On Thu, Aug 26, 2004 at 03:37:38PM -0500, joe wrote:
>         1. Alter tables don't seem to require any interaction with
> slonik to be replicated, as long as the alter table statement(s) is/are
> executed on all nodes before any queries (inserts, updates, deletes)
> that use the new schema are run.  This doesn't feel quite safe (or sane)
> to me and I was wondering how other people handle modifications to
> existing tables.

Use the slonik command EXECUTE SCRIPT to do DDL changes.  Commands
executed there will be applied to all nodes in a safe point in
the replication sequence.

Ie.  create one file:

alter table blah add column new_column type;
alter table blah alter column new_column set default good_default;

Execute a slonik script that executes those commands.

psql to your db, 
  update blah set new_column = good_default where new_column is null;

Create another file:

alter table blah alter column new_column set not null;

Execute a slonik script that executes that command.

This will all be done in the correct order for the replication, not
interfering with other writes that may happen.

>         2. adding new tables to an existing replication set (apparantly)
> requires unsubscribing that set, adding the table then re-subscribing
> the set. This forces a full dump of the table (and possibly full set?)
> when it happens. (Not fun when the table/set is several gigs). Do most
> people just make a new table set when they add tables to the
> source/master node or how do they deal with that situation?
> 

Create a new set with the new table.
Subscribe the same slave nodes to the new set that are subscribed to the
old set.
Wait for the subscribe to complete on the slave nodes.
Execute the slonik MERGE SET command to merge the new set with the
existing set.


-- 
How many of you believe in telekinesis? Raise my hand...