On May 19, 2017, at 07:59 PM, Aurelien Bompard wrote:
>> Have you done any other analysis on where the bottlenecks are? Is it CPU,
>> db, I/O?
>I haven't investigated properly but it seems to be the database (which
>is on another host).
That's interesting because most of that runs through SQLAlchemy. I wonder if
it's a general problem with the way we integrate with that, or if there are
specific queries that are inefficient. I have no doubt there's room for
improvement on the latter, but we'd have to do the instrumentation to find
>Interesting. I wonder if we can have SQLAlchemy play well with
>asyncio, I haven't tried that before.
Me neither! asyncio is pretty new, so we're seeing lots of efforts to
async-ify more parts of the ecosystem. I don't know whether the SA folks are
looking into that at all. A general search shows some interesting projects
and discussions on the topic.
>There is still a gunicorn.py file in the contrib directory, but I
>haven't tested it either.
I tried to get it to work but had problems. I'm really interested in seeing
if we can leverage uvloop for improved performance.