Hi!
On Tue, Oct 3, 2023 at 8:35 PM Alexander Korotkov <aekorotkov@gmail.com> wrote:
> Thank you for the interesting ideas. I'd like to try to revive the
> version with the flag in pg_database. Will use other ideas as backup
> if no success.
I've revived the patch version with pg_database.dathasloginevt flag.
I took v32 version [1] and made the following changes.
* Incorporate enchantments made on flagless version of patch.
* Read dathasloginevt during InitPostgres() to prevent extra catalog
access and even more notable StartTransactionCommand() when there are
no login triggers.
* Hold lock during setting of pg_database.dathasloginevt flag (v32
version actually didn't prevent race condition).
* Fix AlterEventTrigger() to check event name not trigger name
* Acquire conditional lock while resetting pg_database.dathasloginevt
flag to prevent new database connection to hang waiting another
transaction to finish.
This version should be good and has no overhead. Any thoughts?
Daniel, could you please re-run the performance tests?
Links
1. https://www.postgresql.org/message-id/CAMEv5_vDjceLr54WUCNPPVsJs8WBWWsRW826VppNEFoLC1LAEw%40mail.gmail.com
------
Regards,
Alexander Korotkov
