Scott Marlowe wrote:

>I'll try to figure out what's making it fail.  Still, this seems like a
>poor failure mode.  It would see to me that since one DDL script is
>failing, the other should be rolled back.  natch?
>
>  
>
Not possible.

Slony-I does asynchronous replication.  That means that the changes are
COMMITted on the provider before anything even _starts_ heading to the
subscriber.

>>In any case, logs for node #2 are vital to the diagnosis.
>>    
>>
>
>Should I turn up the verbosity a bit in the logging and see if I can get
>more data?  I'll try to figure out exactly what in the ddl script is
>breaking the update...
>  
>
Yes, that would presumably be useful.

The clever idea that one of our guys came up with is that you should
test DDL changes in a sort of "two phase" manner...

1.  Submit the scripts directly, inside a BEGIN clause, which is then
rolled back, on the "master."

That shows you if things conspicuously break there.

2.  Submit the script with BEGIN/work/ROLLBACK via Slonik.  That should
behave as a no-op, but show off any errors you might hit with it WITHOUT
having it take effect partially on your nodes.

After those two steps, you run it in full blown fashion.

DDL is pretty scary, all in all...