Tom Lane wrote:
> It's removing potentially-important data without any notice or recourse
> to the user. There seems to be a contingent around here that thinks
> that as soon as xmin is older than GlobalXmin it is no longer of
> interest to anyone, but I have lost count of how often I have found it
> invaluable for forensic purposes. I have resisted having VACUUM freeze
> tuples before they've reached a quite-respectable age, and I object to
> having CLUSTER do it either.
How about freezing anything older than vacuum_freeze_min_age, just like
VACUUM does?
> I could maybe accept a CLUSTER FREEZE option to do this, but that's not
> what's in the patch.
I wouldn't like to add more options to CLUSTER, people are already
confused about the similar VACUUM options.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com