PyMongo bulk_write UpdateOne only runs last operation - mongodb

Question

Got a weird bug that I can't quite figure out.
I have some pymongo code that looks like this:
from pymongo import UpdateOne
client = pymongo.MongoClient()
...
def update_image_locations(user_key, dataset_key, preset_name,
keys_and_coords):
db = docdb_client.db
col = db.col
operations = []
query = {'ownerKey': user_key, 'imageInfo.datasetKey': dataset_key}
for key_and_coords in keys_and_coords:
query['key'] = key_and_coords['key']
operations.append(
pymongo.UpdateOne(
query, {
'$set': {
'imageInfo.presets.%s.coords' % preset_name:
key_and_coords['coords']
}
}))
print(operations)
if len(operations) > 0:
print(col.bulk_write(operations, ordered=False).bulk_api_result)
# This section fails with a KeyError.
cursor = col.find({
'ownerKey': user_key,
'imageInfo.datasetKey': dataset_key
}, {'imageInfo': 1}
)
for doc in cursor:
print(doc['imageInfo']['presets'])
If I print out the bulk_write output, I get the following.
{'writeErrors': [], 'writeConcernErrors': [], 'nInserted': 0, 'nUpserted': 0, 'nMatched': 65, 'nModified': 65, 'nRemoved': 0, 'upserted': []}
which as far as I can tell is exactly what I expect.
However, I get KeyError failures for all but the last document in the collection when I try to iterate through the documents that should ostensibly have the new field. If I then go into the actual mongodb shell, I can confirm that only the last operation from the bulk_write seems to have actually gone off.
Based on the bulk_api_result I would expect that all of the documents would be updated, instead of only the last one. What's going on?
EDIT:
As requested, before and after queries. I'm not showing the full doc because there's a lot of vector embedding info that's going to muddle things.
Query:
> db.user_uploads.find({}, {'imageInfo.presets': 1})
Before:
{ "_id" : ObjectId("6074792104cc23375a8f979a"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979b"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979c"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979d"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979e"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979f"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a0"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a1"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a2"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a3"), "imageInfo" : { } }
After:
{ "_id" : ObjectId("6074792104cc23375a8f979a"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979b"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979c"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979d"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979e"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f979f"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a0"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a1"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a2"), "imageInfo" : { } }
{ "_id" : ObjectId("6074792104cc23375a8f97a3"), "imageInfo" : { "presets" : { "preset_one" : { "coords" : [ 2.229365348815918, 1.4654869735240936 ] } } } }

Answer 1

Turns out the answer has to do with how the query is constructed. Specifically, this works:
for key_and_coords in keys_and_coords:
query = {'key': key_and_coords['key']}
operations.append(
pymongo.UpdateOne(
query, {
'$set': {
'imageInfo.presets.%s.coords' % preset_name:
key_and_coords['coords']
}
}))
and this fails:
query = {}
for key_and_coords in keys_and_coords:
query['key'] = key_and_coords['key']
operations.append(
pymongo.UpdateOne(
query, {
'$set': {
'imageInfo.presets.%s.coords' % preset_name:
key_and_coords['coords']
}
}))
I think what's happening here is some async javascript-esque magic, where the query object is passed by reference to the bulk operation which then executes them once all of the bulk operations are in place. Since the query is passed by reference, the actual key value gets overwritten each time until the last one (which is also why only the last object is updated). Unfortunately this was tough to catch because printing out the queries and the operations both looked fine, but the async kicked in at execution. Still, not really an issue with pymongo after all.
Thanks to everyone who responded!