694221 - Sync server no longer streams application/newlines data

Status: RESOLVED WONTFIX

(Cloud Services Graveyard :: Server: Sync, defect)

Description

[Toby Elliott [:telliott]]

On queries for particularly large sets of ids, when application/newlines was being used, we'd stream the data as we got it back from the database. This meant that the data would start arriving at the user faster, preventing some timeouts. 

Now, we marshal all the data before shipping it off, no matter what the format. Can our wsgi model support streaming responses?

Comment 1

[Ryan Kelly [:rfkelly]]

Reviving this bug a we think streaming may help us with current server load issues.  Some notes for myself:

  * PyMySQL supports an "SSCursor" class for streaming the results out of MySQL.  It's unclear how difficult this will be to use with sqlalchemy.
  * Gunicorn appears to correctly translate generator-based output into a chunked encoding respose.
  * nginx may need some tweaking to avoid buffering the output at that layer as well ("proxy_buffering off" or similar?)
  * not clear yet how much of baseapp assumes a buffered response...

Should we do this only for application/newlines, or for JSON output as well?  It seems simple enough to emit the leading "{" and trailing "}" as separate chunks, streaming all the things in the middle.

Comment 2

[Toby Elliott [:telliott]]

we should definitely do this for json. it's just a matter of putting {} around it, adding a comma before everything but the first one and json-outputting each record.

Comment 3

[Ryan Kelly [:rfkelly]]

Closing out this ancient bug