On Fri, Jul 22, 2011 at 11:37 AM, marmu <mu at forsa.de> wrote:
>
> Hello Guillaume,
>
> ok, that was my mistake. but now I altered the psql command accordingly and
> this is the output:
>
> fquest-b1:~ # psql -U $PGBENCHUSER -h localhost -d $MASTERDBNAME -c "begin;
> alter table pgbench_history add column id serial; update pgbench_history set
> id =
> nextval('history_id_seq'); alter table pgbench_history add primary key(id);
> commit";
> NOTICE:  ALTER TABLE will create implicit sequence "pgbench_history_id_seq"
> for serial column "pgbench_history.id"
> ERROR:  relation "history_id_seq" does not exist
> LINE 2: nextval('history_id_seq'); alter table pgbench_history add p...
>                ^
> fquest-b1:~ # psql -U $PGBENCHUSER -h localhost -d $MASTERDBNAME -c
> "begin;update pgbench_history set id = nextval('pgbench_history_id_seq');
> alter table pgbench_history add primary key(id); commit";
> ERROR:  relation "pgbench_history_id_seq" does not exist
> LINE 1: begin;update pgbench_history set id = nextval('pgbench_histo...
>
> I really don't get it. Where should the relation "pgbench_history_id_seq"
> come from, do I have to create everything manually? Slony is kind of a pain
> in the ass for novices like me, excuse my language.
> My colleage says, "but when it is finally running, it is solid as a rock and
> amazingly fast". I'd like to experience that soon :)

This isn't notably a Slony issue, except insofar as the Slony
documentation describes how pgbench *used* to work, rather than what
it got changed to more recently.

Understanding what's going on is a lot more important than getting a
recipe, because when PostgreSQL or pgbench change, that means the
recipe may need to change, and that is indeed what happened.

Slony requires that all replicated tables have a primary key defined.
The pgbench "history" table doesn't have one of those by default, so,
to replicate the pgbench schema requires adding a primary key to that
table.

On older versions of pgbench, what was necessary was to run a query like:
  alter table history add column id serial;
  alter table history add primary key(id);

And then add the sequence that gets generated for that ID column to replication.

On newer versions of pgbench, the tables have the prefix "pgbench_"
prepended to them, thereby changing the queries to:
 alter table pgbench_history add column id serial;
 alter table pgbench_history add primary key (id);

Since the names of the tables and sequences have all changed, the
Slony configuration needs to account for the new names.  Nobody
updated the documentation, so I have done so.
   <http://git.postgresql.org/gitweb/?p=slony1-engine.git;a=commit;h=928300d1fc0078d2e980aa3f601c351c12495d8a>