On Wed, Jan 27, 2021 at 5:24 PM Peter Geoghegan <pg@bowt.ie> wrote:
> The issue here is that it would also be nice to use a "recently dead"
> bit on the primary, but if you do that then maybe you're back to the
> original problem. OTOH, I also think that we could repurpose the
> LP_NORMAL bit in index AMs, so we could potentially have 3 different
> definitions of dead-ness without great difficulty!
To be clear, what I mean is that you currently have two bits in line
pointers. In an nbtree leaf page we only currently use one -- the
LP_DEAD bit. But bringing the second bit into it allows us to have a
representation for two additional states (not just one), since of
course the meaning of the second bit can be interpreted using the
LP_DEAD bit. You could for example having encodings for each of the
following distinct per-LP states, while still preserving on-disk
compatibility:
"Not known to be dead in any sense" (0), "Unambiguously dead to all"
(what we now simply call LP_DEAD), "recently dead on standby"
(currently-spare bit is set), and "recently dead on primary" (both
'lp_flags' bits set).
Applying FPIs on the standby would have to be careful to preserve a
standby-only bit. I'm probably not thinking of every subtlety, but
"putting all of the pieces of the puzzle together for further
consideration" is likely to be a useful exercise.
--
Peter Geoghegan