On 7/18/2007 3:50 PM, Andrew Sullivan wrote:
> On Wed, Jul 18, 2007 at 11:06:44PM +0400, Dmitry Koterov wrote:
>> Sorry, but I have not undersand this clearly. :-(
>> Could you please answer the single practical question (just "yes" or "no"):
>> if I perform
>> 
>> BEGIN;
>> INSERT INTO tbl(id, parent_id) VALUES(1, 2);    -- child first!
>> INSERT INTO tbl(id, parent_id) VALUES(2, null);  -- parent
>> COMMIT;
>> 
>> on master (assuming that parent_id foreign key is deferrable), DOES Slony
>> GUARANTEE that ANY subscriber will receive and SUCESSFULLY process this
>> data, or there is a probability that a subscriber will fail to update?
> 
> Yes, but I'm not sure whether in your example
> statement1(tbl)==statement2(tbl). 
> 
> Foreign keys are disabled on the replica on the grounds that they're
> enforced on the origin anyway.  But your case isn't a foreign key,
> AFAICT, because the tables are the same one in each statement as I
> read it.

That would be a self referencing table and still a foreign key. And for 
sure can someone use a deferrable constraint in that case to allow 
inserting self referencing rows out of order. Nothing wrong with the 
example.

To answer Dimitry's questions, first all changes done by one transaction 
on the origin will be replicated in the same slon transaction on a 
subscriber. It is possible that multiple origin transactions are grouped 
together into one replication transaction on serializable snapshot 
visibility boundaries. But the actions of a single origin transaction 
are never split into multiple replication transactions. This means that 
there will never be a state on a subscriber that would not have been 
visible under MVCC rules on the origin in precisely a SYNC events 
transaction. Unless one uses READ DIRTY isolation, of course.

Second foreign key constraints are internally implemented with triggers 
in Postgres. Since Slony disables triggers on a replica, the simple 
answer is that foreign keys are not checked on replicated tables on a slave.


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #