Hey Shai --

Generally sounds like a good idea, I don't see why this shouldn't go
in. You might want to wait until someone with more Oracle experience
chimes in, but I'm having a hard time seeing why a number -> str ->
number dance would be a good idea.

I'm generally adverse to introducing more settings -- settings creep
is a real problem -- so a couple of alternate ideas you might consider
in producing a patch:

1. Just pick one correct behavior, probably for each type. My guess
would be that ints and floats should get the fast behavior, and
decimals the slow-but-correct one. If you're using a decimal field
it's because you care about precision, so correctness makes more sense
there IMO.

2. If you must introduce a setting, do it in DATABASES[OPTIONS]
instead of a top-level ORACLE_FAST_NUMBERS or whatever.

Jacob

On Wed, Aug 29, 2012 at 7:14 AM, Shai Berger <s...@platonix.com> wrote:
> Hi Django developers,
>
> I've been working with Oracle on a project with some performance requirements,
> and we've noticed long ago that the Django Oracle backend performs
> significantly worse than other backends; we've been able to profile it and see
> that, indeed, a lot of time is spent in processing the returned rows, while
> the Oracle DBMS itself is not breaking a sweat.
>
> This week, I found what is, I think, the main culprit: The Oracle backend
> prefers accuracy to performance. Since Oracle numbers can have more 
> significant
> digits than C doubles (=CPython floats), and the backend wishes to preserve
> this accuracy, it makes cxOracle (the Python Oracle client library) return all
> numbers as strings, and then converts the strings back to numbers -- ints,
> floats or decimals -- according to the Oracle data type descriptions.
>
> To make things even slower, the conversion decisions are made anew for each
> row; this is done in the function _rowfactory (currently at
> https://github.com/django/django/blob/master/django/db/backends/oracle/base.py#L790).
> The function itself is called less efficiently than possible (on rows returned
> from cxOracle, instead of within cxOracle before the rows are returned).
>
> For our application -- as for many others, I guess -- precision of 38 digits,
> on real numbers or integers, is not important. I made a modified backend which
> returns numbers as numbers and the improvement in performance is very
> significant. A row-factory function is still required to make sure strings are
> returned as unicode, and creating such a function per query to only go over
> the string columns may improve things even more, but I didn't even have to go
> there yet.
>
> Can we put this into Django core? Perhaps with a Settings knob for people who
> really need all the precision?
>
> Thanks,
>         Shai.
>
> --
> You received this message because you are subscribed to the Google Groups 
> "Django developers" group.
> To post to this group, send email to django-developers@googlegroups.com.
> To unsubscribe from this group, send email to 
> django-developers+unsubscr...@googlegroups.com.
> For more options, visit this group at 
> http://groups.google.com/group/django-developers?hl=en.
>

-- 
You received this message because you are subscribed to the Google Groups 
"Django developers" group.
To post to this group, send email to django-developers@googlegroups.com.
To unsubscribe from this group, send email to 
django-developers+unsubscr...@googlegroups.com.
For more options, visit this group at 
http://groups.google.com/group/django-developers?hl=en.

Reply via email to