Horton

RE: [PROPOSAL] Termination of Background Workers for ALTER/DROP DATABASE

Aya Iwata (Fujitsu) <iwata.aya@fujitsu.com>

View thread
From: "Aya Iwata (Fujitsu)" <iwata.aya@fujitsu.com>
To: 'Pavel Stehule' <pavel.stehule@gmail.com>, Michael Paquier <michael@paquier.xyz>
Cc: Peter Smith <smithpb2250@gmail.com>, Chao Li <li.evan.chao@gmail.com>, "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2025-12-25T13:24:33Z
Lists: pgsql-hackers

Attachments

Hi

https://www.postgresql.org/message-id/aPBoXXW3XuwiIsHG%40paquier.xyz

Based on previous discussions, we had configured DROP/ALTER DATABASE to
terminate only background workers with specific parameters set.

To reconsider the default behavior, I created the following patches:
0001 is the patch for the existing implementation.
0002 is the patch modified based on Pavel-san's suggestion.

The 0002 patch changes the default behavior to terminate background workers
when DROP/ALTER DATABASE is executed. It also includes a test set
for this change.
For background workers that should not be terminated, setting
the BGWORKER_PROTECTED parameter prevents this.

The relationship with the FORCE option during DROP DATABASE is as follows:

FORCE option present, flag present... TERMINATE
FORCE option present, flag absent... TERMINATE
FORCE option absent, flag present... Does not TERMINATE due to the flag
FORCE option absent, flag absent... TERMINATE

I would like to know your comments on which implementation is preferable.

Regards,
Aya Iwata