On 17/10/2018 23:11, Peter Eisentraut wrote:
> On 13/10/2018 04:01, Andres Freund wrote:
>> I don't see how this could be argued. It has to be a self-conflicting
>> lockmode, otherwise you'd end up doing renames of tables where you
>> cannot see the previous state. And you'd get weird errors about updating
>> invisible rows etc.
>
>> I don't buy this description. Imo it's a fundamental correctness
>> thing. Without it concurrent DDL would potentially overwrite the rename,
>> because they could start updating while still seeing the old version.
>
> OK, I can refine those descriptions/comments. Do you have any concerns
> about the underlying principle of this patch?
Patch with updated comments to reflect your input.
--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services