Re: Using CTID system column as a "temporary" primary key
Laurenz Albe <laurenz.albe@cybertec.at>
From: Laurenz Albe <laurenz.albe@cybertec.at>
To: Sebastien Flaesch <sebastien.flaesch@4js.com>, Kirk Wolak
<wolakk@gmail.com>
Cc: Geoff Winkless <pgsqladmin@geoff.dj>, pgsql-general <pgsql-general@lists.postgresql.org>
Date: 2023-03-29T15:53:08Z
Lists: pgsql-general
On Wed, 2023-03-29 at 14:23 +0000, Sebastien Flaesch wrote: > From: Laurenz Albe <laurenz.albe@cybertec.at> > > It is safe to assume that the CTID is stable within a single transaction > > only if you use REPEATABLE READ or better transaction isolation level. > > > > With READ COMMITTED, you see updated rows (and consequently changed CTID) > > within a single transaction. And if you use SELECT ... FOR UPDATE, you > > could even see a changed CTID within a single statement. > > > > So don't use CTID to identify rows unless you use REPEATABLE READ or better. > > Thanks for the advice about REPEATABLE READ isolation level! ... but that is only useful in a read-only scenario. If you try to UPDATE the row in a REPEATABLE READ transaction, you will get a serialization error if there was a concurrent update. In short: don't use the CTID to identify a row. Yours, Laurenz Albe