with driver version 2.0 but fails on 2.1 with a BSONObjectTooLarge error.

With driver version 2.0.1, that program prints:

2025/06/10 10:44:29 command started: dropDatabase
2025/06/10 10:44:29 command started: update
2025/06/10 10:44:29 command started: update
2025/06/10 10:44:29 command started: update
9

(Three updates, then a succesful response with 9 upserts)

With driver version 2.1.0 and 2.2.2, it prints

2025/06/10 10:44:43 command started: dropDatabase
2025/06/10 10:44:44 command started: update
2025/06/10 10:44:44 (BSONObjectTooLarge) BSON size limit hit while building Message. Size: 18874889 (0x1200209); maxSize: 16793600(16MB)
exit status 1

(Error on the first and only update.)

It seems like something has gone wrong with splitting the bulk write into batches, but that's a guess.

Definition of done: what must be done to consider the task complete?

collection.BulkWrite with very large documents should not fail with BSONObjectTooLarge errors.

The exact Go version used, with patch level:

Tested on go 1.24.3 and go 1.23.8, same result on both

The exact version of the Go driver used:

• v2.0.1 (good)
• v2.1.0 (broken)
• v2.2.2 (broken)

Describe how MongoDB is set up. Local vs Hosted, version, topology, load balanced, etc.

I'm testing on a local 3-node replica set with MongoDB 8.0. I've also seen the same problem in Mongo 7.0 and with sharded clusters.

The operating system and version (e.g. Windows 7, OSX 10.8, ...)

I'm on macOS 15.5 (but I would be surprised if this were OS related).