Version

24.03

Which installation method(s) does this occur on?

Docker, Conda, Source

Describe the bug.

When using the Milvus service for writing to a vector database, the performance drops when using small batch sizes or infrequent writes. This is because the service wants to reindex the database after each message, or after a set time has elapsed (it is hard coded to 3 seconds). This is inefficient for a few reasons:

• When uploading infrequently, if the reindex takes ~3 seconds, then you can get into a loop where you: add 1 message -> reindex (3 sec) -> add 1 message -> reindex (3 sec). This causes messages to back up and the service cannot keep up.
• When uploading frequently with large batch sizes, reindexing can be triggered by the number of rows. This again can cause issues because the index can take longer than data is generated.

Ideally, we would use something similar to a debounce to update the index. So reindexing only occurs after some set time where no messages have been added.

Minimum reproducible example

milvus_service = MilvusVectorDBService(uri=milvus_server_uri)

# Create the collection
...

# Make a small dataframe with 5 rows
df = cudf.DataFrame({
    "id": list(range(num_input_rows)),
    "age": [random.randint(20, 40) for i in range(num_input_rows)],
    "embedding": [[random.random() for _ in range(3)] for _ in range(num_input_rows)]
})

# Add the rows to the collection in a loop
for _ in range(10000):

    milvus_service.insert_dataframe(collection_name, df)

    # Sleep some amount to allow the data to be inserted (this may need to be tweaked to trigger the bug)
    time.sleep(0.1)

Relevant log output

[Paste the error here, it will be hidden by default]

Full env printout

[Paste the results of print_env.sh here, it will be hidden by default]

Other/Misc.

No response

Code of Conduct

• I agree to follow Morpheus' Code of Conduct
• I have searched the open bugs and have found no duplicates for this bug report