restricts doesn't seem to be uploaded when using a private endpoint for matching engine

Question

I have a matching engine set up in Google Cloud that is on a VPC. I'm uploading vectors and trying to use the restricts attribute to save some metadata for limiting searches later on.

When I create my datapoint, it seems like the restricts attribute is there:

metadata = [
    aiplatform_v1.IndexDatapoint.Restriction(namespace='color', allow_list=['red']),
    aiplatform_v1.IndexDatapoint.Restriction(namespace='size', allow_list=['medium']) 
]

emb_query = [[.5, .5]]

    
# Create datapoint
id = 101
datapoints = [
    aiplatform_v1.IndexDatapoint(
        datapoint_id=str(id),
        feature_vector=emb_query[0],
        restricts=metadata
    )
]

print(id, datapoints[0])

This prints:

(101,
  datapoint_id: "101"
  feature_vector: 0.5
  feature_vector: 0.5
  restricts {
     namespace: "color"
     allow_list: "red"
   }
  restricts {
     namespace: "size"
     allow_list: "medium"
  })

So far so good. Now I create the request and send it:

upsert_request = aiplatform_v1.UpsertDatapointsRequest(
    index=index_name, datapoints=datapoints
)

client.index_client.upsert_datapoints(request=upsert_request)

Now, for whatever reason, when I query it back, the MatchNeighbor doesn't have the restricts attribute set:

# Fetch the endpoint
my_index_endpoint = aiplatform.MatchingEngineIndexEndpoint(
    index_endpoint_name=INDEX_ENDPOINT_NAME
)

# Execute the request
response = my_index_endpoint.match(
    deployed_index_id=DEPLOYED_INDEX_ID,
    queries=[QUERY_EMBEDDING],
    # The number of nearest neighbors to be retrieved
    num_neighbors=1,
)

# print(response)
response[0]
# [MatchNeighbor(id='101', distance=1.0, feature_vector=None, crowding_tag=None, restricts=None, numeric_restricts=None)]

So why is the restricts attribute now None?

Answer 1

Try return_full_datapoint=True