> On Sat, Feb 12, 2005 at 06:07:32PM -0500, cbbrowne at ca.afilias.info wrote:
>> I would further "buy" that when doing "COPY_SET/ENABLE_SUBSCRIPTION",
>> all
>> of the objects should be checked and locked up front _before_ copying in
>> data.  We have had cases where tables were cleaned out of place that
>> should have been there, with the result that subscriptions got 3h in and
>> THEN failed.
>>
>> We have also had cases where COPY_SET failed because vacuum scripts were
>> running and a "deadlock-capable" series of vacuums backed up alongside
>> COPY_SET.  Locking _everything_ would make that either succeed very
>> quickly or fail very quickly.
>>
>> Mind you, it's not exactly a tiny change.
>
> I'd like the option of having it drop ALL the indexes on the target tables
> and rebuild them after the COPY as well.  The initial copy is such a pig
> with even just the primary key in place.

There's a TODO for PostgreSQL there to be a new option on COPY that will
do this, deferring index updates until the end when they are
regenerated...