Обсуждение: Provide a pg_truncate_freespacemap function

Hello,

As we are currently experiencing a FSM corruption issue [1], we need to 
rebuild FSM when we detect it. 

I noticed we have something to truncate a visibility map, but nothing for the 
freespace map, so I propose the attached (liberally copied from the VM 
counterpart) to allow to truncate a FSM without incurring downtime, as 
currently our only options are to either VACUUM FULL the table or stop the 
cluster and remove the FSM manually.

Does that seem correct ?


[1]  https://www.postgresql.org/message-id/flat/
1925490.taCxCBeP46%40aivenlaptop#7ace95c95cab17b6d92607e5362984ac

--
Ronan Dunklau

Greetings,

* Ronan Dunklau (ronan.dunklau@aiven.io) wrote:
> As we are currently experiencing a FSM corruption issue [1], we need to
> rebuild FSM when we detect it.

Ideally, we'd figure out a way to pick up on this and address it without
the user needing to intervene, however ...

> I noticed we have something to truncate a visibility map, but nothing for the
> freespace map, so I propose the attached (liberally copied from the VM
> counterpart) to allow to truncate a FSM without incurring downtime, as
> currently our only options are to either VACUUM FULL the table or stop the
> cluster and remove the FSM manually.

I agree that this would generally be a useful thing to have.

> Does that seem correct ?

Definitely needs to have a 'REVOKE ALL ON FUNCTION' at the end of the
upgrade script, similar to what you'll find at the bottom of
pg_visibility--1.1.sql in the tree today, otherwise anyone could run it.

Beyond that, I'd suggest a function-level comment above the definition
of the function itself (which is where we tend to put those- not at the
point where we declare the function).

Thanks!

Stephen

Le mercredi 6 mars 2024, 20:28:44 CET Stephen Frost a écrit :
> I agree that this would generally be a useful thing to have.

Thanks !

>
> > Does that seem correct ?
>
> Definitely needs to have a 'REVOKE ALL ON FUNCTION' at the end of the
> upgrade script, similar to what you'll find at the bottom of
> pg_visibility--1.1.sql in the tree today, otherwise anyone could run it.
>
> Beyond that, I'd suggest a function-level comment above the definition
> of the function itself (which is where we tend to put those- not at the
> point where we declare the function).

Thank you for the review. Here is an updated patch for both of those.


Best regards,

--
Ronan