Re: Concurrent ALTER SEQUENCE RESTART Regression

On 2017-04-26 21:07:20 -0400, Peter Eisentraut wrote:
> On 4/26/17 19:12, Michael Paquier wrote:
> > Well, there are more DDL commands where it is possible to see "tuple
> > concurrently updated" easily, an example is ALTER ROLE. So nothing is
> > concurrent-proof with this code and I think needs a careful lookup
> > because this error should never be something that is user-visible.
>
> Yeah, it's been like this since time immemorial, so I don't think we
> need a last minute fix now.

Uh. Until v10 this worked in a somewhat weird way, but it worked.

> One thing we could do, since all catalog updates now go through
> CatalogTupleUpdate(), is not use simple_heap_update() there but do the
> heap_update() directly and provide a better user-facing error message.

I think it's unacceptable to regress with an error message here. I've
seen sequence DDL being used while concurrent DML was onging in a number
of production use cases, and just starting to error out instead of
properly blocking doesn't seem acceptable to me.

- Andres