On 1 June 2013 09:22, Simon Riggs si...@2ndquadrant.com wrote:
A much better idea is to hold the xmin epoch on the tuple, in addition
to the xid, if there was a good place to hold this.
This idea is an almost exact duplicate of Heikki Linnakangas' idea for
exactly the same problem.
Given
As we get faster, we'll need to wrap the xid counter more quickly. At
current full speed, we could wrap the xid counter every 72 hours.
This is a concern for very large tables, since currently we have to
rescan the whole table. This has my interest for change in the next
release.
We could solve
