On 2020-Jun-26, Michael Paquier wrote:
> On Thu, Jun 25, 2020 at 11:24:27AM -0400, Alvaro Herrera wrote:
> > I don't understand the proposal. Michael posted a patch that adds
> > pg_wal_oldest_lsn(), and you say we should apply the patch except the
> > part that adds that function -- so what part would be applying?
>
> I have sent last week a patch about only the removal of min_safe_lsn:
> https://www.postgresql.org/message-id/20200619121552.GH453547@paquier.xyz
> So this applies to this part.
Well, I oppose that because it leaves us with no way to monitor slot
limits. In his opening email, Masao-san proposed to simply change the
value by adding 1. How you go from adding 1 to a column to removing
the column completely with no recourse, is beyond me.
Let me summarize the situation and possible ways forward as I see them.
If I'm mistaken, please correct me.
Problems:
i) pg_replication_slot.min_safe_lsn has a weird definition in that all
replication slots show the same value
ii) min_safe_lsn cannot be used with pg_walfile_name, because it returns
the name of the previous segment.
Proposed solutions:
a) Do nothing -- keep the min_safe_lsn column as is. Warn users that
pg_walfile_name should not be used with this column.
b) Redefine min_safe_lsn to be lsn+1, so that pg_walfile_name can be used
and return a useful value.
c) Remove min_safe_lsn; add functions that expose the same value
d) Remove min_safe_lsn; add a new view that exposes the same value and
possibly others
e) Replace min_safe_lsn with a "distance" column, which reports
restart_lsn - oldest valid LSN
(Note that you no longer have an LSN in this scenario, so you can't
call pg_walfile_name.)
The original patch implemented (e); it was changed to its current
definition because of this[1] comment. My proposal is to put it back.
[1] https://postgr.es/m/20171106132050.6apzynxrqrzghb4r@alap3.anarazel.de
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services