How to stream UDF result to client to reduce memory CPU consuption

Could someone confirm/decline if UDF supports streaming data to the client? May I for example run aggregation and pipe partial result to client to process it? From what I see, aerospike doesn’t support it. It first make aggregation and then send the result to the client. In case of very huge data volume, the client cannot process the result because it doesn’t fit the memory. Any way how to improve it?

Have you reviewed our Steam UDF Guide?

Sure, I’ve reviewed it. But my question is about different subject. I’m asking if aql could stream the result of aggregating.

Also, is it good practice to build any more or less ad hoc ETLs on top of aql?

AQL does support aggregations: AQL aggregation docs. AQL is really meant as a user-facing data browser and shouldn’t be used by the application.