Horton

Re: BUG #17504: psql --single-transaction -vON_ERROR_STOP=1 still commits after client-side error

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Michael Paquier <michael@paquier.xyz>
Cc: Kyotaro Horiguchi <horikyota.ntt@gmail.com>, christoph.berg@credativ.de, pgsql-bugs@lists.postgresql.org
Date: 2022-06-10T15:41:49Z
Lists: pgsql-bugs 
Michael Paquier <michael@paquier.xyz> writes:
> On Fri, Jun 03, 2022 at 05:56:03PM +0900, Kyotaro Horiguchi wrote:
>>> It causes psql to issue a BEGIN command before the first such option
>>> and a COMMIT command after the last one, thereby wrapping all the
>>> commands into a single transaction.
>>> If any of the commands fails, a ROLLBACK command is sent instead.
>>> This ensures that either all the commands complete successfully, or
>>> no changes are applied.

> Looks fine to me, so applied down to 10.

While preparing release notes, I had to study this commit to figure
out what it actually did, and I do not think it is right yet.  It
definitely does not do what is claimed above, ie rollback if "any"
of the steps fails.  As coded, it rolls back only if the *last*
step fails.  Now, if you've got ON_ERROR_STOP set, then the loop
terminates after an error so there's no difference ... but if you
do not, I think this behavior is not what's expected.

ISTM that an appropriate fix would be to remember if any command
failed, not just the last one.

			regards, tom lane

Commits

  1. Tweak behavior of psql --single-transaction depending on ON_ERROR_STOP

  2. Doc: remove a04ccf6df from release notes, now that it's reverted.

  3. Revert "Fix psql's single transaction mode on client-side errors with -c/-f switches".

  4. Fix psql's single transaction mode on client-side errors with -c/-f switches