testinherit does a somewhat nontrivial handling of inheritance...

A "master" sales table is created which has three "sub tables" for
three regions (US west, US east, Canada) which inherit from it, and all
four tables are replicated.

Notice that constraints do not automatically propagate to the
inherited tables; they need to be added explicitly if direct
manipulations to the inherited tables are to retain referential
integrity.  (Mind you, inserts to the "master" table are checked at
creation time, so as long as you don't do modifications to the
inherited tables, you at least start with referential integrity...)

Rules are added to cause the data for those three regions to be
automatically directed to the right partition table.

There is also a product table, products...

A stored procedure, purchase_product, is used to generate sales data,
demonstrating that it's fine to use stored procs with Slony-I;
replication will pick up updates even if they weren't expressly
declared.

The queries used to request product purchases use random() to choose
products and regions, demonstrating that Slony-I copes perfectly well
with nondeterministic queries.

Note that slon processes are configured purely via command line, not
via .conf file

This test also performs a subtransaction test; it generates 16
more-or-less concurrent psql sessions that:
 - create lists of products/regions (randomly, in temp tables local to each session)
 - performs purchases on elements of those lists
 - some of those purchases get rolled back, others don't

This test sets up its slonik scripts using
tools/configure-replication.sh, thereby drawing the blanket of
"replication regression tests" over this tool.