On Wed, Sep 02, 2020 at 03:00:44PM +0900, Michael Paquier wrote:
> Yeah, that's true as well. Still, I would like to see first if people
> are fine with changing this code path to be transactional. This way,
> we will have zero history in the tree where there was a risk for an
> inconsistent window.
So, I have begun a thread about that part with a dedicated CF entry:
https://commitfest.postgresql.org/30/2725/
While considering more this point, I think that making this code path
transactional is the correct way to go, as a first step. Also, as
this thread is about adding more tests and that this has been done
with fe7fd4e9, I am marking the CF entry as committed. Let's discuss
the reset of relreplident for the parent relation when its replica
identity index is dropped once the transactional part is fully
discussed, on a new thread.
--
Michael