| From: | Raja Sai pranav <rajasaipranav0(at)gmail(dot)com> |
|---|---|
| To: | alpkeles99(at)gmail(dot)com, pgsql-bugs(at)lists(dot)postgresql(dot)org |
| Subject: | Re: BUG #19790: "cannot start a subtransaction ... inval messages" after "ALTER DOMAIN ... DROP CONSTRAINT" |
| Date: | 2026-11-21 07:28:14 |
| Message-ID: | CAPHduZCbAvqw3J6yMYL5qkaZiHaM592+16RW+ZkD70V7mP82fA@mail.gmail.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-bugs |
Hi,
I reproduced this on master (4a26e36f87). BUG #19919 is the same problem:
https://postgr.es/m/19919-c286b9e822bf23b9@postgresql.org
AlterDomainDropConstraint() queues an invalidation for the domain's pg_type
row after the constraint is deleted. The deletion's
CommandCounterIncrement() has already run, and the one at the end of the
command does nothing because no catalog change follows it, so the message
stays pending until the next subtransaction start trips over it. DROP
CONSTRAINT IF EXISTS on a missing constraint hits it too.
The attached patch queues the invalidation before the deletion, so the
deletion's CommandCounterIncrement() processes it, and only when a
constraint is actually dropped. Plans that depend on the constraint are
still rebuilt. The test in domain.sql fails without the fix and passes with
it; regress and pg_upgrade pass.
v1 is for master. The REL_18 version is the same change (it cherry-picks
cleanly), tested the same way on REL_18_STABLE.
Regards,
Sai Pranav
| Attachment | Content-Type | Size |
|---|---|---|
| REL_18-v1-0001-Fix-unprocessed-inval-messages-error-after-ALTER-.patch | application/octet-stream | 4.5 KB |
| v1-0001-Fix-unprocessed-inval-messages-error-after-ALTER-.patch | application/octet-stream | 4.5 KB |
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Raja Sai pranav | 2026-11-21 08:27:21 | Re: BUG #19917: nbtree skip scan leaks one copy of a pass-by-reference skip array element per rescan |
| Previous Message | Raja Sai pranav | 2026-11-21 07:14:19 | Re: BUG #19881: CLUSTER uses a disabled sort path when enable_sort is off |