Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> On Tue, Jul 19, 2005 at 10:44:35PM -0400, Tom Lane wrote:
>> If it's a matter of whether VACUUM removes rows or not, that's expected:
>> open-transaction bookkeeping is done across the whole cluster. (This is
>> not something simple to change, since we may have to consider
>> transaction IDs in shared catalogs.)
> I wonder if this restriction could be lifted if we tracked last-vacuum-
> Xid per relation?
No, that's unrelated. What is of concern is the open-transaction
horizon. We could tighten things up by keeping a more complicated set
of information in the PGPROC array --- rather than advertising a single
transaction XMIN, each transaction would have to calculate and advertise
both a "global" XMIN and a "local" XMIN (the latter counting only open
transactions within your own database). This would then allow a more
refined calculation in GetOldestXmin().
I'm inclined to think that the benefit wouldn't be worth the distributed
cost of doing this in every GetSnapshotData() call ...
regards, tom lane
