Horiguchi-San
On Tuesday, November 10, 2020 10:00 AM Kyotaro Horiguchi <horikyota....@gmail.com> wrote: > FWIW, the following is that, I think it works not only when wal_level=minimal > for SET UNLOGGED, and only works when minimal for SET LOGGED. > > https://www.postgresql.org/message-id/20201002.100621.166891875652013 > 6893.horikyota....@gmail.com > > | For fuel(?) of the discussion, I tried a very-quick PoC for in-place > | ALTER TABLE SET LOGGED/UNLOGGED and resulted as attached. After > some > | trials of several ways, I drifted to the following way after poking > | several ways. > | > | 1. Flip BM_PERMANENT of active buffers 2. adding/removing init fork 3. > | sync files, 4. Flip pg_class.relpersistence. > | > | It always skips table copy in the SET UNLOGGED case, and only when > | wal_level=minimal in the SET LOGGED case. Crash recovery seems > | working by some brief testing by hand. > I agree to this aspect of the in-place flipping of UNLOGGED. > > Couldn't we have something like the following? > ALTER TABLE table1, table2, table3 SET UNLOGGED; > > That is, multiple target object specification in ALTER TABLE sttatement. I *really* appreciate the 1st patch. Did you register this patch with the current or next commitfest ? I cannot find it. We should avoid that this patch misses people's attention. When I'm available, I'd like to join the review for the patch as well. Also, I supposed that as long as something impossible to implement for wal_level=none is not founded, we could develop both patches in parallel, because both have good points described in the previous e-mails of this thread. Best, Takamichi Osumi
