'code': 225, 'codeName': 'TransactionTooOld', 'errmsg': 'Cannot start transaction 3 on session ab828da0-dead-4cbd-beef-ac612334a5c1 - ugh4jez5/+Zo0w7yt4WMrZ1cJoa3zmk86txJfzwiQ18= because a newer transaction 4 has already started.

I'm running a sharded cluster, where each replica set is a PSA (primary, secondary, arbiter).

I'm processing a very large csv file where the data looks like this:

I need to group the tags by _id, so i use bulk operations in pymongo like:

UpdateOne({"_id": row["_id"]}, {"$addToSet": {"tag": row["tag"]}}, upsert=True)

which i run in batches of 5000.

If i run only one thread, i get no errors. If i split the csv file into 8 and run 8 parallel processes, i start getting the error above after a while, but it runs successfully for a few minutes. I'm suspecting that i hit a region of the csv file where i have the same _id over and over again. This looks similar to #14322, which i also had problems with, a few years ago, in an identical scenario.

What does that error even mean? What workaround can i try?