Alvaro Herrera <alvherre@commandprompt.com> wrote:
> If these limitations become a problem, you can always change them.
> A couple of zeroes at the start of the pg_clog filenames aren't
> going to bother anyone, I don't think. Not so sure about your new
> proposed design's space usage.
I guess that's a call the community can make now -- if a
serializable transaction which is not flagged as read only remains
open long enough for over a billion other transactions to commit, is
it OK for the old transaction to be automatically canceled? Is it
worth messing with the SLRU limits to double that?
Beyond a certain point you have transaction ID wrap-around, so at
that point this would be the least of your troubles -- canceling
the old transaction might even be helpful. I thought that was at 2
billion, but Heikki was saying it's at 1 billion in an earlier post.
-Kevin