Open Stack

On Thu, May 7, 2015 at 5:05 PM, Adam Lawson <alawson at aqorn.com> wrote:

> what sort of limitations have you discovered that had to do specifically
> with the fact we're using Python?
>

Python is great.  Conscious decision to optimize for developer wall time
over cpu cycles has made it a great language for 20 years - and probably
will for another 20 at *least* (IMHO).

I don't think you would pick out anything to point at as a "limitation" of
python that you couldn't point at any dynamic interpreted language, but my
list is something like this:

   - Dynamic Interpreted Runtime overhead
   - Eventlet non-blocking hub is NOT OK for blocking operations (cpu, disk)
   - OTOH, dispatch to threads has overhead AND GIL
   - non-byte-aligned buffers restricts access to O_DIRECT and asyncio

*So often* this kinda stuff just doesn't matter.  Or even lots of times
even when it *does* matter - it doesn't matter that much in the grand
scheme of things.  Or maybe it matters a non-trivial amount, *but* there's
still other things that just mater more *right now*.  I think Swift has
been in that last case for a long time, maybe we still are - great thing
about open-source is redbo can publish an experiment on a feature branch in
gerrit and in-between the hard work of testing it - we can pontificate
about it on the mailing list!  ;)

FWIW, I don't think anyone should find it particularly surprising that a
mature data-path project would naturally gravitate closer to the metal in
the critical paths - it shouldn't be a big deal - unless it all works out -
and it's $%^&! tons faster - then BOOYAH! ;)

But I'd suggest you be very careful not to draw any assumptions in general
about a great language like python even if this one time this one project
thought maybe they should find out if some part of the distributed system
might be better by some measure in something not-python.  ;)

Cheers,

-Clay
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20150507/669b67c8/attachment.html>