On Thu, Mar 8, 2018 at 12:34 AM, Pavan Deolasee
<pavan.deolasee@gmail.com> wrote:
> I am actually very surprised that 0001-Invalidate-ip_blkid-v5.patch does not
> do anything to deal with the fact that t_ctid may no longer point to itself
> to mark end of the chain. I just can't see how that would work. But if it
> does, it needs good amount of comments explaining why and most likely
> updating comments at other places where chain following is done. For
> example, how's this code in heap_get_latest_tid() is still valid? Aren't we
> setting "ctid" to some invalid value here?
So the general idea of the patch is that this new kind of marking
marks the CTID chain as "broken" and that code which cares about
following CTID chains forward can see that it's reached a point where
the chain is broken and throw an error saying "hey, I can't do the
stuff we normally do in concurrency scenarions, because the CTID chain
got broken by a cross-partition update".
I don't think it's practical to actually make CTID links across
partitions work. Certainly not in time for v11. If somebody wants to
try that at some point in the future, cool. But that's moving the
goalposts an awfully long way. When this was discussed about a year
ago, my understanding is that there was a consensus that doing nothing
was not acceptable, but that throwing an error in the cases where
anomalies would have happened was good enough. I don't think anyone
argued that we had to be able to perfectly mimic the usual EPQ
semantics as a condition of having update tuple routing. That's
setting the bar at a level that we're not going to be able to reach in
the next couple of weeks. I suppose we could still decide that if we
can't have that, we don't want update tuple routing at all, but I
think that's an overreaction.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
