Are Bulk Operations in MongoDB atomic?

Bulk operations in MongoDB are a way to perform multiple write operations (like inserts, updates, or deletes) in a single database command, which can significantly improve network efficiency. However, contrary to a common misconception, a bulk operation is not an atomic transaction. If your bulk operation contains writes to multiple documents, it does not have the 'all-or-nothing' property. Some writes can succeed while others fail, and these partial results are committed to the database .

It's important to understand that MongoDB's atomicity guarantee still applies at the single-document level. This means that within a bulk operation, each individual insertOne, updateOne, or deleteOne operation is atomic for the document it affects . For example, if you have a bulk operation that updates Document A and then inserts Document B, the update to Document A is an atomic operation. If it succeeds, the change to Document A is persisted, even if the subsequent insertion of Document B fails .

For use cases that require true atomicity across multiple documents (an 'all-or-nothing' guarantee), MongoDB has supported multi-document ACID transactions since version 4.0 . By wrapping your bulk operation logic within a transaction, you ensure that if any operation fails, the entire set of changes is rolled back. However, it's worth noting that transactions come with a performance overhead. The MongoDB documentation suggests that a large majority (80-90%) of applications using the document model can achieve data consistency without them by properly structuring their data and leveraging the atomicity of single-document operations . For your use case of scraping and upserting 1 million products, an unordered bulkWrite with upserts remains the most efficient approach, as the occasional failure of a single product upsert likely does not necessitate rolling back the entire batch.