Just realised I didn't cc: the list on this. Was obviously having a good
evening.
Richard Huxton wrote:
> Glen Parker wrote:
> > I still don't think it will work. Two concurrent transactions could
>> still screw the data up.
>>
>> Before: 9 rows.
>> Tx 1: See 9 rows, issue insert, see 10 rows.
>> Tx 2: See 9 rows, issue insert, see 10 rows.
>> After: 11 rows.
>
> True, and well corrected.
>
> Serves me right for posting at the end of a long day. Particularly
> irritating as its almost identical to the example I've used myself when
> showing the limitations of PG's low-locking MVCC model.
>
--
Richard Huxton
Archonet Ltd