On Sat, 2009-01-10 at 11:45 +0000, Simon Riggs wrote:
> On Sat, 2009-01-10 at 09:40 +0000, Simon Riggs wrote:
>
> > This leads to an error when we SubTransSetParent(child_xid, top_xid);
> > since this assumes that the top_xid is the parent, which it is not.
> > Mostly you wouldn't notice unless you were looking up the subtrans
> > status for an xid that had committed but was the child of an aborted
> > subtransaction, with the top level xid having > 64 subtransactions.
> > It's possible the confusion leads to other bugs in UnobservedXid
> > processing, but I didn't look too hard at that.
> >
> > AFAICS we need both parent and top xids.
>
> I wonder if its possible to derive the parent by looking at the
> highest/most newly assigned xid? Abort records would remove aborted
> subtransactions and AFAIK we currently assign a new subtransaction only
> ever from the latest current subtransaction. (This wouldn't be
> necessarily true if supported true branch-anywhere subtransactions, but
> we don't). Sounds correct, but not really sure.
Starting to sound like a do-me-later-if-ever optimisation and certainly
nothing I want to rely on in court.
I'm progressing with parent_xid added to the xlog record header, for
now.
-- Simon Riggs www.2ndQuadrant.comPostgreSQL Training, Services and Support
