Re: pgsql: Fix a couple of bugs in MultiXactId freezing

On 2013-12-03 15:40:44 -0300, Alvaro Herrera wrote:
> Andres Freund wrote:
>
> > I wondered about that myself. How would you suggest the format to look
> > like?
> > ISTM per tuple we'd need:
> >
> > * OffsetNumber off
> > * uint16 infomask
> > * TransactionId xmin
> > * TransactionId xmax
> >
> > I don't see why we'd need infomask2, but so far being overly skimpy in
> > that place hasn't shown itself to be the greatest idea?

> I don't see that we need the xmin; a simple bit flag indicating whether
> the Xmin was frozen should be enough.

Yea, that would work as well.

> For xmax we need more detail, as you propose. In infomask, are you
> proposing to store the complete infomask, or just the flags being
> changed? Note we have a set of bits used in other wal records,
> XLHL_XMAX_IS_MULTI and friends, which perhaps we can use here.

Tentatively the complete one. I don't think we'd win enough by using
compute_infobits/fix_infomask_from_infobits and we'd need to extend the
bits stored in there unless we are willing to live with not transporting
XMIN/XMAX_COMMITTED which doesn't seem like a good idea.

Btw, why is it currently ok to modify the tuple in heap_freeze_tuple()
without being in a critical section?

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services