On Mon, Mar 3, 2025 at 9:40 AM Melanie Plageman
<melanieplageman@gmail.com> wrote:
>
> I'd certainly like to merge it in 18. The patch itself needs more
> work, which I'm planning on doing this week. I think we have a
> reasonable amount of consensus on using GUC_LIST_INPUT.
>
> I am less sure about merging log_disconnections into log_connections.
> I think it makes sense in principle. But, in 18, when people satisfied
> with existing behavior change log_connections from 'on' to 'all' to
> comply with the new format, they will get the disconnections messages.
>
> I am still somewhat nervous about merging the patch in general and
> people being surprised because they didn't pay attention to the
> discussion and now log_connections is different and they would have
> -1'd if they'd been paying attention.
Okay, I got cold feet today working on this. I actually think we
probably want some kind of guc set (set like union/intersection/etc
not set like assign a value) infrastructure for gucs that can be equal
to any combination of predetermined values. See my musings on this
here [1].
It is too late in the cycle for me to get something like that done,
but I am interested in working on it next release. IIRC you did not
deprecate log_connections in your patch, but adding another guc that
is a staged version of it seems not as good as turning log_connections
into a more modular GUC. But I just don't think we want to move
log_connections/disconnections toward a string/list.
- Melanie
[1] https://www.postgresql.org/message-id/CAAKRu_a5-7sUP%2BQ6YD5emQYS1w7ffBDUNf-NMbcxR-dpOdGehw%40mail.gmail.com
