Enrico Weigelt wrote:
> Bruno Wolff III wrote:
>>
>> This gets brought up a lot. The problem is that the index doesn't include
>> information about whether the current transaction can see the referenced
>> row. Putting this information in the index will add significant overhead
>> to every update and the opinion of the developers is that this would be
>> a net loss overall.
>
> wouldn't it work well to make this feature optionally for each
> index ? There could be some flag on the index (ie set at create
> time) which tells postgres whether to store mvcc information.
There is no reason to assume it can't work.
There is little reason to assume that it will be the best
solution in many circumstances.
There is a big reason why people are sceptical: there is no patch.
The issue has been debated and beaten to death. People have
formed their opinions and are unlikely to change their position.
If you want to convince people, your best bet is to submit a
patch and have OSDL measure the performance improvement.
Jochem
