Re: [HACKERS] pg_dump and thousands of schemas

On Tue, Nov 13, 2012 at 7:12 PM, Denis <socsam(at)gmail(dot)com> wrote:
> Jeff Janes wrote
>> On Thu, Nov 8, 2012 at 1:04 AM, Denis &lt;
>
>> socsam@
>
>> &gt; wrote:
>>>
>>> Still I can't undesrtand why pg_dump has to know about all the tables?
>>
>> Strictly speaking it probably doesn't need to. But it is primarily
>> designed for dumping entire databases, and the efficient way to do
>> that is to read it all into memory in a few queries and then sort out
>> the dependencies, rather than tracking down every dependency
>> individually with one or more trips back to the database. (Although
>> it still does make plenty of trips back to the database per
>> table/sequence, for acls, defaults, attributes.
>>
>> If you were to rewrite pg_dump from the ground up to achieve your
>> specific needs (dumping one schema, with no dependencies between to
>> other schemata) you could probably make it much more efficient. But
>> then it wouldn't be pg_dump, it would be something else.
>>
>
> Please don't think that I'm trying to nitpick here, but pg_dump has options
> for dumping separate tables and that's not really consistent with the idea
> that "pg_dump is primarily designed for dumping entire databases".

I think it is compatible. From my reading of pg_dump, those other
options seem to have been bolted on as an afterthought, not as part of
its primary design.

Cheers,

Jeff