On Tue, Nov 8, 2011 at 11:28 AM, Vick Khera <vivek@khera.org> wrote:
> On Tue, Oct 18, 2011 at 6:21 PM, David Pirotte <dpirotte@gmail.com> wrote:
>> The underlying purpose is to get Londiste to acknowledge the table's key,
>> and this strategy seems to work without any problems. Londiste doesn't seem
>> to care that the "primary key" is only reflected in pg_index and isn't
>> accompanied by the relevant pg_constraint entry. Is modifying the
>> underlying pg_catalog tables like this "Very Bad"? Will it have mysterious
>> and unintended consequences, or can I get away with it? Thanks!
>
> The badness I see that will eventually come back to bite you is that
> your unique constraint is lacking "NOT NULL" and a PK by definition
> has NOT NULL. Therefore some other parts of the system is permitted
> to make that assumption, and when stuff fails because you lied to the
> system, you will probably never ever figure out or even know.
>
Agreed. I'd be more inclined to change londiste, or just ditch it for
something else that will recognize the unique index as a unique enough
identifier to enable replication. That limitation is kind of lame.
Robert Treat
conjecture: xzilla.net
consulting: omniti.com