Yes, I thought about it when I saw the interspersed commas were send on their own. A quick benchmark could tell how much entries should be buffered before sending each chunk. But our solution is already good enough for @quda I guess 
lud
0
quda
Tried with Stream.intersperse/2. For very big files is about 30% slower than the initial solution.
0
quda
derek-zhou
Something like this, untested:
defp delay_chunk(conn, data, buffer, byte_counter) when byte_counter >= 65536 do
{chunk(conn, Enum.reverse(buffer)), [data], 0}
end
defp delay_chunk(conn, data, buffer, byte_counter) do
{conn, [data | buffer], byte_counter+IO.iodata_length(data)}
end
You may need to add error handling and a final flush at the end, but you get the idea.
2
quda
My petty contribution to the solution of @lud with Stream.intersperse/2:
file
...
|> Stream.intersperse(",")
|> Stream.chunk_every(2)
By applying Stream.every/2 with 2 as argument new chunks are created into the stream by combining the previous records with the comma from intersperse.
Proud of my finding! 
5
APB9785
Creator of ECSx
What version of Elixir/Erlang are you running? I recently had an issue with File.stream being very slow, and upgrading Erlang from 24.0 to 24.1 fixed it.
1
quda
I’ve just upgraded it to 24.1 (from 22). Same speed for stream, don’t see any difference. But it’s quite fast, IMO, 60MB/4.5s
2