Simon Riggs wrote:
> On Mon, 2008-08-18 at 16:22 -0400, Tom Lane wrote:
>> David Fetter <david@fetter.org> writes:
>>> What would need to happen for the next jump up from where varlena is
>>> now, to 8 bytes?
>> Dealing with upwards-of-4GB blobs as single Datums isn't remotely sane,
>> and won't become so in the near (or even medium) future. So I don't
>> see the point of doing all the work that would be involved in making
>> this go.
>>
>> What would make more sense is to redesign the large-object stuff to be
>> somewhat modern and featureful, and provide stream-access APIs (think
>> lo_read, lo_seek, etc) that allow offsets wider than 32 bits. The main
>> things I think we'd need to consider besides just the access API are
>>
>> - permissions features (more than "none" anyway)
>> - better management of orphaned objects (obsoleting vacuumlo)
>> - support > 16TB of large objects (maybe partition pg_largeobject?)
>> - dump and restore probably need improvement to be practical for such
>> large data volumes
>
> Sounds like a good list.
>
> Probably also using a separate Sequence to allocate numbers rather than
> using up all the Oids on LOs would be a good plan.
The ability to partition the large object store would not suck either...
For backup/recovery purposes mainly.
//Magnus
