On 2021-Oct-06, Bossart, Nathan wrote:
> On 10/6/21, 3:44 PM, "Tom Lane" <tgl@sss.pgh.pa.us> wrote:
> > The situation for "ALTER some-other-relation-kind" is a bit more
> > confused, because some cases throw errors and some don't; but I really
> > doubt that tightening things up here will earn you anything but
> > brickbats. I *definitely* don't agree with discarding the policy
> > about ALTER TABLE, especially if it's only done for RENAME.
>
> I think we should at least consider adding this check for ALTER INDEX
> since we choose a different lock level in that case.
I agree -- letting ALTER INDEX process relations that aren't indexes is
dangerous, with its current coding that uses a reduced lock level. But
maybe erroring out is not necessary; can we instead loop, locking the
object with ShareUpdateExclusive first, assuming it *is* an index, and
if it isn't then we release and restart using the stronger lock this
time?
--
Álvaro Herrera 39°49'30"S 73°17'W — https://www.EnterpriseDB.com/
Syntax error: function hell() needs an argument.
Please choose what hell you want to involve.