Re: Multi-Master Logical Replication

On Fri, Apr 29, 2022 at 07:05:11PM +1000, Peter Smith wrote:
> This MMLR proposal was mostly just to create an interface making it
> easier to use PostgreSQL core logical replication CREATE
> PUBLICATION/SUBSCRIPTION for table sharing among a set of nodes.
> Otherwise, this is difficult for a user to do manually. (e.g.
> difficulties as mentioned in section 2.2 of the original post [1] -
> dealing with initial table data, coordinating the timing/locking to
> avoid concurrent updates, getting the SUBSCRIPTION options for
> copy_data exactly right etc)
>
> At this time we have no provision for HA, nor for transaction
> consistency awareness, conflict resolutions, node failure detections,
> DDL replication etc. Some of the features like DDL replication are
> currently being implemented [2], so when committed it will become
> available in the core, and can then be integrated into this module.

Uh, without these features, what workload would this help with? I think
you made the mistake of jumping too far into implementation without
explaining the problem you are trying to solve. The TODO list has this
ordering:

https://wiki.postgresql.org/wiki/Todo
Desirability -> Design -> Implement -> Test -> Review -> Commit

--
Bruce Momjian <bruce(at)momjian(dot)us> https://momjian.us
EDB https://enterprisedb.com

Indecision is a decision. Inaction is an action. Mark Batterson