Nicolas Barbier wrote:
> On 10/21/05, Richard Huxton <dev@archonet.com> wrote:
>
>
>>jeff sacksteder wrote:
>>
>>
>>>Are there known limits to how many rows can be inserted by one transaction,
>>>or does that just reflect the already documented row, table and database
>>>limits?
>>
>>Well, the system will need to be able to roll back the transaction, so
>>at some point your system will grind to a halt. I shouldn't be surprised
>>if there was some counter that couldn't cope beyond 2^31 rows too but
>>no-one's found it yet.
>
>
> Just by not indicating that a transaction did commit, others will keep
> ignoring its rows. There is nothing to rollback here, thanks to MVCC.
> Of course, those rows will still be physically present until the next
> VACUUM.
D'oh - thanks Jeff. Due to brain malfunction I'd typed "roll back"
instead of "commit". What I was trying to get at was that if you commit
a zillion rows on your laptop you can sit there all day with your I/O
saturated while the WAL writes it out.
--
Richard Huxton
Archonet Ltd