Understanding Transactions

On Wed, 12 Dec 2001, Joe Koenig wrote:

inserts faster in groups of 1,000 or 5,000? I know that letting each
insert be in its own transaction creates a lot of overhead, but I didn't
know if putting 5,000 inserts into a transaction created overhead for
that transaction. Hopefully my question makes sense. Thanks,

Have you tried putting them all in one transaction?
That may be the easiest approach.

I would think that a group of 1000 has a higher overhead than 5000.

We have made up to 8 tables with a maximum of 6Million tuples, and copy
from a file for each table where are all stored is the only way I think is
possible, If possible disabling fsync.
Hope it helps

On Wed, 12 Dec 2001, Joe Koenig wrote:

I've been reading through the archive and I see that when doing a large
amount of inserts it is much faster to wrap a bunch of them in a
transaction. But here's my question. Say I need to do about 100,000
inserts and using COPY isn't an option. Is postgres going to do the
inserts faster in groups of 1,000 or 5,000? I know that letting each
insert be in its own transaction creates a lot of overhead, but I didn't
know if putting 5,000 inserts into a transaction created overhead for
that transaction. Hopefully my question makes sense. Thanks,

Well, it depends on the schema to some extent probably. If the table
has foreign keys, there was a problem (it's been fixed but I don't
know in what version) with the deferred trigger manager on long
transactions. 1k or 5k rows is probably okay in any case.