Simon Riggs <simon@2ndQuadrant.com> writes:
>> We should just use MyPgXact->xid
>> rather than add more to the transaction path
>>
>> I'll simplify the patch and commit.
> Committed, but forgot to give appropriate credit. Sorry about that.
This patch didn't fix things, it broke things. The former guarantee
that age's reference point would hold still throughout a transaction
just disappeared.
What I read your previous suggestion to be was that age() would keep
local state and use inspection of the current VXID to tell if its
cache was stale. That would keep the fix localized (which I agree
is a good idea) without losing the stability guarantee.
regards, tom lane