Alvaro Herrera <alvherre@commandprompt.com> writes:
> Erik Jones wrote:
>> And, to finish up, is there any reason that pg_restore couldn't
>> already work with separate processes working in parallel?
> The problem is that the ordering of objects in the dump is the only
> thing that makes the dump consistent with regards to the dependencies of
> objects. So pg_restore cannot make any assumptions of parallelisability
> of the restoration process of objects in the dump.
That's true at the level of DDL operations, but AFAIK we could
parallelize table-loading and index-creation steps pretty effectively
--- and that's where all the time goes.
A more interesting question is what sort of hardware you need for that
actually to be a win, though. Loading a few tables in parallel sounds
like an ideal recipe for oversaturating your disk bandwidth...
regards, tom lane
