Update March 2021
This post still gets a fair few hits, so I want to preface this by saying that I wrote this some time ago, and probably wouldn’t do the same thing today. The main problem is that I ignored the client! Sending a single whole JSON object in such a way means the client has to reassemble the whole thing in-memory anyway, which somewhat defeats the point, unless you have fat clients and very thin servers (or processing is slow and you have a very short socket timeout). You might have a valid use-case for a hack like this, but if I were to solve this problem again, I’d send newline-delimited json instead (with metadata in headers or on a separate endpoint if necessary).
I have a SQLAlchemy query (able to behave as an iterator) which could return a large result set. First version of the code was very simple. Release objects have a to_dict() function which returns a dictionary, so I append to a list and jsonify the result:
# releases = <SQLAlchemy query object> output =  for r in releases: output.append(r.to_dict()) return jsonify(releases=output), 200
(context on github)
This result set could potentially grow to a point that fitting it memory would be impractical – with only a thousand releases there is already a significant lag before we start getting results. Continue reading