On Tue, 2008-01-22 at 20:53 +0100, Pavel Stehule wrote:
> >
> > Agreed. I think Pavel Stehule was doing some experiments with them, I
> > don't know if he got anywhere.
> >
>
> I did only first research. Any hack is possible - you can stack
> current transaction, but real implementation needs similar work like
> nested transaction :( and it is too low level for me. And some code
> cleaning is necessary. There are global variables.
>
> And there is most important question about data visibility - is
> autonomous transaction independent on main transaction (isolation)?
> You have to thing about deadlock, about reference integrity, etc. This
> task isn't simple.
Yes, I think autonomous transactions should be on the TODO. They're
useful for
- error logging
- auditing
- creating new partitions automatically
Plus I think we'd be able to improve the code for CREATE INDEX under
HOT, and probably a few other wrinkly bits of code.
-- Simon Riggs 2ndQuadrant http://www.2ndQuadrant.com