Spark pipelines that keep MongoDB in sync with another dataset need to delete documents that fall out of a join. Today that is not possible through the connector write path.

operationType is insert, replace, and update only. Replace/update upsert the matches and leave everyone else in the collection. overwrite drops the collection. Spark SQL DELETE FROM (SPARK-414) is a different API and does not take a DataFrame of ids from a join. Users are forced onto the Java driver.

This should work:

unmatched = mongo_df.join(other, "id", "left_anti")

unmatched.select("_id").write.format("mongodb")
.option("operationType", "delete").mode("append").save()

Each row is a DeleteOneModel on idFieldList (default _id), same pattern as replace/update.

Out of scope: SQL DELETE (SPARK-414 / SPARK-465), overwrite, UPDATE/MERGE.

ross@mongodb.com I want to implement this. Is operationType=delete the right shape, or do you want this only via SQL DELETE?