Alex Shulgin wrote:
> OK, I think I have now all bases covered, though the updated patch is
> not that "pretty".
>
> The problem is that we don't know in advance if the (sub)transaction is
> going to succeed or abort, and in case of aborted truncate we need to
> use the stats gathered prior to truncate. Thus the need to track
> insert/update/deletes that happened before first truncate separately.
Ugh, this is messy indeed. I grant that TRUNCATE is a tricky case to
handle correctly, so some complexity is expected. Can you please
explain in detail how this works?
> To the point of making a dedicated pgstat testing tool: let's have
> another TODO item?
Sure.
--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services