On 01/04/2014 01:22 AM, Merlin Moncure wrote:
> Long term, I'd rather see an optimized 'ORM flush' assemble the data
> into a structured data set (perhaps a JSON document) and pass it to
> some receiving routine that decomposed it into records.
The same is true on the input side. I'd much rather be sending an ORM
client a big JSON / YAML / whatever graph than a horrible,
duplication-filled chained LEFT JOIN projection like they currently rely
on. When they're not just doing n+1 selects, which is worse.
I think that's really a side-issue though. ORMs aren't going to change
in a hurry, and batching / fire-and-forget support is good for all sorts
of other jobs too.
-- Craig Ringer http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services