Argh. How embarrassing. I saw the ENABLE_SUBSCRIPTION and copy_set
messages come out in the log, but I didn't see the data coming through.
I was in the middle of setting up the pgbench example when I sent my
brain-dead email. Of course it works fine in the simple pgbench
database, so I obviously screwed something up in my application test db.
Apologies for the noise.

>It does this automatically when you add a node.  In fact, what 
>people would like is to _prevent_ that, because the current 
>approach is considerably less efficient than pg_dump | psql.

Are there any plans to allow this flexibility? 

Thanks.

- DAP