Removal of unnecessary CommandCounterIncrement() when doing ON COMMIT DELETE ROWS
Hi all,
When doing a set of ON COMMIT DELETE ROWS actions for relations, there
is a CCI happening after each truncation:
@@ -13334,10 +13334,8 @@ PreCommit_on_commit_actions(void)
* exists at truncation time.
*/
if (oids_to_truncate != NIL)
- {
heap_truncate(oids_to_truncate);
- CommandCounterIncrement(); /* XXX needed? */
- }This has been visibly introduced by f9b5b41 back in 2002 which reworked
how ON COMMIT works.
Alvaro has mentioned that it would not be needed here:
20181106125337.o23kjdv546bu2tei@alvherre.pgsql
And I think that I agree with that, because visibly this applies to
index rebuilds but in those cases CCIs are happening locally. So I
think that we can get rid of that, and I suggest to remove it only on
HEAD only of course.
Any objections?
--
Michael
Attachments:
oncommit-cci-remove.patchtext/x-diff; charset=us-asciiDownload+1-3
On 2018-Nov-09, Michael Paquier wrote:
Hi all,
When doing a set of ON COMMIT DELETE ROWS actions for relations, there
is a CCI happening after each truncation:
Right, but note that this is not a loop so it's only one CCI, not one
per relation. (This doesn't seem a big deal either way, but I see no
reason to have that line there.)
And I think that I agree with that, because visibly this applies to
index rebuilds but in those cases CCIs are happening locally. So I
think that we can get rid of that, and I suggest to remove it only on
HEAD only of course.
+1
--
�lvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
On Mon, Nov 12, 2018 at 01:19:31PM -0300, Alvaro Herrera wrote:
On 2018-Nov-09, Michael Paquier wrote:
When doing a set of ON COMMIT DELETE ROWS actions for relations, there
is a CCI happening after each truncation:Right, but note that this is not a loop so it's only one CCI, not one
per relation. (This doesn't seem a big deal either way, but I see no
reason to have that line there.)
This looks like a lack of caffeine from my side.
And I think that I agree with that, because visibly this applies to
index rebuilds but in those cases CCIs are happening locally. So I
think that we can get rid of that, and I suggest to remove it only on
HEAD only of course.+1
Thanks Álvaro for the input, I have committed the change.
--
Michael